CVE-2026-43434: Linux kernel · rust_binder: check ownership before using vma
Source GCVE Database
Not known to be exploited · CVSS 7.8 High.
Data as of 27 Oct 2026, 00:07 UTC
CVSS:3.1/AV:L/AC:L/PR:L/UI:N/S:U/C:H/I:H/A:HExploitation evidence
Disclosed: evidence found.
CVE published 8 May 2026
Detectable: no evidence on record.
No public nuclei template indexed
Weaponised: no evidence on record.
No public exploit code indexed in Metasploit or Exploit-DB
Exploited in the wildExploited: no evidence on record.
Not in CISA KEV and no source reports exploitation
Observed by OffSeq MirageSeen by Mirage: no evidence on record.
No attack traffic recorded by OffSeq Mirage sensors
A missing step is not evidence of safety. Sources: cve.org (opens in a new tab), the Metasploit, Exploit-DB and nuclei-templates index as of 26 Oct 2026.
Description
In the Linux kernel, the following vulnerability has been resolved: rust_binder: check ownership before using vma When installing missing pages (or zapping them), Rust Binder will look up the vma in the mm by address, and then call vm_insert_page (or zap_page_range_single). However, if the vma is closed and replaced with a different vma at the same address, this can lead to Rust Binder installing pages into the wrong vma. By installing the page into a writable vma, it becomes possible to write to your own binder pages, which are normally read-only. Although you're not supposed to be able to write to those pages, the intent behind the design of Rust Binder is that even if you get that ability, it should not lead to anything bad. Unfortunately, due to another bug, that is not the case. To fix this, store a pointer in vm_private_data and check that the vma returned by vma_lookup() has the right vm_ops and vm_private_data before trying to use the vma. This should ensure that Rust Binder will refuse to interact with any other VMA. The plan is to introduce more vma abstractions to avoid this unsafe access to vm_ops and vm_private_data, but for now let's start with the simplest possible fix. C Binder performs the same check in a slightly different way: it provides a vm_ops->close that sets a boolean to true, then checks that boolean after calling vma_lookup(), but this is more fragile than the solution in this patch. (We probably still want to do both, but the vm_ops->close callback will be added later as part of the follow-up vma API changes.) It's still possible to remap the vma so that pages appear in the right vma, but at the wrong offset, but this is a separate issue and will be fixed when Rust Binder gets a vm_ops->close callback.
References
- nvd.nist.gov/vuln/detail/CVE-2026-43434 (opens in a new tab)
- git.kernel.org/stable/c/20a01f20d1f4064d90a8627aa41b5987f0220bb9 (opens in a new tab)
- git.kernel.org/stable/c/5a472d04fb4b9115fb7d1535bd885cea450f14db (opens in a new tab)
- git.kernel.org/stable/c/8ef2c15aeae07647f530d30f6daaf79eb801bcd1 (opens in a new tab)