CVE-2021-47498

What it actually does

Race condition in device-mapper request queuing allows blk-mq requests to be dispatched during DM suspend, triggering kernel panic when nr_requests updates or external unquiesce events occur.

Why it matters to the business

Unexpected system crashes and storage path instability during maintenance or load changes can cause unplanned downtime and interrupt critical workloads.

How an attack unfolds

An attacker with local access triggers concurrent DM suspend/resume operations alongside nr_requests updates or elevator switches, causing a request to be queued while the device is suspended and resulting in a panic.

How to fix it

Upgrade the Linux kernel to a version containing the fix for CVE-2021-47498; for pkp/ojs (GIT) apply commit 99a9f393190383454aa5ddffedffc89596f6c682 or later.

How to confirm the fix worked

Confirm the kernel version contains the dm rq requeue change, reproduce the nr_requests update plus dm-mpath suspend/resume stress test, and verify absence of panic.

Typical attack path

1. Obtain local access to the system
2. Initiate DM multipath suspend
3. Trigger nr_requests update or elevator switch
4. Issue I/O requests during the suspend window
5. Observe kernel panic on resume

How to detect it

Review kernel logs for panic traces referencing dm rq or blk-mq during suspend/resume operations; correlate with nr_requests sysfs changes or elevator switches.

• kernel panic in dm_rq_queue_rq
• blk_mq quiesce/unquiesce mismatch during suspend

What your security team should do

• Apply the kernel fix immediately on affected systems
• Monitor for storage path stability during maintenance windows
• Restrict local access that could trigger the race