On Oct 20, 2022 / 16:18, Jaegeuk Kim wrote:

...

> Thanks, I think that fix looks good to me. I applied into the original patch.
> https://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs-tools.git/commit/?h=dev&id=281d3e72370f6c39c0d57acaf37a7f0e003ddd28

Oh, happy to know that the fix is good. And thank you for adding my SoB tag.

One more thing, my fix missed care for fsck/resize.c. I suggest to apply one
more hunk below to the commit in same manner as mkfs/f2fs_format.c.

diff --git a/fsck/resize.c b/fsck/resize.c
index c048b16..79945e1 100644
--- a/fsck/resize.c
+++ b/fsck/resize.c
@@ -476,8 +476,8 @@ static void rebuild_checkpoint(struct f2fs_sb_info *sbi,
                        get_cp(rsvd_segment_count)) *
                        c.new_overprovision / 100);

-       if (get_cp(rsvd_segment_count) > get_cp(overprov_segment_count))
-               ASSERT_MSG("Cannot support wrong overprovision ratio\n");
+       if (get_cp(overprov_segment_count) < get_cp(rsvd_segment_count))
+               set_cp(overprov_segment_count, get_cp(rsvd_segment_count));

        DBG(0, "Info: Overprovision ratio = %.3lf%%\n", c.new_overprovision);
        DBG(0, "Info: Overprovision segments = %u (GC reserved = %u)\n",

Without this change, the assert message "Cannot support wrong overprovision
ratio" was printed when I ran resize.f2fs -t X command to the f2fs formatted
with mkfs.f2fs -s Y. With the change above, the assert message is not printed.

> 
> > 
> > FYI, I tried to fix and created a patch which allows reserved segments 
> > larger
> > than overprovisioning segments [1]. It compares those two, and take larger 
> > one
> > to subtract from usable segments to get the segments for users. I confirmed 
> > it
> > keeps small number of overprovisioning segments for no -s option case, and
> > avoids the mkfs.f2fs failure for the -s option and zoned block device cases.
> > However, it increases runtime of my test script which fills f2fs and do file
> > overwrites to test f2fs GC on zoned block devices. It takes +60% longer 
> > runtime.
> > Then GC performance looks worse than before, and this fix does not look good
> > for me.
> 
> I think you can try to avoid that by tuning /sys/fs/f2fs/xx/reserved_blocks?

Thanks, I was able to shorten the runtime using the sysfs attribute. So the
longer GC time was just caused by the smaller reserved segments size. Good.

-- 
Shin'ichiro Kawasaki

_______________________________________________
Linux-f2fs-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/linux-f2fs-devel

Reply via email to