I manually relocated a few bad blocks on a bad IBM drive I had when I replaced the drive. It took a lot of time and effort. I had to run the dd command many times very carefully to make it work. One big problem for me was that read-ahead obscured which actual sectors were in error. I needed a 'raw' LVM device, but I don't think such a thing exists for LVM1 on Linux 2.4.x. What I did was used pvmove to move the PE containing the bad block to a different spot on the hard drive, then allocated a new LV that was one LE long, and forced it to allocate the PE containing the bad block. Then I used dd to carefully copy over the LE in sections, narrowing down the location of the bad sectors until I had copied everything that could possibly be read. After that, I ran fsck on the filesystem that had originally contained the bad block, and I was fine. I checked carefully, and it didn't even seem that I had lost any data. Long, time consuming process though. Actually, it may have been even ickier than I first thought. It could be that pvmove wouldn't work, and I had to shorten the LV containing the bad block (the BLV) to contain all PEs prior to the bad one, allocate a new LV (the NLV) containing all the bad PE, lengthen the BLV by 1 PE, using a brand new PE, then lengthen it to its original length so it would contain all the PEs after that bad PE, the do the procedure I outlined above. Now that I think of it, I'm nearly positive that pvmove didn't work. I had dearly wished for some kind of option to pvmove that would force it to try as hard as it could to get good reads of all the sectors in a PE, then move the LE to a new PE, even if there were errors. Have fun (if at all possible), -- Eric Hopper <hopper@omnifarious.org> Omnifarious Software
Attachment:
signature.asc
Description: This is a digitally signed message part
