Adrian Huang, profile picture
Uploaded byAdrian Huang
PDF, PPTX3,030 views

Page cache in Linux kernel

The document provides a detailed overview of page cache and buffer cache in the Linux kernel, focusing on their roles in optimizing disk data access and interaction with the file system. It discusses techniques for locating existing page caches, the implications of different buffer structures, and scenarios for I/O operations with or without buffer headers. Additionally, it elaborates on the relationship between file system block size, sector size, and their influence on caching mechanisms.

Embed presentation

Download as PDF, PPTX
Page Cache Adrian Huang | Jan, 2022 * Based on kernel 5.11 (x86_64) – QEMU * SMP (4 CPUs) and 8GB memory * Kernel parameter: nokaslr norandmaps * Userspace: ASLR is disabled * EXT4 file system * Legacy BIOS
Agenda • What is page cache? • Page cache & buffer cache (struct buffer_head) • How to find an existed page cache? • Interaction with generic block layer: methods for bio construction 1. Based on buffer_head 2. [Without buffer_head] Based on page descriptor & file system • File system block size & sector size • [Detail Discussion] With or without buffer_head • File system-based IO • Block device-based IO
What is page cache? • page cache (stored in physical memory) = cache disk data • Speed up disk data access • Linux kernel refers to the page cache for disk R/W • If there is enough free memory, • the page cache is kept for an indefinite time • can be reused by other processes without accessing the disk • Open a file with the O_DIRECT flag → Bypass page cache • Application: Some database applications use their own disk cache algorithm • Especially for large data access • Example: Using direct I/O with Oracle Reference from: Chapter 15. The Page Cache, Understanding the Linux Kernel, Third Edition
Page cache & buffer cache 1. Page cache: Interaction with VFS. (Upper layer) 2. Buffer cache: Interaction with the disk. (Lower layer) 4KB 4KB 512B 512B 512B 512B [file] file->f_pos (continuous file position) Page cache and buffer cache 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B sector .. Disk 4KB Page Cache Buffer Cache (Buffer head)
Page Descriptor Buffer Head Buffer Head Buffer Head Buffer Head Buffer Buffer Buffer Buffer Page Frame page_address(page) Disk .. b_page b_data b_this_page private b_dev + b_blocknr: submit_bh()->submit_bio() b_this_page buffer_head b_state b_blocknr b_page b_data b_size = 1024 b_bdev Legend Page cache & buffer cache: relationship 1. Block size = file system-based unit 2. Page cache might *NOT* include buffer_head struct. (File system specific: file data) b_page b_this_page b_this_page b_this_page b_this_page private [page->flags] • PG_private: page cache (fs-private data)
task_struct files dentry d_parent d_name d_inode qstr name = “mnt” u32 hash u32 len = 3 u64 hash_len union files_struct fd_array[] file f_inode f_pos f_mapping . . file inode *i_mapping i_atime i_mtime i_ctime mnt dentry f_path address_space i_data host page_tree i_mmap page mapping index radix_tree_root height = 2 rnode radix_tree_node count = 2 63 0 1 … page 2 3 radix_tree_node count = 1 63 0 1 … 2 3 page page slots[0] slots[3] slots[1] slots[3] slots[2] index = 1 index = 3 index = 194 radix_tree_node count = 1 63 0 1 … 2 3 Radix Tree (or XArray): How to find an existed page cache? [v4.20] XArray replaced radix tree
task_struct files dentry d_parent d_name d_inode qstr name = “mnt” u32 hash u32 len = 3 u64 hash_len union files_struct fd_array[] file f_inode f_pos f_mapping . . file inode *i_mapping i_atime i_mtime i_ctime mnt dentry f_path address_space i_data host page_tree i_mmap page mapping index Radix Tree (or XArray): How to find an existed page cache?
Page Cache/Buffer Cache (fs/buffer.c, mm/readahead.c, mm/filemap.c) Disk filesystem Disk filesystem Block Device File Mapping Layer Generic Block Layer I/O Scheduler Layer Block Device Driver Disk Disk Disk submit_bio() sys_mount()/__x64_sys_mount() VFS mount(…) Buffer b_this_page buffer_head b_state b_blocknr b_page b_data b_size b_bdev Buffer buffer_head … private Page Frame . . ext4_mount->ext4_fill_super fs/buffer.c: ext4_fill_super -> ext4_sb_bread_unmovable bi_io_vec bio bi_iter bi_size bvec_iter bi_sector: 512-byte sector bv_len bio_vec bv_page bv_offset ext4_read_bh_lock-> … ->submit_bh -> submit_bio bio_add_page(…) Userspace application page mapping index XArray Interaction with generic block layer: bio construction based on buffer_head
Interaction with generic block layer: buffer_head • buffer_head cache: per-cpu variable • Scenarios oFile system metadata (file) ▪ superblock ▪ inode info ▪ extent tree oFile hole oBlock device: page cache is not up-to-date
