Incremental backup fails on a 3 TB qcow2 disk: value is out of range ... <= 4294967295 in computeVhdBatAndFileSize

[Razielh]

Hi,

An incremental (delta) backup job fails on a VM that has a 3 TB qcow2 disk. The run is a “full” (base) export, and the transfer fails after about 4 minutes.

Environment

XCP-ng 8.3.0, sm-3.2.12-25.1.xcpng8.3
Xen Orchestra built from sources (ronivay installer), commit 3923756ff (2026-10-22, latest master at the time of testing)
Backup type: incremental / delta (this run: Type: full)
Remote: SMB share on a Synology NAS
VM: 1 VDI of 3 TB in qcow2 format

What happens

The snapshot succeeds.
The transfer to the remote fails after about 4 minutes. The same error is reported at the transfer, remote and job levels:
Error: The value of "value" is out of range. It must be >= 0 and <= 4294967295. Received 4_294_971_137

Stack trace (xo-server)

xo:backups:AbstractVmRunner WARN writer step failed {
error: RangeError [ERR_OUT_OF_RANGE]: The value of "value" is out of range. It must be >= 0 and <= 4294967295. Received 4_294_971_137
at checkInt (node:internal/buffer:78:11)
at writeU_Int32BE (node:internal/buffer:808:3)
at Buffer.writeUInt32BE (node:internal/buffer:821:10)
at DiskConsumerVhdStream.computeVhdBatAndFileSize (packages/vhd-lib/disk-consumer/BaseVhd.mjs:73:11)
at DiskConsumerVhdStream.toStream (packages/vhd-lib/disk-consumer/DiskConsumerVhdStream.mjs:62:36)
at toVhdStream (packages/vhd-lib/disk-consumer/index.mjs:31:19)
at RemoteAdapter.writeVhd (@xen-orchestra/backups/RemoteAdapter.mjs:695:28)
at asyncEach.concurrency (@xen-orchestra/backups/_runners/_writers/IncrementalRemoteWriter.mjs:284:45)
at next (@vates/async-each/index.js:90:37) {
code: 'ERR_OUT_OF_RANGE'
},
step: 'writer.transfer()',
writer: 'IncrementalRemoteWriter'
}

This is followed by VDI_IN_USE (destroy) retries on the snapshot cleanup, which look like a side effect of the failed transfer.

Analysis
The crash happens while computing the VHD BAT. The BAT stores block offsets as 32-bit sector numbers (512-byte units), so a VHD cannot address data beyond ~2 TiB. The qcow2 disk is 3 TB, and the stream’s block offsets go past that limit (4,294,971,137 sectors ≈ 2 TiB + 1.9 MiB).

In the meantime, I’m protecting this VM’s data with a file-level backup agent, but I’d like to go back to XO for the whole VM.

Thanks!