On Sat, Mar 29, 2025 at 1:22 PM Rick Macklem <rick.mack...@gmail.com> wrote: > > On Sat, Mar 29, 2025 at 1:09 PM Shawn Webb <shawn.w...@hardenedbsd.org> wrote: > > > > On Sat, Mar 29, 2025 at 01:04:08PM -0700, Rick Macklem wrote: > > > On Sat, Mar 29, 2025 at 12:50 PM Shawn Webb <shawn.w...@hardenedbsd.org> > > > wrote: > > > > > > > > On Sat, Mar 29, 2025 at 12:39:02PM -0700, Rick Macklem wrote: > > > > > > I had added filesystem extended attribute support to libarchive, > > > > > > which > > > > > > is what FreeBSD's tar(1) is based off of. I upstreamed that, so > > > > > > that's > > > > > > taken care of. FreeBSD's tar(1) has supported extended attributes > > > > > > since 2020 (see libarchive PR 1409: > > > > > > https://github.com/libarchive/libarchive/pull/1409) > > > > > Ok, thanks for the info. If this stuff goes into FreeBSD, it probably > > > > > needs > > > > > to be tweaked to use the different syscall API so that it can handle > > > > > large > > > > > attributes and maybe the attribute's mode. (someday, maybe?) > > > > > > > > I believe libarchive has been updated in FreeBSD since October 2020, > > > > so the vendored libarchive in FreeBSD should already support it. But, > > > > yeah, if FreeBSD makes changes to how extended attributes work, I or > > > > someone else would need to update libarchive to account for that. > > > > > > > > Since HardenedBSD follows FreeBSD closely (we sync every six hours), I > > > > would probably volunteer to update the libarchive code. > > > > > > > > > > Just one data point here: HardenedBSD uses filesystem extended > > > > > > attributes to toggle certain exploit mitigations on a > > > > > > per-application > > > > > > basis. That's why we added support to libarchive: so we can ship > > > > > > certain packages with exploit mitigations pre-toggled. > > > > > Just curious. Does it use "system" or "user" attribute space? > > > > > > > > We use the system namespace, though the userland tool (hbsdcontrol) > > > > was recently taught about the user namespace. The kernel side only > > > > supports system namespace. So the user namespace support in > > > > hbsdcontrol is somewhat meaningless. I do plan to eventually get to > > > > the kernel side, but my TODO list continues growing. :-) > > > Ok, this wouldn't be affected by the patches I've been doing, since they > > > handle user space only. (system space will still work, but only via the > > > extattr_XXX() APIs. > > > > Cool. I have another project that uses user namespaces: > > https://git.hardenedbsd.org/shawn.webb/altfs > > > > AltFS is a fusefs driver that stores file payload in filesystem > > extended attributes, using the user namespace. It only partially works > > and again is bitten by more important items on my TODO list. It mainly > > serves as a proof-of-concept for a weird data exfiltration technique. > > Not at all meant for actual production use. > > > > Do you already have a patch for review in Phabric? I might want to add > > myself to it so I can more easily keep informed. > Not yet. I am still cleaning things up and testing. I have put the VFS/syscall changes up in phabricator under D49583. I listed a few reviewers, but anyone is welcome to review/comment on it.
rick > Also, there ahs not been much response related to the question "should this > go in FreeBSD?". Dennis doesn't sounds like a "no" and the two posters on > freebsd-hackers@ I assume are a"yes", but I haven't heard from anyone else. > (Good technical comments, but not related to "should it be in FreeBSD?".) > > rick > > > > > Thanks, > > > > -- > > Shawn Webb > > Cofounder / Security Engineer > > HardenedBSD > > > > Tor-ified Signal: +1 303-901-1600 / shawn_webb_opsec.50 > > https://git.hardenedbsd.org/hardenedbsd/pubkeys/-/raw/master/Shawn_Webb/03A4CBEBB82EA5A67D9F3853FF2E67A277F8E1FA.pub.asc
