© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File System Modules
2© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
What to Expect?
W's of a File System
Three levels of File Systems
Design a File System
FS relation with a Hard Disk Partition
FS relation with /
Writing a FS module
3© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Why we need a File System?
Preserve Data for Later
Access
Update
Discard
Tag Data for
Categorizing
Easy & Fast Lookup
Fast & Reliable Data Management by
Optimized Operations
4© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
What is a File System?
Three things at three levels
Hardware Space – The Physical Organization of Data
on the Storage Devices
Kernel Space – Drivers to decode & access the data
from the Physical Organization
User Space – All what you see from / - The User View
Which ever it be, it basically does
Organize our data
For easy access by tagging
For fast maintenance by optimizing
5© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
The Three File Systems
Hardware (Space) File System - Storer
Kernel (Space) File System - Decoder
User (Space) File System - Show-er
Storer: Partition for the File System (/dev/*)
Creating the File System (mkfs.*)
Decoder: Driver for the File System (*.ko)
Show-er: Root File System (/xyz)
Connector of the 3: mount
Let's try an example
6© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Recall
7© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Understanding a Hard Disk
Example (Hard Disk)
Heads (or Platters): 0 – 9
Tracks (or Cylinders): 0 – 24
Sectors: 1 – 64
Size of the Hard Disk
10 x 25 x 64 x 512 bytes = 8000KiB
Device independent numbering
(h, t, s) → 64 * (10 * t + h) + s → (1 - 16000)
8© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Partitioning a Hard Disk
First Sector – Master Boot Record (MBR)
Contains Boot Info
Contains Physical Partition Table
Maximum Physical Partitions: 4
At max 1 as Extended Partition
Rest as Primary Partition
Extended could be further partitioned into
Logical Partitions
In each partition
First Sector – Boot Record (BR)
Remaining for File System / Format
Extended Partition BR contains the Logical Partition Table
9© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Placement of a Hardware FS
Each Partition contains a File System
Raw (no FS)
Organized
Organized one is initialized by the corresponding Formating
“Partition Type is to OS”
W95*, DOS*, BSD*, Solaris*, Linux, ...
“Format Type is to File System”
ext2, ext3, vfat, ntfs, jffs2, …
Particular Partition Types support only Particular File
System Types
10© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Design a FS
Let's Take 1 Partition
With 2N blocks of 1 sector each
Size = 2N x 512 bytes = N KiB
Given 3 data pieces
Source Code – 384 bytes
Image – 768 bytes
Document – 500 bytes
Place optimally for further operations
11© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File System Design
C
384b
Img
768b
Doc
500b
12© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File System Design
C
384b Img ...
512b
Doc
500b
...Img
256b
Img
768b
Doc
500b
13© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File System Design
C
384b Img ...
512b
Doc
500b
...Img
256b
C 384 1 1
Img 768 2 3
Doc 500 4 4
XXX X X X
XXX X X X
...
14© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File System Design
C
384b Img ...
512b
Doc
500b
...Img
256b
C 384 1 1
Img 768 2 3
Doc 500 4 4
0 0 0 0
0 0 0 0
...
15© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File System Design
C ...
512b
Img ...
512b
Doc
500b
...Img
256b
C 384 1 1
Img 768 2 3
Doc 500 4 4
0 0 0 0
0 0 0 0
... … C
488b
16© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File System Design
C ...
512b
Img ...
512b
Doc
500b
...Img
256b
C 1000 1 5 0 0 0
Img 768 2 3 0 0 0
Doc 500 4 0 0 0 0
0 0 0 0 0 0 0
...
… C
488b
17© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File System Design
C ...
512b
Img ...
512b
Doc
500b
...Img
256b
C 1000 4 8 0 0 0
Img 768 5 6 0 0 0
Doc 500 7 0 0 0 0
0 0 0 0 0 0 0
...
… C
488b
18© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File System Design
C ...
512b
Img ...
512b
Doc
500b
...Img
256b … C
488b4
FST
...
C 1000 5 9 0 0 0
Img 768 6 7 0 0 0
Doc 500 8 0 0 0 0
0 0 0 0 0 0 0
...
19© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File System Design
C ...
512b
Img ...
512b
Doc
500b
...Img
256b … C
488b4
FST
...
C 1000 5 9 0 0 0TS rw-
Img 768 6 7 0 0 0TS rw-
Doc 500 8 0 0 0 0TS rw-
0 0 0 0 0 0 00 ---
...
20© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Kernel Space File System
Also called the File System modules
Examples: ext2, ext3, vfat, ntfs, jffs2, …, real_sfs, ...
