Microkernel-based Operating Systems - Introduction

Transcription

1 Faculty of Computer Science Institute for System Architecture, Operating Systems Group Microkernel-based Operating Systems - Introduction Nils Asmussen Dresden, Oct

2 Lecture Goals Provide deeper understanding of OS mechanisms Illustrate alternative OS design concepts Promote OS research at TU Dresden Make you all enthusiastic about OS development in general and microkernels in particular TU Dresden, MOS - Introduction Slide 2 von 47

3 Organization: Lecture Lecture every Tuesday, 4:40 PM, INF/E01 Slides: -> Teaching -> Microkernel-based Operating Systems Subscribe to our mailing list: This lecture is not: Microkernel construction (in summer term) TU Dresden, MOS - Introduction Slide 3 von 47

4 Organization: Exercises Exercises (roughly) bi-weekly Tuesday, 2:50 PM, INF/E08 Practical exercises in the computer lab Paper reading exercises Read a paper beforehand. Sum it up and prepare 3 questions. We expect you to actively participate in discussion. First exercise: next week Practical Exercise: Booting Room: to be announced TU Dresden, MOS - Introduction Slide 4 von 47

5 More Practical Stuff: Complex lab Complex lab in parallel to lecture Build several components of an OS Komplexpraktikum for (Media) Computer Science students Starts in 2 weeks, 2:50 PM, INF/E08 TU Dresden, MOS - Introduction Slide 5 von 47

6 Schedule Date Oct 09 Oct 16 Oct 23 Oct 30 Nov 06 Nov 13 Nov 30 Nov 27 Dec 04 Dec 11 Jan 8 Jan 15 Jan 22 Jan 30 Lecture Intro Threads & Synchronization IPC Memory Management Real-Time Device Drivers Resource Management Heterogeneous Systems Legacy Reuse Virtualization Secure Systems Trusted Computing Faults, Failures & Resilience Spare slot TU Dresden, MOS - Introduction Slide 6 von 47

7 Purpose of Operating Systems Manage the available resources Hardware (CPU, memory,...) Software (file systems, networking stack,...) Provide easier-to-use interface to access resources Unix: read/write data from/to sockets instead of fiddling with TCP/IP packets on your own Perform privileged / HW-specific operations x86: ring 0 vs. ring 3 Device drivers Provide separation and collaboration Isolate users / processes from each other Allow cooperation if needed (e.g., sending messages between processes) TU Dresden, Microkernels - Intro

8 Monolithic kernels: Linux User mode Application Application Application Application Kernel mode System-Call Interface Linux Kernel File Systems VFS File System Impl. Device Drivers Networking Sockets Protocols Processes Scheduling IPC Memory Management Page allocation Address spaces Swapping Hardware Access Hardware CPU, Memory, PCI, Devices TU Dresden, MOS - Introduction Slide 8 von 47

9 What's the problem? Security issues All components run in privileged mode. Direct access to all kernel-level data. Module loading easy living for rootkits. Resilience issues Faulty drivers can crash the whole system. 75% of today's OS kernels are drivers. Software-level issues Complexity is hard to manage. Custom OS for hardware with scarce resources? TU Dresden, Microkernels - Intro

10 The microkernel vision Minimal OS kernel less error prone small Trusted Computing Base suitable for verification System services in user-level servers flexible and extensible Protection between individual components More resilient crashing component does not (necessarily...) crash the whole system More secure inter-component protection TU Dresden, Microkernels - Intro

11 The microkernel vision User mode Application Application Application Application File Systems VFS File System Impl. Device Drivers Networking Sockets Protocols Memory Management Page allocation Swapping Kernel mode Microkernel System-Call Interface Hardware Access Address Spaces Threads Scheduling IPC Hardware CPU, Memory, PCI, Devices TU Dresden, MOS - Introduction Slide 11 von 47

12 What microkernels can give us... OS personalities Customizability Servers may be configured to suit the target system (small embedded systems, desktop PCs, SMP systems, ) Remove unneeded servers Enforce reasonable system design Well-defined interfaces between components No access to components besides these interfaces Improved maintainability TU Dresden, MOS - Introduction Slide 12 von 47

13 The mother of all microkernels Mach developed at CMU, Rick Rashid (former head of MS Research) Avie Tevanian (former Apple CTO) Brian Bershad U. of Washington) Foundation for several real systems Single Server Unix (BSD4.3 on Mach) MkLinux (OSF) IBM Workplace OS NeXT OS Mac OS X TU Dresden, MOS - Introduction Slide 13 von 47

14 Mach: Technical details Simple, extensible communication kernel Everything is a pipe. ports as secure communication channels Multiprocessor support Message passing by mapping Multi-server OS personality POSIX-compatibility Shortcomings performance drivers still in the kernel TU Dresden, MOS - Introduction Slide 14 von 47