Disk filesystem Mapping Layer Generic Block Layer I/O Scheduler Layer Block Device Driver Disk Disk Disk submit_bio() sys_read()/__x64_sys_read() VFS Page Frame ext4_file_read_iter -> generic_file_read_iter bi_io_vec bio bi_iter bi_size bvec_iter bi_sector: 512-byte sector bv_len bio_vec bv_page bv_offset ext4_mpage_readpages -> ext4_map_blocks -> submit_bio m_lblk ext4_map_blocks m_pblk m_len m_flags reference page_address(page) 1. Allocate a page descriptor 2. Add it to XArray XArray page mapping index file f_pos reference Page Cache/Buffer Cache (fs/buffer.c, mm/readahead.c, mm/filemap.c) Disk filesystem Block Device File bio_add_page(…) Interaction with generic block layer: bio construction based on page & file system
Disk filesystem Mapping Layer Generic Block Layer I/O Scheduler Layer Block Device Driver Disk Disk Disk submit_bio() sys_read()/__x64_sys_read() VFS Page Frame ext4_file_read_iter -> generic_file_read_iter bi_io_vec bio bi_iter bi_size bvec_iter bi_sector: 512-byte sector bv_len bio_vec bv_page bv_offset ext4_mpage_readpages -> ext4_map_blocks -> submit_bio m_lblk ext4_map_blocks m_pblk m_len m_flags reference page_address(page) 1. Allocate a page descriptor 2. Add it to XArray XArray page mapping index file f_pos reference Page Cache/Buffer Cache (fs/buffer.c, mm/readahead.c, mm/filemap.c) Disk filesystem Block Device File bio_add_page(…) Interaction with generic block layer: bio construction based on page & file system 1. No need to allocate buffer_head struct 2. [Scenario] readahead mechanism A. File read/write (file system) B. Block device read/write: corresponding page caches are not available yet.
File system block size & sector size 4KB Page Cache Disk file system block size Mapping Layer: file system Generic Block Layer sector size
File system block size & sector size 4KB Page Cache Disk file system block size Mapping Layer: file system Generic Block Layer sector size
File system block size & sector size: file system block size = 1024 4KB Page Cache Disk Mapping Layer: file system Generic Block Layer sector size bi_size = 1024 bvec_iter bi_sector bv_len = 1024 bio_vec bv_page bv_offset bio Kernel User
File system block size & sector size: file system block size = 1024 4KB Disk Mapping Layer: file system Generic Block Layer sector size bi_size = 1024 bvec_iter bi_sector bv_len = 1024 bio_vec bv_page bv_offset bio
File system block size & sector size: file system block size = 4096 4KB Page Cache Disk Mapping Layer: file system Generic Block Layer sector size bi_size = 4096 bvec_iter bi_sector bv_len = 4096 bio_vec bv_page bv_offset bio Kernel User
File system block size & sector size: file system block size = 4096 4KB Page Cache Disk Mapping Layer: file system Generic Block Layer sector size bi_size = 4096 bvec_iter bi_sector bv_len = 4096 bio_vec bv_page bv_offset bio
File system access & raw block device access 4KB Page Cache Disk file system block size Mapping Layer: file system Generic Block Layer sector size Kernel User VFS sys_read()/__x64_sys_read() 4KB Page Cache Disk block size Mapping Layer: block device file Generic Block Layer sector size Kernel User VFS sys_read()/__x64_sys_read() Submit IO with or without buffer_head struct?
With or without buffer_head Kernel User page cache available? Allocate/init page struct(s) file hole? Submit IO with buffer_head struct Submit IO without buffer_head struct N Is page up-to-date? Return the page Invoke mapping->a_ops->readpage() N Y Y Y N readahead path readpage path [block device: blkdev_readpage()] Submit IO with buffer_head struct ext4 file system: ext4_readpage()
File hole detection – file system implementation read_pages aops->readahead blkdev_readahead mpage_readahead do_mpage_readpage blkdev_get_block ext4_readahead ext4_mpage_readpages ext4_map_blocks reiserfs_readahead do_mpage_readpage reiserfs_get_block [special case] cannot detect file hole ext4 block device reiserfs vfs_read
read_pages aops->readahead blkdev_readahead mpage_readahead do_mpage_readpage blkdev_get_block ext4_readahead ext4_mpage_readpages ext4_map_blocks reiserfs_readahead do_mpage_readpage reiserfs_get_block [special case] cannot detect file hole ext4 block device reiserfs vfs_read Detect file hole Check file hole for specific blocks. Do not set MAPPED flag if blocks are holes File hole detection – file system implementation
• ext4 file system • block size: 1024 bytes • sector size: 512 bytes Test Configuration Sector Size ext4 file system: block size mount command With or without buffer_head: system configuration
File size = 1023 bytes 4KB Page Cache Mapping Layer: file system Generic Block Layer sector size = 512 bi_size = 1024 bvec_iter bi_sector bv_len = 1024 bio_vec bv_page bv_offset bio page_address(page) 1. Allocate a page descriptor 2. Add it to XArray XArray page mapping index file f_pos reference Page Frame read.c
