16-Year-Old Linux KVM Flaw Lets Guest VMs Escape to Host on Intel and AMD x86 Systems

CVE-2026-43284 is a write-what-where memory-safety flaw (CWE-123) in the Linux kernel's xfrm/ESP (IPsec) code: the IPv4/IPv6 UDP datagram send paths fail to mark pages spliced into a UDP skb with SKBFL_SHARED_FRAG when MSG_SPLICE_PAGES attaches pipe pages, so the ESP input path wrongly treats the packet as privately owned and decrypts it in place over memory the skb does not exclusively own. It is triggered when locally generated UDP traffic built from spliced pipe pages is processed by the ESP decryption path, typically in IPsec NAT-traversal (ESP-in-UDP, UDP port 4500) configurations; the in-place decrypt can corrupt or expose data still referenced by the pipe or other owners of those shared pages. A local, low-privileged attacker can thereby cause high-severity confidentiality, integrity, and availability impact that extends beyond the packet itself (CVSS 3.1: 8.8, scope changed); no remote or unauthenticated trigger is described. Any Linux system running a kernel with the vulnerable UDP splice and ESP-in-UDP paths is affected, most importantly IPsec VPN gateways and clients; the advisory lists no affected version ranges. No public proof-of-concept or confirmed in-the-wild exploitation is known (not on CISA KEV), but EPSS assigns a 93.2% probability of exploitation within 30 days (100th percentile), so patching should not be deferred.

Do: Apply your distribution's kernel security update containing the xfrm/esp fix for CVE-2026-43284 and reboot into the patched kernel, prioritizing IPsec VPN gateways and clients that use ESP-in-UDP NAT-T (UDP/4500). Until patched, restrict untrusted local users from sending splice-based (MSG_SPLICE_PAGES) UDP traffic through IPsec-protected paths, or temporarily avoid ESP-in-UDP encapsulation where feasible. Verify the fix via your vendor's package changelog, as the advisory does not name specific fixed version numbers.

CVE-2026-43500 is a memory-corruption flaw in the Linux kernel's AF_RXRPC (rxrpc) subsystem: the DATA and RESPONSE packet handlers only copy received skbs to a linear buffer before decryption when the skb is marked cloned, so non-cloned skbs that still carry externally owned paged fragments (SKBFL_SHARED_FRAG, e.g. set by splice() into a UDP socket, or skbs with a frag_list) fall through to the in-place decryption path. In that path the shared fragment pages are bound directly into the AEAD/skcipher scatter-gather list via skb_to_sgvec(), so decryption writes into pages shared with other kernel or user contexts, causing out-of-bounds/write-what-where corruption (CWE-787/CWE-123), notably via the splice-into-UDP-socket loopback vector. A local low-privileged attacker who can get the machine to process such rxrpc packets can corrupt kernel memory and escalate to root (CVSS 3.1: 7.8 high, local vector, high confidentiality/integrity/availability impact). Affected systems are Linux kernels with the AF_RXRPC subsystem available — the transport used by the AFS/kAFS network filesystem — which mainstream distribution kernels ship as a module or built-in; no specific affected or fixed version ranges are provided in the data. It is not on CISA's KEV list and the CVE record lists no public PoC, but news reports describe a 'Dirty Frag' local privilege-escalation exploit giving root across major distributions and a released PoC, and EPSS puts the 30-day exploitation probability at about 93% (100th percentile).

Do: Apply the kernel update containing the rxrpc fix as soon as your distribution ships it — the fix extends the unshare gate to also unshare skbs with frag lists or shared fragments before in-place decryption; no fixed version numbers are given in the available data, so follow your distro's security advisory. Until patched, check whether rxrpc/AFS is in use (e.g. 'lsmod | grep rxrpc') and avoid or restrict local splice()-into-UDP/loopback AFS traffic on multi-user hosts.

