Ocfs2 devel - Apr 2023 - [PATCH] ocfs2: reduce ioctl stack usage

On Tue, Apr 18, 2023 at 05:37:06PM +0800, Joseph Qi
wrote:
> Andrew picked ocfs2 patches into -mm tree before.
Yup and that's fine obviously, but this belongs to fs/ and we're aiming
to take fs/ stuff through the dedicated fs trees going forward.

Thanks!
Christian
> 
> Thanks,
> Joseph
> 
> On 4/18/23 5:17 PM, Christian Brauner wrote:
> > 
> > On Mon, 17 Apr 2023 22:56:24 +0200, Arnd Bergmann wrote:
> >> On 32-bit architectures with KASAN_STACK enabled, the total stack
usage
> >> of the ocfs2_ioctl function grows beyond the warning limit:
> >>
> >> fs/ocfs2/ioctl.c: In function 'ocfs2_ioctl':
> >> fs/ocfs2/ioctl.c:934:1: error: the frame size of 1448 bytes is
larger than 1400 bytes [-Werror=frame-larger-than=]
> >>
> >> Move each of the variables into a basic block, and mark
ocfs2_info_handle()
> >> as noinline_for_stack, in order to have the variable share stack
slots.
> >>
> >> [...]
> > 
> > Going by git log, ocfs2 patches don't go through a separate tree.
> > So unless there are objections I'm taking this through fs.misc,
> > 
> > tree: git://git.kernel.org/pub/scm/linux/kernel/git/vfs/idmapping.git
> > branch: fs.misc
> > [1/1] ocfs2: reduce ioctl stack usage
> >       commit: 85ef56bc2d65215f43ceb7377ca14a779468928d
> > 
> > Christian

On Tue, Apr 18, 2023 at 02:56:38PM +0200, Christian Brauner
wrote:
> On Tue, Apr 18, 2023 at 05:37:06PM +0800, Joseph Qi wrote:
> > Andrew picked ocfs2 patches into -mm tree before.
> 
> Yup and that's fine obviously, but this belongs to fs/ and we're
aiming
> to take fs/ stuff through the dedicated fs trees going forward.
The challenge here is that there isn't an active ocfs2 tree at the
moment (git/jlbec/ocfs2.git was last updated 11 years ago), and it's
not clear ocfs2 has an active maintainer, although as discussed at a
recent ext4 telechat, there still are some enterprise users of it
hanging on.

We're starting considering changes to fs/jbd2 so we can get off of
using the struct bh for jbd2 (still very early days; it's more in the
early concept brain-storming stage.)  And since ocfs2 also uses
fs/jbd2, the fact that ocfs2 isn't under active maintenance is going
to be more of a challenge moving forward.

Something we should discuss in the hallway track of LSF/MM, perhaps...

							- Ted

On 4/18/23 8:56 PM, Christian Brauner wrote:
> On Tue, Apr 18, 2023 at 05:37:06PM +0800, Joseph Qi wrote:
>> Andrew picked ocfs2 patches into -mm tree before.
> Yup and that's fine obviously, but this belongs to fs/ and we're
aiming
> to take fs/ stuff through the dedicated fs trees going forward.
Either is fine for me.
Hi Andrew, what's your opinion?

Thanks,
Joseph

On Tue, Apr 18, 2023 at 02:56:38PM +0200, Christian Brauner
wrote:
> On Tue, Apr 18, 2023 at 05:37:06PM +0800, Joseph Qi wrote:
> > Andrew picked ocfs2 patches into -mm tree before.
> 
> Yup and that's fine obviously, but this belongs to fs/ and we're
aiming
> to take fs/ stuff through the dedicated fs trees going forward.
Er...  Assuming that there *is* an active fs tree for filesystem
in question.  Do you really want dedicated e.g. affs, adfs, etc.
git trees - one for each filesystem in there?