DRBD 9 edges toward mainline Linux with 7 prep patches

Christoph Böhmwalder posted a 7-patch series to the Linux kernel mailing lists on September 23, 2026. It is titled "drbd: preparation for DRBD 9". The series reshapes the kernel's old DRBD code so that the much newer DRBD 9 can follow it into mainline. That matters to anyone running replicated storage on Linux, because the version shipped inside the kernel is still on the older 8.4 branch.

DRBD stands for Distributed Replicated Block Device. It copies a disk, block by block, to one or more other machines over the network. If one server dies, another holds an identical copy of the data. LINBIT, the company behind it, says the project has been maintained for more than two decades.

What the 7 patches change

The cover letter describes the approach in one line. "Each patch moves a piece of the in-tree DRBD 8.4 code to the form it has in the out-of-tree DRBD 9 code," Böhmwalder wrote.

"In-tree" means code that lives in the official kernel source. "Out-of-tree" means code maintained separately and built as an add-on module. DRBD 9 has only ever been out-of-tree.

The cover letter lists these changes:

• Higher ceilings for fast networks: socket buffers up to 128 MiB, and a maximum resync rate of 8 GiB/s.
• The file drbd_worker.c is renamed to drbd_sender.c, matching DRBD 9's layout.
• The drbd_interval data structure's bitfields are replaced.
• The table of network packet names moves to a new location.
• Log messages gain rate limiting per object, so one failing device cannot flood the kernel log.

The series went to both the linux-kernel and linux-block lists. The cover letter stresses that nothing changes for current users: "Defaults are unchanged, all existing configurations remain valid."

Why DRBD 9 is worth the effort

LINBIT explained the goal in an April 27, 2026 blog post by Michael Troutman. The kernel carries DRBD 8.4.11, which the post calls stagnant. Active development happens on DRBD 9.3, outside the kernel.

The gap between the two is large.

Capability	DRBD 8.4 (in the kernel)	DRBD 9 (out-of-tree)
Nodes per volume	2, for failover	Up to 31 peers
Split-brain protection	Relies on fencing	Built-in quorum for 3 or more nodes
Failed-node handling	Manual STONITH or fencing setup	Automatic

Split brain is the failure where two machines both believe they hold the live copy and start writing different data. Quorum prevents it by letting a node write only while it can see a majority of the cluster. STONITH, short for "shoot the other node in the head", is the older fix: power off a suspect node before it can do damage. DRBD 9 also adds safeguards for syncing a node that rejoins the cluster.

LINBIT's post named Linux 7.2 as its target for the DRBD 9.3 patches. This week's series is described as preparation, so the full DRBD 9 code is a later, separate submission.

What this means for developers

If you run DRBD today, check which one you have. You may already load LINBIT's out-of-tree DRBD 9 module rather than the kernel's 8.4 driver. cat /proc/drbd prints the loaded version. This series changes nothing for either group yet, and the cover letter promises existing configurations stay valid.

If you rely on the in-kernel 8.4 driver, the upstream work is a reason to plan a test of DRBD 9 now. Once it lands, the in-kernel driver will gain multi-peer volumes and quorum. Those change how a cluster must be configured and fenced. Trying them on a staging cluster first beats discovering the differences after a distribution kernel upgrade.

If you build storage on fast networks, note the new limits. A resync ceiling of 8 GiB/s and 128 MiB socket buffers leave room for fast data-center links. LINBIT's site notes DRBD can replicate over TCP/IP or RDMA, the low-latency transport used on InfiniBand and RoCE networks. The defaults stay the same, so you still have to raise them yourself.

Watch the linux-block list for the next round. Review comments on these 7 patches will show how quickly maintainers accept a large, long-running out-of-tree driver. That answer decides when a stock kernel can replace a separately packaged DRBD module.