Linux 3442 Published by

Linus Torvalds released Linux 7.3-rc7 on Sunday, admitting it arrived about six hours late because he mixed up the day after traveling to the Kernel Maintainer Summit and Linux Plumbers Conference. It's a fixes-only snapshot that leans heavily on VFS maintainer Christian Brauner's mount-namespace hardening, credited with 58 entries. The cycle also deprecates ext4's data=journal mode for removal in 2028 and pulls in an AMD TLBI Erratum #1718 workaround. Stable Linux 7.3 is still on track for next weekend, around 18 October 2026.



Linux 7.3-rc7 Released, and Torvalds Admits It Arrived Six Hours Late

The seventh release candidate is heavy on VFS mount hardening and carries a long-term ext4 deprecation, with stable 7.3 still on track for next weekend.

Linus Torvalds shipped Linux 7.3-rc7 on Sunday, about six hours past his normal time, and he doesn't seem too bothered. He'd mixed up Sunday with Saturday while shaking off travel from the Kernel Maintainer Summit and the Linux Plumbers Conference, and only caught the slipover at dinner.

The result is a fixes-only weekly snapshot, the seventh in the 7.3 cycle. It bundles regressions and bug fixes across the whole tree, with a heavy tilt toward one area: mount and filesystem hardening. Keep in mind that rc7 is often the last candidate before a kernel goes stable, so this one carries more weight than it might otherwise.

Stable 7.3 is still targeted for next weekend, around 18 October. Torvalds said the batch gives no reason to delay, unless something new shows up.

The delay was entirely self-inflicted. "I was mentally off by a whole day and was thinking it was Saturday the whole day," he wrote. "Until it struck me over dinner that no, it's Sunday." Six hours late. But on the right day, with the weekly cadence intact.

Tuxwk

Mount hardening leads the list

By a wide margin, the largest contribution came from Christian Brauner, the VFS maintainer, credited with 58 entries in the shortlog. That number is part core fixes and part the selftests needed to prove those fixes actually hold up.

The patches get into the gnarly territory: mount propagation, bind mounts, automounts, namespaces, and filesystem objects. Brauner is keeping copied mounts properly marked unbindable, blocking synchronous unmounts of busy propagated copies, and stopping mount namespace loops that a recursive bind mount could create.

It's not housekeeping, then. Clean mount isolation is a security property, and it matters enormously for containers and sandboxing. The selftests under selftests/filesystems and selftests/move_mount_set_group check behaviors like busy submounts surviving a synchronous umount, exactly the kind of subtle regression you want caught in test rather than in production.

Torvalds called the diffstat "slightly strange" because roughly half of rc7 is test code living in tools/testing/. Strip that out and you're left with a near-even split: graphics and networking drivers on one side, filesystems, core networking, and per-architecture maintenance on the other.

Everything else

Two changes deserve a longer look. First, Theodore Ts'o deprecated ext4's data=journal mode, with formal removal slated for January 2028. It journals both data and metadata, which is ext4's strongest crash-consistency guarantee and its slowest by a comfortable margin. If you lean on it for safety-critical workloads, start planning a migration before the clock runs out.

On the hardware side, rc7 pulls in the AMD TLBI Erratum #1718 workaround by Borislav Petkov. Certain AMD Zen CPUs can leave stale translations behind, and this kernel-side guard protects machines that haven't received the microcode fix yet. It came through Ingo Molnar's x86/urgent tree and should get backported to stable kernels.

The rest is a grab-bag: RISC-V fixes, an arm64 KCOV adjustment, Bluetooth concurrency work from Chengfeng Ye, an io_uring patch that stops rings charging against RLIMIT_MEMLOCK, and DRM fixes spread across AMD, Intel, and VMware. Controllers get two new IDs as well, the 8BitDo Pro 2 Wired Controller and the Corsair Novablade Pro, plus new ALSa quirks for an ASRock NUC and an HP Pavilion.

For developers and testers who want to stress-test before stable 7.3 ships, the source tarball is available below. Keep in mind that rc kernels are for hardware vendors, distro maintainers, and testers who can report problems early. They're not for production machines. Unless something new crops up, 7.3 should land on schedule.

Linux kernel 7.3-rc7 released

Linux kernel version 7.3-rc7 is now available:

Full source: https://git.kernel.org/torvalds/t/linux-7.3-rc7.tar.gz
Patch: https://git.kernel.org/torvalds/p/v7.3-rc7/v7.2

You can view the summary of the changes at the following URL:
https://git.kernel.org/torvalds/ds/v7.3-rc7/v7.3-rc6