4KB Page Cache Mapping Layer: file system Generic Block Layer sector size = 512 bi_size = 2048 bvec_iter bi_sector bv_len = 2048 bio_vec bv_page bv_offset bio page_address(page) 1. Allocate a page descriptor 2. Add it to XArray XArray page mapping index file f_pos reference Page Frame File size = 2047 bytes
4KB Page Cache Mapping Layer: file system Generic Block Layer sector size = 512 bi_size = 4096 bvec_iter bi_sector bv_len = 4096 bio_vec bv_page bv_offset bio page_address(page) 1. Allocate a page descriptor 2. Add it to XArray XArray page mapping index file f_pos reference Page Frame File size = 4095 bytes
File size = 5119 bytes Mapping Layer: file system Generic Block Layer sector size = 512 bi_size = 5120 bvec_iter bi_sector bv_len = 4096 bio_vec bv_page bv_offset bio page_address(page) XArray page mapping index Page Frame / # /read /adrian/mnt/files/5119.txt 512 bv_len = 1024 bio_vec bv_page bv_offset bi_vcnt = 2 4KB Page Cache page mapping index • Spatial locality • Default readahead pages: 32 o If file size < “default readahead pages”, read the number of pages of the file. Readahead mechanism Kernel User
Readahead mechanism: default pages = 32
File size = 5119 bytes Mapping Layer: file system Generic Block Layer sector size = 512 bi_size = 5120 bvec_iter bi_sector bv_len = 4096 bio_vec bv_page bv_offset bio bv_len = 1024 bio_vec bv_page bv_offset bi_vcnt = 2 4KB Page Cache
file-hole-2 (ext4 file system) hole data hole data block # 0 1 2 3 Disk Buffer b_this_page buffer_head b_state b_blocknr b_page b_data b_size b_bdev Buffer buffer_head … Page Frame page mapping index XArray Buffer Buffer buffer_head … b_this_page buffer_head b_state b_blocknr b_page b_data b_size b_bdev bi_io_vec bio bi_iter bi_io_vec bio bi_iter private bio_add_page bio_add_page submit_bio submit_bio File hole – Use buffer_head
file-hole-2 (ext4 file system) hole data hole data block # 0 1 2 3 Disk Buffer b_this_page buffer_head b_state b_blocknr b_page b_data b_size b_bdev Buffer buffer_head … Page Frame page mapping index XArray Buffer Buffer buffer_head … b_this_page buffer_head b_state b_blocknr b_page b_data b_size b_bdev bi_io_vec bio bi_iter bi_io_vec bio bi_iter private bio_add_page bio_add_page submit_bio submit_bio File hole – Use buffer_head xxd file-hole-2
File hole – Use buffer_head hole data hole data block # 0 1 2 3
File hole – Use buffer_head hole data hole data block # 0 1 2 3
Call path with/without buffer_head struct Call path without buffer_head Call path with buffer_head
task_struct files dentry d_parent d_name d_inode qstr name = “mnt” u32 hash u32 len = 3 u64 hash_len union files_struct fd_array[] file f_inode f_pos f_mapping . . file inode *i_mapping i_atime i_mtime i_ctime mnt dentry f_path address_space i_data host page_tree i_mmap page mapping index a_ops address_space_operations readpage readahead … mapping layer: file system or raw block disk Interaction between VFS/mm and mapping layer Disk filesystem Mapping Layer Generic Block Layer I/O Scheduler Layer Block Device Driver Disk Disk Disk submit_bio() sys_read()/__x64_sys_read() VFS Page Cache/Buffer Cache (fs/buffer.c, mm/readahead.c, mm/filemap.c) Disk filesystem Block Device File mm
task_struct files dentry d_parent d_name d_inode qstr name = “mnt” u32 hash u32 len = 3 u64 hash_len union files_struct fd_array[] file f_inode f_pos f_mapping . . file inode *i_mapping i_atime i_mtime i_ctime mnt dentry f_path address_space i_data host page_tree i_mmap page mapping index a_ops address_space_operations readpage readahead … mapping layer: file system or raw block disk Interaction between VFS/mm and mapping layer
Block device access – without buffer_head (full-page access) Page Frame blkdev_read_iter -> generic_file_read_iter bi_io_vec bio bi_iter bi_size = 4096 bvec_iter bi_sector = 512 bv_len bio_vec bv_page bv_offset read_pages -> blkdev_readahead -> mpage_readahead -> do_mpage_readpage b_blocknr = 256 buffer_head b_bdev b_size = 4096 reference page_address(page) 1. Allocate a page descriptor 2. Add it to XArray XArray page mapping index file f_pos reference get_block -> map_bh bio_add_page(…)
Block device access – without buffer_head (full-page access) bi_io_vec bio bi_iter bi_size = 4096 bvec_iter bi_sector = 512 bv_len bio_vec bv_page bv_offset read_pages -> blkdev_readahead -> mpage_readahead -> do_mpage_readpage b_blocknr = 256 buffer_head b_bdev b_size = 4096 reference get_block -> map_bh
Block device access – with buffer_head Unread super block data block # 0 1 2 3 Unread /dev/loop0 Page cache (not up-to-update)
Backup
[File system] page cache without buffer_head
[Block device] page cache with buffer_head
[Block device] page cache with buffer_head
[Block device] page cache without buffer_head

