Introduction to Pseudo and
Stackable File Systems
under BSD
Allan Fields
Afields Research / AFRSL
http://afields.ca
[email protected]
2
Outline
● Introduction ● Templated Base Filesystems
● User-space and Kernel-space ● FiST and fistgen
● What is the VFS? ● Supported Platforms
● What are Vnodes? ● The FiST language
● Vnode Operations ● Pros & Cons of template
● Mount points & the mount filesystems
structure ● Developing a template fs in
● Root Vnode FiST
● Traditional File Systems
● Namespace ● User Filesystems
● "Next Generation" Filesystems
● Pseudo File Systems ● Extending Filesystem
● Portal File Systems Semantics
● Adapting Filesystem
● Stackable File Systems Namespaces
● What is Vnode Stacking?
● How does it work? ● Conclusions
● Applications of Vnode ● Q&As
Stacking
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
3
Introduction
● The BSD file system was first developed at
University of California at Berkeley
● Significant portions of Free/Net/OpenBSD
file system based on 4.4BSD code
● BSD Operating Systems default to UFS file
system with FFS filestore
● Recent changes to UFS in FreeBSD include:
● Snapshots: ability to save state of mounted file
system at point in time while continuing use
● UFS2 inode format - supports: Extended
Attributes; 64-bit fields
● Access Control Lists using EAs
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
4
Userspace and Kernelspace
● BSD Operating Systems use virtual memory
- hardware protection of address space
● User space: User process in the BSD
Operating Systems run in user-space or
user land
● Each process has own address space;
process structure, user+kernel stack, heap;
text and data pages
● Kernel space (kernel land): the kernel has
own address space
●
System Calls (syscalls) interface to kernel
from userland processes
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
5
What is the VFS?
● VFS is kernel layer, provides object oriented
interface to file systems
● The vnode interface provides unified set of
routines called by kernel
● Abstracts interfacing details from
underlying file systems
● Supports multiple file system types
● VFS provides a common set of routines for
working with file systems: mounting, sync,
quota, etc.
● Standard/default setof vfsops
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
6
What are Vnodes?
● File system in Unix allows uniform access to
multiple objects types
●
Vnode: generalized file system object -
struct vnode
● Vnodes can represent different types of
objects
● VREG: regular files
● VDIR: directories
● VBLK: block devices
● VCHR: character devices
● VLNK: link devices
● VSOCK: sockets
● VFIFO: named pipe
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
7
Vnode Operations
● Vnode Operations (vnops) are basic file
system primitives for operating on vnodes
● vnode operation vector: structure contains
function pointers to routines
● VOP_* Macros: provide object oriented
calling of vop on vnode
● Default vnode ops: Standard routines
defined by system; vnode operations not
defined by file system fall back on defaults
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
8
The Big Picture
● User processes performing I/O or lookups
call standard library functions such as:
● read(), write(), readdir()
● This initiates syscall into kernel
● Kernel then references vnode and calls
appropriate vnode operation
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
9
Namespace
● Namespaces composed of symbolic
identifiers, usually representative of objects
●
Namespaces have two primary elements:
● Names or identifiers:
● point to underlying objects
● usually unique at any given scope
● Spaces:
● flat / single-level namespaces
● hierarchical directories (trees)
● BSD filesy stem namespace is hierarchical,
path based; Element of namespace is path
component
● Lookup recursively resolves vnode
referenced by path
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
10
Pseudo File Systems
● Pseudo File Systems behave like normal file
systems with some primary differences
●
Allow creation of arbitrary file hierarchies
accessible by standard user binaries
● Provide uniform access to objects with
name and data components, convenient way
to represent hierarchical information
● Can expose non-filesystem objects through
VFS/vnode layers
● Names may not correspond 1:1 to files
● May not have backing storage at all
● Concerned with “top-half”: namespace, file
system semantics
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
11
Pseudo File Systems Include..
