On April 13, linux@xxxxxxxxxxx wrote: > I'm looking forward to testing the new SATA NCQ support that the linux > IDE developers have working, but of course that opens me to the risk of > disk corruption. > > So I'd like to be able to do clever things with the existing RAID arrays > to mitigate the damage. > > I have a lot of useful-but-not-essential data (old backups) on a RAID 5 > partition. I'm willing to risk the data but I'd rather not delete it > preemptively. I also don't want silent corruption. > > What I'd like to be able to do is, after a few experiments, verify the > RAID 5 parity to see if I've scrogged things. Is there some option to > mdadm to do that? I could just md5sum the component partitions, but it > would be useful information to also know *where* the damage occurred. > Relative to the LBA28 limit, for example. Try echo check > /sys/block/mdX/md/sync_action It doesn't tell to where it finds errors (though maybe it could). It will report a count of errors in /sys/block/mdX/md/mismatch_cnt > > And similarly for RAID-1 mirrors. I have the most essential data > (vmlinuz and a basic test mode installation) on a 6-way RAID-1, so if > something doesn't match, it's easy enough to find a valid copy. But I > need to notice that things aren't right. > > > The other thing I'd like to do is to be able to reintegrate a mirrored > pair (particularly RAID-10) in the "wrong" direction. That is: > - I split the mirror and literally unplug half so it can't get > corrupted. > - I boot with a test kernel and run my tests. Oops! It scribbled on > the disk. > - I want to reboot with a known-good kernel and regenerate the mirrors > from the old, good data. > > One thing I could do is remove the experimental partitions from the mirror > set and generate a fresh RAID-0 md to experiment on. Then I can mount > the old md without fear of it getting overwritten. But if I forget, > is there something I can do before plugging in the old, good drives that > will ensure that the kernel doesn't copy the new, corrupted data on top > of the old, good data as soon as I reboot? mdadm --zero-superblock /dev/experimental-device Then the old/good data will be all that is found. NeilBrown - To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