Role: Decode the Hardware FS Layout to Access Data
from the User Space (by implementing the decode logic)
Through the following 2 Interactions:
Interface with the Hardware FS Layout in the hardware partition
Interface with the User Space FS to give a unified view, as at /
First part done through Block Driver
Second part done through VFS
Kernel provides VFS to achieve the virtual user view
21© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
VFS Specifics
Inherited from the Linux specific FS
Promoted to a super set of various FS
Providing a unified view to the User
Default values for various attributes
owner, group, permissions, …
File Types
Same seven types as in ext2/ext3
22© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Virtual File System Interactions
complete
vfat
VFS
ext2
Partition 1
....
Partition 2 ... Partition N ...
User Space File System View @ /
....
Block Driver Block Driver.... ....
real_sfs
Block Driver
23© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Virtual File System Internals
Super Block
Directory Entry Table
Inode Table
d
> File Name
> Inode Number
> File Meta Data
> File Data Block Nos
> FS Meta Data
> Root Inode Number
d
Data Block
r
24© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
VFS Interaction Internals
vfat
VFS
ext2
Buffer
Cache
Block
Drivers
Directory
Cache
Inode
Cache
....
Block
Drivers
25© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
VFS Translation Callbacks
Five categories
File System level
Super Block Operations
Super Block level (Meta Meta Data)
Inode Operations
Inode level (Meta Data)
Directory Entry Operations
File level (Data)
File Operations through the Buffer Cache
Buffer Cache level
Actual Data Operations
26© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File System Registration
Header: <linux/fs.h>
APIs
int register_filesystem(file_system_type *)
int unregister_filesystem(file_system_type *)
File System Operations
get_sb/mount: (Invoked by mount)
get_sb_bdev/mount_bdev -> Fill the Super Block
kill_sb: (Invoked by umount)
kill_block_super
27© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Fill the Super Block
Block Size & Bits
Magic Number / File System Type
Super Block Operations
Root Inode
Inode Operations
Inode Mode
File Operations
28© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Super Block Operations
Header: <linux/fs.h>
APIs
write_inode: Update File Meta Data
statfs: Get the File Status (Invoked by stat)
simple_statfs (Handler from libfs)
29© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Inode Operations
Header: <linux/fs.h>
APIs
lookup
Search the file name by absolute path
Traverse the inode chain
if found matching inode, then populate the dentry with the inode
information
Inode Information
Size
Mode
File Operations
...
30© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File Operations
Header: <linux/fs.h>
APIs
open → generic_file_open
read → do_sync_read → generic_file_aio_read →
do_generic_file_read → readpage
write → do_sync_write → generic_file_aio_write →
generic_file_buffered_write → write_begin & write_end
release
readdir: Change Directory (cd). List Directory (ls)
Directory Entry
Fill Function
fsync: Sync the File (sync)
simple_sync_file
31© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Address Space Operations
The Buffer / Page Cache level Operations
Header: <linux/fs.h>
Page Operations
readpage (Invoked by read)
write_begin, write_end (Invoked by write)
writepage (Invoked by pdflush kernel threads)
Page Functions
PageUptodate, PageDirty, PageWriteback, PageLocked
SetPageUptodate, ClearPageDirty, unlock_page, ...
page_address, ...
32© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
ext2/3 Data Block Organization
size ... db10 ib dib tib... db0uidmode
DB0 ... DB10 ...
DB11
... ...
... ...
...
DB...
DB...
DB...
33© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Dive into the Real Driver
© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File System Pieces – Set 1
Hardware Space
File System Design (Logic) – real_sfs_ds.h
File System Creator (Application) – format_real_sfs.c /
mkfs.sfs.c
Kernel Space
File System Decoder (Driver) – real_sfs_ops.[hc]
File System Translator (Driver) – real_sfs.c
File System Show-er (VFS)
User Space
File System Connector (mount)
© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
File System Pieces – Set 2
Hardware Space
File System Design (Logic) – ddk_fs_ds.h
File System Creator (Application) – mkfs.ddkfs.c
Kernel Space
File System Decoder (Driver) – ddk_fs_ops.[hc]
File System Translator (Driver) – ddk_fs.c
File System Show-er (VFS)
User Space
File System Connector (mount)
36© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
What all have we learnt?
W's of a File System
Three levels of File Systems
And relation between them
Design a File System
FS relation with a Hard Disk Partition
FS relation with /
Role of VFS
Writing a FS module
Simplified Interaction with the Block Driver
5 Operation Sets with VFS
37© 2010-19 SysPlay Workshops <workshop@sysplay.in>
All Rights Reserved.
Any Queries?

File System Modules

  • 1.
