Linux 7.3 Preview: Btrfs Gets 5x Faster as Key Optimizations Land

Linux 7.3 Preview: Btrfs Gets 5x Faster as Key Optimizations Land

📋 Summary: Linux 7.3 is approaching its expected October 18 release and brings improvements across storage, memory management, scheduling, security, and hardware support. Btrfs and EXT4 receive notable performance work, while fixes to KSM, zsmalloc, and scheduling reduce latency under specific workloads. The release also contains substantial NTFS3, ksmbd, and BPF security hardening, with several important optimizations contributed by engineers from Chinese technology companies.

Linux 7.3 is entering the final stages of development, and this release is shaping up to be more than a routine kernel update.

The biggest changes are spread across several layers of the kernel: storage performance, memory management, scheduling, security, and hardware enablement.

Some of the most eye-catching numbers are significant. Certain Btrfs operations are reported to become roughly 5× faster, direct I/O throughput can exceed 90% of theoretical peak in the relevant workload, and one KSM-related latency problem falls from 705 ms to just 1.44 ms after the fix.

There is also a notable amount of engineering work from Chinese companies, including contributions from ZTE, Xiaomi, and ByteDance.

At the same time, Linux 7.3 contains a substantial collection of security fixes, including NTFS3 hardening, multiple ksmbd use-after-free fixes, and continued strengthening of the BPF verifier.

For most desktop users, these changes will not suddenly transform the Linux experience overnight. But for storage servers, virtualization hosts, containers, ARM devices, and systems exposed to network services, Linux 7.3 contains several improvements worth paying attention to.

🗓️ Linux 7.3 Release Timeline

#

Before looking at the technical changes, it is worth establishing the current development schedule.

The Linux 7.3 merge window closed on August 30, after which the kernel entered the weekly release-candidate cycle.

The current timeline is:

If a serious regression requires another release candidate, the final release could move to approximately October 25.

This distinction matters because downstream Linux distributions can freeze their kernels before the upstream kernel reaches final release.

Ubuntu 26.10 creates an interesting timing gap

#

Ubuntu 26.10 is scheduled to release on October 15, three days before the expected upstream Linux 7.3 release.

That means the kernel included in the initial Ubuntu 26.10 release is likely to be based on the latest upstream release candidate available at the distribution freeze point rather than the final Linux 7.3 release.

In other words, seeing a Linux 7.3-based development kernel in Ubuntu 26.10 does not necessarily mean Ubuntu is shipping the final upstream 7.3 release on day one.

This is a normal consequence of distribution development schedules and upstream kernel release timing.

💾 Btrfs and EXT4 Both Receive Major Performance Work

#

Storage is one of the most interesting areas of Linux 7.3.

Btrfs logging and fsync improvements

#

Btrfs receives two particularly notable optimizations, targeting logging and fsync behavior.

The reported improvements can reach approximately 5× performance gains in the relevant workloads.

Direct I/O performance also receives attention.

In the tested scenario, throughput moves from approximately half of theoretical peak performance to more than 90% of the theoretical limit.

These numbers should not be interpreted as a universal 5× improvement for every Btrfs workload. Filesystem performance is highly dependent on workload characteristics, hardware, queue depth, I/O pattern, and filesystem configuration.

Nevertheless, the changes target operations that can matter significantly to databases, virtualization, servers, and other workloads that rely heavily on direct I/O and filesystem synchronization.

EXT4 gets faster on high-speed NVMe

#

EXT4 is also receiving performance improvements.

After refactoring the IOmap layer, the referenced benchmark shows performance on high-speed NVMe storage increasing from approximately:

1.92 million IOPS → 2.19 million IOPS

Parallel direct I/O writes also receive improvements.

This is important because EXT4 remains one of the most widely deployed Linux filesystems. Even relatively specialized kernel optimizations can therefore affect a large number of real-world systems.

Btrfs changes its free-space-tree default

#

There is also a smaller but technically interesting Btrfs configuration change.

Free-space tree v2 has been the mkfs default since Linux 5.15. Linux 7.3 further aligns the kernel-side default behavior by moving away from free-space tree v1 as the default.

