Virtio validation, kmemleak RCU, and mm/BPF series

Kernel development on this day centered on virtio device-value validation with a use-after-free fix, plus memory-management and BPF proposals spanning kmemleak, binfmt_misc loaders, and inode labeling. Private NUMA nodes, rmap reindexing, and page-table fixes completed the set of active threads.

Virtio drivers add device-value validation

A four-patch series introduces validation of device-reported values across virtio drivers and includes a use-after-free fix affecting virtio-mem, input, and console. Participants examined the underlying trust model and the role of LLM-assisted checks. The work matters for anyone shipping or auditing virtio front-ends that consume untrusted device data. (discussion)

kmemleak reports Tasks RCU quiescent states

A patch replaces cond_resched() with cond_resched_tasks_rcu_qs() inside kmemleak_scan so that Tasks RCU quiescent states are reported during the scan. The change prevents stalls that otherwise block BPF struct_ops frees. BPF and memory-debugging users who hit delayed reclaim of struct_ops objects have a direct interest. (discussion)

binfmt_misc gains transparent interpreters and loader substitution

Christian Brauner posted a 21-patch series that adds an 'L' loader-substitution flag and transparent interpreter modes to binfmt_misc, together with BPF_BINPRM_LOADER support. The series enables PT_INTERP loader substitution under the new interfaces. Binary-format and BPF loader developers should examine the proposed ABI. (discussion)

bpf_init_inode_xattr kfunc for atomic inode labeling

A v5 three-patch series on bpf-next adds the bpf_init_inode_xattr kfunc so BPF LSM programs can label inodes atomically at creation time through inode_init_security. Four participants discussed the interface and its placement. LSM authors who need creation-time labeling without races gain a new primitive. (discussion)

mm/rmap reindexes MAP_PRIVATE file-backed folios

A v2 fifteen-patch series reindexes MAP_PRIVATE file-backed anonymous folios by virtual pgoff rather than the prior scheme. Andrew Morton accepted the work into mm-new. Reverse-mapping accuracy for private file mappings is the practical benefit. (discussion)

x86 PAT fix for split large-page tables

A patch corrects direct PTE allocation inside __split_large_page() that previously skipped kernel ptdesc marking and the corresponding IOTLB invalidation on free. The short thread treated the omission as a security-relevant correctness issue. Proper kernel page-table accounting avoids stale IOTLB entries after splits. (discussion)

Private Memory NUMA Nodes reach v5

A 36-patch v5 series proposes private NUMA nodes that isolate device memory and expose per-capability opt-ins for reclaim, migration, and related operations. Two participants carried the discussion. Driver and isolation work that needs distinct memory nodes without full system NUMA visibility is the intended audience. (discussion)

Watermark boost on GFP_ATOMIC failure

The v11 patch to mm/page_alloc boosts watermarks after GFP_ATOMIC order-0 allocation failures. Andrew Morton noted missing review and questioned networking relevance in a brief exchange. Paths that repeatedly hit atomic allocation pressure, including some networking code, are the stated motivation. (discussion)