KEMBAR78
Linux Kernel - Virtual File System | PDF
Adrian Huang, profile picture
Uploaded byAdrian Huang
1,705 views

Linux Kernel - Virtual File System

The document provides an in-depth examination of the virtual file system (VFS) in the Linux kernel, focusing on its inner workings in relation to the ext4 file system format. It covers key components such as directory entries (dentry), inodes, and the anatomy of system calls like 'mount' and 'ls', explaining object relationships within VFS. Various data structures and their roles in file operations, memory management, and file system interactions are also analyzed.

Downloaded 67 times
Virtual File System (VFS) in Linux Kernel Adrian Huang | Nov, 2021 * Based on kernel 5.11 (x86_64) – QEMU * SMP (4 CPUs) and 8GB memory * Kernel parameter: nokaslr norandmaps * Userspace: ASLR is disabled * Legacy BIOS
Agenda • Kernel components affected by an IO request • Object relationship in VFS • Ext4 file system format • dentry (Directory Entry) and inode (Index Node) • Example: ext4 • Anatomy of `mount` system call • Anatomy of `ls` system call
Kernel components affected by an IO request VFS Page Cache/Buffer Cache (fs/buffer.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()
open(“/mnt/a.txt”, …) open(“/mnt/b.txt”, …) Process A File Object File Object File Object open file table dentry dentry dentry cache inode inode inode cache inode table Kernel - VFS dentry in data blocks Disk Object relationship in VFS dentry cache list • Linked via hlist_bl_{head,node} (hash table with bit lock) inode cache list • Linked via hlist_{head,node} (hash table) Note
Super Block Group Descriptors Reserved GDT blocks Block bitmap inode bitmap inode table: inode descriptor Data Block Data Block . . . 0 1 Total inode count Total block count Free blocks Free inodes … Block Group 0 . . . Block Group N Boot Sector Partition mode Owner UID, Owner GID i_atime (last access time) i_ctime (last inode change time) i_mtime (last data modification time) size ext4_extent_idx ei_block ei_leaf_lo ei_leaf_hi ei_unused ext4_extent_header eh_magic = 0xF30A eh_entries eh_max eh_depth eh_generation ext4_extent_idx ei_block ei_leaf_lo ei_leaf_hi ei_unused ext4_extent ee_block ee_len ei_start_hi ei_start_lo ext4_extent ee_block ee_len ei_start_hi ei_start_lo . . . . . . root index node leaf node Disk blocks Extent Tree Ext4 file system format Storage size addressed by a block group: 4096 (bytes) * 8 * 4096 (block size) = 128MB
Super Block Group Descriptors Reserved GDT blocks Block bitmap inode bitmap inode table: inode descriptor Data Block Data Block . . . 0 1 Total inode count Total block count Free blocks Free inodes … Block Group 0 . . . Block Group N Boot Sector Partition mode Owner UID, Owner GID i_atime (last access time) i_ctime (last inode change time) i_mtime (last data modification time) size ext4_extent_idx ei_block ei_leaf_lo ei_leaf_hi ei_unused ext4_extent_header eh_magic = 0xF30A eh_entries eh_max eh_depth eh_generation ext4_extent_idx ei_block ei_leaf_lo ei_leaf_hi ei_unused ext4_extent ee_block ee_len ei_start_hi ei_start_lo ext4_extent ee_block ee_len ei_start_hi ei_start_lo . . . . . . root index node leaf node Disk blocks Extent Tree Ext4 file system format
inode# Purpose 0 Doesn't exist; there is no inode 0 1 List of defective blocks 2 Root directory 3 User quota 4 Group quota 5 Boot loader 6 Undelete directory 7 Reserved group descriptors inode. ("resize inode") 8 Journal inode 9 The "exclude" inode 10 Replica inode 11 First non-reserved inode. Usually this is the lost+found directory. See s_first_ino in the superblock. Relationship between inode and directory entry (dentry) inode number link count
inode bitmap: 19 * 4096 = 77824 (0x13000) inode 1-16 inode 17 `dumpe2fs 16MiB.img` Relationship between inode and directory entry (dentry)
inode entry: 35 * 4096 + 128 * (2 – 1) [inode #2] = 0x23080 `dumpe2fs 16MiB.img` inode structure data of inode #2 Relationship between inode and directory entry (dentry)
mode Owner UID, Owner GID i_atime i_ctime i_mtime size ext4_extent_header eh_magic = 0xF30A eh_entries = 1 eh_max = 4 eh_depth = 0 eh_generation = 0 root Disk blocks ext4_extent ee_block = 0 ee_len = 0x0001 ei_start_hi = 0x0000 ei_start_lo = 0x00000004 Relationship between inode and dentry – root (inode #2) ext4_inode (inode #2) leaf node . 2 name inode Directory Entry (dentry) Block #4: A list of struct ext4_dir_entry_2 .. 2 -> 2646185 lost+found 11 a123 12 a789 17 Block #0-3 Block #5 inode structure data of inode #2 Root Directory dentry in data block of inode #2
type = directory i_links_count = 5 Relationship between inode and dentry – ‘b456’ (inode #13) i-node table . 2 name inode Directory Entry (dentry) - / .. 2 -> 2646185 lost+found 11 a123 12 a789 17 2 12 . 12 name inode .. 2 b456 13 hello.c 15 hello-hard-link.c 15 Directory Entry (dentry) - /a123 13 . 13 name inode .. 12 test.c 16 test-hard-link.c 16 test-soft-link.c 14 Directory Entry (dentry) - /a123/b456 14 16 Data block pointers type = directory i_links_count = 3 Data block pointers type = directory i_links_count = 2 Data block pointers type = symbolic link i_links_count = 1 Data block pointers = “test.c” type = file i_links_count = 2 Data block pointers File Data /a123/b456/test.c size = 6 size = 100 inode #