● procfs
● Provides a file system mount with processes
● Process status and control through standard file
entries
● devfs
● Provides dynamicaly generated tree w/ device
nodes
● Old approach was manually created device
entries in user-space
● Interface directly to kernel device drivers
● Now standard in FreeBSD; use in jails
● portalfs
● Establish TCP connections through FS
● Socket semantics in file system
● Pipes and naïve tools
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
12
Practical Benefits
● Dynamically generated hierarchies can
accurately represent changing structures,
entities w/o need to maintain corresponding
file entries
● Use existing tools, standar library file
functions: open(), read(), write(), etc..
● Exploits *nix file system closure: accesible
from shell
● FreeBSD PseudoFS
● Generic code for constructing pseudo file
system trees
IntroductionInterfaces VFS and under
provides API for building
●
to Pseudo and Stackable File Systems BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
13
Stackable File Systems
● Stackable file systems provide modular
object-oriented approach to building file
system
● Mounting stackable file systems on top of
each other allows build complex behaviour
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
14
Applications of Vnode Stacking
● I/O Request Transformation
● Replication
● Caching
● Fail-over
● Transformation of Lookup Requests
● Overlays/Union File Systems
● Security, Logging and Auditing
● User and attribute mapping
● Encryption, Compression
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
15
Template Based File Systems
● Template Based File Systems are file
systems which reuse templates to generate
file system code
● Designed for portability between Operating
Systems with differing vnode interfaces
● Templates take care of vnode interface
details.
●
Template file systems aren't just for file
system developers:
● System administrators have easy access to
cross-platform file systems
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
16
FiST and fistgen
● FiST or the File System Translator:
language for creating template based file
systems
●
Fistgen: Tool to generate file system code
based on FiST code and file system
templates
● Supports vnode stacking, fan-in and fan-out
● Enables:
● portability of code
● reduced development time and effort
● Minimal performance overhead
● Many have expressed interest using FiST
under BSD
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
17
Supported Platforms
● fistgen provides templates for several
platforms:
● Linux
● FreeBSD
● Solaris
● Linux support is most mature, latest
development efforts support Linux kernels
● Current state of FreeBSD support:
● FreeBSD support vastly improved in recent
releases
● FreeBSD templates still require work to support
wider range of releases
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
18
Work on fistgen FreeBSD port
● Problems w/ templates & different releases
● Keeping templates synchronized with
changes to VFS, vnode interfaces (requires
periodic maintenance)
● Linux-supported features needing poring:
● SCAs: Size Change Algorithm support
● Attach-mode mounts
●
Fistgen has potential for true cross-platform
file system on major free nixes
● Porting to additional platforms requires
more templates to be built: Net/OpenBSD
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
19
FiST Dev. Flowchart
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
20
The FiST language
● FiST is intermediary language: contains
declarations, rules and C code-blocks
● fistgen fills-out template, produces C code,
sets up Makefile
● Supports pre-call, call & Post-call operation
● Hooks into vnode ops: before, during &
after call to vnop, call any further vnops
●
insert arbitrary blocks of code (doesn't have
to be vnode specific)
● Define filter functions to {en,de}code data
● Read-only (readops), writeonly (writeops)
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
21
Pros of Template File Systems
● Portability: cross-platform support
● Ease of Use: potentially shallower developer
learning curve
● Frees developer to focus on file system
specific code, semantics
● Maintainability - easier to maintain:
● 1 set of templates than 7-8 different file systems
● 4 sets of templates than 4 ports of same file
system
● Extensible approach to constructing file
systems
