Kernel fixes, BPF Rust work, and process debates

Kernel traffic on 2026-10-27 mixed long-standing bug fixes with large subsystem proposals and heated process arguments. Security patches for IPv6 and procfs sat alongside BPF verifier changes aimed at Rust and a dispute over AI contribution tags.

IPv6 skb_over_panic on narrow MTU fragments

A long-standing skb_over_panic in __ip6_append_data can trigger on raw IPv6 sockets when mtu minus fragheaderlen falls between 1 and 7 during a single sendmsg. The condition has existed since 2.6.12 and was isolated by code audit on netdev, where a one-line fix was proposed. Readers monitoring network stack stability should note the age and simplicity of the trigger.

Kernel patch drops Assisted-by tags

A documentation patch removing Assisted-by references drew 82 messages from 21 participants on linux-kernel. Discussion turned to AI-generated patch volume, maintainer burnout, and whether existing review practices can absorb the load. The exchange highlights growing tension over how AI contributions are acknowledged and filtered.

BPF arena stack and state-joining verifier

An 81-patch series on bpf proposes stack support inside the arena together with a verifier pass that joins states. The combination is intended to let Rust programs compile directly to BPF. The scale of the change and its Rust focus make it relevant to anyone tracking eBPF evolution.

PAGEMAP_SCAN use-after-free fix

A patch pins mm_users inside do_pagemap_scan to close a use-after-free in the PAGEMAP_SCAN ioctl. The race appears when hugetlb VMAs lack mmget_not_zero under concurrent process exit. The fix matters for anyone relying on pagemap interfaces in the presence of short-lived processes.

Rust SPDM requester reaches v4

Version 4 of a 22-patch series adds a Rust implementation of SPDM to the kernel library. The requester is meant to parse complex untrusted authentication data more safely than prior C code. The series continues the kernel's selective adoption of Rust for security-sensitive protocols.

TAINT_FORCED_BIND proposal meets resistance

Greg KH posted a patch adding TAINT_FORCED_BIND when userspace forces driver binds or unbinds through sysfs, aiming to cut syzbot noise. VFIO and PCI developers objected strongly and urged a revert. The heated thread underscores friction between automated testing needs and driver core policy.

Missed VMAs in per-VMA-locked proc maps

A patch corrects a race in the per-VMA-locked paths that read /proc/PID/maps and related files. Concurrent VMA splits or merges can cause some mappings to be omitted. The change restores completeness for tools that parse process address space under load.

Proxy execution gains sleeping-owner handling

Version 33 of the proxy execution series introduces handling for sleeping owners. Reviewers focused on cycle detection, PSI accounting, and activation races. The incremental progress keeps the long-running scheduler work visible to those following lock and priority inheritance changes.