type = directory i_links_count = 5 Relationship between inode and dentry – ‘test.c’ (inode #16) i-node table . 2 name inode Directory Entry (dentry) - / .. 2 -> 2646185 lost+found 11 a123 12 a789 17 2 12 . 12 name inode .. 2 b456 13 hello.c 15 hello-hard-link.c 15 Directory Entry (dentry) - /a123 13 . 13 name inode .. 12 test.c 16 test-hard-link.c 16 test-soft-link.c 14 Directory Entry (dentry) - /a123/b456 14 16 Data block pointers type = directory i_links_count = 3 Data block pointers type = directory i_links_count = 2 Data block pointers type = symbolic link i_links_count = 1 Data block pointers = “test.c” type = file i_links_count = 2 Data block pointers File Data /a123/b456/test.c size = 6 size = 100 inode structure data of inode #16 inode entry: 35 * 4096 + 128 * (16 – 1) [inode #16] = 0x23780 address inode structure of inode #16 Data block pointer represented by struct ext4_extent i_links_count inode #
type = directory i_links_count = 5 Relationship between inode and dentry – ‘test-soft-link.c’ (inode #14) i-node table . 2 name inode Directory Entry (dentry) - / .. 2 -> 2646185 lost+found 11 a123 12 a789 17 2 12 . 12 name inode .. 2 b456 13 hello.c 15 hello-hard-link.c 15 Directory Entry (dentry) - /a123 13 . 13 name inode .. 12 test.c 16 test-hard-link.c 16 test-soft-link.c 14 Directory Entry (dentry) - /a123/b456 14 16 Data block pointers type = directory i_links_count = 3 Data block pointers type = directory i_links_count = 2 Data block pointers type = symbolic link i_links_count = 1 Data block pointers = “test.c” type = file i_links_count = 2 Data block pointers File Data /a123/b456/test.c size = 6 size = 100 inode structure data of inode #14 inode entry: 35 * 4096 + 128 * (14 – 1) [inode #14] = 0x23680 address inode structure of inode #14 i_links_count Data block pointer Hard Link 1. Refer to an inode number 2. Must reside on the same file system 3. Cannot be made to a directory Soft Link 1. Refer to a filename instead of an inode number 2. Can be linked to a file in a different file system 3. Can be made to a directory inode #
mount – mount a new ext4 file system • stat() – get file status from inode o Name resolution → get dentry → get inode • openat() o Name resolution o Update open file table • mount() o Name resolution o Add a mount entry to ‘mount’ hash table o Parse file system
stat("/adrian/img/src.img“, …) user_path_at(): fill a path struct (dentry and mount info) after resolving a pathname vfsmount mnt_root mnt_sb dentry d_name path mnt dentry d_inode path struct • stat() – get file status from inode o Name resolution → get dentry → get inode
stat("/adrian/img/src.img“, …) 1. namei: convert a “name” to an “inode” 2. nameidata: store the current status when walking a path nameidata last name root inode path task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” local variable
stat("/adrian/img/src.img“, …) path_lookupat(): 1. Configure members of nameidata struct based on absolute pathname or relative pathname 2. Name resolution vfsmount mnt_root mnt_sb dentry d_name qstr name = “/” u32 hash u32 len u64 hash_len union nameidata last (S*) name root (S*) inode path (S*) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” local variable path mnt dentry current->fs->root d_inode * Data storage (not a data pointer) Note
link_path_walk(): name resolution – round 1 vfsmount mnt_root mnt_sb dentry d_name qstr name = “/” u32 hash u32 len u64 hash_len union nameidata last (S*) name root (S*) inode path (S*) * Data storage (not a data pointer) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” path mnt dentry current->fs->root d_inode qstr name = “adrian/img/src.img” u32 hash u32 len = 6 u64 hash_len union variable ‘name’ = “adrian/img/src.img” dentry d_name d_inode path mnt dentry variable ‘name’ = “img/src.img” hash 1 2 3 4 5 6 Note local variable link_path_walk() • stop until the last dentry is resolved • for-loop in this function qstr name = “adrian” u32 hash u32 len = 6 u64 hash_len union
link_path_walk(): name resolution – round 1 nameidata last (S*) name root (S*) inode path (S*) * Data storage (not a data pointer) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” path mnt dentry current->fs->root qstr name = “adrian/img/src.img” u32 hash u32 len = 6 u64 hash_len union variable ‘name’ = “adrian/img/src.img” dentry d_name d_inode path mnt dentry variable ‘name’ = “img/src.img” hash 1 2 3 4 5 6 Note local variable lookup_slow()
vfsmount mnt_root mnt_sb dentry d_name qstr name = “/” u32 hash u32 len u64 hash_len union nameidata last (S*) name root (S*) inode path (S*) * Data storage (not a data pointer) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” local variable path mnt dentry current->fs->root d_inode qstr name = “img/src.img” u32 hash u32 len = 3 u64 hash_len union variable ‘name’ = “img/src.img” dentry d_name d_inode path mnt dentry variable ‘name’ = “src.img” hash 1 2 3 4 5 6 Note link_path_walk(): name resolution – round 2 qstr name = “img” u32 hash u32 len = 3 u64 hash_len union link_path_walk() • stop until the last dentry is resolved • for-loop in this function
vfsmount mnt_root mnt_sb dentry d_name qstr name = “/” u32 hash u32 len u64 hash_len union nameidata last (S*) name root (S*) inode path (S*) * Data storage (not a data pointer) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” local variable path mnt dentry current->fs->root d_inode qstr name = “src.img” u32 hash u32 len = 3 u64 hash_len union variable ‘name’ = “src.img” dentry d_name d_inode path mnt dentry variable ‘name’ = “” 2 Note Function return link_path_walk(): name resolution – round 3 1 hash qstr name = “img” u32 hash u32 len = 3 u64 hash_len union link_path_walk() • stop until the last dentry is resolved • for-loop in this function
