On June 14th, Linus Torvalds released the stable Linux 7.1 kernel. After the busy 7.0 release this is a calmer cycle, but two changes stand out from the rest: a completely new NTFS driver and Intel FRED enabled by default. The rest is a solid round of optimizations, hardening and fresh hardware support.

A From-Scratch NTFS Driver

The headline change of this release. Until now access to Windows partitions relied on the ntfs3 driver contributed by Paragon Software, which could be unreliable on writes. Linux 7.1 ships a new NTFS driver written from the ground up, with full write support, delayed allocation and integration with iomap and folios.

In practice this means faster and safer writes to NTFS partitions, which matters most for Windows/Linux dual-boot users and anyone plugging in removable drives formatted under Windows. The old ntfs3 stays around for now, but the direction is clear.

Intel FRED Enabled by Default

FRED (Flexible Return and Event Delivery) reworks how the CPU handles transitions between user space and the kernel: system calls, interrupts and exceptions. It replaces the IDT-based mechanism that had been carried over largely unchanged since the 1980s. The result is lower overhead on those transitions, cleaner CPU state management and the elimination of several fragile edge cases that had been a source of bugs.

Support had been landing across several releases, and in 7.1 it is on by default. The benefit is immediate on Panther Lake and newer CPUs.

New clone3() Flags

The clone3() system call gained three new flags for managing child processes. The most interesting one enables automatic child reaping without needing a SIGCHLD handler, while another ties process termination to a pidfd handle. This is a real win for supervisors, container runtimes and anything that juggles large numbers of child processes and used to fight with zombie processes.

sched_ext: Sub-Schedulers

sched_ext, the extensible scheduler class that lets you write CPU scheduling policy in BPF, gained initial sub-scheduler support. It is the foundation for separate scheduling policies per control group (cgroup). Eventually you will be able to run one scheduler for a database and another for batch jobs on the same machine.

io_uring and BPF

The io_uring dispatch loop can now be replaced with custom logic in BPF. This gives you control over how and in what order operations run, without bouncing back to user space. It continues the direction set in 7.0, which introduced cBPF filtering for io_uring.

Security

  • Landlock got a new access right for pathname UNIX domain sockets, wired in through a new LSM hook. Sandboxed applications can now control more precisely which sockets they are allowed to connect to.
  • The default permissions on /proc/PID/mem were tightened, making it harder to peek into the memory of other processes.
  • New LSM hooks were added for the overlay filesystem and UNIX domain sockets.

Filesystems

  • Btrfs the shutdown operation moved from experimental to stable
  • exFAT cluster preallocation without zeroing via fallocate(), less fragmentation
  • CIFS/SMB O_TMPFILE support for creating temporary files
  • Ceph complete infrastructure for collecting per-subvolume I/O metrics
  • ublk zero-copy I/O, less unnecessary data copying

Performance and Memory

The swap subsystem was modernized with the old swap map removed. The result is lower memory consumption and less legacy code. It continues the cleanup started in 7.0, which consolidated the swapin paths.

Drivers and Hardware

CPU:

  • Intel FRED by default (above), benefiting Panther Lake
  • amd-pstate and intel_idle power management improvements

GPU:

  • Intel begins Nova Lake P graphics enablement
  • Faster Intel Arc Battlemage graphics
  • Fixes for older AMD Radeon cards

Other:

  • Apple SMC power driver, meaning battery level reporting on Apple Silicon MacBooks on the mainline kernel
  • Lenovo Legion Go controller support
  • 12 new ARM and RISC-V SoCs, including Qualcomm and Rockchip

Removed

  • Intel i486 CPU support begins to be retired, starting with the build options
  • The old ISDN network driver code, removed partly to cut down the noise of automated bug reports against dead hardware
  • The UDP-Lite protocol
  • More than 140,000 lines of obsolete code in total

Linux 7.1 is a non-LTS release, so it will land first in rolling distros (Arch, openSUSE Tumbleweed) and Fedora, and reach servers as an optional HWE kernel. If you dual-boot or work with NTFS drives, this is a release worth watching.

Sources: Phoronix, Phoronix feature overview, 9to5Linux, Linuxiac