Daily update · Aug 19–20, 2026

NTFS security fixes, virtio stability, and new hardware support

A large batch of NTFS fixes addresses security vulnerabilities and data corruption, while virtio and vhost get reliability improvements, and new PCIe and vDPA hardware support lands.

Hardware support

#cifs #kvm #ntfs #nvdimm #pcie #virtio

122 meaningful commitsDaily snapshot

In brief

This update brings a comprehensive set of NTFS fixes covering security, data integrity, performance, and compatibility. Virtio and vhost subsystems receive multiple stability and security fixes, including a critical infinite loop in virtio-net and an out-of-bounds read in virtio-crypto. SMB/CIFS gets a privilege escalation fix and several correctness improvements. NVDIMM and virtio-pmem reliability is improved, KVM arm64 interrupt handling is hardened, and new hardware support for Tegra264 PCIe and Intel VMD is added.

NTFS

Security hardening for NTFS

A series of fixes closes out-of-bounds read/write flaws in NTFS attribute validation, index root handling, cluster allocation, and boot sector parsing, and prevents unprivileged writes to reserved $LX* extended attributes. Malformed NTFS images can no longer trigger memory corruption or bypass permission checks.

Why it matters: Users mounting untrusted NTFS drives are protected from kernel crashes or potential privilege escalation via crafted filesystem images.

ffc755828de1 dc09bf79b76f 19cac7902a8a a83e82b0ec3a fea9e4488f38 cc9d09fef784 ea6a67ef6445

Data integrity and error handling fixes

Several write paths could corrupt data or leave invalid on-disk structures: compressed blocks rewritten as all zeros kept stale data, failed compressed writes could lose old clusters, and EA updates could leave malformed attribute records. Additional fixes address races, error handling, and large-page support, including serializing resident iomap reads with MFT record modifications and propagating compression allocation errors correctly.

Why it matters: Prevents data loss and filesystem corruption for users writing to NTFS volumes, especially with compressed files or extended attributes, and improves stability on ARM64 with 64K pages.

2bd734ee2cc8 eb67671cd814 0ab6dc1ac460 41e1cf4ec10d 772e4c8f2555 650a2751c9ae 7ed0028126b5 b54559689a00 d9e00c457d4a f2b2aeeaadce 60038154b40e 9badcfb91bc3 6e03fbd5f772 8a81803f8750 3e497af63725

Performance and compatibility improvements

Compressed write handling now reuses compression contexts and output workspaces, submits one bio per compression unit instead of up to sixteen, and skips read-modify-write cycles for full overwrites. Windows filename restrictions are now applied only when the windows_names mount option is set, and per-file chmod modes stored in $LXMOD take precedence over fmask/dmask mount masks.

Why it matters: Faster write performance on NTFS volumes with compressed files, and NTFS volumes behave more like native Linux filesystems for filenames and permissions.

33200693eb22 8cd2bf1d7af9 a8a4681f2bfc 5b676a77f3e4 71feaa7686fb 4d730a4dd78c

Virtio and vhost

Virtio transport and device fixes

Virtio MMIO transport version 3 is now supported, the balloon driver stops workers before shutdown and handles failed buffer adds, virtio-crypto clamps device-reported result lengths to prevent an out-of-bounds read, the PCI admin queue index is fixed for INTx mode, vsock restores iterator state after failed transmissions, and virtio-net no longer loops forever when the backend stops responding.

Why it matters: Improves stability and security for virtual machines using virtio devices, preventing hangs, memory corruption, and data corruption.

fa8833c085b6 7e17eef04600 198eda395067 d62fb5cc8a43 281eb4732aae f77a956f6a19 dc3f1eef9ab6 5ab3b28c5b3b 0d0eff39ceb3

vhost and VDUSE improvements

The vhost-net features array no longer overwrites returned data, zero-size IOTLB invalidations are rejected, and VDUSE gains feature negotiation ioctls, a queue-ready flag, and suspend support for live migration.

Why it matters: Fixes corrupted feature data in vhost-net, prevents a denial of service via malformed IOTLB messages, and enables better live migration and device configuration for VDUSE-based virtio devices.

92a7b138f245 9059c62f1197 a596238c2ab6 3b441820cb61 4c318d91cc60 b282418bc366

SMB/CIFS

Security and correctness fixes

The SMB client now clears setuid/setgid bits on write when using cifsacl, modefromsid, or posix extensions, fixes a use-after-free in the DFS cache, corrects link counts for overwritten open files on posix mounts, and guards fscache invalidation and resize under the inode lock.

