{"dataType":"CVE_RECORD","dataVersion":"5.2","cveMetadata":{"cveId":"CVE-2026-99715","assignerOrgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","state":"PUBLISHED","assignerShortName":"Linux","dateReserved":"2026-09-25T10:25:14.430Z","datePublished":"2026-11-20T05:50:53.826Z","dateUpdated":"2026-11-22T11:09:50.333Z"},"containers":{"cna":{"providerMetadata":{"orgId":"416baaa9-dc9f-4396-8d5f-8c081fb06d67","shortName":"Linux","dateUpdated":"2026-11-22T11:09:50.333Z"},"descriptions":[{"lang":"en","value":"In the Linux kernel, the following vulnerability has been resolved:\n\nnvme-rdma: fix NULL queue->ctrl dereference on reconnect\n\nSince commit 2a8513091d2f (\"nvme-rdma: parallelize I/O queue allocation\nand startup\"), a reconnect that gets more I/O queues than before grows\nthe tag set before the new queues are set up. queue->ctrl is only set\nby nvme_rdma_setup_io_queues(), so I/O submitted in between reaches\nnvme_rdma_queue_rq() on a queue that was never allocated and\ndereferences a NULL queue->ctrl:\n\n  BUG: kernel NULL pointer dereference, address: 000000000000034c\n  RIP: 0010:nvme_rdma_queue_rq+0x52/0xb90\n\nSet queue->ctrl for every queue when allocating the controller. Such\nrequests are then requeued like on any other queue that is not live."}],"metrics":[{"cvssV3_1":{"version":"3.1","vectorString":"CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:N/I:N/A:H","baseScore":7.5,"baseSeverity":"HIGH"},"scenarios":[{"lang":"en","value":"AV:N - The target at the IP in nvme_rdma_alloc_ctrl (RDMA port 4420) sends RDMA_CM_EVENT_DISCONNECTED into nvme_rdma_cm_handler and, on reconnect, a Set Features completion for NVME_FEAT_NUM_QUEUES whose cdw0 nvme_set_queue_count stores as a larger queue_count. nvme_rdma_configure_io_queues calls nvme_unfreeze before nvme_rdma_setup_io_queues, so blk-mq enters nvme_rdma_queue_rq on a new hctx whose queue->ctrl is still NULL.\nAC:L - The target chooses the NUM_QUEUES cdw0, so an early connect can leave higher kzalloc'd nvme_rdma_queue slots with a NULL ctrl and a later reconnect can raise tag_set.nr_hw_queues past them. It drops the RDMA session once it sees namespace I/O; nvme_unfreeze then wakes those submitters into nvme_rdma_queue_rq before setup stores queue->ctrl, and ctrl_loss_tmo lets it repeat.\nPR:N - nvme_rdma_cm_handler and nvme_set_queue_count run on the initiator's outbound RDMA connection to the configured target and do not consult a host uid, capability, or DH-HMAC identity of a local user. The peer needs no access to /dev/nvme-fabrics.\nUI:N - nvmf_should_reconnect schedules nvme_rdma_reconnect_ctrl_work after the disconnect (NVMF_DEF_RECONNECT_DELAY is 10s) with no mount or nvme-connect step in the trigger. The target itself emits the CM disconnect and the NUM_QUEUES completion.\nS:U - nvme_rdma_queue_rq faults on a NULL queue->ctrl in the initiator kernel that owns that controller. The pointer is a zeroed slot in the host's queues[] array, and the oops is confined to that kernel.\nC:N - nvme_rdma_alloc_ctrl kzallocs queues[], so an unsetup queue->ctrl is NULL. nvme_rdma_queue_rq feeds &queue->ctrl->ctrl to nvme_check_ready, which faults on READ_ONCE(ctrl->state) (commit oops at 000000000000034c) before any CQE or data buffer is returned to the peer.\nI:N - The crashing access is READ_ONCE of nvme_ctrl.state through the NULL queue->ctrl of a slot nvme_rdma_setup_io_queues has not initialized. nvme_rdma_queue_rq takes that path from nvme_check_ready and returns before nvme_rdma_map_data or nvme_rdma_post_send.\nA:H - The commit's trace is a kernel NULL-pointer oops at nvme_rdma_queue_rq+0x52/0xb90. That oops hits the block I/O dispatch path of the controller and panics hosts with panic_on_oops; another reconnect reopens the same window."}]}],"affected":[{"product":"Linux","vendor":"Linux","defaultStatus":"unaffected","repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/nvme/host/rdma.c"],"versions":[{"version":"2a8513091d2f0b9a1e94b1843c48c712e7c17301","lessThan":"ed9e1f64bc79c92340b3d57c7294542ed13eb280","status":"affected","versionType":"git"},{"version":"2a8513091d2f0b9a1e94b1843c48c712e7c17301","lessThan":"3ee50a2d76f278ea0ced45716a62730059283787","status":"affected","versionType":"git"}]},{"product":"Linux","vendor":"Linux","defaultStatus":"affected","repo":"https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git","programFiles":["drivers/nvme/host/rdma.c"],"versions":[{"version":"7.3","status":"affected"},{"version":"0","lessThan":"7.3","status":"unaffected","versionType":"semver"},{"version":"7.3.5","lessThanOrEqual":"7.3.*","status":"unaffected","versionType":"semver"},{"version":"7.4-rc2","lessThanOrEqual":"*","status":"unaffected","versionType":"original_commit_for_fix"}]}],"cpeApplicability":[{"nodes":[{"operator":"OR","negate":false,"cpeMatch":[{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"7.3","versionEndExcluding":"7.3.5"},{"vulnerable":true,"criteria":"cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*","versionStartIncluding":"7.3","versionEndExcluding":"7.4-rc2"}]}]}],"references":[{"url":"https://git.kernel.org/stable/c/ed9e1f64bc79c92340b3d57c7294542ed13eb280"},{"url":"https://git.kernel.org/stable/c/3ee50a2d76f278ea0ced45716a62730059283787"}],"title":"nvme-rdma: fix NULL queue->ctrl dereference on reconnect","x_generator":{"engine":"bippy-1.2.0"}}}}