{"id":"AZL-101210","summary":"CVE-2026-90015 affecting package kernel 6.6.150.1-1","details":"In the Linux kernel, the following vulnerability has been resolved:\n\nxhci: fix lost bounce buffers on TDs spanning several ring segments\n\nWhen a TD reaches a link TRB with data that is not aligned to the\nendpoint's wMaxPacketSize, xhci_align_td() stages the unalignable tail\nthrough the bounce buffer of the ring segment holding that link TRB.\nxhci_unmap_td_bounce_buffer() later unmaps it and, for IN transfers,\ncopies the data back into the URB's buffer.\n\nThe enqueue path records the segment that was bounced in td-\u003ebounce_seg,\nunder the assumption that a TD never spans more than two ring segments.\nThat assumption does not hold: a TD large enough to span three or more\nsegments crosses several link TRBs and can be bounced at each of them.\nOnly the last one survives in td-\u003ebounce_seg, so every earlier bounce\nbuffer is neither copied back nor DMA unmapped.\n\nThe URB still completes with actual_length equal to the requested length\nand no error, so the transfer looks successful while a wMaxPacketSize\nsized hole in the destination buffer silently keeps its previous\ncontents. It also leaks a DMA mapping per dropped bounce.\n\nAny sufficiently large and fragmented bulk transfer can hit this. It was\nfound with a USB mass storage device behind xHCI backing a dm-verity\ntarget with 512 byte hash blocks, where the stale data is detected rather\nthan silently consumed. The device enumerates as SuperSpeed, so\nwMaxPacketSize is 1024, while dm-bufio issues one 512 byte bio per hash\nblock. verity_prefetch_io() makes the block layer merge hundreds of them\ninto a single request of up to 512 scatterlist entries of 512 bytes each.\nAt 256 TRBs per ring segment such a TD spans three segments, and every\nsegment boundary falls on an odd multiple of 512, i.e. unaligned to\nwMaxPacketSize. dm-bufio then caches a hash block holding stale data and\ndm-verity declares the metadata block corrupted:\n\n  device-mapper: verity: 8:2: metadata block 10850 is corrupted\n\nA reproducer running this under qemu is available at\nhttps://github.com/baloo/xhci-verity\n\nThe bounce state (bounce_buf, bounce_dma, bounce_len, bounce_offs)\nalready lives on the ring segment, so there is nothing extra to track.\nKeep recording the last bounced segment in td-\u003ebounce_seg and, on\ncompletion, walk the segments from td-\u003estart_seg up to it, unmapping\nevery segment that still has a pending bounce.\n\nStopping at td-\u003ebounce_seg rather than td-\u003eend_seg matters: a bounce\nimplies the TD continues past that segment's link TRB, so bounce_seg is\nalways strictly before end_seg, and a later TD may already have started\nin end_seg and been bounced there. Walking that far would copy a foreign\nbounce buffer into this URB and unmap it twice. It also keeps the walk\ncorrect if a TD ever wraps the whole ring so that end_seg == start_seg.\n\n[mn: Add ring-\u003enum_segs check to prevent unlikely infinite for loop.]","modified":"2026-09-18T05:37:08Z","published":"2026-09-16T11:17:14Z","upstream":["CVE-2026-90015"],"references":[{"type":"WEB","url":"https://nvd.nist.gov/vuln/detail/CVE-2026-90015"}],"affected":[{"package":{"name":"kernel","ecosystem":"Azure Linux:3","purl":"pkg:rpm/azure-linux/kernel"},"ranges":[{"type":"ECOSYSTEM","events":[{"introduced":"0"},{"last_affected":"6.6.150.1-1"}]}],"database_specific":{"source":"https://github.com/microsoft/AzureLinuxVulnerabilityData/blob/main/osv/AZL-101210.json"}}],"schema_version":"1.9.0"}