Has anyone else seen TRIM stop reclaiming space from a Windows 11 25H2 VM after the May 2026 updates? I'm running Proxmox VE 9.2-1 with a single VM on LVM-thin storage. Windows reports that TRIM completed successfully through Defragment and PowerShell, but Proxmox still considers the released blocks allocated. Running fstrim on the host does not reclaim the space. SDelete and GParted cause the thin pool usage to increase instead, eventually consuming all available thin space. Removing the virtual disk immediately frees the space, and adding it back makes the pool appear full again. I found reports that a Windows storage-driver change in the May updates may be responsible. Is this a confirmed issue, and is there a reliable workaround besides rolling back updates, changing the virtual disk bus, or rebuilding the VM?
2 Answers
There is a known Windows virtual-storage issue tracked by the VirtIO/KVM guest-driver project that appears related to TRIM or discard handling after the May 2026 updates. The workaround described there involves using the affected driver’s correction or reverting to a pre-update driver/software combination, rather than simply running fstrim again on the Proxmox host. Changing the disk from SCSI to SATA is not guaranteed to help, because the problem can be in how Windows and the guest storage driver pass discard information through to the thin-provisioned device.
I saw nearly the same behavior with a Windows Server 2019 VM. Creating a new virtual disk seemed to restore reclaiming, although I’m not completely sure why. On the original disk, running SDelete followed by TRIM would release some space, but ordinary filesystem free space was never reclaimed. Be careful with SDelete on a thin-provisioned disk: writing the entire free area can consume the remaining pool space and freeze the VM.
That matches what happened here. SDelete eventually filled the thin pool and froze the VM. It seems to free only the blocks it actively overwrites, while Proxmox continues treating other Windows free space as allocated. Removing and reattaching the disk changes the reported usage but doesn’t fix the underlying behavior.
I found that issue as well, but the practical workaround wasn’t clear. Rolling back the update or replacing the virtual disk seems more like avoiding the problem than repairing TRIM. For now, keeping the VM disk smaller and leaving enough headroom in the thin pool is the safest option.