15 Case study: IBM Workplace OS Main goals: multiple OS personalities run on multiple HW architectures Win Apps Unix Apps OS/2 Apps Windows Personality Unix Personality OS/2 Personality Network Processes Power... Files OS base services Mach microkernel ARM PPC x86 MIPS Alpha TU Dresden, MOS - Introduction Slide 15 von 47

16 IBM Workplace OS: Why did it fail? Never finished (but spent 1 billion $) Failure causes: Underestimated difficulties in creating OS personalities Management errors, forced divisions to adopt new system without having a system Second System Effect : too many fancy features Too slow Conclusion: Microkernel worked, but system atop the microkernel did not TU Dresden, MOS - Introduction Slide 16 von 47

17 IBM Workplace OS: Lessons learned OS personalities did not work Flexibility but monolithic kernels became flexible, too (Linux kernel modules) Better design but monolithic kernels also improved (restricted symbol access, layered architectures) Maintainability still very complex Performance matters a lot TU Dresden, MOS - Introduction Slide 17 von 47

18 Microkernels: Proven advantages Subsystem protection / isolation Code size Microkernel-based OS Fiasco kernel: ~ 34,000 LoC HelloWorld (+boot loader +root task): ~ 10,000 LoC Linux kernel (3.0.4., x86 architecture): Kernel: ~ 2.5 million LoC +drivers: ~ 5.4 million LoC (generated using David A. Wheeler's 'SLOCCount') Customizability Tailored memory management / scheduling / algorithms Adaptable to embedded / real-time / secure / systems TU Dresden, MOS - Introduction Slide 18 von 47

19 Challenges We need fast and efficient kernels covered in the Microkernel construction lecture in the summer term We need fast and efficient OS services Memory and resource management Synchronization Device Drivers File systems Communication interfaces Subject of this lecture TU Dresden, MOS - Introduction Slide 19 von 47

20 Who is (or was) out there? FU Amsterdam (Andrew Tanenbaum) MS Research Johns Hopkins University The L4 Microkernel Family Originally developed by Jochen Liedtke at IBM and GMD 2 nd generation microkernel Several kernel ABI versions TU Dresden, MOS - Introduction Slide 20 von 47

21 The L4 family a timeline (or tree...) N1 N2 SeL4 SeL4 University of New South Wales / NICTA / Open Kernel Labs NICTA:: Pistachio-embedded OKL4v2 OKL4 L4/x86 L2, L3 v2 x0 x2/v4 ABI Specification Implementation University of Karlsruhe L4Ka::Hazelnut L4Ka::Pistachio Fiasco/L4v2 Nova Fiasco OC Nova TU Dresden L4.Sec Fiasco/L4.Fiasco Fiasco.OC TU Dresden, MOS - Introduction Slide 21 von 47

22 L4 concepts Jochen Liedtke: A microkernel does no real work. Kernel only provides inevitable mechanisms. Kernel does not enforce policies. But what is inevitable? Abstractions Threads Address spaces (tasks) Mechanisms Communication Resource mapping (Scheduling) TU Dresden, Microkernels - Intro

23 Taking a closer look at L4 Case study: L4/Fiasco.OC TU Dresden, MOS - Introduction Slide 23 von 47

24 Case study: L4/Fiasco.OC Everything is an object Task Address spaces Thread Activities, scheduling IPC Gate Communication, resource mapping IRQ Communication Factory Create other objects, enforce resource quotas One system call: invoke_object() Parameters passed in UTCB Types of parameters depend on type of object TU Dresden, MOS - Introduction Slide 24 von 47

25 L4/Fiasco.OC: Types of Objects Kernel-provided objects Threads Tasks IRQs Generic communication object: IPC gate Send message from sender to receiver Used to implement new objects in user-level applications TU Dresden, Microkernels - Intro

26 L4/Fiasco.OC: User-level objects Everything above kernel built using user-level objects that provide a service Networking stack File system call()... Client Service 1 call() Kernel provides Object creation/management Object interaction: Inter-Process Communication (IPC) Service 2 call() TU Dresden, Microkernels - Intro

27 L4/Fiasco.OC: How to call objects? To call an object, we need an address: Telephone number Postal address IP address... call(service1.id) Client Service 1 Kernel Simple idea, right? ID is wrong? Kernel returns ENOTEXIST But not so fast! This scheme is insecure: Client could simply guess IDs brute-force. Existence/non-existence can be used as a covert channel TU Dresden, Microkernels - Intro

28 L4/Fiasco.OC: Local names for objects Global object IDs are insecure (forgery, covert channels). inconvenient (programmer needs to know about partitioning in advance) Solution in Fiasco.OC Task-local capability space as an indirection Object capability required to invoke object Per-task name space Maps names to object capabilities. Configured by task's creator TU Dresden, MOS - Introduction Slide 28 von 47

29 Capabilities / Local Names Indirection allows for security and flexibility. Address Space Address Space Address Space TU Dresden, Fiasco/OC & L4Re Slide 29 von 47