More Related Content

PDF
Memory Mapping Implementation (mmap) in Linux Kernel
byAdrian Huang
 
PDF
Physical Memory Management.pdf
byAdrian Huang
 
PDF
Process Address Space: The way to create virtual address (page table) of user...
byAdrian Huang
 
PDF
Physical Memory Models.pdf
byAdrian Huang
 
PDF
Memory Compaction in Linux Kernel.pdf
byAdrian Huang
 
PDF
Linux Kernel - Virtual File System
byAdrian Huang
 
PDF
Decompressed vmlinux: linux kernel initialization from page table configurati...
byAdrian Huang
 
PDF
Reverse Mapping (rmap) in Linux Kernel
byAdrian Huang
 
Memory Mapping Implementation (mmap) in Linux Kernel
byAdrian Huang
 
Physical Memory Management.pdf
byAdrian Huang
 
Process Address Space: The way to create virtual address (page table) of user...
byAdrian Huang
 
Physical Memory Models.pdf
byAdrian Huang
 
Memory Compaction in Linux Kernel.pdf
byAdrian Huang
 
Linux Kernel - Virtual File System
byAdrian Huang
 
Decompressed vmlinux: linux kernel initialization from page table configurati...
byAdrian Huang
 
Reverse Mapping (rmap) in Linux Kernel
byAdrian Huang
 

What's hot

PPTX
Slab Allocator in Linux Kernel
byAdrian Huang
 
PDF
qemu + gdb: The efficient way to understand/debug Linux kernel code/data stru...
byAdrian Huang
 
PDF
Memory Management with Page Folios
byAdrian Huang
 
PDF
malloc & vmalloc in Linux
byAdrian Huang
 
PDF
semaphore & mutex.pdf
byAdrian Huang
 
PPTX
qemu + gdb + sample_code: Run sample code in QEMU OS and observe Linux Kernel...
byAdrian Huang
 
PDF
spinlock.pdf
byAdrian Huang
 
PDF
Vmlinux: anatomy of bzimage and how x86 64 processor is booted
byAdrian Huang
 
PDF
Anatomy of the loadable kernel module (lkm)
byAdrian Huang
 
PDF
Linux dma engine
bypradeep_tewani
 
PPTX
Linux Initialization Process (2)
byshimosawa
 
PDF
Linux Performance Profiling and Monitoring
byGeorg Schönberger
 
PDF
Kdump and the kernel crash dump analysis
byBuland Singh
 
PPTX
Linux Initialization Process (1)
byshimosawa
 
PDF
Arm device tree and linux device drivers
byHoucheng Lin
 
PDF
Ixgbe internals
bySUSE Labs Taipei
 
PPTX
Linux Kernel Booting Process (1) - For NLKB
byshimosawa
 
PDF
The Linux Block Layer - Built for Fast Storage
byKernel TLV
 
PPTX
Linux Memory Management
byNi Zo-Ma
 
ODP
eBPF maps 101
bySUSE Labs Taipei
 
Slab Allocator in Linux Kernel
byAdrian Huang
 
qemu + gdb: The efficient way to understand/debug Linux kernel code/data stru...
byAdrian Huang
 
Memory Management with Page Folios
byAdrian Huang
 
malloc & vmalloc in Linux
byAdrian Huang
 
semaphore & mutex.pdf
byAdrian Huang
 
qemu + gdb + sample_code: Run sample code in QEMU OS and observe Linux Kernel...
byAdrian Huang
 
spinlock.pdf
byAdrian Huang
 
Vmlinux: anatomy of bzimage and how x86 64 processor is booted
byAdrian Huang
 
Anatomy of the loadable kernel module (lkm)
byAdrian Huang
 
Linux dma engine
bypradeep_tewani
 
Linux Initialization Process (2)
byshimosawa
 
Linux Performance Profiling and Monitoring
byGeorg Schönberger
 
Kdump and the kernel crash dump analysis
byBuland Singh
 
Linux Initialization Process (1)
byshimosawa
 
Arm device tree and linux device drivers
byHoucheng Lin
 
Ixgbe internals
bySUSE Labs Taipei
 
Linux Kernel Booting Process (1) - For NLKB
byshimosawa
 
The Linux Block Layer - Built for Fast Storage
byKernel TLV
 
Linux Memory Management
byNi Zo-Ma
 
eBPF maps 101
bySUSE Labs Taipei
 

Similar to Page cache in Linux kernel

PDF
Page Cache in Linux 2.6.pdf
byycelgemici1
 
ODP
Memory management in Linux
byRaghu Udiyar
 
PPT
kerch04.ppt
byKalimuthuVelappan
 
PDF
Understand
byKalimuthu Velappan
 
PDF
Java File I/O Performance Analysis - Part I - JCConf 2018
byMichael Fong
 
PPT
8 1-os file system implementation
byGol D Roger
 
PPT
Windows memory manager internals
bySisimon Soman
 
PPTX
Ioppt
byarun sharma
 
PPTX
Linux memory-management-kamal
byKamal Maiti
 
PPT
Driver development – memory management
byVandana Salve
 
PDF
ASPLOS2011 workshop RESoLVE "Effect of Disk Prefetching of Guest OS "
byKuniyasu Suzaki
 
PDF
009709863.pdf
byKalsoomTahir2
 
PPT
Memory comp
byMohansonale1
 
ODP
Introduction to file system and OCFS2
byGang He
 
PPT
Unix file systems 2 in unix internal systems
bysenthilamul
 
PDF
Ugif 10 2012 beauty ofifmxdiskstructs ugif
byUGIF
 
PPT
file management_part2_os_notes.ppt
byHelalMirzad
 
PPTX
OS Unit5.pptx
byDHANABALSUBRAMANIAN
 
PPT
Memory
byMuhammed Mazhar Khan
 
PDF
lvm logical volume management is a tool in Linux that allows you to manage st...
byBhupeshKanire1
 
