The Road to a Functional Design. The Team  Peter Scott  Principal Consultant and founding partner at Kernel Drivers, LLC  Microsoft MVP  Jeffrey Altman.

Slides:



Advertisements
Similar presentations
Chapter 12: File System Implementation
Advertisements

More on File Management
Chapter 4 : File Systems What is a file system?
Allocation Methods - Contiguous
OpenAFS for Windows Status Report AFS & Kerberos Best Practice Workshop 2010.
File Management Chapter 12. File Management File management system is considered part of the operating system Input to applications is by means of a file.
OpenAFS for Windows Deep Dive: Reparse Points, Path Processing, and Implications for Namespace Design Jeffrey Altman Your File System Inc European.
File Management Systems
File System Implementation
Other File Systems: AFS, Napster. 2 Recap NFS: –Server exposes one or more directories Client accesses them by mounting the directories –Stateless server.
6/24/2015B.RamamurthyPage 1 File System B. Ramamurthy.
OPERATING SYSTEMS Introduction
File System Variations and Software Caching May 19, 2000 Instructor: Gary Kimura.
1 I/O Management in Representative Operating Systems.
NFS. The Sun Network File System (NFS) An implementation and a specification of a software system for accessing remote files across LANs. The implementation.
File Management Chapter 12.
7/15/2015B.RamamurthyPage 1 File System B. Ramamurthy.
Operating Systems Concepts 1. A Computer Model An operating system has to deal with the fact that a computer is made up of a CPU, random access memory.
MCTS Guide to Microsoft Windows Server 2008 Network Infrastructure Configuration Chapter 7 Configuring File Services in Windows Server 2008.
File System. NET+OS 6 File System Architecture Design Goals File System Layer Design Storage Services Layer Design RAM Services Layer Design Flash Services.
CSE 486/586, Spring 2012 CSE 486/586 Distributed Systems Distributed File Systems Steve Ko Computer Sciences and Engineering University at Buffalo.
 AFS name space managed as mini-file systems (aka Volumes)  AFS Mount Points are the equivalent of NTFS Junctions and DFS Referrals which are stored.
Windows RDMA File Storage
File Management Chapter 12. File Management File management system is considered part of the operating system Input to applications is by means of a file.
Networked File System CS Introduction to Operating Systems.
CS 6560 Operating System Design Lecture 13 Finish File Systems Block I/O Layer.
Chapter Oracle Server An Oracle Server consists of an Oracle database (stored data, control and log files.) The Server will support SQL to define.
Operating Systems (CS 340 D) Dr. Abeer Mahmoud Princess Nora University Faculty of Computer & Information Systems Computer science Department.
Distributed Systems. Interprocess Communication (IPC) Processes are either independent or cooperating – Threads provide a gray area – Cooperating processes.
Announcing U.S. Dept of Energy SBIR Grant Supporting Development of Next Generation OpenAFS Jeffrey Altman, President Your File System Inc. 13 September.
COMP 3438 – Part I - Lecture 4 Introduction to Device Drivers Dr. Zili Shao Department of Computing The Hong Kong Polytechnic Univ.
OpenAFS on Windows: A Status Report Jeffrey Altman The OpenAFS Project 16 October 2012.
1 Lecture 20: I/O n I/O hardware n I/O structure n communication with controllers n device interrupts n device drivers n streams.
A Comparative Study of the Linux and Windows Device Driver Architectures with a focus on IEEE1394 (high speed serial bus) drivers Melekam Tsegaye
The Native AFS Client on Windows The Road to a Functional Design Jeffrey Altman, President Your File System Inc. 14 September 2010.
OpenAFS on Windows: A Status Report Jeffrey Altman The OpenAFS Project 26 March 2014.
File Management Chapter 12. File Management File management system is considered part of the operating system Input to applications is by means of a file.
Unit OS8: File System 8.6. Lab Manual. 2 Copyright Notice © David A. Solomon and Mark Russinovich These materials are part of the Windows Operating.
Module 4.0: File Systems File is a contiguous logical address space.
Chapter 11: File System Implementation Silberschatz, Galvin and Gagne ©2005 Operating System Concepts Chapter 11: File System Implementation Chapter.
Silberschatz, Galvin and Gagne ©2009 Operating System Concepts – 8 th Edition File System Implementation.
GLOBAL EDGE SOFTWERE LTD1 R EMOTE F ILE S HARING - Ardhanareesh Aradhyamath.
System Components ● There are three main protected modules of the System  The Hardware Abstraction Layer ● A virtual machine to configure all devices.
File Systems cs550 Operating Systems David Monismith.
© 2006 EMC Corporation. All rights reserved. The Host Environment Module 2.1.
4P13 Week 9 Talking Points
2010 Storage Developer Conference. Kernel Drivers, LLC. All Rights Reserved. Building Windows File Systems: A Look at the OpenAFS Client for Windows Peter.
Introduction to Operating Systems Concepts
Jonathan Walpole Computer Science Portland State University
Chapter 11: File System Implementation
Chapter 11: Implementing File Systems
Chapter 12: File System Implementation
File System Implementation
Unit OS8: File System 8.6. Quiz
CSE451 NTFS Variations and other File System Issues Autumn 2002
Operating Systems (CS 340 D)
Filesystems.
Chapter 3: Windows7 Part 4.
Chapter 3: Windows7 Part 2.
Chapter 2: System Structures
File System B. Ramamurthy B.Ramamurthy 11/27/2018.
Chapter 3: Windows7 Part 2.
Distributed File Systems
Distributed File Systems
Distributed File Systems
Chapter 16 File Management
Chapter 15: File System Internals
Distributed File Systems
Distributed File Systems
Presentation transcript:

The Road to a Functional Design

The Team  Peter Scott  Principal Consultant and founding partner at Kernel Drivers, LLC  Microsoft MVP  Jeffrey Altman  OpenAFS Gatekeeper and Elder  President of YFS, Inc.

SMB …  The old AFS model leveraged the Microsoft Loopback Adapter for passing requests to the AFS Service in user mode  Easier to implement but reliant on system components such as SMB and MSLoopBack  Hard to get bugs fixed in these modules  Not very performance focused  Generic solution to fit all situations  Typical Microsoft interface … minimal documentation

Goals of the Design  Need to leverage as much functionality within the AFS Service as possible  Keep all server communication in service  Data retrieval  Callback registration and notification  Metadata management  Complete integration into the Microsoft IFS (Installable File System) API  Stability and performance

Windows File System Model  Windows IFS Interface  IRP Based model  ‘Fast IO’ Interface used for more than just IO  Network Provider Interface for Network Redirectors only  A network file system is not much different from a local file system, in Windows  MUP (Multiple UNC Provider) Registration  Pre-Vista uses different model  IOCTL_REDIR_QUERY_PATH(_EX)  \\afs\Dementia.org\User\Foo.txt \\afs\Dementia.org\User\Foo.txt  Path Parsing  \Device\MUP\;AFS\Redirector\;C:\\AFS\Dementia.org\User\Foo.txt  \;C:\\AFSDementia.org\User\Foo.txt  \device\MUP\AFS\Dementia.org\User\Foo.txt  \AFS\Dementia.org\User\Foo.txt  Network Provider Library  User mode interface for Wnet API

Windows Internals IO Manager NTFS Filter FDO C: MUP LanMan Filter AFS Redir WNetxxx Network Provider AFS Service App Foo CM and MM MUP LanMan Filter AFS Redir Filter Old MUP Stack AFS Cache AFS Mount Mgr

Windows Internals  Windows Vista Changes  Memory Manger and Cache Manager changes  Theoretical limit of 4GB paging IO requests but have not seen anything larger than 256MB  Pre-Vista had a maximum of 64KB  ‘Dummy’ pages in Memory Manager – does not effect redirector  MUP Changes  Tons of new ‘features’ – Bitlocker, built in AV, Indexer, Single Instance Storage, etc.

MUP Registration  MUP – Handles mappings between the UNC name space and the file systems which manage them  MUP changes in Windows Vista  Old model  Register with MUP using a named device object  Prefix resolution and IRP_MJ_CREATE requests handled by MUP, all others sent to file system  New Model  Register with MUP using an unnamed device object and a name of the file system control device

Old MUP Design  Registration with MUP used a named device object  Prefix resolution by MUP used the IOCTL_REDIR_QUERY_PATH request  Cache entries for 15 minutes unless flushed  IO Manager would send all requests, post IRP_MJ_CREATE, directly to file system  Network redirectors would register, separately, as a file system resulting in filter attachment issues

Old MUP Design  Only prefix resolution and IRP_MJ_CREATE requests handled through MUP  All subsequent requests issued to redirector IO Manager WNetxxx Network Provider AFS Service App Foo CM and MM MUP (DFS) LanMan Filter AFS Redir Filter AFS AFS Cache

New MUP Design  Register with MUP using a device name and an unnamed device object  Results in MUP creating a symbolic link from the device name to \Device\MUP  Prefix resolution using IOCTL_REDIR_QUERY_PATH_EX  All requests go through MUP  Single attachment point for filters

New MUP Design  All requests go through MUP  Single point access – Better? IO Manager MUP LanMan Filter AFS Redir WNetxxx Network Provider AFS Service App Foo CM and MM AFS AFS Cache

Path Parsing in Windows  2 forms can be sent – drive letter or not …  Drive letter names come into MUP as \Device\MUP\;AFS\Redirector\;C:\\AFS\Dementia.org\User Which are mapped by MUP into \;C:\\AFSDementia.org\User  UNC names come into MUP as \device\MUP\AFS\Dementia.org\User Which are mapped by MUP into \AFS\Dementia.org\User\Foo.txt

Network Provider Interface  User mode library with supporting interface in file system  Used to support WNet API in user mode  Implements drive letter mapping  Communicates with file system for state and connection information  Maintains per user information on mappings