    © 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File System Modules
  • 2.
    2© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. What to Expect? W's of a File System Three levels of File Systems Design a File System FS relation with a Hard Disk Partition FS relation with / Writing a FS module
  • 3.
    3© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Why we need a File System? Preserve Data for Later Access Update Discard Tag Data for Categorizing Easy & Fast Lookup Fast & Reliable Data Management by Optimized Operations
  • 4.
    4© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. What is a File System? Three things at three levels Hardware Space – The Physical Organization of Data on the Storage Devices Kernel Space – Drivers to decode & access the data from the Physical Organization User Space – All what you see from / - The User View Which ever it be, it basically does Organize our data For easy access by tagging For fast maintenance by optimizing
  • 5.
    5© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. The Three File Systems Hardware (Space) File System - Storer Kernel (Space) File System - Decoder User (Space) File System - Show-er Storer: Partition for the File System (/dev/*) Creating the File System (mkfs.*) Decoder: Driver for the File System (*.ko) Show-er: Root File System (/xyz) Connector of the 3: mount Let's try an example
  • 6.
    6© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Recall
  • 7.
    7© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Understanding a Hard Disk Example (Hard Disk) Heads (or Platters): 0 – 9 Tracks (or Cylinders): 0 – 24 Sectors: 1 – 64 Size of the Hard Disk 10 x 25 x 64 x 512 bytes = 8000KiB Device independent numbering (h, t, s) → 64 * (10 * t + h) + s → (1 - 16000)
  • 8.
    8© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Partitioning a Hard Disk First Sector – Master Boot Record (MBR) Contains Boot Info Contains Physical Partition Table Maximum Physical Partitions: 4 At max 1 as Extended Partition Rest as Primary Partition Extended could be further partitioned into Logical Partitions In each partition First Sector – Boot Record (BR) Remaining for File System / Format Extended Partition BR contains the Logical Partition Table
  • 9.
    9© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Placement of a Hardware FS Each Partition contains a File System Raw (no FS) Organized Organized one is initialized by the corresponding Formating “Partition Type is to OS” W95*, DOS*, BSD*, Solaris*, Linux, ... “Format Type is to File System” ext2, ext3, vfat, ntfs, jffs2, … Particular Partition Types support only Particular File System Types
  • 10.
    10© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Design a FS Let's Take 1 Partition With 2N blocks of 1 sector each Size = 2N x 512 bytes = N KiB Given 3 data pieces Source Code – 384 bytes Image – 768 bytes Document – 500 bytes Place optimally for further operations
  • 11.
    11© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File System Design C 384b Img 768b Doc 500b
  • 12.
    12© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File System Design C 384b Img ... 512b Doc 500b ...Img 256b Img 768b Doc 500b
  • 13.
    13© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File System Design C 384b Img ... 512b Doc 500b ...Img 256b C 384 1 1 Img 768 2 3 Doc 500 4 4 XXX X X X XXX X X X ...
  • 14.
    14© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File System Design C 384b Img ... 512b Doc 500b ...Img 256b C 384 1 1 Img 768 2 3 Doc 500 4 4 0 0 0 0 0 0 0 0 ...
  • 15.
    15© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File System Design C ... 512b Img ... 512b Doc 500b ...Img 256b C 384 1 1 Img 768 2 3 Doc 500 4 4 0 0 0 0 0 0 0 0 ... … C 488b
  • 16.
    16© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File System Design C ... 512b Img ... 512b Doc 500b ...Img 256b C 1000 1 5 0 0 0 Img 768 2 3 0 0 0 Doc 500 4 0 0 0 0 0 0 0 0 0 0 0 ... … C 488b
  • 17.
    17© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File System Design C ... 512b Img ... 512b Doc 500b ...Img 256b C 1000 4 8 0 0 0 Img 768 5 6 0 0 0 Doc 500 7 0 0 0 0 0 0 0 0 0 0 0 ... … C 488b
  • 18.
    18© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File System Design C ... 512b Img ... 512b Doc 500b ...Img 256b … C 488b4 FST ... C 1000 5 9 0 0 0 Img 768 6 7 0 0 0 Doc 500 8 0 0 0 0 0 0 0 0 0 0 0 ...
  • 19.
    19© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File System Design C ... 512b Img ... 512b Doc 500b ...Img 256b … C 488b4 FST ... C 1000 5 9 0 0 0TS rw- Img 768 6 7 0 0 0TS rw- Doc 500 8 0 0 0 0TS rw- 0 0 0 0 0 0 00 --- ...
  • 20.