vfsmount mnt_root mnt_sb dentry d_name qstr name = “/” u32 hash u32 len u64 hash_len union nameidata last (S*) name root (S*) inode path (S*) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” local variable path mnt dentry current->fs->root d_inode qstr name = “src.img” u32 hash u32 len = 6 u64 hash_len union variable ‘name’ = “src.img” dentry d_name qstr name = “src.img” u32 hash u32 len = 6 u64 hash_len union d_inode path mnt dentry variable ‘name’ = “” 3 4 5 6 Function return lookup_last(): name resolution hash 1 2 * Data storage (not a data pointer) Note lookup_last() • get the corresponding dentry based on the filename • “while loop”: Follow the symlink in the final component
vfsmount mnt_root mnt_sb dentry d_name qstr name = “/” u32 hash u32 len u64 hash_len union nameidata last (S*) name root (S*) inode path (S*) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” local variable path mnt dentry current->fs->root d_inode qstr name = “src.img” u32 hash u32 len = 6 u64 hash_len union variable ‘name’ = “src.img” dentry d_name qstr name = “src.img” u32 hash u32 len = 6 u64 hash_len union d_inode path mnt dentry variable ‘name’ = “” 3 4 5 6 Function return lookup_last(): name resolution hash 1 2 * Data storage (not a data pointer) Note
mount – mount a new ext4 file system
hlist_head.first hlist_head.first . . mount_hashtable mount mnt_parent struct dentry *mnt_mountpoint struct vfsmount mnt **pprev *next const char *mnt_devname = “none” mount mnt_parent struct dentry *mnt_mountpoint **pprev struct hlist_node mnt_hash *next const char *mnt_devname = “none” struct hlist_node mnt_hash dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union super_block s_blocksize s_maxbytes s_op = ramfs_sops s_root s_bdev s_id[32] = “rootfs” s_fs_info s_type = rootfs_fs_type d_parent mnt_root mnt_sb struct vfsmount mnt mount – before mounting a new ext4 file system d_sb
hlist_head.first hlist_head.first . . mount_hashtable mount mnt_parent struct dentry *mnt_mountpoint struct vfsmount mnt **pprev *next const char *mnt_devname = “none” mount mnt_parent struct dentry *mnt_mountpoint **pprev struct hlist_node mnt_hash *next const char *mnt_devname = “none” struct hlist_node mnt_hash dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union super_block s_blocksize s_maxbytes s_op = ramfs_sops s_root s_bdev s_id[32] = “rootfs” s_fs_info s_type = rootfs_fs_type d_parent mnt_root mnt_sb struct vfsmount mnt mount – after mounting a new ext4 file system mount mnt_parent struct dentry *mnt_mountpoint **pprev *next const char *mnt_devname = “/dev/loop0” mnt_root mnt_sb dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union struct vfsmount mnt super_block s_op = ext4_sops s_id[32] = “loop0” s_type = ext4_fs_type d_parent struct hlist_node mnt_hash dentry d_name d_parent qstr name = “mnt” u32 hash u32 len = 3 u64 hash_len union /adrian/mnt d_sb d_sb
hlist_head.first hlist_head.first . . mount_hashtable mount mnt_parent struct dentry *mnt_mountpoint struct vfsmount mnt **pprev *next const char *mnt_devname = “none” mount mnt_parent struct dentry *mnt_mountpoint **pprev struct hlist_node mnt_hash *next const char *mnt_devname = “none” struct hlist_node mnt_hash dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union super_block s_blocksize s_maxbytes s_op = ramfs_sops s_root s_bdev s_id[32] = “rootfs” s_fs_info s_type = rootfs_fs_type d_parent mnt_root mnt_sb struct vfsmount mnt mount – after mounting a new ext4 file system mount mnt_parent struct dentry *mnt_mountpoint **pprev *next const char *mnt_devname = “/dev/loop0” mnt_root mnt_sb dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union struct vfsmount mnt super_block s_op = ext4_sops s_id[32] = “loop0” s_type = ext4_fs_type d_parent struct hlist_node mnt_hash dentry d_name d_parent qstr name = “mnt” u32 hash u32 len = 3 u64 hash_len union /adrian/mnt d_sb d_sb
`ls –color mnt`
`ls –color mnt`: openat(AT_FDCWD, "/adrian/mnt“, …)
`ls –color mnt`: openat(AT_FDCWD, "/adrian/mnt“, …)
`ls –color mnt`: openat(AT_FDCWD, "/adrian/mnt“, …)
path mnt dentry vfsmount mnt_root mnt_sb qstr name = “mnt” u32 hash u32 len = 3 u64 hash_len union dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union super_block s_blocksize s_maxbytes s_op = ramfs_ops s_root s_bdev s_id[32] = “rootfs” s_fs_info d_inode dentry d_name d_inode d_parent s_type = rootfs_fs_type path mnt dentry vfsmount mnt_root mnt_sb qstr name = “mnt /” u32 hash u32 len = 3 1 u64 hash_len union dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union super_block s_blocksize s_maxbytes s_op = ext4_sops s_root s_bdev s_id[32] = “loop0” s_fs_info d_inode dentry d_name d_inode d_parent s_type = ext4_fs_type /adrian/mnt data structures in the original file system /adrian/mnt data structures in newly mounted file system traverse_mounts() `ls –color mnt`: openat(AT_FDCWD, "/adrian/mnt“, …) The original mount point is changed the newly one when running name resolution
path mnt dentry vfsmount mnt_root mnt_sb qstr name = “mnt /” u32 hash u32 len = 3 1 u64 hash_len union dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union super_block s_blocksize s_maxbytes s_op = ext4_sops s_root s_bdev s_id[32] = “loop0” s_fs_info d_inode dentry d_name d_inode d_parent s_type = ext4_fs_type /adrian/mnt data structures in newly mounted file system flags: path->dentry->d_flags `ls –color mnt`: openat(AT_FDCWD, "/adrian/mnt“, …) REF-walk [Two concurrent mechanisms for name resolution] 1. REF-walk: concurrent management with refcounts and spinlock 2. RCU-walk: faster name resolution lookup

