Linux kernel nvme-tcp: a controller can force a host WARN by answering a write with C2HData

Impact

nvme_tcp_handle_c2h_data() never checks that the named command actually asked to read data. A controller - or anyone who can impersonate one on the storage network - that answers a write command with a C2HData PDU drives the host into _copy_to_iter() and trips WARN_ON_ONCE(i->data_source). No data is copied, so this is not memory corruption; what the controller gets is a kernel warning it can raise at will, plus a controller reset each time. On hosts booted with panic_on_warn, which is common in fleets that want crash-on-anomaly behaviour, that warning is a remote node kill. GPU nodes that mount datasets or checkpoints over NVMe/TCP can be knocked out repeatedly by a single misbehaving or hostile target.

Who can reach it

Network, unauthenticated at the NVMe layer - anyone who can present as the NVMe/TCP controller the host connects to, which in practice means a compromised storage target or a position on the storage network path. No tenant credentials required; tenants inside a GPU pod cannot reach this directly.

What to do

Apply a stable kernel that rejects a C2HData PDU whose command is not a read. Kernel-image change: drain and reboot each NVMe/TCP client node. Until then, the exposure is bounded by keeping the storage network segregated from tenant traffic, and by not running NVMe/TCP hosts with panic_on_warn, which is what turns this from noise into an outage.

References

Related entries

• Linux kernel (net/tls): When a NIC with active kTLS offload goes down, the offload teardown freed the TLS context whileCVE-2021-47131 · Linux kernel (net/tls)High
• VMware ESXi / Workstation / Fusion (storage controller out-of-bounds read/write): A malicious actor inside a VMCVE-2024-22273 · VMware ESXi / Workstation / Fusion (storage controller out-of-bounds read/write)High
• Linux kernel (net/xfrm): The error path of xfrm_input leaves the secpath entry pointing at poisoned memory, and theCVE-2024-43878 · Linux kernel (net/xfrm)High
• Linux kernel NVMe target authentication (nvmet-auth DH group setup): CtrlCVE-2024-50215 · Linux kernel NVMe target authentication (nvmet-auth DH group setup)High
• OpenSSH (sshd): regreSSHion: signal-handler race in sshd giving unauthenticated remote root on glibc LinuxCVE-2024-6387 · OpenSSH (sshd)High
• Linux kernel (net/smc): The CLC prefix-match check on the listen path dereferences the destination cache entry'sCVE-2025-40168 · Linux kernel (net/smc)High