    20© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Kernel Space File System Also called the File System modules Examples: ext2, ext3, vfat, ntfs, jffs2, …, real_sfs, ... Role: Decode the Hardware FS Layout to Access Data from the User Space (by implementing the decode logic) Through the following 2 Interactions: Interface with the Hardware FS Layout in the hardware partition Interface with the User Space FS to give a unified view, as at / First part done through Block Driver Second part done through VFS Kernel provides VFS to achieve the virtual user view
  • 21.
    21© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. VFS Specifics Inherited from the Linux specific FS Promoted to a super set of various FS Providing a unified view to the User Default values for various attributes owner, group, permissions, … File Types Same seven types as in ext2/ext3
  • 22.
    22© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Virtual File System Interactions complete vfat VFS ext2 Partition 1 .... Partition 2 ... Partition N ... User Space File System View @ / .... Block Driver Block Driver.... .... real_sfs Block Driver
  • 23.
    23© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Virtual File System Internals Super Block Directory Entry Table Inode Table d > File Name > Inode Number > File Meta Data > File Data Block Nos > FS Meta Data > Root Inode Number d Data Block r
  • 24.
    24© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. VFS Interaction Internals vfat VFS ext2 Buffer Cache Block Drivers Directory Cache Inode Cache .... Block Drivers
  • 25.
    25© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. VFS Translation Callbacks Five categories File System level Super Block Operations Super Block level (Meta Meta Data) Inode Operations Inode level (Meta Data) Directory Entry Operations File level (Data) File Operations through the Buffer Cache Buffer Cache level Actual Data Operations
  • 26.
    26© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File System Registration Header: <linux/fs.h> APIs int register_filesystem(file_system_type *) int unregister_filesystem(file_system_type *) File System Operations get_sb/mount: (Invoked by mount) get_sb_bdev/mount_bdev -> Fill the Super Block kill_sb: (Invoked by umount) kill_block_super
  • 27.
    27© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Fill the Super Block Block Size & Bits Magic Number / File System Type Super Block Operations Root Inode Inode Operations Inode Mode File Operations
  • 28.
    28© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Super Block Operations Header: <linux/fs.h> APIs write_inode: Update File Meta Data statfs: Get the File Status (Invoked by stat) simple_statfs (Handler from libfs)
  • 29.
    29© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Inode Operations Header: <linux/fs.h> APIs lookup Search the file name by absolute path Traverse the inode chain if found matching inode, then populate the dentry with the inode information Inode Information Size Mode File Operations ...
  • 30.
    30© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File Operations Header: <linux/fs.h> APIs open → generic_file_open read → do_sync_read → generic_file_aio_read → do_generic_file_read → readpage write → do_sync_write → generic_file_aio_write → generic_file_buffered_write → write_begin & write_end release readdir: Change Directory (cd). List Directory (ls) Directory Entry Fill Function fsync: Sync the File (sync) simple_sync_file
  • 31.
    31© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Address Space Operations The Buffer / Page Cache level Operations Header: <linux/fs.h> Page Operations readpage (Invoked by read) write_begin, write_end (Invoked by write) writepage (Invoked by pdflush kernel threads) Page Functions PageUptodate, PageDirty, PageWriteback, PageLocked SetPageUptodate, ClearPageDirty, unlock_page, ... page_address, ...
  • 32.
    32© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. ext2/3 Data Block Organization size ... db10 ib dib tib... db0uidmode DB0 ... DB10 ... DB11 ... ... ... ... ... DB... DB... DB...
  • 33.
    33© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Dive into the Real Driver
  • 34.
    © 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File System Pieces – Set 1 Hardware Space File System Design (Logic) – real_sfs_ds.h File System Creator (Application) – format_real_sfs.c / mkfs.sfs.c Kernel Space File System Decoder (Driver) – real_sfs_ops.[hc] File System Translator (Driver) – real_sfs.c File System Show-er (VFS) User Space File System Connector (mount)
  • 35.
    © 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. File System Pieces – Set 2 Hardware Space File System Design (Logic) – ddk_fs_ds.h File System Creator (Application) – mkfs.ddkfs.c Kernel Space File System Decoder (Driver) – ddk_fs_ops.[hc] File System Translator (Driver) – ddk_fs.c File System Show-er (VFS) User Space File System Connector (mount)
  • 36.
    36© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. What all have we learnt? W's of a File System Three levels of File Systems And relation between them Design a File System FS relation with a Hard Disk Partition FS relation with / Role of VFS Writing a FS module Simplified Interaction with the Block Driver 5 Operation Sets with VFS
  • 37.
    37© 2010-19 SysPlayWorkshops <workshop@sysplay.in> All Rights Reserved. Any Queries?