More Related Content

PDF
Vmlinux: anatomy of bzimage and how x86 64 processor is booted
byAdrian Huang
 
PDF
Process Address Space: The way to create virtual address (page table) of user...
byAdrian Huang
 
PDF
Decompressed vmlinux: linux kernel initialization from page table configurati...
byAdrian Huang
 
PDF
qemu + gdb: The efficient way to understand/debug Linux kernel code/data stru...
byAdrian Huang
 
PDF
Page cache in Linux kernel
byAdrian Huang
 
PDF
Physical Memory Models.pdf
byAdrian Huang
 
PDF
Memory Compaction in Linux Kernel.pdf
byAdrian Huang
 
PDF
spinlock.pdf
byAdrian Huang
 
Vmlinux: anatomy of bzimage and how x86 64 processor is booted
byAdrian Huang
 
Process Address Space: The way to create virtual address (page table) of user...
byAdrian Huang
 
Decompressed vmlinux: linux kernel initialization from page table configurati...
byAdrian Huang
 
qemu + gdb: The efficient way to understand/debug Linux kernel code/data stru...
byAdrian Huang
 
Page cache in Linux kernel
byAdrian Huang
 
Physical Memory Models.pdf
byAdrian Huang
 
Memory Compaction in Linux Kernel.pdf
byAdrian Huang
 
spinlock.pdf
byAdrian Huang
 

What's hot

PDF
Physical Memory Management.pdf
byAdrian Huang
 
PDF
Memory Mapping Implementation (mmap) in Linux Kernel
byAdrian Huang
 
PDF
Reverse Mapping (rmap) in Linux Kernel
byAdrian Huang
 
PPTX
Linux Initialization Process (2)
byshimosawa
 
PPTX
Slab Allocator in Linux Kernel
byAdrian Huang
 
PDF
Memory Management with Page Folios
byAdrian Huang
 
PDF
Anatomy of the loadable kernel module (lkm)
byAdrian Huang
 
PDF
malloc & vmalloc in Linux
byAdrian Huang
 
PPTX
Linux Memory Management
byNi Zo-Ma
 
PPTX
qemu + gdb + sample_code: Run sample code in QEMU OS and observe Linux Kernel...
byAdrian Huang
 
PDF
Kernel_Crash_Dump_Analysis
byBuland Singh
 
PDF
Linux Internals - Part II
byEmertxe Information Technologies Pvt Ltd
 
PPTX
Linux kernel debugging
byHao-Ran Liu
 
PDF
Kernel Recipes 2017 - Understanding the Linux kernel via ftrace - Steven Rostedt
byAnne Nicolas
 
PDF
Process Scheduler and Balancer in Linux Kernel
byHaifeng Li
 
PPTX
Linux Kernel Booting Process (1) - For NLKB
byshimosawa
 
PPTX
Linux Memory Management with CMA (Contiguous Memory Allocator)
byPankaj Suryawanshi
 
PDF
Linux device drivers
byEmertxe Information Technologies Pvt Ltd
 
PPT
Linux Crash Dump Capture and Analysis
byPaul V. Novarese
 
PDF
LISA2019 Linux Systems Performance
byBrendan Gregg
 
Physical Memory Management.pdf
byAdrian Huang
 
Memory Mapping Implementation (mmap) in Linux Kernel
byAdrian Huang
 
Reverse Mapping (rmap) in Linux Kernel
byAdrian Huang
 
Linux Initialization Process (2)
byshimosawa
 
Slab Allocator in Linux Kernel
byAdrian Huang
 
Memory Management with Page Folios
byAdrian Huang
 
Anatomy of the loadable kernel module (lkm)
byAdrian Huang
 
malloc & vmalloc in Linux
byAdrian Huang
 
Linux Memory Management
byNi Zo-Ma
 
qemu + gdb + sample_code: Run sample code in QEMU OS and observe Linux Kernel...
byAdrian Huang
 
Kernel_Crash_Dump_Analysis
byBuland Singh
 
Linux Internals - Part II
byEmertxe Information Technologies Pvt Ltd
 
Linux kernel debugging
byHao-Ran Liu
 
Kernel Recipes 2017 - Understanding the Linux kernel via ftrace - Steven Rostedt
byAnne Nicolas
 
Process Scheduler and Balancer in Linux Kernel
byHaifeng Li
 
Linux Kernel Booting Process (1) - For NLKB
byshimosawa
 
Linux Memory Management with CMA (Contiguous Memory Allocator)
byPankaj Suryawanshi
 
Linux device drivers
byEmertxe Information Technologies Pvt Ltd
 
Linux Crash Dump Capture and Analysis
byPaul V. Novarese
 
LISA2019 Linux Systems Performance
byBrendan Gregg
 

Similar to Linux Kernel - Virtual File System

PPT
Vfs
byWaqas !!!!
 
PPTX
UNIT III.pptx
byYogapriyaJ1
 
PPTX
UNIX_Linux_File_Subsystem_Detail ved.pptx
bypriyankamurotia137
 
PDF
TLPI Chapter 14 File Systems
byShu-Yu Fu
 
PPT
Linux
bykeydak11
 
PPTX
Ext filesystem4
byNeha Kulkarni
 
PDF
009709863.pdf
byKalsoomTahir2
 
ODP
4. linux file systems
byMarian Marinov
 
PPT
Unix file systems 2 in unix internal systems
bysenthilamul
 
PPTX
Examining Linux File Structures
byprimeteacher32
 
PDF
Ch11 file system implementation
byWelly Dian Astika
 
PDF
Part 03 File System Implementation in Linux
byTushar B Kute
 
PDF
Course 102: Lecture 27: FileSystems in Linux (Part 2)
byAhmed El-Arabawy
 
PPTX
Ubuntu File System
byBhagyesh Patel
 
PPTX
Files and directories in Linux 6
byMeenakshi Paul
 
PDF
Linux fundamental - Chap 10 fs
byKenny (netman)
 
PPTX
file system overview in oerating system .
bypavanm6360
 
PDF
Operating Systems - Implementing File Systems
byMukesh Chinta
 
PDF
1 04 rao
bydimitar9
 
PDF
File systems for Embedded Linux
byEmertxe Information Technologies Pvt Ltd
 
Vfs
byWaqas !!!!
 
