{ "dataType": "CVE_RECORD", "dataVersion": "5.2", "cveMetadata": { "cveId": "CVE-2025-38209", "assignerOrgId": "416baaa9-dc9f-4396-8d5f-8c081fb06d67", "state": "PUBLISHED", "assignerShortName": "Linux", "dateReserved": "2025-04-16T04:51:23.994Z", "datePublished": "2025-07-04T13:37:28.853Z", "dateUpdated": "2026-08-05T12:00:22.452Z" }, "containers": { "cna": { "providerMetadata": { "orgId": "416baaa9-dc9f-4396-8d5f-8c081fb06d67", "shortName": "Linux", "dateUpdated": "2026-08-05T12:00:22.452Z" }, "descriptions": [ { "lang": "en", "value": "In the Linux kernel, the following vulnerability has been resolved:\n\nnvme-tcp: remove tag set when second admin queue config fails\n\nCommit 104d0e2f6222 (\"nvme-fabrics: reset admin connection for secure\nconcatenation\") modified nvme_tcp_setup_ctrl() to call\nnvme_tcp_configure_admin_queue() twice. The first call prepares for\nDH-CHAP negotitation, and the second call is required for secure\nconcatenation. However, this change triggered BUG KASAN slab-use-after-\nfree in blk_mq_queue_tag_busy_iter(). This BUG can be recreated by\nrepeating the blktests test case nvme/063 a few times [1].\n\nWhen the BUG happens, nvme_tcp_create_ctrl() fails in the call chain\nbelow:\n\nnvme_tcp_create_ctrl()\n nvme_tcp_alloc_ctrl() new=true ... Alloc nvme_tcp_ctrl and admin_tag_set\n nvme_tcp_setup_ctrl() new=true\n nvme_tcp_configure_admin_queue() new=true ... Succeed\n nvme_alloc_admin_tag_set() ... Alloc the tag set for admin_tag_set\n nvme_stop_keep_alive()\n nvme_tcp_teardown_admin_queue() remove=false\n nvme_tcp_configure_admin_queue() new=false\n nvme_tcp_alloc_admin_queue() ... Fail, but do not call nvme_remove_admin_tag_set()\n nvme_uninit_ctrl()\n nvme_put_ctrl() ... Free up the nvme_tcp_ctrl and admin_tag_set\n\nThe first call of nvme_tcp_configure_admin_queue() succeeds with\nnew=true argument. The second call fails with new=false argument. This\nsecond call does not call nvme_remove_admin_tag_set() on failure, due to\nthe new=false argument. Then the admin tag set is not removed. However,\nnvme_tcp_create_ctrl() assumes that nvme_tcp_setup_ctrl() would call\nnvme_remove_admin_tag_set(). Then it frees up struct nvme_tcp_ctrl which\nhas admin_tag_set field. Later on, the timeout handler accesses the\nadmin_tag_set field and causes the BUG KASAN slab-use-after-free.\n\nTo not leave the admin tag set, call nvme_remove_admin_tag_set() when\nthe second nvme_tcp_configure_admin_queue() call fails. Do not return\nfrom nvme_tcp_setup_ctrl() on failure. Instead, jump to \"destroy_admin\"\ngo-to label to call nvme_tcp_teardown_admin_queue() which calls\nnvme_remove_admin_tag_set()." } ], "metrics": [ { "cvssV3_1": { "version": "3.1", "vectorString": "CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H", "baseScore": 9.8, "baseSeverity": "CRITICAL" }, "scenarios": [ { "lang": "en", "value": "AV:N - The trigger — failure of the second admin-queue configuration — is entirely controlled by the remote NVMe-over-TCP target across the network (TCP reset, failed TLS-PSK handshake, or an error status to Connect/Enable/Identify). A malicious, compromised, or impersonated storage target attacks the initiator host's kernel purely over TCP.\nAC:L - A rogue target reproduces the condition on demand and 100% reliably by simply refusing or aborting the second connection; nothing depends on timing, memory layout, or state outside the attacker's control, and the bug reproduces after a few blktests nvme/063 iterations. The required kernel options (CONFIG_NVME_TCP_TLS, CONFIG_NVME_HOST_AUTH) are enabled in mainstream distribution kernels.\nPR:N - The attacker needs no privileges or credentials on the victim host, and the failing second connection happens before any successful mutual authentication. Because DH-CHAP is unidirectional by default (host authenticates to controller), an impersonating target can pass the first handshake without knowing the secret and then simply fail the TLS reconnect.\nUI:N - NVMe-oF controllers are created automatically by nvmf-autoconnect systemd/udev tooling at boot and after controller-loss timeouts, so no human action is required at exploitation time. The attacker only has to be the endpoint the host is already configured to connect to.\nS:U - The use-after-free is on a kernel slab object and its consequences stay entirely within the kernel's own security authority. No VM, IOMMU, or sandbox boundary is crossed.\nC:H - The freed nvme_tcp_ctrl containing the live blk_mq_tag_set is subsequently dereferenced by blk_mq_queue_tag_busy_iter(), which follows set->tags[] pointers out of reclaimed heap memory. Reoccupying the slab yields an arbitrary kernel read primitive and leaks kernel pointers.\nI:H - The dangling tag set is used to call set->ops->timeout(), an indirect call through a function-pointer table read from freed, sprayable heap memory, giving control-flow hijack and arbitrary write potential. Each failed connect attempt leaks another dangling object, allowing repeated heap grooming.\nA:H - The reported failure mode is a KASAN slab-use-after-free in blk_mq_queue_tag_busy_iter() leading to kernel oops/panic, and the corrupted structure is walked by the block-layer timeout worker. A remote attacker can induce it repeatedly on every connection attempt." } ] } ], "affected": [ { "product": "Linux", "vendor": "Linux", "defaultStatus": "unaffected", "repo": "https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git", "programFiles": [ "drivers/nvme/host/tcp.c" ], "versions": [ { "version": "104d0e2f622233477ef7e57e59e8a4c3bb062c82", "lessThan": "db1da838b6012e4570c6f81e28ffe1d0ff595948", "status": "affected", "versionType": "git" }, { "version": "104d0e2f622233477ef7e57e59e8a4c3bb062c82", "lessThan": "e7143706702a209c814ed2c3fc6486c2a7decf6c", "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/tcp.c" ], "versions": [ { "version": "6.15", "status": "affected" }, { "version": "0", "lessThan": "6.15", "status": "unaffected", "versionType": "semver" }, { "version": "6.15.4", "lessThanOrEqual": "6.15.*", "status": "unaffected", "versionType": "semver" }, { "version": "6.16", "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": "6.15", "versionEndExcluding": "6.15.4" }, { "vulnerable": true, "criteria": "cpe:2.3:o:linux:linux_kernel:*:*:*:*:*:*:*:*", "versionStartIncluding": "6.15", "versionEndExcluding": "6.16" } ] } ] } ], "references": [ { "url": "https://git.kernel.org/stable/c/db1da838b6012e4570c6f81e28ffe1d0ff595948" }, { "url": "https://git.kernel.org/stable/c/e7143706702a209c814ed2c3fc6486c2a7decf6c" } ], "title": "nvme-tcp: remove tag set when second admin queue config fails", "x_generator": { "engine": "bippy-1.2.0" } } } }