On Fri, Nov 01, 2019 at 04:09:08PM +0100, David Sterba wrote: > The raid1c34 patches are not intrusive and could be backported on top of > 5.3 because all the preparatory work has been merged already. Indeed, that's how I ended up testing them. I couldn't get the 5.4-rc kernels to run long enough to do meaningful testing before they locked up. I tested with 5.3.8 + patches. I left out the last patch that removes the raid1c3 incompat flag because 5.3 didn't have the block group tree code to apply it to. I ran my raid1 and raid56 corruption recovery tests modified for raid1c3. The first test is roughly: mkfs.btrfs -draid1c3 -mraid1c3 /dev/vd[bcdef] mount /dev/vdb /test cp -a 9GB_data /test sync sysctl vm.drop_caches=3 diff -r 9GB_data /test head -c 9g /dev/urandom > /dev/vdb head -c 9g /dev/urandom > /dev/vdc sync sysctl vm.drop_caches=3 diff -r 9GB_data /test btrfs scrub start -Bd /test sysctl vm.drop_caches=3 diff -r 9GB_data /test btrfs scrub start -Bd /test sysctl vm.drop_caches=3 diff -r 9GB_data /test First scrub reported a lot of corruption on /dev/vdb and /dev/vdc. Second scrub reported no errors. diff (all instances) reported no differences. Second test is: mkfs.btrfs -draid6 -mraid1c3 /dev/vd[bcdef] # rest as above... Similar results: first scrub reported many errors as expected. Second scrub reported no errors. No diffs.
Attachment:
signature.asc
Description: PGP signature