Page Cache in Linux 2.6.pdf
byycelgemici1
 
Memory management in Linux
byRaghu Udiyar
 
kerch04.ppt
byKalimuthuVelappan
 
Understand
byKalimuthu Velappan
 
Java File I/O Performance Analysis - Part I - JCConf 2018
byMichael Fong
 
8 1-os file system implementation
byGol D Roger
 
Windows memory manager internals
bySisimon Soman
 
Ioppt
byarun sharma
 
Linux memory-management-kamal
byKamal Maiti
 
Driver development – memory management
byVandana Salve
 
ASPLOS2011 workshop RESoLVE "Effect of Disk Prefetching of Guest OS "
byKuniyasu Suzaki
 
009709863.pdf
byKalsoomTahir2
 
Memory comp
byMohansonale1
 
Introduction to file system and OCFS2
byGang He
 
Unix file systems 2 in unix internal systems
bysenthilamul
 
Ugif 10 2012 beauty ofifmxdiskstructs ugif
byUGIF
 
file management_part2_os_notes.ppt
byHelalMirzad
 
OS Unit5.pptx
byDHANABALSUBRAMANIAN
 
Memory
byMuhammed Mazhar Khan
 
lvm logical volume management is a tool in Linux that allows you to manage st...
byBhupeshKanire1
 

Recently uploaded

PDF
DSD-INT 2025 The Sinterklaas Storm in the Southwest of the Netherlands - Wope...
byDeltares
 
PPTX
Understanding-ROM-RAM-Cache-and-Buffer-Memory.pptx.pptx
bylata2850
 
PDF
An Agent Walks Into a Bar... and Meets Another Agent: Multi-Agent Systems in ...
byMaxim Salnikov
 
PDF
DSD-INT 2025 Century-Scale Impacts of Sediment-Based Coastal Restoration unde...
byDeltares
 
PDF
DSD-INT 2025 Building-Aware Flood and Lifeline Scour Modeling with Delft3D FM...
byDeltares
 
PDF
BCA 1st Semester Fundamentals Solved Question Paper 44121
byKuvempu University
 
PPTX
Future of Software Testing: AI-Powered Open Source Testing Tools
bySophie Lane
 
PDF
Bring AI and build AI agents into your Jakarta EE Apps with LangChain4J-CDI
byBuhake Sindi
 
PDF
DSD-INT 2025 MeshKernel and D-Grid Editor - Stout
byDeltares
 
PDF
Data Integration with Salesforce Bootcamp
byDele Amefo
 
PPTX
wAIred_LearnwithoutAI_KotlinDevDay_27112025.pptx
bySimonedeGijt
 
PDF
Internship Project Training Report-3.pdf
byPawank36
 
PDF
DSD-INT 2025 Reviving a Lost Meander - Deen
byDeltares
 
PPTX
operating sys about shells and terminals commands .pptx
byYazeedAbdullah6
 
PPTX
Spec-driven Development with Kiro - AWS User Group Melbourne
byMalinda Kapuruge
 
PDF
DSD-INT 2025 Modernizing Hydrodynamics in Large Flood Forecasting System - Mi...
byDeltares
 
PDF
7 Levels of a Web Performance Journey @ Google DevFest Brooklyn 2025
bySergeyChernyshev
 
PDF
DSD-INT 2025 Hydrodynamic and Morphodynamic Modeling with Delft3D FM for an I...
byDeltares
 
PPTX
CRM Development Company | Custom CRM Development Services
byyugababu033
 
PDF
DSD-INT 2025 Validation of SFINCS on Historical River Floods at the Global Sc...
byDeltares
 
DSD-INT 2025 The Sinterklaas Storm in the Southwest of the Netherlands - Wope...
byDeltares
 
Understanding-ROM-RAM-Cache-and-Buffer-Memory.pptx.pptx
bylata2850
 
An Agent Walks Into a Bar... and Meets Another Agent: Multi-Agent Systems in ...
byMaxim Salnikov
 
DSD-INT 2025 Century-Scale Impacts of Sediment-Based Coastal Restoration unde...
byDeltares
 
DSD-INT 2025 Building-Aware Flood and Lifeline Scour Modeling with Delft3D FM...
byDeltares
 
BCA 1st Semester Fundamentals Solved Question Paper 44121
byKuvempu University
 
Future of Software Testing: AI-Powered Open Source Testing Tools
bySophie Lane
 
Bring AI and build AI agents into your Jakarta EE Apps with LangChain4J-CDI
byBuhake Sindi
 
DSD-INT 2025 MeshKernel and D-Grid Editor - Stout
byDeltares
 
Data Integration with Salesforce Bootcamp
byDele Amefo
 
wAIred_LearnwithoutAI_KotlinDevDay_27112025.pptx
bySimonedeGijt
 
Internship Project Training Report-3.pdf
byPawank36
 
DSD-INT 2025 Reviving a Lost Meander - Deen
byDeltares
 
operating sys about shells and terminals commands .pptx
byYazeedAbdullah6
 
Spec-driven Development with Kiro - AWS User Group Melbourne
byMalinda Kapuruge
 
DSD-INT 2025 Modernizing Hydrodynamics in Large Flood Forecasting System - Mi...
byDeltares
 
