On Fri, Mar 2, 2018 at 10:25 AM, Miklos Szeredi <miklos@xxxxxxxxxx> wrote: > Hi, > > I get xfstests failure of overlay/017 on a kvm guest. The symptom is > that the chardev turns into a whiteout (i.e. the 1/1 device number > magically turns to 0/0 after flushing the caches) but then turns back > into the old chardev on remount or another cache flush. So the > underlying xfs is not currupted but the cache is (verified that i_rdev > was indeed zeroed out on the relevant inode). > > Btw. i_rdev on the blkdev also turned zero, but the test doesn't > respond to that, because it's only checking the inum and not other > attributes. > > The test triggers reliably, but shows signs of a heisenbug (e.g. > adding printks lessens or eliminates the chance of triggering) > > Attaching the full xfs trace, the output of the testcase and guest > kernel config. > > Haven't been able to reproduce without overlayfs, so possibly > overlayfs is the culprit. > > Any ideas? > That is supposed to be fixed by acd1d71598f7 xfs: preserve i_rdev when recycling a reclaimable inode It fixes a v4.15 regression. Cheers, Amir. -- To unsubscribe from this list: send the line "unsubscribe linux-xfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