UNIT III.pptx
byYogapriyaJ1
 
UNIX_Linux_File_Subsystem_Detail ved.pptx
bypriyankamurotia137
 
TLPI Chapter 14 File Systems
byShu-Yu Fu
 
Linux
bykeydak11
 
Ext filesystem4
byNeha Kulkarni
 
009709863.pdf
byKalsoomTahir2
 
4. linux file systems
byMarian Marinov
 
Unix file systems 2 in unix internal systems
bysenthilamul
 
Examining Linux File Structures
byprimeteacher32
 
Ch11 file system implementation
byWelly Dian Astika
 
Part 03 File System Implementation in Linux
byTushar B Kute
 
Course 102: Lecture 27: FileSystems in Linux (Part 2)
byAhmed El-Arabawy
 
Ubuntu File System
byBhagyesh Patel
 
Files and directories in Linux 6
byMeenakshi Paul
 
Linux fundamental - Chap 10 fs
byKenny (netman)
 
file system overview in oerating system .
bypavanm6360
 
Operating Systems - Implementing File Systems
byMukesh Chinta
 
1 04 rao
bydimitar9
 
File systems for Embedded Linux
byEmertxe Information Technologies Pvt Ltd
 

Recently uploaded

PDF
The new Volto Form Block, Plone Conference 2025
byRob Gietema
 
PDF
Where Does Your Data Go When You Put In Into A Database_.pdf
byDave Stokes
 
PPT
SAP ERP Training For Senior Managers.ppt
byBalanathanVirupasan
 
PDF
Security Architecture Views the FDA Expects — And How to Get Them Right
byICS
 
PDF
Practical Performance Tuning for Serverless Java on AWS- InfoQ Dev Summit
byVadym Kazulkin
 
PDF
The Hidden Cost of Complicated PMO Tools - Using Gartner’s Guidance and OnePl...
byOnePlan Solutions
 
PDF
Safe Confined Space Entry Monitoring_ Singapore Experts.pdf
byprasadexiga0
 
PDF
Scraping Amazon Prime Watchlist Data for Viewing Patterns.pdf
byyashpatric
 
PPTX
AI-Powered Intelligent Agents for Fraud Detection
byleolucus1995
 
PDF
Best Tableau Alternatives for Data Analytics and Reporting.pdf
byVarsha Nayak
 
DOCX
What Is the Best BI Software Other Than Tableau.docx
byVarsha Nayak
 
PPTX
Moving Plone Blob Storage to S3 Object Storage - Plone Conference 2025
byAlin Voinea
 
PDF
SAP & CTRM Testing Like Never Before
byZoe Gilbert
 
PPTX
Financial Literacy for Software Developers
byBalanathanVirupasan
 
PPTX
Cyber-Security-Essentials (21AK1A0217).pptx
byGaganGagan44
 
PDF
The Rise of AI Agents: Powering the Next Industrial Era
byMaxim Salnikov
 
PDF
Driving Stress Detection Using Multimodal CNN with Nonlinear Representation o...
byPranjal Kumar
 
PDF
Privacy-first in-browser Generative AI web apps: offline-ready, future-proof,...
byMaxim Salnikov
 
PPSX
Domain Centered Architecture for complex business domains
byRob Vens
 
PDF
Austin Marketo User Group: Ground Control 2 MOPS: AI-Ignite your Map
bybbedford2
 
The new Volto Form Block, Plone Conference 2025
byRob Gietema
 
Where Does Your Data Go When You Put In Into A Database_.pdf
byDave Stokes
 
SAP ERP Training For Senior Managers.ppt
byBalanathanVirupasan
 
Security Architecture Views the FDA Expects — And How to Get Them Right
byICS
 
Practical Performance Tuning for Serverless Java on AWS- InfoQ Dev Summit
byVadym Kazulkin
 
The Hidden Cost of Complicated PMO Tools - Using Gartner’s Guidance and OnePl...
byOnePlan Solutions
 
Safe Confined Space Entry Monitoring_ Singapore Experts.pdf
byprasadexiga0
 
Scraping Amazon Prime Watchlist Data for Viewing Patterns.pdf
byyashpatric
 
AI-Powered Intelligent Agents for Fraud Detection
byleolucus1995
 
Best Tableau Alternatives for Data Analytics and Reporting.pdf
byVarsha Nayak
 
What Is the Best BI Software Other Than Tableau.docx
byVarsha Nayak
 
Moving Plone Blob Storage to S3 Object Storage - Plone Conference 2025
byAlin Voinea
 
SAP & CTRM Testing Like Never Before
byZoe Gilbert
 
Financial Literacy for Software Developers
byBalanathanVirupasan
 
Cyber-Security-Essentials (21AK1A0217).pptx
byGaganGagan44
 
The Rise of AI Agents: Powering the Next Industrial Era
byMaxim Salnikov
 
Driving Stress Detection Using Multimodal CNN with Nonlinear Representation o...
byPranjal Kumar
 
Privacy-first in-browser Generative AI web apps: offline-ready, future-proof,...
byMaxim Salnikov
 