7 Levels of a Web Performance Journey @ Google DevFest Brooklyn 2025
bySergeyChernyshev
 
DSD-INT 2025 Hydrodynamic and Morphodynamic Modeling with Delft3D FM for an I...
byDeltares
 
CRM Development Company | Custom CRM Development Services
byyugababu033
 
DSD-INT 2025 Validation of SFINCS on Historical River Floods at the Global Sc...
byDeltares
 

Page cache in Linux kernel

  • 1.
    Page Cache Adrian Huang| Jan, 2022 * Based on kernel 5.11 (x86_64) – QEMU * SMP (4 CPUs) and 8GB memory * Kernel parameter: nokaslr norandmaps * Userspace: ASLR is disabled * EXT4 file system * Legacy BIOS
  • 2.
    Agenda • What ispage cache? • Page cache & buffer cache (struct buffer_head) • How to find an existed page cache? • Interaction with generic block layer: methods for bio construction 1. Based on buffer_head 2. [Without buffer_head] Based on page descriptor & file system • File system block size & sector size • [Detail Discussion] With or without buffer_head • File system-based IO • Block device-based IO
  • 3.
    What is pagecache? • page cache (stored in physical memory) = cache disk data • Speed up disk data access • Linux kernel refers to the page cache for disk R/W • If there is enough free memory, • the page cache is kept for an indefinite time • can be reused by other processes without accessing the disk • Open a file with the O_DIRECT flag → Bypass page cache • Application: Some database applications use their own disk cache algorithm • Especially for large data access • Example: Using direct I/O with Oracle Reference from: Chapter 15. The Page Cache, Understanding the Linux Kernel, Third Edition
  • 4.
    Page cache &buffer cache 1. Page cache: Interaction with VFS. (Upper layer) 2. Buffer cache: Interaction with the disk. (Lower layer) 4KB 4KB 512B 512B 512B 512B [file] file->f_pos (continuous file position) Page cache and buffer cache 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B 512B sector .. Disk 4KB Page Cache Buffer Cache (Buffer head)
  • 5.
    Page Descriptor Buffer Head BufferHead Buffer Head Buffer Head Buffer Buffer Buffer Buffer Page Frame page_address(page) Disk .. b_page b_data b_this_page private b_dev + b_blocknr: submit_bh()->submit_bio() b_this_page buffer_head b_state b_blocknr b_page b_data b_size = 1024 b_bdev Legend Page cache & buffer cache: relationship 1. Block size = file system-based unit 2. Page cache might *NOT* include buffer_head struct. (File system specific: file data) b_page b_this_page b_this_page b_this_page b_this_page private [page->flags] • PG_private: page cache (fs-private data)
  • 6.
    task_struct files dentry d_parent d_name d_inode qstr name = “mnt” u32hash u32 len = 3 u64 hash_len union files_struct fd_array[] file f_inode f_pos f_mapping . . file inode *i_mapping i_atime i_mtime i_ctime mnt dentry f_path address_space i_data host page_tree i_mmap page mapping index radix_tree_root height = 2 rnode radix_tree_node count = 2 63 0 1 … page 2 3 radix_tree_node count = 1 63 0 1 … 2 3 page page slots[0] slots[3] slots[1] slots[3] slots[2] index = 1 index = 3 index = 194 radix_tree_node count = 1 63 0 1 … 2 3 Radix Tree (or XArray): How to find an existed page cache? [v4.20] XArray replaced radix tree
  • 7.
    task_struct files dentry d_parent d_name d_inode qstr name = “mnt” u32hash u32 len = 3 u64 hash_len union files_struct fd_array[] file f_inode f_pos f_mapping . . file inode *i_mapping i_atime i_mtime i_ctime mnt dentry f_path address_space i_data host page_tree i_mmap page mapping index Radix Tree (or XArray): How to find an existed page cache?
  • 8.
    Page Cache/Buffer Cache (fs/buffer.c,mm/readahead.c, mm/filemap.c) Disk filesystem Disk filesystem Block Device File Mapping Layer Generic Block Layer I/O Scheduler Layer Block Device Driver Disk Disk Disk submit_bio() sys_mount()/__x64_sys_mount() VFS mount(…) Buffer b_this_page buffer_head b_state b_blocknr b_page b_data b_size b_bdev Buffer buffer_head … private Page Frame . . ext4_mount->ext4_fill_super fs/buffer.c: ext4_fill_super -> ext4_sb_bread_unmovable bi_io_vec bio bi_iter bi_size bvec_iter bi_sector: 512-byte sector bv_len bio_vec bv_page bv_offset ext4_read_bh_lock-> … ->submit_bh -> submit_bio bio_add_page(…) Userspace application page mapping index XArray Interaction with generic block layer: bio construction based on buffer_head
  • 9.
    Interaction with genericblock layer: buffer_head • buffer_head cache: per-cpu variable • Scenarios oFile system metadata (file) ▪ superblock ▪ inode info ▪ extent tree oFile hole oBlock device: page cache is not up-to-date
  • 10.