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
22
Cons of Template File Systems
● Templates can break easily with changes to
vnode interface (also true w/ non-template fs)
●
Some templates aren't complete: FreeBSD
templates (more developers needed)
● Slight overhead incurred (minimal impact)
● Coding Trade-off: easier to develop with,
but loose some control over code (can be
"tweaked" afterward, address some
problems manually)
● Some loss of versatility: Not all things are
possible in FiST (support could be added)
● Exporting FS over NFS can be problematic
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
23
Developing a Template fs in FiST
● Process for creating a file system with
fistgen:
● Fetch fistgen tarball
fetch ftp://ftp.filesystems.org/fist/fistgen-0.0.n.tar.gz
tar xzf fistgen-0.0.n.tar.gz
● Build fistgen cd fistgen-0.0.n; ./configure&&make
● Build test filesystems make genall # Test target only works on Linux
● Create a .fist file cp testfs/wrapfs.fist testfs.fist
● (Add code here) $EDITOR testfs.fist
● Generate code fistgen testfs.fist
● Run make cd out/testfs; make
● Test kernel module kldload ./testfs.ko
● Attempt mount mkdir /mnt/test
● Sanity Checks ./mount_testfs /tmp /mnt/test
mkdir /mnt/test/1; rmdir /mnt/test/1
touch /mnt/test/2; rm /mnt/test/2
dd if=/dev/random of=/mnt/test/3; dd if=/mnt/test/3 of=/dev/null
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
24
User File Systems
● User file systems such a cfsd: user-process
● NFS interface, kernel-stub
● Problems w/ user file systems under BSD:
● Slow: context-switch necessary to service
I/O request
● Scalability limited: single-process user-file
systems create bottlenecks
● Hurd: microkernel based, provides API for
filesystem sevices in userspace
● DragonFlyBSD: proposing various changes:
message passing, VFS; could make user file
systems practical
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
25
"Next Generation" File Systems
● Latest file systems to hit the scene provide
journaling or balanced-tree implementations
●
Differing on-disk structure (filestore layout)
● Include enhanced semantics
● Mostly monolithic development model:
● Separate/”full” file systems (unit-wise
replacement for existing FS)
● Not the Stacking Model
● Implement both namespace and filestore
components
● BSD focus remains on extending UFS and
supporting additional file systems
● Linux has three major next-gen file systems
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
26
Nextgen includes..
● XFS (SGI) [Popular journaling File System]
● Some BSD ports available or in progress
● Ext3fs
● Next generation of Linux Extended File System
● ReiserFS (Namesys) [B+Tree Based]
● currently Linux-only development; major distros
● significant departure from traditional file
systems: reiser4 creates new syscalls,
significant changes in semantics
● supports plug-ins
● BeFS (Be, Inc.)
● BeOS file system, was clean reimplementation
of traditional inode-based file system
● Provided native [extended] attributes and
indexing (“database” like features)
● BeOS/BeFS being reimplemented: OpenBeOS
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
27
Adapting File System Namepsaces
● pseudo and stackable file systems provide
ability to adapt namespaces to specific
purposes
● Examples:
● procfs (pseudo) - creates a namespace modeled
on running processes
● unionfs (stackable) - combines namespaces of
two file systems; overlays
● VFS vs. other kernel APIs
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
28
Conclusions
● Many possibilities exist for pseudo and
stackable file systems
●
Pseudo file systems provide
● Wide-ranging applications for advanced file
systems
● Going beyond a simple 1:1 filename:object
mapping
● Accompanying paper: Introduction to
Pseudo and Stackable File systems under
BSD
●
See:
● More info: http://afields.ca/bsdcan/2004
● FiST homepage: http://www.filesystems.org
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
29
Q&As
● Questions? Comments?
● Speaking of name spaces: "file system" vs.
"filesystem"?
● Liberals: filesystem is a new word!
● Conservatives: file system (two words)
● Traditionalists: Don't you mean: microcomputer
disk filing system?
● Pragmatists: Randomly placed use of both
forms, doesn't matter
● Foldoc: filesystem syn. file system
Introduction to Pseudo and Stackable File Systems under BSD fs_pseudo_stack.sxi,v 1.1 2004/05/14 afields
