LINBIT/windrbd

BSOD on drbdadm down / drbdadm disconnect

johannesthoma opened this issue · 2 comments

When disconnecting while sync is still in progress sometimes there is a BSOD in drbd_free_peer_req(). The version is 1.0.0-rc5. This has also been reported by a user (but this might be a different BSOD) HenryKellner.

Some patches to the workqueue layer made happening this less frequently but the bug is still there in 1.0.0-rc7 (about 75 iterations were neccessary to reproduce it). Will keep on trying to fix this.

The relevant test setup to reproduce it is:

A Primary (can be Linux) doing I/O.
A Secondary Windows running disconnect-connect-wait-connected-loop-check-if-disconnect-succeeded-and-attach.sh
(see the windrbd-test folder in the repo).

We now wait for all IRPs on the backing device to complete (which takes some milliseconds) and this seems to solve the issue (no BSOD any more on disconnect / down)

Closing this issue.