    Disk filesystem Mapping Layer Generic BlockLayer I/O Scheduler Layer Block Device Driver Disk Disk Disk submit_bio() sys_read()/__x64_sys_read() VFS Page Frame ext4_file_read_iter -> generic_file_read_iter bi_io_vec bio bi_iter bi_size bvec_iter bi_sector: 512-byte sector bv_len bio_vec bv_page bv_offset ext4_mpage_readpages -> ext4_map_blocks -> submit_bio m_lblk ext4_map_blocks m_pblk m_len m_flags reference page_address(page) 1. Allocate a page descriptor 2. Add it to XArray XArray page mapping index file f_pos reference Page Cache/Buffer Cache (fs/buffer.c, mm/readahead.c, mm/filemap.c) Disk filesystem Block Device File bio_add_page(…) Interaction with generic block layer: bio construction based on page & file system
  • 11.
    Disk filesystem Mapping Layer Generic BlockLayer I/O Scheduler Layer Block Device Driver Disk Disk Disk submit_bio() sys_read()/__x64_sys_read() VFS Page Frame ext4_file_read_iter -> generic_file_read_iter bi_io_vec bio bi_iter bi_size bvec_iter bi_sector: 512-byte sector bv_len bio_vec bv_page bv_offset ext4_mpage_readpages -> ext4_map_blocks -> submit_bio m_lblk ext4_map_blocks m_pblk m_len m_flags reference page_address(page) 1. Allocate a page descriptor 2. Add it to XArray XArray page mapping index file f_pos reference Page Cache/Buffer Cache (fs/buffer.c, mm/readahead.c, mm/filemap.c) Disk filesystem Block Device File bio_add_page(…) Interaction with generic block layer: bio construction based on page & file system 1. No need to allocate buffer_head struct 2. [Scenario] readahead mechanism A. File read/write (file system) B. Block device read/write: corresponding page caches are not available yet.
  • 12.
    File system blocksize & sector size 4KB Page Cache Disk file system block size Mapping Layer: file system Generic Block Layer sector size
  • 13.
    File system blocksize & sector size 4KB Page Cache Disk file system block size Mapping Layer: file system Generic Block Layer sector size
  • 14.
    File system blocksize & sector size: file system block size = 1024 4KB Page Cache Disk Mapping Layer: file system Generic Block Layer sector size bi_size = 1024 bvec_iter bi_sector bv_len = 1024 bio_vec bv_page bv_offset bio Kernel User
  • 15.
    File system blocksize & sector size: file system block size = 1024 4KB Disk Mapping Layer: file system Generic Block Layer sector size bi_size = 1024 bvec_iter bi_sector bv_len = 1024 bio_vec bv_page bv_offset bio
  • 16.
    File system blocksize & sector size: file system block size = 4096 4KB Page Cache Disk Mapping Layer: file system Generic Block Layer sector size bi_size = 4096 bvec_iter bi_sector bv_len = 4096 bio_vec bv_page bv_offset bio Kernel User
  • 17.
    File system blocksize & sector size: file system block size = 4096 4KB Page Cache Disk Mapping Layer: file system Generic Block Layer sector size bi_size = 4096 bvec_iter bi_sector bv_len = 4096 bio_vec bv_page bv_offset bio
  • 18.
    File system access& raw block device access 4KB Page Cache Disk file system block size Mapping Layer: file system Generic Block Layer sector size Kernel User VFS sys_read()/__x64_sys_read() 4KB Page Cache Disk block size Mapping Layer: block device file Generic Block Layer sector size Kernel User VFS sys_read()/__x64_sys_read() Submit IO with or without buffer_head struct?
  • 19.
    With or withoutbuffer_head Kernel User page cache available? Allocate/init page struct(s) file hole? Submit IO with buffer_head struct Submit IO without buffer_head struct N Is page up-to-date? Return the page Invoke mapping->a_ops->readpage() N Y Y Y N readahead path readpage path [block device: blkdev_readpage()] Submit IO with buffer_head struct ext4 file system: ext4_readpage()
  • 20.
    File hole detection– file system implementation read_pages aops->readahead blkdev_readahead mpage_readahead do_mpage_readpage blkdev_get_block ext4_readahead ext4_mpage_readpages ext4_map_blocks reiserfs_readahead do_mpage_readpage reiserfs_get_block [special case] cannot detect file hole ext4 block device reiserfs vfs_read
  • 21.
    read_pages aops->readahead blkdev_readahead mpage_readahead do_mpage_readpage blkdev_get_block ext4_readahead ext4_mpage_readpages ext4_map_blocks reiserfs_readahead do_mpage_readpage reiserfs_get_block [special case] cannot detectfile hole ext4 block device reiserfs vfs_read Detect file hole Check file hole for specific blocks. Do not set MAPPED flag if blocks are holes File hole detection – file system implementation
  • 22.
    • ext4 filesystem • block size: 1024 bytes • sector size: 512 bytes Test Configuration Sector Size ext4 file system: block size mount command With or without buffer_head: system configuration
  • 23.
    File size =1023 bytes 4KB Page Cache Mapping Layer: file system Generic Block Layer sector size = 512 bi_size = 1024 bvec_iter bi_sector bv_len = 1024 bio_vec bv_page bv_offset bio page_address(page) 1. Allocate a page descriptor 2. Add it to XArray XArray page mapping index file f_pos reference Page Frame read.c
  • 24.