In the Linux kernel, the following vulnerability has been resolved: KVM: x86: Fix shadow paging use-after-free due to unexpected GFN The shadow MMU computes GFNs for direct shadow pages using sp->gfn plus the SPTE index. This assumption breaks for shadow paging if the guest page tables are modified between VM entries (similar to commit aad885e77496, "KVM: x86/mmu: Drop/zap existing present SPTE even when creating an MMIO SPTE", 2026-03-27). The flow is as follows: - a PDE is installed for a 2MB mapping, and a page in that area is accessed. KVM creates a kvm_mmu_page consisting of 512 4KB pages; the kvm_mmu_page is marked by FNAME(fetch) as direct-mapped because the guest's mapping is a huge page (and thus contiguous). - the PDE mapping is changed from outside the guest. - the guest accesses another page in the same 2MB area. KVM installs a new leaf SPTE and rmap entry; the SPTE uses the "correct" GFN (i.e. based on the new mapping, as changed in the previous step) but that GFN is outside of the [sp->gfn, sp->gfn + 511] range; therefore the rmap entry cannot be found and removed when the kvm_mmu_page is zapped. - the memslot that covers the first 2MB mapping is deleted, and the kvm_mmu_page for the now-invalid GPA is zapped. However, rmap_remove() only looks at the [sp->gfn, sp->gfn + 511] range established in step 1, and fails to find the rmap entry that was recorded by step 3. - any operation that causes an rmap walk for the same page accessed by step 3 then walks a stale rmap and dereferences a freed kvm_mmu_page. This includes dirty logging or MMU notifier invalidations (e.g., from MADV_DONTNEED). The underlying issue is that KVM's walking of shadow PTEs assumes that if a SPTE is present when KVM wants to install a non-leaf SPTE, then the existing kvm_mmu_page must be for the correct gfn. Because the only way for the gfn to be wrong is if KVM messed up and failed to zap a SPTE... which shouldn't happen, but *actually* only happens in response to a guest write. That bug dates back literally forever, as even the first version of KVM assumes that the GFN matches and walks into the "wrong" shadow page. However, that was only an imprecision until 2032a93d66fa ("KVM: MMU: Don't allocate gfns page for direct mmu pages") came along. Fix it by checking for a target gfn mismatch and zapping the existing SPTE. That way the old SP and rmap entries are gone, KVM installs the rmap in the right location, and everyone is happy.

In the Linux kernel, the following vulnerability has been resolved: KVM: arm64: vgic-its: Drop the translation cache reference only for the erased entry vgic_its_invalidate_cache() walks the per-ITS translation cache with xa_for_each() and drops the cache's reference on each entry with vgic_put_irq(). It puts the iterated pointer, though, rather than the value returned by xa_erase(). The function is called from contexts that do not exclude one another: the ITS command handlers hold its_lock, the GITS_CTLR write path holds cmd_lock, and the path that clears EnableLPIs in a redistributor's GICR_CTLR holds neither. Two or more of them can drain the same cache concurrently, and if each one observes the same entry, erases it and then puts it, the single reference the cache holds on that entry is dropped more than once. The entry can then be freed while an ITE still maps it. xa_erase() is atomic and returns the previous entry, so put only the entry that this context actually removed. The cache reference is then dropped exactly once per entry even when the invalidations run concurrently, and the behavior is unchanged when only one context runs.

In the Linux kernel, the following vulnerability has been resolved: KVM: x86: Fix shadow paging use-after-free due to unexpected role Commit 0cb2af2ea66ad ("KVM: x86: Fix shadow paging use-after-free due to unexpected GFN") fixed a shadow paging mismatch between stored and computed GFNs; the bug could be triggered by changing a PDE mapping from outside the guest, and then deleting a memslot. The rmap_remove() call would miss entries created after the PDE change because the GFN of the leaf SPTE does not match the GFN of the struct kvm_mmu_page. A similar hole however remains if the modified PDE points to a non-leaf page. In this case the gfn can be made to match, but the role does not match: the original large 2MB page creates a kvm_mmu_page with direct=1, while the new 4KB needs a kvm_mmu_page with direct=0. However, kvm_mmu_get_child_sp() does not compare the role, and therefore reuses the page. The next step is installing a leaf (4KB) SPTE on the new path which records an rmap entry under the gfn resolved by the walk. But when that child is zapped its parent kvm_mmu_page has direct=1 and kvm_mmu_page_get_gfn() computes the gfn for the 4KB page as sp->gfn + index instead of using sp->shadowed_translation[] (or sp->gfns[] in older kernels). It therefore fails to remove the recorded entry. When the memslot is dropped the shadow page is freed but the rmap entry survives, as in the scenario that was already fixed. Code that later walks that gfn (dirty logging, MMU notifier invalidation, and so on) dereferences an sptep that lies in the freed page, causing the use-after-free.