makefs: Use timestamps provided by -T when adding RockRidge TF records

[bnovkov]

Summary

makefs' implementation of the RockRidge TF record ignores timestamps
passed using the -T flag, making the resulting ISO image
nonreproducible. Fix this by selecting the appropriate stat struct
before creating the TF record.

PR: 285630
Sponsored by: Klara, Inc.
Sponsored by: The FreeBSD Foundation

[emaste]

As an aside the way makefs uses stampst.st_ino as a flag feels kind of hacky to me, but is consistent with what's already done.

[emaste]

Also, makefs has some reasonable existing tests; it should be straightforward to add a test for -T with Rock Ridge TF.

[bnovkov]

In D49488#1128372, @emaste wrote:

Also, makefs has some reasonable existing tests; it should be straightforward to add a test for -T with Rock Ridge TF.

Thanks for pointing this out, I've added a test case for this in D49492

In D49488#1128369, @emaste wrote:

As an aside the way makefs uses stampst.st_ino as a flag feels kind of hacky to me, but is consistent with what's already done.

I agree, I'll see if this can be rearranged to be a bit cleaner, makefs already has a couple of places that could use a refactor.

[kevans]

usr.sbin/makefs/cd9660/iso9660_rrip.c

759

You could (before pushing) unwrap these lines now that st->st_[mac]time would fit cleanly onto the line before, but I do not insist.

[emaste]

usr.sbin/makefs/cd9660/iso9660_rrip.c

759

Yeah, that's preferable.

[markj]

The ZFS backend will need this as well. I'll be back the week after next and can work on it then, but please feel free to do it in the meantime if you like. I believe it should just be a matter of adjusting the fs_populate_time() calls in fs_populate_sattrs().

[bnovkov]

While working on porting this check for the ZFS backend, I realized that this issue only comes up when building images with an mtree manifest.
It turns out that read_mtree does not properly populate the fsnode struct if -T is passed, unlike walk_dir which behaves correctly in this situation.
Rather than plugging the same check into all backends, I have drafted a patch for mtree.c that should provide a backend-agnostic fix for this issue.
Please see D49531 for more information.

I'll abandon this revision if the other one is deemed to be more suitable.