Existing v1 filesystems remain compatible.

The change is therefore less about suddenly converting existing filesystems and more about making the default configuration consistent with the newer free-space-tree implementation.

🧠 KSM Fix Removes a Serious Latency Spike

#

One of the more surprising changes is in memory management.

Engineers from ZTE identified a problem involving Kernel Samepage Merging (KSM).

KSM allows the kernel to identify identical memory pages and merge them, reducing physical memory consumption. It can be useful on systems running multiple virtual machines or workloads containing large amounts of duplicated memory.

The problem was related to lock contention under memory pressure.

In the problematic scenario, KSM could hold a lock long enough to cause applications to stall for hundreds of milliseconds.

The reported worst-case latency drops from:

705 ms → 1.44 ms

after the fix.

That is an enormous difference.

Who benefits from the KSM fix?

#

The change is particularly relevant to systems that actively use memory deduplication.

Potentially affected environments include:

• Virtualization hosts
• KVM deployments
• Container-heavy systems
• Memory-constrained servers
• Systems running many similar workloads

A normal desktop user is much less likely to encounter this specific problem.

However, if a Linux workstation or server suddenly experiences a long application freeze under heavy memory pressure, kernel memory-management behavior can be one of the less obvious causes.

🧊 zsmalloc Gets Faster on Small Devices

#

Another memory-management optimization targets zsmalloc, the compressed-memory allocator used in several Linux subsystems.

The improvement, contributed by engineers from Xiaomi, produces up to approximately 1.83× the performance on the referenced Raspberry Pi 4B testing.

This is especially relevant for low-cost ARM systems, edge devices, and other hardware with constrained memory resources.

Unlike a general-purpose desktop performance optimization, this is a targeted improvement whose benefits depend heavily on whether the workload actually uses the affected allocator.

For Raspberry Pi-class devices, however, memory efficiency can have an outsized impact because CPU and RAM resources are much more limited than on a modern desktop.

⚡ ByteDance Refactors SMP Scheduling

#

Scheduling is another area receiving substantial attention.

Engineers from ByteDance contributed an SMP-related refactoring that improves latency under network-heavy workloads.

In the referenced test, worst-case inter-process communication latency falls from approximately:

17 ms → 1.5 ms

That’s roughly a 90% reduction.

Again, this should be understood as a workload-specific result rather than a universal claim that every Linux application will suddenly experience 90% lower latency.

Desktop impact

#

For desktop systems, improvements in scheduler latency can potentially reduce contention between foreground applications and background workloads.

That can translate into:

• Fewer visible stalls
• More consistent application responsiveness
• More stable frame delivery under CPU contention
• Better behavior when background tasks are consuming system resources

The difference may be subtle during normal workloads.

Under heavy CPU and network activity, however, scheduler latency becomes considerably more important.

Server impact

#

Servers are even more sensitive to tail latency.

A small number of extremely slow scheduling events can have a disproportionate effect on response-time percentiles.

Reducing worst-case latency is therefore particularly useful for high-concurrency network services.

🧩 sched_ext Pushes Programmable Scheduling Further

#

Linux’s sched_ext framework is also progressing rapidly.

sched_ext allows developers to experiment with custom scheduling policies using BPF-based mechanisms.

In Linux 7.3 development, sched_ext and its associated self-test infrastructure account for a significant portion of the non-driver changes in the referenced rc2 development cycle.

The broader direction is important.

Traditional Linux scheduling policies are implemented deep inside the kernel, while sched_ext creates a framework through which developers can experiment with alternative scheduling strategies without rewriting the entire scheduler subsystem.

For ordinary users, this is not yet a feature they need to configure.

For kernel developers, researchers, and performance engineers, however, programmable scheduling represents a significant direction for future Linux development.

🔐 Linux 7.3 Is Also a Security-Focused Release

#

Performance improvements tend to generate headlines, but Linux 7.3 also contains a substantial amount of security hardening.

Three areas are particularly notable:

• NTFS3
• ksmbd
• BPF verification

These changes matter even if you never notice them during normal desktop use.

