CVE-2021-47214

What it actually does

Incorrect reservation handling in hugetlb_mcopy_atomic_pte during userfaultfd is_continue error paths causes restore_reserve_on_error to delete reservations and remove_inode_hugepages to leak h->resv_hugepages counts.

Why it matters to the business

Long-running services on affected systems can experience gradual huge-page reservation accounting errors that degrade memory management and may lead to availability issues under sustained load.

How an attack unfolds

An attacker triggers specific error conditions (idx >= size or !huge_pte_none) inside the is_continue path of hugetlb_mcopy_atomic_pte; the resulting call to restore_reserve_on_error followed by remove_inode_hugepages produces a reproducible reservation leak without requiring elevated privileges beyond the ability to exercise userfaultfd.

How to fix it

Upgrade owncast/owncast to commit 7860fe78dc56dc5b2b6e11011dc1e842fff57890 or later; rebuild and redeploy containers or hosts using the patched kernel/userspace combination.

How to confirm the fix worked

Confirm the commit hash is present in the running kernel or owncast build; verify that new_pagecache_page (now page_in_pagecache) is set to true in the is_continue case and that restore_reserve_on_error is no longer invoked on those paths.

Typical attack path

1. Obtain local or container access to a system running the vulnerable kernel and userfaultfd-enabled workloads
2. Exercise hugetlb page-copy operations that hit the error paths in is_continue handling
3. Repeat to accumulate reservation leaks until accounting corruption or resource pressure appears

How to detect it

Inspect kernel logs for repeated hugetlb reservation warnings, audit resv_hugepages counters via /proc/meminfo or similar interfaces, and correlate spikes with userfaultfd activity in affected processes.

What your security team should do

• Apply the listed commit to all owncast deployments within standard change windows
• Monitor hugetlb reservation counters for unexpected growth until patches are deployed
• Review any workloads that rely on userfaultfd with huge pages for alternative mitigations if immediate upgrade is delayed