Domain Centered Architecture for complex business domains
byRob Vens
 
Austin Marketo User Group: Ground Control 2 MOPS: AI-Ignite your Map
bybbedford2
 

Linux Kernel - Virtual File System

  • 1.
    Virtual File System(VFS) in Linux Kernel Adrian Huang | Nov, 2021 * Based on kernel 5.11 (x86_64) – QEMU * SMP (4 CPUs) and 8GB memory * Kernel parameter: nokaslr norandmaps * Userspace: ASLR is disabled * Legacy BIOS
  • 2.
    Agenda • Kernel componentsaffected by an IO request • Object relationship in VFS • Ext4 file system format • dentry (Directory Entry) and inode (Index Node) • Example: ext4 • Anatomy of `mount` system call • Anatomy of `ls` system call
  • 3.
    Kernel components affectedby an IO request VFS Page Cache/Buffer Cache (fs/buffer.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()
  • 4.
    open(“/mnt/a.txt”, …) open(“/mnt/b.txt”,…) Process A File Object File Object File Object open file table dentry dentry dentry cache inode inode inode cache inode table Kernel - VFS dentry in data blocks Disk Object relationship in VFS dentry cache list • Linked via hlist_bl_{head,node} (hash table with bit lock) inode cache list • Linked via hlist_{head,node} (hash table) Note
  • 5.
    Super Block Group Descriptors Reserved GDT blocks Block bitmap inode bitmap inode table: inodedescriptor Data Block Data Block . . . 0 1 Total inode count Total block count Free blocks Free inodes … Block Group 0 . . . Block Group N Boot Sector Partition mode Owner UID, Owner GID i_atime (last access time) i_ctime (last inode change time) i_mtime (last data modification time) size ext4_extent_idx ei_block ei_leaf_lo ei_leaf_hi ei_unused ext4_extent_header eh_magic = 0xF30A eh_entries eh_max eh_depth eh_generation ext4_extent_idx ei_block ei_leaf_lo ei_leaf_hi ei_unused ext4_extent ee_block ee_len ei_start_hi ei_start_lo ext4_extent ee_block ee_len ei_start_hi ei_start_lo . . . . . . root index node leaf node Disk blocks Extent Tree Ext4 file system format Storage size addressed by a block group: 4096 (bytes) * 8 * 4096 (block size) = 128MB
  • 6.
    Super Block Group Descriptors Reserved GDT blocks Block bitmap inode bitmap inode table: inodedescriptor Data Block Data Block . . . 0 1 Total inode count Total block count Free blocks Free inodes … Block Group 0 . . . Block Group N Boot Sector Partition mode Owner UID, Owner GID i_atime (last access time) i_ctime (last inode change time) i_mtime (last data modification time) size ext4_extent_idx ei_block ei_leaf_lo ei_leaf_hi ei_unused ext4_extent_header eh_magic = 0xF30A eh_entries eh_max eh_depth eh_generation ext4_extent_idx ei_block ei_leaf_lo ei_leaf_hi ei_unused ext4_extent ee_block ee_len ei_start_hi ei_start_lo ext4_extent ee_block ee_len ei_start_hi ei_start_lo . . . . . . root index node leaf node Disk blocks Extent Tree Ext4 file system format
  • 7.
    inode# Purpose 0 Doesn'texist; there is no inode 0 1 List of defective blocks 2 Root directory 3 User quota 4 Group quota 5 Boot loader 6 Undelete directory 7 Reserved group descriptors inode. ("resize inode") 8 Journal inode 9 The "exclude" inode 10 Replica inode 11 First non-reserved inode. Usually this is the lost+found directory. See s_first_ino in the superblock. Relationship between inode and directory entry (dentry) inode number link count
  • 8.
    inode bitmap: 19* 4096 = 77824 (0x13000) inode 1-16 inode 17 `dumpe2fs 16MiB.img` Relationship between inode and directory entry (dentry)
  • 9.
    inode entry: 35* 4096 + 128 * (2 – 1) [inode #2] = 0x23080 `dumpe2fs 16MiB.img` inode structure data of inode #2 Relationship between inode and directory entry (dentry)
  • 10.
    mode Owner UID, OwnerGID i_atime i_ctime i_mtime size ext4_extent_header eh_magic = 0xF30A eh_entries = 1 eh_max = 4 eh_depth = 0 eh_generation = 0 root Disk blocks ext4_extent ee_block = 0 ee_len = 0x0001 ei_start_hi = 0x0000 ei_start_lo = 0x00000004 Relationship between inode and dentry – root (inode #2) ext4_inode (inode #2) leaf node . 2 name inode Directory Entry (dentry) Block #4: A list of struct ext4_dir_entry_2 .. 2 -> 2646185 lost+found 11 a123 12 a789 17 Block #0-3 Block #5 inode structure data of inode #2 Root Directory dentry in data block of inode #2
  • 11.
    type = directory i_links_count= 5 Relationship between inode and dentry – ‘b456’ (inode #13) i-node table . 2 name inode Directory Entry (dentry) - / .. 2 -> 2646185 lost+found 11 a123 12 a789 17 2 12 . 12 name inode .. 2 b456 13 hello.c 15 hello-hard-link.c 15 Directory Entry (dentry) - /a123 13 . 13 name inode .. 12 test.c 16 test-hard-link.c 16 test-soft-link.c 14 Directory Entry (dentry) - /a123/b456 14 16 Data block pointers type = directory i_links_count = 3 Data block pointers type = directory i_links_count = 2 Data block pointers type = symbolic link i_links_count = 1 Data block pointers = “test.c” type = file i_links_count = 2 Data block pointers File Data /a123/b456/test.c size = 6 size = 100 inode #
  • 12.
    type = directory i_links_count= 5 Relationship between inode and dentry – ‘test.c’ (inode #16) i-node table . 2 name inode Directory Entry (dentry) - / .. 2 -> 2646185 lost+found 11 a123 12 a789 17 2 12 . 12 name inode .. 2 b456 13 hello.c 15 hello-hard-link.c 15 Directory Entry (dentry) - /a123 13 . 13 name inode .. 12 test.c 16 test-hard-link.c 16 test-soft-link.c 14 Directory Entry (dentry) - /a123/b456 14 16 Data block pointers type = directory i_links_count = 3 Data block pointers type = directory i_links_count = 2 Data block pointers type = symbolic link i_links_count = 1 Data block pointers = “test.c” type = file i_links_count = 2 Data block pointers File Data /a123/b456/test.c size = 6 size = 100 inode structure data of inode #16 inode entry: 35 * 4096 + 128 * (16 – 1) [inode #16] = 0x23780 address inode structure of inode #16 Data block pointer represented by struct ext4_extent i_links_count inode #
