Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB
CVE-2026-98478 Overview
CVE-2026-98478 is a Linux kernel vulnerability in the Universal Flash Storage (UFS) core driver. The flaw causes a circular wait deadlock during SCSI error recovery, rendering affected systems unable to complete UFS error handling. The regression was introduced by commit 08b12cda6c44 ("scsi: ufs: core: Switch to scsi_get_internal_cmd()"), which routed internal UFS commands through a request queue sharing the host tagset with regular Logical Units (LUNs). When the error handler quiesces the tagset, internal commands such as the NOP OUT UPIU can no longer be dispatched, blocking recovery.
Critical Impact
Affected Linux kernels can enter an unrecoverable deadlock during UFS error recovery, blocking storage I/O and preventing the host from completing device initialization or recovery.
Affected Products
- Linux kernel versions containing commit 08b12cda6c44 ("scsi: ufs: core: Switch to scsi_get_internal_cmd()")
- Systems using the UFS (Universal Flash Storage) host controller driver (scsi/ufs/core)
- Devices relying on UFS storage for boot and runtime I/O, including mobile and embedded Linux platforms
Discovery Timeline
- 2026-10-18 - CVE-2026-98478 published to NVD
- 2026-10-19 - Last updated in NVD database
Technical Details for CVE-2026-98478
Vulnerability Analysis
The defect lives in the UFS error-handling path of the Linux SCSI stack. During recovery, ufshcd_err_handling_prepare() calls blk_mq_quiesce_tagset(&hba->host->tag_set). This marks every request queue in the tagset as quiesced, including hba->host->pseudo_sdev->request_queue, which the driver uses for internal commands.
When ufshcd_verify_dev_init() subsequently issues an internal command such as a NOP OUT UFS Protocol Information Unit (UPIU) through blk_execute_rq(), blk_mq_run_hw_queue() skips the quiesced queue. The internal command never dispatches, error handling never progresses, and the host enters a circular wait deadlock. This is a race condition and resource-management flaw rather than a memory-corruption bug.
Root Cause
The root cause is a design regression. Commit 08b12cda6c44 moved internal command allocation onto the pseudo SCSI device's request queue, which shares the host tagset with regular LUNs. The error handler assumed quiescing the full tagset would only stop user I/O, but it also stops the internal commands required to complete recovery.
Attack Vector
This issue is a reliability and availability defect triggered by legitimate error-recovery flows on UFS hardware. There is no evidence of remote exploitation. The deadlock manifests when the kernel enters UFS error handling, for example after a device timeout or link error, resulting in a Denial of Service against storage I/O on the affected host.
No verified public exploit code exists for CVE-2026-98478. The upstream fix is available in kernel commits 9f6f484 and b52d695.
Detection Methods for CVE-2026-98478
Indicators of Compromise
- Kernel task hung messages involving ufshcd_verify_dev_init, blk_execute_rq, or blk_mq_run_hw_queue in dmesg and /var/log/kern.log.
- Repeated UFS error-handler invocations (ufshcd_err_handler) that never complete, followed by I/O stalls on UFS-backed block devices.
- System watchdog resets or boot failures on devices where the root filesystem resides on UFS storage.
Detection Strategies
- Inventory Linux hosts and identify kernels that include commit 08b12cda6c44 but lack fix commits 9f6f484 or b52d695.
- Correlate storage I/O stalls with concurrent UFS error-handler activity in kernel logs to distinguish this deadlock from hardware faults.
- Monitor block layer metrics such as in_flight counters on UFS queues for sustained non-zero values after an error event.
Monitoring Recommendations
- Forward kernel ring buffer output to a central log platform and alert on hung_task warnings referencing UFS or blk_mq symbols.
- Track UFS controller error counters exposed through sysfs to catch recovery loops early.
- Baseline normal UFS recovery duration and alert when ufshcd_err_handler runtime exceeds that baseline.
How to Mitigate CVE-2026-98478
Immediate Actions Required
- Identify Linux systems running UFS storage with kernels containing the regression commit 08b12cda6c44 and schedule them for patching.
- Apply the stable kernel update that includes commits 9f6f4840def931814c106fc800f57a2d26befee6 and b52d695d062095327b944acf7daabbc816ab319b.
- Prioritize mobile, embedded, and automotive Linux devices where UFS is the primary storage medium.
Patch Information
The fix keeps the host tagset quiesced during error recovery but unquiesces the pseudo SCSI device on top of it, so internal commands remain dispatchable while logical units stay quiesced. The patch also re-quiesces the pseudo device before unquiescing the tagset so that quiesce_depth stays balanced. Clock scaling and ufshcd_pause_command_processing() are unaffected and continue to quiesce the whole tagset, internal commands included. Patches are available at Kernel Git Commit 9f6f484 and Kernel Git Commit b52d695.
Workarounds
- No supported workaround exists; the deadlock is inherent to the error-handling path and must be resolved by applying the upstream patch.
- Reduce exposure by minimizing conditions that trigger UFS error handling, such as unstable power, marginal UFS cards, or aggressive clock scaling, until patches are deployed.
- On fleets that cannot be patched immediately, prepare recovery procedures including watchdog-triggered reboots to restore service after a deadlock.
# Verify the running kernel and check for the fix commits in the source tree
uname -r
git -C /usr/src/linux log --oneline | grep -E '9f6f4840def9|b52d695d0620'
Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.
Sources: National Vulnerability Database, CISA KEV, Exploit-DB

