{"id":"AZL-77151","summary":"CVE-2026-23086 affecting package kernel for versions less than 6.6.126.1-1","details":"In the Linux kernel, the following vulnerability has been resolved:\n\nvsock/virtio: cap TX credit to local buffer size\n\nThe virtio transports derives its TX credit directly from peer_buf_alloc,\nwhich is set from the remote endpoint's SO_VM_SOCKETS_BUFFER_SIZE value.\n\nOn the host side this means that the amount of data we are willing to\nqueue for a connection is scaled by a guest-chosen buffer size, rather\nthan the host's own vsock configuration. A malicious guest can advertise\na large buffer and read slowly, causing the host to allocate a\ncorrespondingly large amount of sk_buff memory.\nThe same thing would happen in the guest with a malicious host, since\nvirtio transports share the same code base.\n\nIntroduce a small helper, virtio_transport_tx_buf_size(), that\nreturns min(peer_buf_alloc, buf_alloc), and use it wherever we consume\npeer_buf_alloc.\n\nThis ensures the effective TX window is bounded by both the peer's\nadvertised buffer and our own buf_alloc (already clamped to\nbuffer_max_size via SO_VM_SOCKETS_BUFFER_MAX_SIZE), so a remote peer\ncannot force the other to queue more data than allowed by its own\nvsock settings.\n\nOn an unpatched Ubuntu 22.04 host (~64 GiB RAM), running a PoC with\n32 guest vsock connections advertising 2 GiB each and reading slowly\ndrove Slab/SUnreclaim from ~0.5 GiB to ~57 GiB; the system only\nrecovered after killing the QEMU process. That said, if QEMU memory is\nlimited with cgroups, the maximum memory used will be limited.\n\nWith this patch applied:\n\n  Before:\n    MemFree:        ~61.6 GiB\n    Slab:           ~142 MiB\n    SUnreclaim:     ~117 MiB\n\n  After 32 high-credit connections:\n    MemFree:        ~61.5 GiB\n    Slab:           ~178 MiB\n    SUnreclaim:     ~152 MiB\n\nOnly ~35 MiB increase in Slab/SUnreclaim, no host OOM, and the guest\nremains responsive.\n\nCompatibility with non-virtio transports:\n\n  - VMCI uses the AF_VSOCK buffer knobs to size its queue pairs per\n    socket based on the local vsk-\u003ebuffer_* values; the remote side\n    cannot enlarge those queues beyond what the local endpoint\n    configured.\n\n  - Hyper-V's vsock transport uses fixed-size VMBus ring buffers and\n    an MTU bound; there is no peer-controlled credit field comparable\n    to peer_buf_alloc, and the remote endpoint cannot drive in-flight\n    kernel memory above those ring sizes.\n\n  - The loopback path reuses virtio_transport_common.c, so it\n    naturally follows the same semantics as the virtio transport.\n\nThis change is limited to virtio_transport_common.c and thus affects\nvirtio-vsock, vhost-vsock, and loopback, bringing them in line with the\n\"remote window intersected with local policy\" behaviour that VMCI and\nHyper-V already effectively have.\n\n[Stefano: small adjustments after changing the previous patch]\n[Stefano: tweak the commit message]","modified":"2026-08-28T17:48:01.977494389Z","published":"2026-02-04T17:16:19Z","upstream":["CVE-2026-23086"],"references":[{"type":"WEB","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-23086"}],"affected":[{"package":{"name":"kernel","ecosystem":"Azure Linux:3","purl":"pkg:rpm/azure-linux/kernel"},"ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"fixed":"6.6.126.1-1"}]}],"database_specific":{"source":"https://github.com/microsoft/AzureLinuxVulnerabilityData/blob/main/osv/AZL-77151.json"}}],"schema_version":"1.9.0"}