{"id":104123,"date":"2022-05-23T15:36:40","date_gmt":"2022-05-23T13:36:40","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/reliz-yadra-linux-5-18"},"modified":"2022-05-23T15:36:40","modified_gmt":"2022-05-23T13:36:40","slug":"reliz-yadra-linux-5-18","status":"publish","type":"post","link":"https:\/\/prohoster.info\/en\/blog\/news\/reliz-yadra-linux-5-18","title":{"rendered":"Release of Linux Kernel 5.18","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
After two months of development, Linus Torvalds announced the release of Linux kernel 5.18. Among the most notable changes are: a significant cleanup of legacy functionality, the deprecation of the Reiserfs filesystem, the implementation of trace events for user processes, added support for Intel IBT exploit mitigation, enabled buffer overflow detection mode when using the memcpy() function, introduced fprobe function call tracking, improved CPU scheduler performance on AMD Zen, included a driver to manage Intel CPU functionality (SDS), integrated parts of patches for restructuring header files, and approved the adoption of the C11 standard. <\/p>\n
The new version includes 16,206 bug fixes from 2,127 developers (compared to 14,203 fixes from 1,995 developers in the previous release), with a patch size of 108 MB (the changes affected 14,235 files, added 1,340,982 lines of code, and removed 593,836 lines). About 44% of all changes in 5.18 are related to device drivers, approximately 16% pertain to updates to architecture-specific code, 11% are related to the network stack, 3% involve file systems, and 3% concern internal kernel subsystems. <\/p>\n
Key innovations in kernel 5.18: <\/p>\n
\n
Disk subsystem, input\/output, and file systems<\/b>\n
\n
In the Btrfs filesystem, support for passing compressed data during send and receive operations has been added. Previously, when using send\/receive, the sending side would uncompress the data stored in a compressed format, while the receiving side would recompress it before writing. In kernel 5.18, user-space applications using send\/receive calls are now enabled to transmit compressed data without re-packaging. This functionality is made possible through new ioctl operations BTRFS_IOC_ENCODED_READ and BTRFS_IOC_ENCODED_WRITE, allowing direct reading and writing of information in extents.\n
Additionally, Btrfs has improved fsync performance. Support for deduplication and reflink (cloning file metadata with the creation of a link to already existing data without actual duplication) has been added for the entire storage, not limited to mount points. <\/p>\n
In Direct I\/O mode, access to encrypted files is enabled when using fscrypt inline encryption, where encryption and decryption operations are performed by the storage controller rather than the kernel. In traditional kernel-based encryption, access to encrypted files using Direct I\/O remains impossible, as the file access bypasses the buffering mechanism in the kernel.\n
By default, the NFS server supports the NFSv3 protocol, which no longer requires separate activation and is available with general NFS activation. NFSv3 is considered the primary and always-supported version of NFS, while support for NFSv2 may be discontinued in the future. The efficiency of reading directory contents has significantly improved.\n
The ReiserFS filesystem has been classified as obsolete and is expected to be removed in 2025. Decommissioning ReiserFS will reduce maintenance efforts related to shared changes across filesystems concerning support for the new API for mounting, iomap, and volumes.\n
For F2FS, the ability to map user IDs of mounted filesystems has been implemented, which is used to associate files of a specific user on a mounted foreign partition with a different user in the current system.\n
The code for calculating statistics in Device-mapper handlers has been reworked, significantly improving accounting accuracy in handlers such as dm-crypt.\n
Support for 64-bit checksums for integrity verification has been implemented for NVMe devices.\n
A new mount option \"keep_last_dots\" has been proposed for the exfat file system, which prevents the cleaning of dots at the end of file names (in Windows, dots at the end of a name are removed by default).\n
In EXT4, the performance of the fast_commit mode has been improved, and scalability has been increased. The mount option \"mb_optimize_scan,\" which enhances performance under high filesystem fragmentation, has been adapted to work with extent files.\n
Support for write streams in the subsystem managing block devices has been discontinued. This feature was proposed for SSDs but did not gain traction, and there are currently no devices that support this mode, making it unlikely that they will appear in the future. <\/ul>\n
Memory and system services<\/b>\n
\n
Integration of a patch set has begun, significantly reducing the time for kernel rebuilds by restructuring the hierarchy of header files and minimizing cross-dependencies. The 5.18 kernel includes patches that optimize the structure of the task scheduler's header files (kernel\/sched). Compared to the previous release, the CPU time consumption during the build of the kernel\/sched code has decreased by 61%, and the actual time has reduced by 3.9% (from 2.95 to 2.84 seconds).\n
The kernel code now allows the use of the C11 standard, published in 2011. Previously added code to the kernel had to comply with the ANSI C (C89) specification established in 1989. In the build scripts for kernel 5.18, the option \u2018--std=gnu89\u2019 has been replaced with \u2018--std=gnu11 -Wno-shift-negative-value.\u2019 The possibility of using the C17 standard was considered, but it would have required raising the minimum supported version of GCC; however, supporting C11 fits within the current requirements for GCC version (5.1).\n
Task scheduling performance on AMD processors with Zen microarchitecture has been improved, where multiple last-level caches (LLCs) are provided for each node with local memory channels. The new version addresses LLC balance issues between NUMA nodes, resulting in significant performance gains under certain types of workloads.\n
