Kernel: ext4 iomap, PCS fwnode, and memory hardening

Kernel traffic on 2026-08-18 focused on large conversion series for ext4 buffered I/O and networking PCS handling, plus several memory and verifier fixes that close out-of-bounds paths. Reviewers advanced work that modernizes core subsystems while tightening behavior on malformed inputs.

ext4 buffered I/O converts to iomap

A 32-patch series at version 5 proposes moving ext4 regular file buffered I/O from the buffer_head path onto iomap. The series introduces new disksize ordering and zeroing logic as part of the conversion. Filesystem developers gain a shared modern I/O stack and a clearer path away from legacy buffer_head usage.

fwnode PCS provider API reaches v14

A 12-patch series adds a fwnode PCS provider-consumer API together with dynamic PCS attach and detach handling inside phylink. The work targets device-tree and ACPI based systems that need flexible physical coding sublayer management. Network driver authors obtain a cleaner way to bind and rebind PCS instances at runtime.

Swap code stops spinning on bad entries

A three-patch series changes the swap path so malformed swap entries no longer trigger infinite retries or console log floods. Andrew Morton and other reviewers examined the changes. Operators avoid hangs and log spam when swap metadata is corrupt.

TDX guest quote buffers become dynamic

Version 3 of a four-patch series makes TDX guest quote buffer sizes dynamic rather than fixed. Discussion with KVM and Intel participants covers attestation scoping, DICE, and migration uAPI questions raised by the change. Confidential computing stacks gain more flexible attestation buffer handling.

Stable IPVS fix for out-of-bounds write

A stable backport for the 6.6 kernel fixes an out-of-bounds write in ip_vs_nat_icmp. The flaw permitted a local packet modification attack. Sites running IPVS on long-term kernels should take the update.

migrate_device guards compound folio ranges

A patch corrects out-of-bounds writes in migrate_device_range when compound folios extend past the caller-supplied range. The bug was found with KASAN during HMM testing. Device memory migration becomes safer for large folios.

BPF verifier rejects oversized callback args

A BPF patch rejects callback subprograms that place more than five arguments on the stack. The prior verifier gap allowed x86-64 out-of-bounds reads from helper frames, visible as KASAN slab writes. Complex BPF callback programs can no longer trigger that path.

Neighbour tables move per network namespace

Version 4 of a 15-patch net-next series converts the global arp_tbl and nd_tbl into per-netns tables. The change enables neighbour sysctl settings on a per-namespace basis. Container and multi-tenant networking setups gain isolated neighbour table control.