Kernel
After two months of development, Linus Torvalds unveiled the release of the new Linux kernel version 5.10, version that arrives with the status of a branch with a long support period, whose updates will be published for at least two years.
Notable changes include this new version includes compatibility with the MemTag protection mechanism for ARM64 systems, "nosymfollow" mounting option, significant Ext4 optimizations, XFS 2038 fix, new process_madvise system call, improved AMD SEV support through CPU register encryption, ability to pause BPF programs.
The new version received 17470 fixes from 2062 developers, Patch size: 64MB (changes affected 15101 files, added 891932 lines of code, removed 619716 lines). About 42% of all changes introduced in 5.10 are related to device drivers, approximately 16% of the changes are related to updating specific code for hardware architectures, 13% are related to the network stack, 3% are related to file systems and 3% are related to internal kernel subsystems.
Main news in Linux 5.10
Of the main changes that occur, we can find that for ext4 the quick confirmations mode has been added (fast_commit), which significantly reduces delays in many file operations due to faster flushing of metadata to disk when executing an fsync () call. Under normal circumstances, running fsync () synchronizes a redundant set of metadata. In fast_commit mode, only the metadata required to recover the file system in the event of a crash is transferred to the registry, speeding up calls to fsync () and improving the performance of operations that actively manipulate metadata.
While for Btrfs includes important performance optimizations related to fsync () operations. The reduction in log mutex contention resulted in a 4% increase in performance and a 14% reduction in latency when running the dbench benchmark with 32 clients. Eliminating additional commits for links and name changes increased bandwidth by 6% and reduced latency by 30%. Limiting fsync to wait only on rewrites increased performance by 10-40%.
In addition, Btrfs implementation of direct I / O (direct io) has been moved to the iomap framework.
XFS adds inode metadata changes to address data type overflow issues 32-bit time_t in 2038. Added similar changes, moving the timer overflow to year 2468, to the code for calculating disk quota times. XFS V4 format is deprecated, user is advised to update FS to V5 format, but there is more than enough time for the update as V4 support will remain until 2030. XFS has also changed the size of the inode input btree, enabling more redundancy checks and faster mount times.
For the FUSE subsystem implemented the DAX operations support to directly access the file system, bypassing the page cache without application-level locking device that is used to avoid the double cache virtiofs the organization of joint access to guest systems , directories and files. Virtiofs also adds support for separate mounting of partitions with different mount points on the host system.
The file system F2FS adds a new garbage collection mode ATGC (Age Threshold Garbage Collection), improved support for zoned NVMe devices, and faster decompression of compressed data.
In F2FS and Ext4, the way to work with file names has been redesigned without taking into account the capital letters; It has been decided to unify the implementation of case-insensitive file names by moving the associated code to a common library.
Another important change is in the asynchronous I / O interface io_uring that adds the ability to create restricted rings that can be securely shared with an untrusted process. This feature allows the base application to selectively restrict access only to its descriptors of individual files for use in third-party applications via io_uring, plus the PIDFD_NONBLOCK flag was added to the pidfd_open () system call to create a non-locking file descriptor (analogous to O_NONBLOCK for pidfd).