On Mon, 2019-07-01 at 22:05 +0200, Helge Deller wrote: > On 01.07.19 21:54, James Bottomley wrote: > > I recently had the pleasure of a complete reinstall and found that > > I > > can't set up palo on the new system (at least not palo that loads > > from > > an ext2/ext3 filesystem). The problem doesn't seem to be within > > palo > > itself, but the fact that mke2fs is no longer working for us. The > > symptom I see is that after a couple of uses the palo partition > > corrupts itself and then e2fsck destroys the iplboot stored in the > > badblocks list. > > > > The problem seems to be with the badblock list because if I do > > > > palo -I /dev/sdb > > > > Followed by a fsck -f on the palo partition, it finds errors and > > wants > > to clear the badblock inode. > > > > I can reproduce this simply by doing > > > > dd if=/dev/zero of=bbtest.img bs=1M count=100 > > losetup /dev/loop0 bbtest.img > > a=237; while [ $a -le 450 ]; do echo $a >> bblist.txt; a=$[$a+1]; > > done > > mke2fs -b 1024 -l /home/jejb/bblist.txt /dev/loop0 > > e2fsck -f /dev/loop0 > > > > With no palo involvement, so I don't think it's our fault. > > > > I'll take this to the ext2 tools development list. Whatever it is > > seems to be really old because I unearthed an mke2fs from 2011 that > > still has the problem. > > Thanks for trying to fix it. > I'm not using this boot-from-extfs feature on any of my machines. > All boot from a palo partition, so I agree, that hasn't been tested > much. Heh, I suppose I'm the only one who keeps tons of kernels in the boot partition and needs a filesystem to select them. I suppose I did write the feature, it makes sense I'm the only one using it. Empirically, I've found that as long as the badblocks don't include any block from 0-250 ext2/3 seems to be happy creating a valid filesystem. Unfortunately with the label being 64k (so even if the palo partition is first it starts at an offset of 64k) and our alignment being on 256k absolute, we usually end up starting the badblocks at around block 224. I also dumped an older functional palo partition and found that even there the bad blocks start at 224, so it definitely is something that changed in mke2fs. James