🪟 NTFS3 Receives Extensive Hardening

#

The NTFS3 driver receives a substantial series of fixes spanning approximately 21 commits and 922 lines of code in the referenced changes.

The fixes address multiple classes of problems, including:

• Slab out-of-bounds writes
• Out-of-bounds reads during reparse handling
• Out-of-bounds conditions during object ID initialization
• Information leaks involving ntfs_rename()
• Integer overflow issues during MFT cluster validation

Basic support for NTFS Alternate Data Streams (ADS) is also included.

For users who regularly exchange storage devices between Windows and Linux, these improvements are particularly relevant.

NTFS3 sits directly on the filesystem boundary, so correctness and security problems in the driver can potentially affect any Linux system mounting untrusted or externally supplied NTFS volumes.

🌐 ksmbd Fixes Multiple Use-After-Free Bugs

#

Linux’s in-kernel SMB server, ksmbd, also receives several security fixes.

The changes address use-after-free problems involving areas such as:

• Tree connections
• Oplock break notifications
• Session management

The release also adds response name-length validation and improves session draining during unregistration.

Because ksmbd provides network-facing SMB functionality directly from the kernel, bugs in this area deserve particular attention.

Administrators running ksmbd should therefore pay close attention to the stable-kernel versions containing these fixes rather than waiting indefinitely for a major distribution upgrade.

🛡️ BPF Verifier Continues to Harden the Kernel

#

The BPF verifier is another major security boundary inside modern Linux.

Linux 7.3 continues strengthening it in several areas.

Among the changes are improvements for:

• Inferring whether registers are NULL after conditional branches
• Rejecting invalid pseudo-function references to the main program
• Blocking unsafe direct tail calls originating from callback frames
• Correctly identifying certain syscall helpers as sleepable
• Preventing out-of-bounds pointer exposure on speculative execution paths

These changes may not produce visible improvements for desktop users, but they are important for the security model of systems relying heavily on BPF.

BPF is increasingly used for networking, observability, tracing, security tooling, and system-level customization, making verifier correctness increasingly important.

What administrators should do

#

The practical recommendation is simple:

Do not replace a production distribution kernel with a self-compiled mainline development kernel just to obtain these fixes early.

Security fixes are routinely backported into stable Linux branches and distribution kernels.

For production environments, monitor the security advisories and kernel updates provided by your Linux distribution.

🦀 Rust Support Expands to More Architectures

#

Rust support in the Linux kernel continues to expand.

Linux 7.3 adds support for Rust on:

• PPC32BE
• PPC64LE

These architectures join existing Rust-enabled targets including:

• ARM32
• ARM64
• LoongArch
• RISC-V
• x86_64

The PowerPC Rust support initially entered the kernel in an experimental state and has now progressed toward maintained status following additional patch refinement.

This is another indication that Rust is becoming an increasingly important part of the Linux kernel development ecosystem.

🍎 Apple Silicon Gains More USB4 and Thunderbolt Support

#

The Asahi Linux community continues pushing Apple Silicon hardware support forward.

A 19-patch series introduces initial USB4 and Thunderbolt-related support for Apple’s M1, M2, and M3 platforms.

The current work includes:

• XDomain
• USB 3 over USB4 tunneling

However, several areas remain incomplete, including:

• PCIe tunneling
• DisplayPort
• Suspend/resume

The associated Apple PMGR power-management driver also provides an interesting efficiency improvement, reducing power consumption by approximately 1 watt during s2idle in the referenced testing.

For Linux users running Apple Silicon hardware, these incremental kernel improvements can be more important than raw benchmark gains because hardware compatibility remains one of the platform’s primary challenges.

🧬 Linux 7.3 Prepares for New AMD and Intel Hardware

#

The kernel also continues preparing for upcoming processor and GPU generations.

Initial support for AMD’s next-generation Zen 6 platform has landed, alongside a substantial amount of AMD GPU code.

The AMD GPU changes account for approximately one-third of the Linux 7.3-rc1 changes in the cited development statistics.

Intel also receives additional work covering:

• Hybrid CPU scheduling and optimization
• Xe3P graphics
• Nova Lake GPU support

This type of early kernel support is essential for new hardware because operating-system support often has to land before the hardware reaches a broad consumer audience.

