{ "schema_version": "1.6.1", "id": "RPI-CVE-2026-98360", "modified": "2026-10-09T11:39:44Z", "upstream": [ "CVE-2026-98360" ], "published": "2026-10-06T08:47:10Z", "summary": "RDMA/rxe: insert mcg into mcg_tree only after rxe_mcast_add() succeeds", "details": "In the Linux kernel, the following vulnerability has been resolved:\n\nRDMA/rxe: insert mcg into mcg_tree only after rxe_mcast_add() succeeds\n\nrxe_get_mcg() publishes a newly allocated multicast group in\nrxe->mcg_tree before programming the backing Ethernet multicast address\nwith rxe_mcast_add(), which runs outside mcg_lock. A local userspace\nRDMA client reaches this path with ATTACH_MCAST on a UD QP; if\nrxe_mcast_add() then returns an error (for example -ENODEV when the\nbacking netdev has been removed, or a propagated dev_mc_add() error),\nthe unwind frees the published group without removing it from the tree.\nA later lookup of the same MGID dereferences the freed struct rxe_mcg\nfrom __rxe_lookup_mcg().\n\nFix this by keeping the new mcg private until rxe_mcast_add() succeeds.\nSplit the tree publication into __rxe_publish_mcg(), call rxe_mcast_add()\nbefore taking the tree reference, and free the still-private mcg on\nfailure. Because the group is never visible in mcg_tree until the\nmulticast address is programmed, no concurrent caller can look it up or\nattach a QP to a group that is about to be torn down, so the error path\nneeds no conditional unwind. If another caller publishes the same MGID\nwhile the address is being programmed, the post-add re-check under\nmcg_lock finds the winner; this caller then drops its private object and\nbalances its own rxe_mcast_add() with rxe_mcast_del() before returning\nthe winner.\n\nReproduced by forcing the rxe_mcast_add() error return under KASAN:\nwithout the change the next attach to the same MGID reports a\nslab-use-after-free in __rxe_lookup_mcg(); with it the forced failure\nreturns cleanly. A no-injection attach/detach regression, including a\ntwo-QP shared join/leave and re-attach, stays KASAN- and leak-clean.", "severity": [ { "type": "CVSS_V3", "score": "CVSS:3.1/AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H" } ], "affected": [ { "package": { "ecosystem": "Raspberry Pi OS:12", "name": "linux" }, "versions": [ "1:6.1.47-1+rpt4" ], "ecosystem_specific": { "binaries": [ { "binary_name": "linux-image-6.1.0-rpi3-rpi-2712", "binary_version": "1:6.1.47-1+rpt4" }, { "binary_name": "linux-image-6.1.0-rpi3-rpi-v6", "binary_version": "1:6.1.47-1+rpt4" }, { "binary_name": "linux-image-6.1.0-rpi3-rpi-v7", "binary_version": "1:6.1.47-1+rpt4" }, { "binary_name": "linux-image-6.1.0-rpi3-rpi-v7l", "binary_version": "1:6.1.47-1+rpt4" }, { "binary_name": "linux-image-6.1.0-rpi3-rpi-v8", "binary_version": "1:6.1.47-1+rpt4" } ] }, "database_specific": { "per_version": { "1:6.1.47-1+rpt4": { "linux_commit": "655fc658a15ae7a6f37103754adb39ba52a9a14e", "upstream_version": "6.1.47", "assessment": "commit ancestry (strak rule)", "upstream_fixed": "6.1.189", "compiled_filter": false } } } } ], "references": [ { "type": "ADVISORY", "url": "https://www.cve.org/CVERecord?id=CVE-2026-98360" }, { "type": "FIX", "url": "https://git.kernel.org/stable/c/ddb43ac0926d4a931bc9b7744b93627f627457e5" }, { "type": "FIX", "url": "https://git.kernel.org/stable/c/c79a789aa15180a1543b5db49c343d12e3ec214d" }, { "type": "FIX", "url": "https://git.kernel.org/stable/c/faae1fb4ccf8205806a8802c008798dabeb0205b" }, { "type": "FIX", "url": "https://git.kernel.org/stable/c/02c0a2fa69c16248a7432af8a6d64ab2a73a5283" }, { "type": "FIX", "url": "https://git.kernel.org/stable/c/d4fc4e37f8a143b0fe83b42c8fb48cf542154fee" }, { "type": "FIX", "url": "https://git.kernel.org/stable/c/1caceeb2d74bbe88223aea55eb8626b4c5f076fd" } ], "database_specific": { "source": "https://git.kernel.org/pub/scm/linux/security/vulns.git", "upstream_modified": "2026-10-07T06:50:27Z", "severity": { "source": "kernel CNA", "cvss_v3_score": 7.0, "cvss_v3_severity": "High", "scenario": "AV:L - The faulty publish-then-free sequence is in rxe_get_mcg(), reached from the uverbs ATTACH_MCAST command (ib_uverbs_attach_mcast -> ib_attach_mcast -> rxe_attach_mcast) on a local UD QP. A remote RoCEv2 multicast packet (rxe_rcv_mcast_pkt -> rxe_lookup_mcg) can only hit the stale node after a local caller has already caused the failure.\nAC:H - The dangling mcg is only left behind when rxe_mcast_add() fails after the group is published. That needs ib_device_get_netdev() to return NULL while the rxe netdev is being unregistered (an admin or hardware event), or a GFP_ATOMIC failure in __hw_addr_create(). An unprivileged user cannot produce either on demand.\nPR:L - ib_uverbs_attach_mcast() and rxe_attach_mcast() have no capability check, so any user who can open the rxe uverbs device and create a UD QP can drive rxe_get_mcg(). Creating the rxe link itself (RDMA_NL_ADMIN_PERM, checked with netlink_capable against the init namespace) is admin setup the attacker does not perform.\nUI:N - The attacker issues ATTACH_MCAST for the failing MGID and then the follow-up attach/detach/lookup themselves; no other user has to act.\nS:U - The freed rxe_mcg is kernel heap memory in the same kernel security authority; there is no VM, IOMMU or sandbox boundary crossed.\nC:H - The freed struct rxe_mcg stays linked in rxe->mcg_tree. __rxe_lookup_mcg() walks its rb_node and mgid, and later code follows its qp_list, so reallocating that slab slot with attacker-chosen data gives a heap read primitive.\nI:H - After the free, __rxe_lookup_mcg() does kref_get() on the stale object, and rxe_attach_mcg() list_add()s an mca into its qp_list and updates qp_num. These are writes into reallocated memory, and the rb-tree insert/erase relinks through the freed node.\nA:H - The next ATTACH_MCAST, detach, or received multicast lookup for that MGID dereferences the freed rxe_mcg (KASAN slab-use-after-free in __rxe_lookup_mcg()), corrupting the rb-tree and crashing the kernel." } } }