CVE-2026-64533

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> fs/ntfs3: validate lcns_follow in log_replay conversion<br /> <br /> log_replay() converts DIR_PAGE_ENTRY_32 records into DIR_PAGE_ENTRY<br /> records when replaying version 0 restart tables.<br /> <br /> During this conversion, the memmove() length is derived directly from<br /> the on-disk lcns_follow field:<br /> <br /> memmove(&amp;dp-&gt;vcn, &amp;dp0-&gt;vcn_low,<br /> 2 * sizeof(u64) +<br /> le32_to_cpu(dp-&gt;lcns_follow) * sizeof(u64));<br /> <br /> check_rstbl() validates restart table structure, but does not constrain<br /> per-entry lcns_follow values relative to the entry size. A malformed<br /> filesystem image can provide an oversized lcns_follow value, causing<br /> the conversion memmove() to access memory beyond the bounds of the<br /> allocated restart table buffer.<br /> <br /> The same field is later used to bound iteration over page_lcns[],<br /> so validating lcns_follow during conversion also prevents downstream<br /> out-of-bounds access from the same malformed metadata.<br /> <br /> Compute the maximum valid lcns_follow from the already-validated<br /> restart table entry size and reject entries that exceed this bound.<br /> Reuse the existing t16/t32 scratch variables already declared in<br /> log_replay() to avoid introducing new declarations.<br /> <br /> [almaz.alexandrovich@paragon-software.com: fixed the conflicts]