BPF unwind pads, ext4 iomap, Rust pin-init

BPF exception handling, ext4 I/O conversion, and Rust pin-init work led kernel list traffic. Filesystem, networking, and VFS fixes rounded out the day.

BPF exception cleanup on bpf_unwind

A v10 series for bpf-next adds verifier and JIT support so exception cleanup landing pads run when bpf_unwind() unwinds. The change enables Rust panic paths inside BPF programs. BPF and Rust-in-kernel developers gain a clearer exception model for unwind-safe code.

ext4 buffered I/O moves to iomap

A 31-patch v7 series converts ext4 regular-file buffered I/O onto the iomap path and adds disksize-pending handling. Discussion also noted a syzbot WARNING. The work aligns ext4 with current kernel I/O infrastructure used by other filesystems.

Safe self-references in pin-init

A 20-patch v2 series extends the kernel pin-init crate to create self-referential fields safely via pinning and existential lifetimes. Rust kernel code that needs stable self-references can avoid additional unsafe patterns. Drivers and subsystems adopting pin-init are the primary audience.

XFRM netlink attribute size accounting

A patch accounts for XFRMA_IF_ID and XFRMA_OFFLOAD_DEV inside XFRM msgsize helpers. Undercounted attributes can hit BUG_ON in message builders, and the issue is described as a sibling to CVE-2026-43107. IPsec and netlink maintainers should treat it as a correctness fix.

XFS pNFS multi-extent map_blocks

An 8-patch series lets XFS map a whole pNFS block layout in one ->map_blocks call by returning multiple extents. The change cuts lock, flush, and log overhead and corrects range and allocation checks. NFSd and XFS deployments using pNFS see lower mapping cost.

VFS superblock freeze and reconfigure fixes

Christian Brauner posted 15 patches against the VFS superblock freeze, thaw, reconfigure, and fserror paths. The series addresses deadlocks and ordering problems. Stable trees and filesystem maintainers are the natural reviewers.

plist_requeue order fix

A patch corrects plist_requeue() so its shortcut no longer inserts a node at the list head. The bug corrupts plist order for the last priority run when multiple runs exist. Scheduler and other plist users need the ordering restored.

DMA_PMD PMD-backed IO pools

A Google engineer posted an RFC series for DMA_PMD pools backed by PMD-sized pages. The goal is lower IOMMU and IOTLB pressure for networking and confidential-computing VMs. High-throughput DMA paths are the intended beneficiaries.