{"id":"AZL-110984","summary":"CVE-2026-99081 affecting package kernel 6.6.157.1-1","details":"In the Linux kernel, the following vulnerability has been resolved:\n\nbnxt_en: fix DMA mapping length for padded small packets\n\nStefan Fleischmann reported Intel IOMMU DMA Read faults on BCM57412\nNetXtreme-E NICs when transmitting packets on VLAN/macvlan interfaces:\n\n  DMAR: [DMA Read NO_PASID] Request device [18:00.0] fault addr 0xfc499000\n        [fault reason 0x06] PTE Read access is not set\n  bnxt_en 0000:18:00.0 eno1np0: Abandoning msg {0xb4 0x41a} len: 0 due to firmware status: 0x2000001\n  ...\n  NETDEV WATCHDOG: eno1np0 (bnxt_en): transmit queue 0 timed out\n\nThe fault address (0xfc499000) is on an exact 4KB page boundary,\npointing to a DMA read buffer overrun.\n\nIn bnxt_start_xmit(), packets smaller than BNXT_MIN_PKT_SIZE (52 bytes),\nsuch as 42-byte untagged ARP frames, are padded:\n\n    if (length \u003c BNXT_MIN_PKT_SIZE) {\n        pad = BNXT_MIN_PKT_SIZE - length;\n        if (skb_pad(skb, pad))\n            goto tx_kick_pending;\n        length = BNXT_MIN_PKT_SIZE;\n    }\n\n    mapping = dma_map_single(&pdev-\u003edev, skb-\u003edata, len, DMA_TO_DEVICE);\n    ...\n    dma_unmap_len_set(tx_buf, len, len);\n\nHowever, 'len' was initialized earlier to skb_headlen(skb) (e.g. 42 bytes)\nand is left unadjusted after padding. Consequently, dma_map_single() and\ndma_unmap_len_set() map and track only 42 bytes.\n\nLater, the hardware TX buffer descriptor is programmed with the padded length:\n\n    txbd-\u003etx_bd_len_flags_type =\n        cpu_to_le32(((len + pad) \u003c\u003c TX_BD_LEN_SHIFT) | flags |\n                    TX_BD_FLAGS_PACKET_END);\n\nThe NIC DMA engine is thus instructed to read 52 bytes from a region where\nonly 42 bytes were DMA-mapped. If skb-\u003edata ends near the boundary of a 4KB\npage (within 'pad' bytes of the next page), the hardware DMA read overruns\ninto the unmapped adjacent page, triggering an IOMMU fault.\n\nThis issue was exposed after commit 447cbe95ebb9 (\"vlan: fix skb_under_panic\nand races when toggling HW VLAN offload\") because reserving extra VLAN\nheadroom rounded LL_RESERVED_SPACE from 48 up to 64 bytes, shifting skb-\u003edata\noffsets and potentially causing small frames to land right against page\nboundaries.\n\nFix this by using skb_put_padto(skb, BNXT_MIN_PKT_SIZE) in\nbnxt_start_xmit(), before skb_shinfo(skb)-\u003enr_frags is sampled,\nbecause padding might linearize the skb.\nThis ensures skb-\u003elen and skb_headlen(skb) reflect the padded size so\nthat dma_map_single() maps the full buffer and the descriptor length is\nconsistent. This also removes the temporary 'pad' variable and masking logic.\n\nThe padding is done after the SW USO branch, otherwise\nbnxt_sw_udp_gso_xmit() would account the padding as UDP payload\nand send it in extra segments.\n\nNote that SW USO segments are still not padded: with IPv4, a segment\ncarrying less than 10 bytes of UDP payload (small gso_size, or a short\nlast segment) is sent as a frame shorter than BNXT_MIN_PKT_SIZE.\nThis is a separate issue, left for a followup patch.","modified":"2026-10-24T14:16:40.386860604Z","published":"2026-10-23T13:17:32Z","upstream":["CVE-2026-99081"],"references":[{"type":"WEB","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-99081"}],"affected":[{"package":{"name":"kernel","ecosystem":"Azure Linux:3","purl":"pkg:rpm/azure-linux/kernel"},"ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"last_affected":"6.6.157.1-1"}]}],"database_specific":{"source":"https://github.com/microsoft/AzureLinuxVulnerabilityData/blob/main/osv/AZL-110984.json"}}],"schema_version":"1.9.0"}