Skip to main content
Question

VME plug-in: protected VMs abort (QEMU nbd_blockdev_client_closed assertion) when backup sessions close — anyone else seeing this?

  • Nov 15, 2026 5:11 PM UTC
  • 0 comments
  • 3 views

Hi all,

 

Running Veeam Backup & Replication 13.1 (Linux VSA) with the Plug-in for HPE Morpheus VM Essentials against a two-node Morpheus 9.1 cluster (QEMU 8.2.2 on both hosts, Ubuntu builds .10 and .18). Nightly agentless job, NBD transport, one worker per host. The plug-in was 13.2.0.160 when this started in October; a worker update took it to 13.2.1.5, which has run since Nov 1 — the crashes reproduce on both versions.

 

Since mid-October, protected VMs intermittently die right when a backup session closes. The guest's QEMU process aborts on:

blockdev-nbd.c:62: nbd_blockdev_client_closed: Assertion 'qemu_in_main_thread() && nbd_server' failed

 

 

and the VM is gone. We have 130 verified crashes (with core dumps) between Oct 17 and Nov 6 alone, across both hosts, both QEMU builds, and Ubuntu and Debian guests, and it has continued nightly since. From a core backtrace, it looks like the disconnect of a fully negotiated NBD client gets processed in the VM's IO-thread context, which trips the guard assert that the CVE-2024-7409 fix added — so the "fixed" QEMU builds hard-abort the guest where the old code would have misbehaved differently. Survival seems to be a race, which fits the intermittency. Turning guest integration off doesn't help; the export close alone does it.

 

We have since correlated Veeam's own logs for the last two nights against our records: every death coincides with the termination of the guest's NBD backup session (usually in the close phase, right after "NBD client session completed"), 20 of the 30 fatal tasks were scored Success — including tasks whose own logs record `domain is not running` and `Failed to stop the backup job` — and the job's automatic retries re-ran the failed tasks and killed restarted VMs a second and third time the same night. Every victim task also falls back to a full backup after a checkpoint inconsistency, because the previous crash broke the chain, while the VMs that never crash just run clean incrementals.

 

We've now disabled the job until this is resolved, which leaves the affected VMs with no backups at all. A support case is being opened with full logs, but I wanted to ask here first:

 

- Is anyone else on the VME plug-in seeing guests crash at backup teardown?

- Has anyone found a plug-in/QEMU build, transport setting, or job option that avoids it (other than excluding VMs from the job, which leaves them with no backups)?

 

Happy to share sanitized timelines and the backtrace if useful.

 

Thanks!