  • 13.
    type = directory i_links_count= 5 Relationship between inode and dentry – ‘test-soft-link.c’ (inode #14) i-node table . 2 name inode Directory Entry (dentry) - / .. 2 -> 2646185 lost+found 11 a123 12 a789 17 2 12 . 12 name inode .. 2 b456 13 hello.c 15 hello-hard-link.c 15 Directory Entry (dentry) - /a123 13 . 13 name inode .. 12 test.c 16 test-hard-link.c 16 test-soft-link.c 14 Directory Entry (dentry) - /a123/b456 14 16 Data block pointers type = directory i_links_count = 3 Data block pointers type = directory i_links_count = 2 Data block pointers type = symbolic link i_links_count = 1 Data block pointers = “test.c” type = file i_links_count = 2 Data block pointers File Data /a123/b456/test.c size = 6 size = 100 inode structure data of inode #14 inode entry: 35 * 4096 + 128 * (14 – 1) [inode #14] = 0x23680 address inode structure of inode #14 i_links_count Data block pointer Hard Link 1. Refer to an inode number 2. Must reside on the same file system 3. Cannot be made to a directory Soft Link 1. Refer to a filename instead of an inode number 2. Can be linked to a file in a different file system 3. Can be made to a directory inode #
  • 14.
    mount – mounta new ext4 file system • stat() – get file status from inode o Name resolution → get dentry → get inode • openat() o Name resolution o Update open file table • mount() o Name resolution o Add a mount entry to ‘mount’ hash table o Parse file system
  • 15.
    stat("/adrian/img/src.img“, …) user_path_at(): filla path struct (dentry and mount info) after resolving a pathname vfsmount mnt_root mnt_sb dentry d_name path mnt dentry d_inode path struct • stat() – get file status from inode o Name resolution → get dentry → get inode
  • 16.
    stat("/adrian/img/src.img“, …) 1. namei:convert a “name” to an “inode” 2. nameidata: store the current status when walking a path nameidata last name root inode path task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” local variable
  • 17.
    stat("/adrian/img/src.img“, …) path_lookupat(): 1. Configuremembers of nameidata struct based on absolute pathname or relative pathname 2. Name resolution vfsmount mnt_root mnt_sb dentry d_name qstr name = “/” u32 hash u32 len u64 hash_len union nameidata last (S*) name root (S*) inode path (S*) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” local variable path mnt dentry current->fs->root d_inode * Data storage (not a data pointer) Note
  • 18.
    link_path_walk(): name resolution– round 1 vfsmount mnt_root mnt_sb dentry d_name qstr name = “/” u32 hash u32 len u64 hash_len union nameidata last (S*) name root (S*) inode path (S*) * Data storage (not a data pointer) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” path mnt dentry current->fs->root d_inode qstr name = “adrian/img/src.img” u32 hash u32 len = 6 u64 hash_len union variable ‘name’ = “adrian/img/src.img” dentry d_name d_inode path mnt dentry variable ‘name’ = “img/src.img” hash 1 2 3 4 5 6 Note local variable link_path_walk() • stop until the last dentry is resolved • for-loop in this function qstr name = “adrian” u32 hash u32 len = 6 u64 hash_len union
  • 19.
    link_path_walk(): name resolution– round 1 nameidata last (S*) name root (S*) inode path (S*) * Data storage (not a data pointer) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” path mnt dentry current->fs->root qstr name = “adrian/img/src.img” u32 hash u32 len = 6 u64 hash_len union variable ‘name’ = “adrian/img/src.img” dentry d_name d_inode path mnt dentry variable ‘name’ = “img/src.img” hash 1 2 3 4 5 6 Note local variable lookup_slow()
  • 20.
    vfsmount mnt_root mnt_sb dentry d_name qstr name = “/” u32hash u32 len u64 hash_len union nameidata last (S*) name root (S*) inode path (S*) * Data storage (not a data pointer) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” local variable path mnt dentry current->fs->root d_inode qstr name = “img/src.img” u32 hash u32 len = 3 u64 hash_len union variable ‘name’ = “img/src.img” dentry d_name d_inode path mnt dentry variable ‘name’ = “src.img” hash 1 2 3 4 5 6 Note link_path_walk(): name resolution – round 2 qstr name = “img” u32 hash u32 len = 3 u64 hash_len union link_path_walk() • stop until the last dentry is resolved • for-loop in this function
  • 21.
    vfsmount mnt_root mnt_sb dentry d_name qstr name = “/” u32hash u32 len u64 hash_len union nameidata last (S*) name root (S*) inode path (S*) * Data storage (not a data pointer) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” local variable path mnt dentry current->fs->root d_inode qstr name = “src.img” u32 hash u32 len = 3 u64 hash_len union variable ‘name’ = “src.img” dentry d_name d_inode path mnt dentry variable ‘name’ = “” 2 Note Function return link_path_walk(): name resolution – round 3 1 hash qstr name = “img” u32 hash u32 len = 3 u64 hash_len union link_path_walk() • stop until the last dentry is resolved • for-loop in this function
  • 22.