🐧 Linus Tracks Down a One-Character Intel GPU Bug

#

One of the more entertaining development stories involves an Intel Xe GPU issue affecting Battlemage G21 hardware.

Linus Torvalds personally tracked down a VRAM offset calculation bug that could cause black screens.

The final fix changed only one character:

round_up()

to:

round_down()

Finding the problem was anything but simple.

The debugging process reportedly involved approximately 24 patches, custom debugging code, and 18 reboots before the underlying calculation error was isolated.

Interestingly, the commit message for the final change was generated with AI assistance.

It is a good illustration of how difficult kernel debugging can be: a one-character change can sit at the end of a debugging process involving dozens of iterations and repeated hardware testing.

🧪 How to Test Linux 7.3 Safely

#

If these changes make you want to test Linux 7.3 early, there is one rule worth emphasizing:

Do not install a mainline development kernel directly on your primary production machine.

Development kernels exist precisely because regressions are still being discovered.

There are safer ways to experiment.

Option 1: Use an Ubuntu 26.10 development build

#

Ubuntu 26.10 development and beta builds provide an easier way to experiment with the upcoming kernel environment.

Use a secondary machine or virtual machine rather than your primary workstation.

Once the final Ubuntu release arrives, the development installation can transition toward the stable release depending on the selected development channel.

Option 2: Wait for distribution kernels

#

For production systems, waiting is usually the better approach.

Once Linux 7.3 reaches upstream release, distributions can backport individual fixes or move to the new kernel through their normal update channels.

Ubuntu 26.04 LTS users should especially avoid replacing the production kernel simply to obtain an upstream feature.

The HWE cycle and official distribution updates provide a considerably safer path.

Always have a rollback plan

#

Before experimenting with a development kernel:

1. Create a Timeshift snapshot or equivalent system backup.
2. Keep the existing kernel installed.
3. Test on secondary hardware or a virtual machine whenever possible.
4. Never use an experimental kernel for critical production workloads.

A kernel update is fundamentally different from installing an ordinary desktop application. If something goes wrong, the failure can affect the entire operating system.

📊 Does Linux 7.3 Matter to You?

#

The practical impact depends heavily on what you use Linux for.

For the average desktop user, Linux 7.3 is unlikely to feel like an entirely new operating system.

The more interesting benefits appear under specific workloads.

A storage server may benefit from filesystem improvements. A virtualization host may benefit from the KSM fix. A network-heavy server can potentially benefit from scheduler changes. An Apple Silicon Linux system can gain additional hardware support.

Meanwhile, security fixes in NTFS3, ksmbd, and BPF are relevant even when they produce no visible performance change.

🚀 Linux 7.3 Is More Than a Performance Release

#

Linux 7.3 is shaping up to be a well-rounded kernel release.

The headline numbers are impressive: certain Btrfs operations can reach approximately 5× higher performance, EXT4 benefits from I/O-layer improvements, KSM’s worst-case latency drops from 705 ms to 1.44 ms, and scheduler changes significantly reduce latency in the referenced network-heavy workload.

But the less visible changes may ultimately be just as important.

NTFS3 receives extensive hardening, ksmbd fixes multiple use-after-free issues, and the BPF verifier continues closing subtle security gaps.

At the same time, Linux 7.3 expands Rust support, improves Apple Silicon connectivity, prepares for next-generation AMD and Intel hardware, and continues pushing programmable scheduling through sched_ext.

Several of these improvements also demonstrate the increasingly international nature of Linux kernel development, with meaningful contributions from engineers at companies including ZTE, Xiaomi, and ByteDance, alongside developers from the wider global kernel community.

For ordinary users, there is no reason to rush onto a development kernel.

For administrators, developers, and Linux enthusiasts, however, Linux 7.3 is worth watching closely.

The safest strategy is to let upstream finish the release, allow distributions to integrate and backport the important changes, and then upgrade through the normal package-management path.

That way, you get the performance improvements and security hardening without turning your primary Linux installation into a kernel debugging experiment.