30 L4/Fiasco.OC: Object capabilities Capability: Reference to an object Protected by the Fiasco.OC kernel Kernel knows all capability-object mappings. Managed as a per-process capability table. User processes only use indexes into this table. Client Service 1 invoke(capability(3)) IPC Gate: communication channel for Service 1 Kernel TU Dresden, Microkernels - Intro

31 L4/Fiasco.OC: Communication Kernel object for communication: IPC gate Inter-process communication (IPC) Between threads Synchronous Communication using IPC gate: Sender thread puts message into its UTCB Sender invokes IPC gate, blocks sender until receiver ready (i.e., waits for message) Kernel copies message to receiver thread's UTCB Both continue, knowing that message has been transferred/received TU Dresden, MOS - Introduction Slide 31 von 47

32 More L4 concepts TU Dresden, MOS - Introduction Slide 32 von 47

33 L4/Fiasco.OC: Threads Thread Unit of Execution Implemented as kernel object Properties managed by the kernel: Instruction Pointer (EIP) Stack (ESP) Registers User-level TCB User-level applications need to allocate stack memory provide memory for application binary find entry point... Address Space Code Data UTCBs Stack Stack Threads TU Dresden, MOS - Introduction Slide 33 von 47

34 L4/Fiasco.OC: Interrupts Kernel object: IRQ Used for hardware and software interrupts Provides asynchronous signaling invoke_object(irq_cap, WAIT) invoke_object(irq_cap, TRIGGER) IRQ Kernel invoke_object (irq_cap,...) User-space device driver Image source: TU Dresden, MOS - Introduction Slide 34 von 47

35 Problem: Memory partitioning App1 App2 App3 4 GB 0 Physical Memory TU Dresden, MOS - Introduction Slide 35 von 47

36 Solution: Virtual Memory App1 App2 App3 4 GB 0 Physical Memory TU Dresden, MOS - Introduction Slide 36 von 47

37 L4: Recursive address spaces RAM Physical Address Space Device Memory TU Dresden, MOS - Introduction Slide 37 von 47

38 L4: Resource Mappings If a thread has access to a capability, it can map this capability to another thread Mapping / not mapping of capabilities used for implementing access control Abstraction for mapping: flexpage Flexpages describe mapping location and size of resource receiver's rights (read-only, mappable) type (memory, I/O, communication capability) TU Dresden, MOS - Introduction Slide 38 von 47

39 L4/Fiasco.OC: Object types Summary of object types Task Thread IPC Gate IRQ Factory Each task gets initial set of capabilities for some of these objects at startup TU Dresden, MOS - Introduction Slide 39 von 47

40 Building microkernel-based systems What can we build with all this? TU Dresden, MOS - Introduction Slide 40 von 47

41 Kernel vs. Operating System Fiasco.OC is not a full operating system! No device drivers (except UART + timer) No file system / network stack / A microkernel-based OS needs to add these services as user-level components L4Re uclibc libstdc++... IPC Client/Server Framework User-level libraries Ned Init-style task loader Moe Sigma0 L4Re L4 Runtime Environment User mode Kernel mode Basic Resource Manager(s) Fiasco.OC TU Dresden, MOS - Introduction Slide 41 von 47

42 Linux on L4 Application Application Application Application L4 Task L4 Task L4 Task L4 Task System-Call Interface arch-dep Linux Kernel File Systems VFS File System Impl. Device Drivers Networking Sockets Protocols Processes Scheduling IPC Memory Management Page allocation Address spaces Swapping Archindep. Hardware Access arch-dep L4 Task User mode Runtime Environment (L4Re) Kernel mode Fiasco.OC Hardware TU Dresden, MOS - Introduction Slide 42 von 47

43 The Dresden Real-Time Operating System Apps Time service RT Apps L4Linux Network driver SCSI driver Display driver User Mode Privileged Mode Non-RT World RT World Resource Management Layer (L4Re) Fiasco.OC microkernel TU Dresden, MOS - Introduction Slide 43 von 47

44 Virtual machines Isolate not only processes, but also complete Operating Systems (compartments) Server consolidation Apps Apps Apps L4Linux L4Linux Native Linux VMM User Mode Privileged Mode Virtualization Layer (L4Re) Fiasco.OC microkernel TU Dresden, MOS - Introduction Slide 44 von 47

45 Genode Genode := C++-based OS framework developed here in Dresden Aim: hierarchical system in order to Support resource partitioning Layer security policies on top of each other Genode::Core Parent 1 Parent 2 Child Child Child Child Child TU Dresden, MOS - Introduction Slide 45 von 47

46 Lecture outline Basic mechanisms and concepts Memory management Tasks, Threads, Synchronization Communication Building real systems What are resources and how to manage them? How to build a secure system? How to build a real-time system? How to reuse existing code (Linux, standard system libraries, device drivers)? How to improve robustness and safety? TU Dresden, MOS - Introduction Slide 46 von 47

47 Outlook Next lecture: Threads & Synchronization Next week (Oct 16, 4:40 PM) First exercise: Practical Exercise: Booting Room will be announced on mailing list TU Dresden, MOS - Introduction Slide 47 von 47