Linux Kernel 7.3-rc5 lands: large diffstat, same old fixes underneath
Linux 7.3-rc5 has hit the mainline tree, and it's another big one. Linus Torvalds signed it off on September 27, wrapping up the fifth release candidate of the 7.3 development cycle with 577 commits from 275 unique contributors.
Torvalds framed the release the way he does most weeks: "another week, another large rc." Then he pointed out that the split looked weird. Less than a third of the patches touch drivers, and roughly a quarter are selftests. That shape normally raises eyebrows, because drivers usually dominate.
He noted this is partly misleading. Most driver patches are small, while some of the selftest patches are large. The biggest single patch in the whole release is an arm64 KVM ITS table test. So the blob-size distribution skews the picture more than the actual work does.
Underneath the odd shape, Torvalds' verdict was that it's "the same old, same old." Architecture fixes, largely KVM-related with some perf. Core kernel work in the scheduler and BPF. Networking. Scattered VFS and XFS fixes. His bottom line: nothing looks particularly alarming heading into rc6.
By the numbers
The contributor spread is genuinely healthy. The average sits at just 2.10 commits per author, and 184 of the 275 contributors, or 67%, submitted a single commit. Only 53 authors managed three or more commits in this window.
Darrick J. Wong led the pack with 22 commits, all XFS work focused on scrub and repair hardening. That's a lot of churn for one subsystem in a single release. Kumar Kartikeya Dwivedi came in second with 16 BPF commits, followed by Fuad Tabba's 14 on KVM arm64 and pKVM. Wentao Liang rounded out the top four with 13, mostly leak fixes across gpio, DRM, SCSI, and mlx5.
Sean Christopherson and Mickaël Salaün each logged 10, on KVM x86/SEV/perf and landlock respectively. Sean Christopherson is one of those names you'll keep seeing if you watch KVM fixes closely.
Where the work actually went
Networking dominates again, because it always does during an rc. It's roughly the largest bucket at about 76 commits, spanning Ethernet drivers like mlx5e, bcmgenet, and stmmac, plus protocol work on ipv6, sctp, udp, and tipc. There's also a sizable ovpn patchset from Ralf Lici and a round of NFC hardening.
Architecture and KVM make up the next cluster at around 60 commits. A lot of it is arm64 KVM and pKVM work, with smattering of s390, RISC-V, and x86 SEV/PMU/mce material.
The DRM/GPU bucket comes in at roughly 58 commits. nouveau got several use-after-free and double-free fixes, which is fairly routine for that driver. amdgpu, i915, xe, and virtio also show up.
Core filesystem and memory management sit near 50, largely carried by Wong's XFS changes plus VFS/dcache, cgroup, hugetlb, and damon. BPF lands at 34 direct commits, or closer to 53 if you fold in the roughly 19 BPF selftests that inflate that number.
The security-relevant stuff
If you care about memory safety, the keyword scan flags around 101 commits as touching use-after-free, out-of-bounds, infoleak, race, or NULL-deref territory. That's a normal bug-fix load for an rc, not an anomaly worth flagging.
The clusters that stand out are in NFC/LCPO, where there's a batch of length-validation and refcount fixes, and in BPF, which picked up multiple bounds-check fixes including a divide-by-zero in btf_struct_walk and OOB reads on sk_protocol and rtt_min.
The SMB client got a coordinated set of out-of-bounds read fixes, and XFS picked up an integer overflow in xbitmap plus a wild-memcpy guard on rtrefcount btree roots. KVM picked up several stage-2 memory-management race fixes across arm64, s390, and x86.
I'll admit that last point is worth a glance. SEV and stage-2 page-table races have a habit of becoming real problems later, so the fact that this cycle shipped three architecture's worth of them is a small silver lining.
The stuff that isn't a bug fix
Not everything here is damage control. The release also brings hardware enablement, including new HID devices like the Logitech G502 X Lightspeed, the SteelSeries Arctis 7 (2018 edition), and the ASUS ROG Z13 Folio.
On the network side you've got the Quectel EG120K-EA and MeiG Smart SRM821 modems, plus the Intel X4D PF in the sfc driver. There are also a handful of MAINTAINERS updates and mailmap corrections that you won't see in any changelog but matter if you ever file a bug and want it routed correctly.
Four reverts made the cut, mostly in the filesystem and namespace area. Christian Braunter pulled two selftests/filesystems changes and the put_mnt_ns work to defuse a syzbot "WARNING in do_new_mount" saga, alongside hardening on d_instantiate_new() and iterate_supers_type().
If you're keeping score, that last bit is a reminder that even the namespace code has a long, well-documented history of tripping up automated fuzzers.
Linux kernel 7.3-rc5 released
Linux kernel version 7.3-rc5 is now available:
Full source: https://git.kernel.org/torvalds/t/linux-7.3-rc5.tar.gz
Patch: https://git.kernel.org/torvalds/p/v7.3-rc5/v7.2
You can view the summary of the changes at the following URL:
https://git.kernel.org/torvalds/ds/v7.3-rc5/v7.3-rc4
