Kernel: BPF unwinding, leaks, ext4, SPDM, KSMBD, swap

BPF exception support, a vhost memory leak, and ext4 inline-data deprecation led kernel discussion, alongside Rust SPDM work and heated swap accounting debates. Security reports on KSMBD and uaccess cleanup also drew attention.

BPF unwind_goto and unwind_exit for exceptions

A v4 patch series on bpf-next adds unwind_goto and unwind_exit instructions, along with verifier state tracking and JIT support, so BPF programs can handle Rust and C++ exceptions and destructors. Earlier v3 discussion from Alexei covered the same instruction and verifier changes. The work matters because it would let higher-level language runtimes clean up correctly inside the BPF sandbox without breaking verification.

vhost/vringh kernel memory leak via missing access_ok

A bug report describes how vhost/vringh can leak kernel memory through vDPA passthrough when __get_user is used without access_ok. The thread moved toward Linus's uaccess-cleanup branch, which removes __get_user entirely. Readers should care because the path is reachable in virtualization setups and the cleanup aims to close an entire class of similar mistakes.

Ext4 RFC to phase out inline data writes

An RFC series proposes deprecating inline_data write paths for regular ext4 files after more than fifteen recent bug fixes and eight CVEs tied to complex locking. The change would leave the feature for directories or other limited cases while simplifying the regular-file code. Filesystem maintainers and distributions need to weigh the compatibility cost against the long string of security and correctness problems.

Rust SPDM requester in the kernel

A v4 series of twenty-two patches supplies a Rust implementation of SPDM for the kernel, focused on safer parsing of complex untrusted authentication data. The requester path is intended to reduce memory-safety risk when handling device attestation and related protocols. Kernel security and driver developers gain a typed alternative to existing C parsers for this protocol surface.

ZDI high-severity KSMBD race RCE report

ZDI disclosed a high-severity race condition in KSMBD that can lead to remote code execution and filed it as ZDI-CAN-33774. Greg KH questioned the public report arriving without an accompanying patch. The exchange highlights ongoing tension between coordinated disclosure norms and the exposure of network-facing kernel SMB code.

Virtual swap space and cgroup semantics debate

A v6 patchset introduces a virtual swap layer so zswap can operate without a physical swapfile, plus related accounting and reclaim changes. A parallel high-volume thread debates redefining memory.swap cgroup semantics for virtualized and zswap entries, with Google warning of breakage for existing v1 and v2 deployments. Memory-management and container operators must track both the technical design and the ABI risk.