MAINTAINERS: add myself as UFS filesystem maintainer
PATCH 1/1 MAINTAINERS: add myself as UFS filesystem maintainer
fs/ufs has no MAINTAINERS entry. get_maintainer.pl returns only the open lists, so patches for it have no clear reviewer and tend to sit unanswered. The archives show the cost of this: a truncate path memory leak fix was posted in 2018 with a stable tag and was never applied [1], and recent mount path hardening patches went weeks without a maintainer to review or apply them [2][3].
The code needs attention. It parses on-disk BSD and Solaris layouts with very little validation, so several crafted-image mount crashes are still open. I have been reading it and fixing real bugs in the normal paths as well, including the 2018 leak above and an rmdir path that treated an unreadable directory block as empty and could drop a non-empty directory [4].
I am willing to take responsibility for fs/ufs: review incoming patches, carry the pending fixes, and keep it building and correct. Add a Maintained entry to reflect that.
--- MAINTAINERS | 6 ++++++ 1 file changed, 6 insertions(+)
MAINTAINERS +6 -0
@@ -27663,6 +27663,12 @@ L: [email protected] S: Maintained F: drivers/hid/hid-udraw-ps3.c +UFS FILESYSTEM +M: Ali Ahmet Memis <[email protected]> +L: [email protected] +S: Maintained +F: fs/ufs/ + UHID USERSPACE HID IO DRIVER M: David Rheinsberg <[email protected]> L: [email protected]
Signature · 1 line
2.54.0
Quoted 3 lines · Ali Ahmet Memis
I am willing to take responsibility for fs/ufs: review incoming patches, carry the pending fixes, and keep it building and correct. Add a Maintained entry to reflect that.
Convince us you aren't Jia Tan. You have almost no history with the Linux kernel (one commit that I can find, and that has a different email address attached to it).
Quoted 3 lines · Matthew Wilcox
Convince us you aren't Jia Tan. You have almost no history with the Linux kernel (one commit that I can find, and that has a different email address attached to it).
Fair question, and after xz it is the right one to ask, so no offence taken.
Proving a negative is hard, so here is what I can actually show.
The commit you found is 607af438e643 ("tools/power/x86/
intel-speed-select: Harden daemon pidfile open"). I wrote it as Ali Ahmet MEMIS <[email protected]> and Srinivas Pandruvada applied it. I still control that address and will confirm from it that [email protected] is the same person if that is useful. I have moved my kernel work to the iusegentoo.com address and will keep to one from here.
Past that, I would rather be judged on this week's work than on my word:
- ufs: do not treat unreadable directory blocks as empty (rmdir data loss)
- ufs: free the buffer head container in ubh_bforget (Luis Henriques posted this fix in 2018 and it was never applied)
- reviews of the pending cgsize and cssize hardening patches
- a two patch hardening series with KASAN reproducers, for a mount teardown NULL deref and an out of bounds write driven by on-disk cg_cgx
On the entry: point taken, adding myself as Maintained on this much history is premature. Consider this patch dropped. I will keep sending fixes and reviews, and if an entry ever looks warranted to you and the other reviewers we can revisit it then. The fixes do not depend on it.
What I care about is ufs bugs not being ignored. If routing my fixes through you or the vfs tree is the better path, that is fine by me.
Jan Kara
reviewerQuoted 2 lines · Ali Ahmet Memis
What I care about is ufs bugs not being ignored. If routing my fixes through you or the vfs tree is the better path, that is fine by me.
I'm a bit curious: Are you using UFS driver in Linux? Or do you know of anybody using it?
Honza
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR
Quoted 2 lines · Jan Kara
I'm a bit curious: Are you using UFS driver in Linux? Or do you know of anybody using it?
Yes. I run FreeBSD on a second machine that I keep as a small server, and its disks are UFS2. I mount those from my Linux box with the in-tree ufs driver to read data off them, which on Linux is read-only for UFS2. So the driver is not theoretical for me, it is how I reach my own data.
More generally that is what the driver is for: reading FreeBSD, NetBSD, OpenBSD and Solaris disks from Linux. Anyone dual booting a BSD, migrating data off one, or doing recovery or forensics on a BSD or Solaris volume ends up here. It is a small user base but not an empty one, and it is the only in-tree way to do it.
That is also why the mount path bugs bother me. The images I feed it are real BSD volumes, sometimes left unclean after a crash, so an unusual but legitimate on-disk layout should not panic the kernel. The fixes I have been sending are all in that parse and mount path.
Thanks,
Ali
Jan Kara
reviewerQuoted 8 lines · Ali Ahmet Memis
Quoted 2 lines · Jan Kara
I'm a bit curious: Are you using UFS driver in Linux? Or do you know of anybody using it?
Yes. I run FreeBSD on a second machine that I keep as a small server, and its disks are UFS2. I mount those from my Linux box with the in-tree ufs driver to read data off them, which on Linux is read-only for UFS2. So the driver is not theoretical for me, it is how I reach my own data.
Cool! When real users are fixing what they are using the results are usually better than if it is just "close a couple of syzbot bugs" :).
Quoted 5 lines
More generally that is what the driver is for: reading FreeBSD, NetBSD, OpenBSD and Solaris disks from Linux. Anyone dual booting a BSD, migrating data off one, or doing recovery or forensics on a BSD or Solaris volume ends up here. It is a small user base but not an empty one, and it is the only in-tree way to do it.
Yep, I was expecting some real users still do exist but I didn't hear about anybody for a long time :).
Quoted 4 lines
That is also why the mount path bugs bother me. The images I feed it are real BSD volumes, sometimes left unclean after a crash, so an unusual but legitimate on-disk layout should not panic the kernel. The fixes I have been sending are all in that parse and mount path.
Sure.
Honza
Signature · 2 lines
Jan Kara <[email protected]> SUSE Labs, CR