    4KB Page Cache Mapping Layer:file system Generic Block Layer sector size = 512 bi_size = 2048 bvec_iter bi_sector bv_len = 2048 bio_vec bv_page bv_offset bio page_address(page) 1. Allocate a page descriptor 2. Add it to XArray XArray page mapping index file f_pos reference Page Frame File size = 2047 bytes
  • 25.
    4KB Page Cache Mapping Layer:file system Generic Block Layer sector size = 512 bi_size = 4096 bvec_iter bi_sector bv_len = 4096 bio_vec bv_page bv_offset bio page_address(page) 1. Allocate a page descriptor 2. Add it to XArray XArray page mapping index file f_pos reference Page Frame File size = 4095 bytes
  • 26.
    File size =5119 bytes Mapping Layer: file system Generic Block Layer sector size = 512 bi_size = 5120 bvec_iter bi_sector bv_len = 4096 bio_vec bv_page bv_offset bio page_address(page) XArray page mapping index Page Frame / # /read /adrian/mnt/files/5119.txt 512 bv_len = 1024 bio_vec bv_page bv_offset bi_vcnt = 2 4KB Page Cache page mapping index • Spatial locality • Default readahead pages: 32 o If file size < “default readahead pages”, read the number of pages of the file. Readahead mechanism Kernel User
  • 27.
  • 28.
    File size =5119 bytes Mapping Layer: file system Generic Block Layer sector size = 512 bi_size = 5120 bvec_iter bi_sector bv_len = 4096 bio_vec bv_page bv_offset bio bv_len = 1024 bio_vec bv_page bv_offset bi_vcnt = 2 4KB Page Cache
  • 29.
    file-hole-2 (ext4 file system) holedata hole data block # 0 1 2 3 Disk Buffer b_this_page buffer_head b_state b_blocknr b_page b_data b_size b_bdev Buffer buffer_head … Page Frame page mapping index XArray Buffer Buffer buffer_head … b_this_page buffer_head b_state b_blocknr b_page b_data b_size b_bdev bi_io_vec bio bi_iter bi_io_vec bio bi_iter private bio_add_page bio_add_page submit_bio submit_bio File hole – Use buffer_head
  • 30.
    file-hole-2 (ext4 file system) holedata hole data block # 0 1 2 3 Disk Buffer b_this_page buffer_head b_state b_blocknr b_page b_data b_size b_bdev Buffer buffer_head … Page Frame page mapping index XArray Buffer Buffer buffer_head … b_this_page buffer_head b_state b_blocknr b_page b_data b_size b_bdev bi_io_vec bio bi_iter bi_io_vec bio bi_iter private bio_add_page bio_add_page submit_bio submit_bio File hole – Use buffer_head xxd file-hole-2
  • 31.
    File hole –Use buffer_head hole data hole data block # 0 1 2 3
  • 32.
    File hole –Use buffer_head hole data hole data block # 0 1 2 3
  • 33.
    Call path with/withoutbuffer_head struct Call path without buffer_head Call path with buffer_head
  • 34.
    task_struct files dentry d_parent d_name d_inode qstr name = “mnt” u32hash u32 len = 3 u64 hash_len union files_struct fd_array[] file f_inode f_pos f_mapping . . file inode *i_mapping i_atime i_mtime i_ctime mnt dentry f_path address_space i_data host page_tree i_mmap page mapping index a_ops address_space_operations readpage readahead … mapping layer: file system or raw block disk Interaction between VFS/mm and mapping layer Disk filesystem Mapping Layer Generic Block Layer I/O Scheduler Layer Block Device Driver Disk Disk Disk submit_bio() sys_read()/__x64_sys_read() VFS Page Cache/Buffer Cache (fs/buffer.c, mm/readahead.c, mm/filemap.c) Disk filesystem Block Device File mm
  • 35.
    task_struct files dentry d_parent d_name d_inode qstr name = “mnt” u32hash u32 len = 3 u64 hash_len union files_struct fd_array[] file f_inode f_pos f_mapping . . file inode *i_mapping i_atime i_mtime i_ctime mnt dentry f_path address_space i_data host page_tree i_mmap page mapping index a_ops address_space_operations readpage readahead … mapping layer: file system or raw block disk Interaction between VFS/mm and mapping layer
  • 36.
    Block device access– without buffer_head (full-page access) Page Frame blkdev_read_iter -> generic_file_read_iter bi_io_vec bio bi_iter bi_size = 4096 bvec_iter bi_sector = 512 bv_len bio_vec bv_page bv_offset read_pages -> blkdev_readahead -> mpage_readahead -> do_mpage_readpage b_blocknr = 256 buffer_head b_bdev b_size = 4096 reference page_address(page) 1. Allocate a page descriptor 2. Add it to XArray XArray page mapping index file f_pos reference get_block -> map_bh bio_add_page(…)
  • 37.
    Block device access– without buffer_head (full-page access) bi_io_vec bio bi_iter bi_size = 4096 bvec_iter bi_sector = 512 bv_len bio_vec bv_page bv_offset read_pages -> blkdev_readahead -> mpage_readahead -> do_mpage_readpage b_blocknr = 256 buffer_head b_bdev b_size = 4096 reference get_block -> map_bh
  • 38.
    Block device access– with buffer_head Unread super block data block # 0 1 2 3 Unread /dev/loop0 Page cache (not up-to-update)
  • 39.
  • 40.
    [File system] pagecache without buffer_head
  • 41.
    [Block device] pagecache with buffer_head
  • 42.
    [Block device] pagecache with buffer_head
  • 43.
    [Block device] pagecache without buffer_head