AFS Service Communication  Inverted call model  Requests from file system  Uses proprietary IOCtl interface  Communication through CDO (Control Device Object) symlink  IOCtl interface  Requests to file system  Proprietary IOCtl interface for service initiated requests  Cancellable interface through CDO handle

AFS Service Communication  All requests issued through CDO symbolic link - \??\AFSRedirector  Request pool state controlled through open handle AFS Service AFS Redir Request Pool IRP_MJ_DEVICE_CONTROL Dispatch Handler

Merging Worlds  Name space convergence  Symbolic Links – Microsoft and AFS  Mount Points  DFS Links  Component substitutions  File data handling  PIOCtl Interface  “Special” share name handling  PIPE\srvsvc  PIPE\wkssvc  Network Provider Interface

Name Space Convergence  Cells and Shares  Share access mapped into cell names  \\AFS\Dementia.org \\AFS\Dementia.org  Dynamic discovery  Reparse points and symbolic links  Must handle all symbolic links internally, they are not understood by Windows  Support the generic reparse point interface through FSCTL_xxx_REPARSE_POINT controls – currently do not support writing of this data  Mount point processing managed internally  DFS Links are supported through reparse processing  Windows concept of reparse processing

Metadata Handling  Redirector caching model  Cache objects based on FID on a per volume basis  Cache directory entries based on hash of name on a per directory basis  Support case insensitive, sensitive and short name lookups  Asynchronous pruning of trees when not in use  Path name parsing in Windows  Path analyzed component by component, walking a specific branch for achieve the target object  Maintains a list of components used to access current target  Need to support relative symbolic links within a pathname

Name Space Implementation AFS Redirector  FID based access is ‘almost’ lockless – Only volume based lock required  Name based access is complex due to symlink, mount point, DFS link and other abstractions not recognized by Windows Name based access layer FID based access layer Windows IFS and NP InterfaceAFS Service

Name Space Implementation Volume Btree (Cell, Volume) Object Btree (Vnode, Unique) Name BTree (Component CRC) AFS Global Root Volume Volume 1Volume 2 Volume 3 Volume 2 Root Object Object 1 Object 2 Object 2 Root Directory 1 File 1

Name Space Implementation  Handle AFS Symbolic Links, Mount Points, etc.  DirEntry nodes are tracked per directory, contain name based information  Object nodes are tracked by FID per volume  FCB nodes are used within the Windows IFS interface, tracked under the FileObject->FsContext pointer, one per Object node  CCB nodes are used within the Windows IFS interface, tracked under the FileObject>FsContext2 pointer, one per open instance of a file DirEntry1DirEntry2DirEntry3 ObjectFCB Ccb1Ccb2Ccb3Ccb4 FileObject

File Data Handling  Windows caching model  Re-entrant model – Need to be careful of locking hierarchy  Side band locking interface for memory and cache manager components – Fast IO interface  Need to observe IRQ levels while processing requests to underlying AFS Cache  File Extent Interface  Extents describe the location of file data within the AFS Cache  Managed by the AFS Service and provided to the redirector upon request

File Data Handling  AFS Caching  AFS Service populates AFS Cache with requested data and flushes dirty data back to server  AFS Redirector talks directly to the underlying AFS Cache through extents retrieved from the AFS Service  Interesting edge cases arise when performing large file copies using small AFS Cache sizes  Windows ‘optimizations’ in flushing  Leverage Windows Read-Ahead and Write Behind features

File Data Handling  Allows for better performance by allowing redirector direct access to cache file  AFS Service still manages cache layout and population AFS Service AFS Redir Extent Interface AFS Cache AFS

PIOCTL Interface  The interface has not changed from the AFS perspective  Implemented within the redirector as ‘special’ file open requests  File information and data management handled within the AFS Service

Special Share Name Handling  \PIPE\IPC$  Used for remote processing – currently not supported within the AFS Redirector  \PIPE\srvsvc  Used for server and share information processing through the Net API  Currently supported through AFS Service  Leverages Microsoft RPC engine for translation  \PIPE\wkssvc  Used for workstation information processing through the Net API

Invalidation Processing  Callback processing and issues in Windows  Callbacks can be made as a result of requests issued from the file system. Need to ensure these re-entrant calls do not lead to dead locks  ‘Almost’ lockless model in the callback routine through FID access layer  Server initiated callbacks have interesting effects, particularly in the directory change notification interface  Callbacks are FID based while notification is name based

Windows Change Notification  Windows model for directory change notification  Objects added, modified or deleted initiate completion of a notification request  Windows support API is named based … not in AFS  Implement layer on top of Windows support API to map names to/from FIDs  Still edge cases that are not correctly handled, particularly in callback invalidation

AFS Redirector Trace System  Command line configurable – Level, subsystem, buffer size, etc.  Persisted configuration for system startup tracing  In memory buffer so recoverable in crash dump  Retrieve buffer through command line as well as dump to debugger

Yet to be Done …  Alternate Data Streams  Extended Attributes  User and process quotas  Enhanced extent processing interface  Dynamically loadable functional driver – eliminates reboot for updates to file system