    vfsmount mnt_root mnt_sb dentry d_name qstr name = “/” u32hash u32 len u64 hash_len union nameidata last (S*) name root (S*) inode path (S*) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” local variable path mnt dentry current->fs->root d_inode qstr name = “src.img” u32 hash u32 len = 6 u64 hash_len union variable ‘name’ = “src.img” dentry d_name qstr name = “src.img” u32 hash u32 len = 6 u64 hash_len union d_inode path mnt dentry variable ‘name’ = “” 3 4 5 6 Function return lookup_last(): name resolution hash 1 2 * Data storage (not a data pointer) Note lookup_last() • get the corresponding dentry based on the filename • “while loop”: Follow the symlink in the final component
  • 23.
    vfsmount mnt_root mnt_sb dentry d_name qstr name = “/” u32hash u32 len u64 hash_len union nameidata last (S*) name root (S*) inode path (S*) task_struct mm nameidata filename name uptr (userland pointer) refcnt aname char iname[] = “/adrian/img/src.img" char __user *name = “/adrian/img/src.img” local variable path mnt dentry current->fs->root d_inode qstr name = “src.img” u32 hash u32 len = 6 u64 hash_len union variable ‘name’ = “src.img” dentry d_name qstr name = “src.img” u32 hash u32 len = 6 u64 hash_len union d_inode path mnt dentry variable ‘name’ = “” 3 4 5 6 Function return lookup_last(): name resolution hash 1 2 * Data storage (not a data pointer) Note
  • 24.
    mount – mounta new ext4 file system
  • 25.
    hlist_head.first hlist_head.first . . mount_hashtable mount mnt_parent struct dentry *mnt_mountpoint structvfsmount mnt **pprev *next const char *mnt_devname = “none” mount mnt_parent struct dentry *mnt_mountpoint **pprev struct hlist_node mnt_hash *next const char *mnt_devname = “none” struct hlist_node mnt_hash dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union super_block s_blocksize s_maxbytes s_op = ramfs_sops s_root s_bdev s_id[32] = “rootfs” s_fs_info s_type = rootfs_fs_type d_parent mnt_root mnt_sb struct vfsmount mnt mount – before mounting a new ext4 file system d_sb
  • 26.
    hlist_head.first hlist_head.first . . mount_hashtable mount mnt_parent struct dentry *mnt_mountpoint structvfsmount mnt **pprev *next const char *mnt_devname = “none” mount mnt_parent struct dentry *mnt_mountpoint **pprev struct hlist_node mnt_hash *next const char *mnt_devname = “none” struct hlist_node mnt_hash dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union super_block s_blocksize s_maxbytes s_op = ramfs_sops s_root s_bdev s_id[32] = “rootfs” s_fs_info s_type = rootfs_fs_type d_parent mnt_root mnt_sb struct vfsmount mnt mount – after mounting a new ext4 file system mount mnt_parent struct dentry *mnt_mountpoint **pprev *next const char *mnt_devname = “/dev/loop0” mnt_root mnt_sb dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union struct vfsmount mnt super_block s_op = ext4_sops s_id[32] = “loop0” s_type = ext4_fs_type d_parent struct hlist_node mnt_hash dentry d_name d_parent qstr name = “mnt” u32 hash u32 len = 3 u64 hash_len union /adrian/mnt d_sb d_sb
  • 27.
    hlist_head.first hlist_head.first . . mount_hashtable mount mnt_parent struct dentry *mnt_mountpoint structvfsmount mnt **pprev *next const char *mnt_devname = “none” mount mnt_parent struct dentry *mnt_mountpoint **pprev struct hlist_node mnt_hash *next const char *mnt_devname = “none” struct hlist_node mnt_hash dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union super_block s_blocksize s_maxbytes s_op = ramfs_sops s_root s_bdev s_id[32] = “rootfs” s_fs_info s_type = rootfs_fs_type d_parent mnt_root mnt_sb struct vfsmount mnt mount – after mounting a new ext4 file system mount mnt_parent struct dentry *mnt_mountpoint **pprev *next const char *mnt_devname = “/dev/loop0” mnt_root mnt_sb dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union struct vfsmount mnt super_block s_op = ext4_sops s_id[32] = “loop0” s_type = ext4_fs_type d_parent struct hlist_node mnt_hash dentry d_name d_parent qstr name = “mnt” u32 hash u32 len = 3 u64 hash_len union /adrian/mnt d_sb d_sb
  • 28.
  • 29.
    `ls –color mnt`:openat(AT_FDCWD, "/adrian/mnt“, …)
  • 30.
    `ls –color mnt`:openat(AT_FDCWD, "/adrian/mnt“, …)
  • 31.
    `ls –color mnt`:openat(AT_FDCWD, "/adrian/mnt“, …)
  • 32.
    path mnt dentry vfsmount mnt_root mnt_sb qstr name = “mnt” u32hash u32 len = 3 u64 hash_len union dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union super_block s_blocksize s_maxbytes s_op = ramfs_ops s_root s_bdev s_id[32] = “rootfs” s_fs_info d_inode dentry d_name d_inode d_parent s_type = rootfs_fs_type path mnt dentry vfsmount mnt_root mnt_sb qstr name = “mnt /” u32 hash u32 len = 3 1 u64 hash_len union dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union super_block s_blocksize s_maxbytes s_op = ext4_sops s_root s_bdev s_id[32] = “loop0” s_fs_info d_inode dentry d_name d_inode d_parent s_type = ext4_fs_type /adrian/mnt data structures in the original file system /adrian/mnt data structures in newly mounted file system traverse_mounts() `ls –color mnt`: openat(AT_FDCWD, "/adrian/mnt“, …) The original mount point is changed the newly one when running name resolution
  • 33.
    path mnt dentry vfsmount mnt_root mnt_sb qstr name = “mnt/” u32 hash u32 len = 3 1 u64 hash_len union dentry d_name qstr name = “/” u32 hash u32 len = 1 u64 hash_len union super_block s_blocksize s_maxbytes s_op = ext4_sops s_root s_bdev s_id[32] = “loop0” s_fs_info d_inode dentry d_name d_inode d_parent s_type = ext4_fs_type /adrian/mnt data structures in newly mounted file system flags: path->dentry->d_flags `ls –color mnt`: openat(AT_FDCWD, "/adrian/mnt“, …) REF-walk [Two concurrent mechanisms for name resolution] 1. REF-walk: concurrent management with refcounts and spinlock 2. RCU-walk: faster name resolution lookup