Tools for tracing user-space applications have been expanded. The new kernel version adds the capability for user processes to create tracing events and record data in a tracing buffer, which can be viewed through typical kernel tracing utilities like ftrace and perf. User-space tracing events are isolated from kernel tracing events. The status of events can be checked via the file \/sys\/kernel\/debug\/tracing\/user_events_status, while event registration and data recording are managed through the file \/sys\/kernel\/debug\/tracing\/user_events_data.\n
A function call tracking mechanism \u2014 fprobe \u2014 has been added. The fprobe API is based on ftrace but is limited to attaching callback handlers to function entry and exit points. Unlike kprobes and kretprobes, the new mechanism allows the use of a single handler for multiple functions at once.\n
Support for older ARM processors (ARMv4 and ARMv5), which lack a memory management unit (MMU), has been discontinued. Support for ARMv7-M systems without an MMU has been retained.\n
Support for the RISC-like NDS32 architecture, used in processors from Andes Technologies, has been discontinued. The code was removed due to lack of maintenance and low demand for NDS32 support in the main Linux kernel (remaining users rely on specialized kernel builds from hardware manufacturers).\n
By default, kernel builds with support for the a.out executable format for the alpha and m68k architectures, where this format is still in use, have been disabled. It is likely that support for the outdated a.out format will be completely removed from the kernel soon. Plans to remove the a.out format have been discussed since 2019.\n
Minimal support for the vDSO (virtual dynamic shared objects) mechanism has been implemented for the PA-RISC architecture, providing a limited set of system calls available in user space without context switching. vDSO support has enabled the ability to run with a non-executable stack.\n
Support for the Intel HFI (Hardware Feedback Interface) mechanism has been added, allowing hardware to convey information about the current performance and energy efficiency of each CPU to the kernel.\n
A driver for the Intel SDSi (Software-Defined Silicon) mechanism has been added, allowing for the management of enabling additional features in the processor (for example, specialized instructions and additional cache memory). The idea is that chips with locked extended features can be supplied at a lower price, which can then be 'purchased' and activated without hardware replacement of the chip.\n
The amd_hsmp driver has been added to support the AMD HSMP (Host System Management Port) interface, providing access to processor management features through a set of special registers available in AMD EPYC server processors starting from the Fam19h generation. For example, HSMP can provide data on power consumption and temperature, set frequency limits, activate various performance boost modes, and manage memory operation parameters.\n
The asynchronous I\/O interface io_uring has implemented the IORING_SETUP_SUBMIT_ALL option for registering a set of file descriptors in the ring buffer, as well as the IORING_OP_MSG_RING operation, which allows signaling from one ring buffer to another.\n
The DAMOS (Data Access Monitoring-based Operation Schemes) mechanism, which allows for memory release based on access frequency, has expanded capabilities for monitoring memory operations from user space.\n
The third series of patches has been integrated, implementing the concept of page folios, which resemble compound pages but offer improved semantics and a clearer organizational approach. Utilizing folios speeds up memory management in certain kernel subsystems. The proposed patches involved transitioning internal memory management functions to folios, including variations of the get_user_pages() function. Support for creating large folios within data prefetching code has been provided.\n
The build system has introduced support for the USERCFLAGS and USERLDFLAGS environment variables, allowing the passing of additional flags to the compiler and linker.\n
In the eBPF subsystem, the BTF (BPF Type Format) mechanism, which provides type checking information in BPF pseudocode, has implemented the ability to add annotations to variables that reference memory areas in user space. Annotations help the BPF code verification system to more effectively identify and validate memory access.\n
A new memory allocation handler for storing loaded BPF programs has been proposed, which allows for more efficient memory usage in situations where a large number of BPF programs are loaded.\n
The system call madvise() has been enhanced with the MADV_DONTNEED_LOCKED flag, which complements the existing MADV_DONTNEED flag. This allows the kernel to be informed in advance about the impending release of a memory block, indicating that it is no longer needed and may be reused by the kernel. Unlike MADV_DONTNEED, the use of MADV_DONTNEED_LOCKED is permitted for locked pages in RAM, which can be evicted without changing their locked status when madvise is called. If there is a subsequent access to the block that generates a 'page fault', it will return while maintaining its binding. Additionally, a modification has been added to allow the MADV_DONTNEED flag to be used with large memory pages in HugeTLB. <\/ul>\n
Virtualization and Security<\/b>\n
\n
Support for the Intel IBT (Indirect Branch Tracking) execution flow protection mechanism has been added for the x86 architecture. This mechanism prevents the use of exploit construction techniques utilizing Return-Oriented Programming (ROP), where an exploit is formed as a chain of calls to existing machine instruction fragments in memory, ending with a return control instruction (typically, function endings). The essence of the implemented protection method lies in blocking indirect jumps into the function body by adding a special ENDBR instruction at the beginning of the function and allowing execution through an indirect jump only if it leads to this instruction (an indirect call through JMP and CALL must always hit the ENDBR instruction, which is placed at the very start of the function).\n
Stricter boundary checks have been enabled for the memcpy(), memmove(), and memset() functions, which are performed at compile time when CONFIG_FORTIFY_SOURCE mode is enabled. The added change checks for out-of-bounds accesses to structure elements whose sizes are known. It is noted that this implemented feature could block all buffer overflows related to memcpy() in the kernel, identified over the past three years at least.\n