Why it matters: Prevents potential privilege escalation on SMB shares, crashes or memory corruption when accessing DFS shares, and stale cached data on CIFS mounts using fscache.

b8e5dc4f95e5 b1b741cf8e7c bf86c08123c6 77d852c76342 364b18323058 32a7af68df73

NVDIMM and persistent memory

Reliability fixes for pmem and virtio-pmem

The pmem driver now runs PREFLUSH synchronously before data writes so a failed flush cannot be overwritten by a later successful FUA flush, and virtio-pmem no longer allocates a child flush bio in the block submit path, avoiding mempool exhaustion. Additional fixes address a use-after-free, missing memory barriers, DMA cache-line sharing, and hangs when the virtqueue breaks or during freeze.

Why it matters: Improves reliability of persistent memory devices, especially virtio-pmem in virtual machines, by preventing lost flush errors, allocation failures, and crashes.

c644a2f8fef5 40f356e610df decd93e2246f 6b7108712a4b 811808761e19 e57140944b5a 3f14003ce7f0 8de2bc46dbea 36e33955aee0 74d7d137f4bb

KVM/ARM

Interrupt handling fixes

The SPI array lookup in vgic is hardened against speculative underflow, and GICv5 PPI validation now rejects out-of-range IDs and validates timer PPIs before claiming ownership.

Why it matters: Hardens KVM on ARM64 against potential out-of-bounds memory access and prevents crashes or incorrect interrupt routing for VMs using GICv5.

93cfad8da716 2962174fdfa6 302f94dc3f8c

Hardware support

New PCIe and vDPA hardware support

Adds a PCIe controller driver for NVIDIA Tegra264, Intel VMD support for Nova Lake and Dunlow processors, and missing MODULE_DEVICE_TABLE entries for Alibaba and Octeon vDPA drivers.

Why it matters: Enables PCIe on Tegra264-based devices, VMD on newer Intel platforms, and automatic module loading for vDPA devices from Alibaba and Octeon.

01c3c27a0ef6 55aa45154fe4 7eeda5e24872 346da62d9f9c

Source commits122 entries +

2187be212179

virtio: Add ID for virtio media

Albert Esteve · Mar 10, 2026 · 1 files

fa8833c085b6

virtio-mmio: add support for transport version 3

Peter Hilber · Jun 5, 2026 · 1 files

6e03fbd5f772

ntfs: fix kmap_local_page() usage in compress

Namjae Jeon · Jun 15, 2026 · 1 files

b07513e7475f

tools/virtio: Remove unsupported --batch option from vhost_net_test

Yichong Chen · Jun 18, 2026 · 1 files

bd670e5dfd2b

vdpa_sim: fix cleanup after worker creation failure

Linfeng Sun · Jun 20, 2026 · 1 files

f77a956f6a19

crypto: virtio - bound the akcipher result length

Bryam Vargas · Jun 22, 2026 · 1 files

8634a92ee595

crypto: virtio - fix missing le64_to_cpu() conversions

Ben Dooks · Jun 22, 2026 · 2 files

bb652c245b5b

iov_iter: export iov_iter_restore

Octavian Purdila · Jun 22, 2026 · 1 files

5ab3b28c5b3b

vsock/virtio: restore msg_iter on transmission failure

Octavian Purdila · Jun 22, 2026 · 1 files

0d8aebe089b4

virtio: add virtio_device_shutdown() helper

Denis V. Lunev · Jun 24, 2026 · 2 files

29536a923a94

virtio_balloon: factor out virtballoon_quiesce()

Denis V. Lunev · Jun 24, 2026 · 1 files

7e17eef04600

virtio_balloon: quiesce balloon work before device shutdown

Denis V. Lunev · Jun 24, 2026 · 1 files

198eda395067

virtio_balloon: warn on failed buffer add in tell_host()

Denis V. Lunev · Jun 24, 2026 · 1 files

d62fb5cc8a43

virtio_balloon: warn on failed buffer add in stats_handle_request()

Denis V. Lunev · Jun 24, 2026 · 1 files

135f0abe3139

vdpa_sim: clear pending_kick on device reset

Xiong Weimin · Jun 26, 2026 · 1 files

7b411448c668

vdpa_sim: hold iommu_lock across dma_unmap passthrough transition

Xiong Weimin · Jun 26, 2026 · 1 files

92a7b138f245

vhost/net: fix clear_user start address in VHOST_GET_FEATURES_ARRAY

Yufeng Wang · Jun 26, 2026 · 1 files

656662dc53e6

virtio_dma_buf: fix typo in kdoc comment: get_uid -> get_uuid

Li RongQing · Jun 29, 2026 · 1 files

dc3f1eef9ab6

virtio_pci: fix wrong queue index for admin vq in intx path

Li RongQing · Jun 29, 2026 · 1 files

ddf8b4388af7