{"id":"AZL-115435","summary":"CVE-2026-99638 affecting package kernel 6.6.161.1-1","details":"In the Linux kernel, the following vulnerability has been resolved:\n\nnet/sched: clamp skb-\u003equeue_mapping before indexing the tx queue\n\nThis issue was caught by Sashiko (nipa). A producer-side path cap is\nalready merged (ea4d4b5dddb5); this patch adds the matching consumer-side\ncaps.\n\nThe issue:\nThe tx queue is picked and capped in netdev_core_pick_tx() before the\n__dev_xmit_skb() qdisc enqueue path. A tc BPF program running inside an\negress qdisc (not clsact) can then change __sk_buff-\u003equeue_mapping; that\nstore is never re-checked. bpf_convert_ctx_access() bounds it only by\nNO_QUEUE_MAPPING, not by the device's queue count.\n\nTwo consumers read the stale value:\n\n  - On dequeue, qdisc_restart() calls netdev_get_tx_queue(), which\n    indexes dev-\u003e_tx[]. netdev_get_tx_queue() guards it with just a\n    DEBUG_NET_WARN_ON_ONCE, so sch_direct_xmit() takes the out-of-bounds\n    txq-\u003e_xmit_lock and writes txq-\u003exmit_lock_owner past the end of the\n    allocation on non-lltx devices such as ifb. Drivers that index their\n    rings from skb_get_queue_mapping() (ex: ifb) go out of range the same\n    way.\n\n  - On enqueue, taprio_enqueue() indexes q-\u003eqdiscs[] with the same value;\n    the array is sized dev-\u003enum_tx_queues. A duplicating netem on a child\n    qdisc clones the skb and re-enqueues the clone at the taprio root\n    (qdisc_root_bh()) through a direct rootq-\u003eenqueue() call, so the value\n    reaches taprio_enqueue() before any dequeue cap runs.\n\nThe Fix:\nCap the mapping at the consumers and write the capped value back, so the\nqdisc's _tx[] lookup, the taprio q-\u003eqdiscs[] lookup and any later driver\nread stay in range.\n\nOn the dequeue side the cap folds into the dequeue path instead of a\nseparate walk over the returned list: dequeue_skb() caps the head, and\neach bulk helper caps the element it appends. For a qdisc that is not\nTCQ_F_ONETXQUEUE, netdev_cap_txqueue() logs a ratelimited notice and\nselects queue 0.\n\nA TCQ_F_ONETXQUEUE qdisc is bound to a single tx queue: dequeue_skb() uses\nq-\u003edev_queue for the frozen check, the BQL limit and the trace point, while\nqdisc_restart() transmits on the skb's queue. Capping an out-of-range\nmapping to queue 0 would therefore send a per-queue child's skb through a\ndifferent queue than the one it was dequeued for. On mq, mqprio, taprio and\nhtb offload children, a stopped queue 0 could then park the list in\nq-\u003egso_skb, and the later wake would schedule queue 0's own child instead\nof this qdisc. Pin a TCQ_F_ONETXQUEUE mapping to the qdisc's own queue to\nkeep the documented invariant.\n\nThe pin covers every mapping, not only an out-of-range one: an in-range\nmapping that a filter sets on one of those per-queue children is rewritten\nto the child's own queue, and this path does not call netdev_cap_txqueue(),\nso it logs no notice. On a single-queue device qdisc_create() sets\nTCQ_F_ONETXQUEUE, so the one-queue trigger described below also takes this\nbranch and is corrected without a notice.\n\nThat also keeps a bulk list on one queue by construction: every element of\na TCQ_F_ONETXQUEUE qdisc, including a follower changed by a filter, is\npinned to q-\u003edev_queue before it is appended. The fast bulk helper\ntherefore needs no head comparison and no skb_bad_txq requeue; the slow\nhelper, used only by non-ONETXQUEUE qdiscs, still compares and requeues a\nmismatched follower, where a wake on any queue reschedules q.\n\nOn the enqueue side taprio_enqueue() caps the mapping against\nreal_num_tx_queues and writes it back before selecting q-\u003eqdiscs[queue], so\na duplicating child cannot steer the clone out of the array.\n\nConditions to recreate the bug:\nThe dequeue path: load a SCHED_CLS or SCHED_ACT program that stores\n__sk_buff-\u003equeue_mapping (a tc BPF action is enough), attach it to a\ntransmit qdisc (for example a matchall filter on a prio root), and send a\npacket. With a 1-queue device a store of 1 indexes one past dev-\u003e_tx[].\nUnder KASAN with panic_on_warn=1 the unfixed kernel faults with \"BUG:\nKASAN: slab-out-\n---truncated---","modified":"2026-11-21T14:17:23.933274495Z","published":"2026-11-20T06:17:57Z","upstream":["CVE-2026-99638"],"references":[{"type":"WEB","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-99638"}],"affected":[{"package":{"name":"kernel","ecosystem":"Azure Linux:3","purl":"pkg:rpm/azure-linux/kernel"},"ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"last_affected":"6.6.161.1-1"}]}],"database_specific":{"source":"https://github.com/microsoft/AzureLinuxVulnerabilityData/blob/main/osv/AZL-115435.json"}}],"schema_version":"1.9.2"}