Introduction to Microkernels

Faculty of Computer Science Institute for System Architecture, Operating Systems Group

Introduction to Microkernel-
Based Operating Systems
Björn Döbel

Lecture Outline

• Microkernels and what we like about them

• The Fiasco.OC microkernel
– Kernel Objects
– Kernel Mechanisms

• OS Services on top of Fiasco.OC
– Device Drivers
– Virtualization

TU Dresden, 2012-07-18 Microkernels - Intro

Purpose of Operating Systems

• Manage the available resources
– Hardware (CPU) and software (file systems)
• Provide users with an easier-to-use interface to access
resources
– Unix: data read/write access to sockets instead of writing
TCP packets on your own
• Perform privileged / HW-specific operations
– x86: ring0 vs. ring3
– Device drivers
• Provide separation and collaboration
– Isolate users / processes from each other
– Allow cooperation if needed (e.g., sending
messages between processes)


Monolithic kernels - Linux

Application Application Application Application
User mode

Kernel
mode
System-Call Interface

File Systems Networking Processes Memory
VFS Sockets Scheduling Management
Linux File System Impl. Protocols IPC Page allocation
Kernel Address spaces
Device Drivers Swapping

Hardware Access

Hardware
CPU, Memory, PCI, Devices


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?


One vision - microkernels

• 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
– systems get
• More secure – inter-component protection
• More resilient – crashing component does not
(necessarily...) crash the whole system


The microkernel vision

Application Application Application Application
User mode

File Systems Networking Memory
VFS Sockets Management
File System Impl. Protocols Page allocation
Swapping
Device Drivers

Kernel
mode
Address Spaces
System-Call Interface
Threads
Scheduling
Hardware Access IPC

Microkernel

Hardware
CPU, Memory, PCI, Devices


Microkernel-Based Systems

• 1st generation: Mach
– developed at CMU, 1985 - 1994
– Foundation for several real systems (e.g.,
NextOS → Mac OS X)

• 2nd generation: Minix3
– Andrew Tanenbaum @ VU Amsterdam
– Focus on restartability

• 2nd/3rd generation:
– Various kernels of the L4 microkernel family


The L4 family – a timeline
SeL4
N1 N2

OKL4v2
Univ. of New South
Wales / NICTA / Open NICTA::
OKL4
Kernel Labs Pistachio-embedded

L4/x86
ABI Specification
L2, L3 v2 x0 x2/v4
Implementation

Univ. of L4Ka::Hazelnut L4Ka::Pistachio
Karlsruhe

Fiasco/L4v2 Nova

Fiasco
OC Nova

TU
Dresden L4.Sec Fiasco/L4.Fiasco Fiasco.OC


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)


Fiasco.OC – Objects

• OC – Object-Capability system
• System designed around objects providing
services:
call()
Client Service 1

call() call()

Service 2

• Kernel provides
– Object creation/management
– Object interaction: Inter-Process Communication
(IPC)


Fiasco.OC – Calling objects

• To call an object, we need an address:
– Telephone number
– Postal address Client Service 1
– IP address

call(service1.ID)
Kernel

• Kernel returns ENOTEXISTENT if ID is wrong.
• Security issues:
– Client could simply “guess” IDs brute-force.
– Existence/non-existence can be used as a covert
channel

Fiasco.OC – 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

Service1
1 Communication
Channel
2
invoke(capability(3))
3
4 Kernel

Fiasco.OC: System Calls

• “Everything is an object.”

• 1 system call: invoke_object()
– Parameters passed in UTCB
– Types of parameters depend on type of object

• 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


Kernel vs. Operating System

• Fiasco.OC is not a full uClibC libstdc++ ...
operating system! IPC Client/Server Framework
– No device drivers User-level libraries
(except UART + timer)
– No file system / network
Ned
stack / …

L4Re
Init-style task loader
• A microkernel-based OS
needs to add these Moe
services as user-level
Sigma0
components
L4 Runtime User Basic Resouce Manager(s)
mode
Environment
Kernel
(L4Re) mode
Fiasco.OC


Outline for the Next Lectures

• Fiasco.OC's mapping from managed resources
to kernel objects:
– CPU → threads
– Memory → tasks (address spaces)
– Communication → Inter-Process
Communication (IPC)

• L4 Runtime Environment
– Device Drivers
– L4Linux


L4 - Threads
Address Space
• Thread ::= abstraction of execution
– Unit of CPU scheduling
– Threads are temporally isolated

• Properties managed by the kernel: Threads

– Instruction Pointer (EIP) Code
– Stack Pointer (ESP)
Data
– CPU Registers / flags
– (User-level) TCB

• User-level applications need to Stack
– allocate stack memory Stack
– provide memory for application binary
– find entry point
– ...

L4 Threads and the Kernel

• Threads run in userland and enter the kernel
– Through a system call (sysenter / INT 0x30)
– Forced by HW interrupts or CPU exceptions

• Kernel Info Page
– Magic memory page mapped into every task
– Contains kernel-related information
• Kernel version
• Configured kernel features
• System call entry code (allows the kernel to
decide whether sysenter or INT 0x30 are better
for a specific platform)


Thread Control Block (TCB)

• Kernel storage for thread-related information

• One TCB per thread

• Stores user state while thread is inactive

• Extension: User-level Thread Control Block
(UTCB)
– Holds data the kernel does not need to trust
– Mapped into address space
– Most prominent use: system call parameters


Thread Scheduling

• Whenever a thread enters the kernel, a
scheduling decision is made.

• Fiasco.OC: priority- based round-robbin
– Every thread has a priority assigned.
– The thread with the highest priority runs until
• Its time quantum runs out (timer interrupt),
• Thread blocks (e.g., in a system call), or
• A higher-priority thread becomes ready
– Then, the next thread is selected.


L4Re and Threads

• Fiasco provides thread-related system calls
– thread_control → modify properties
– thread_stats_time → get thread runtime
– thread_ex_regs → modify EIP and ESP

• But most L4Re applications don't need to
bother:
– L4Re provides full libpthread including
• pthread_create
• pthread_mutex_*
• pthread_cond_*
• ...


L4Re Applications

• Every L4Re application starts with
– An empty address space
• Memory managed by parent
– One initial thread
• EIP set to binary's entry point by ELF loader
– An initial set of capabilities – the environment
• Parent
• Memory allocator
• Main thread
• Log
• ...


Performing System Calls

• All Fiasco.OC system calls are performed using
IPC with different sets of parameters.
– Functions are called l4_ipc_*()
– Colloquially: invoke
• Generic parameters (in registers):
– Capability to invoke
– Timeout (how long do I want to block at most? –
let's assume L4_IPC_NEVER for now.)
– Message tag describing the rest of the message
• Protocol
• Number of words in UTCB
• Message-specific parameters in UTCB message
registers

Writing Output

• L4Re environment passes a LOG capability

– Implements the L4_PROTO_LOG protocol
• By default implemented in kernel and
printed out to serial console

– UTCB content:
• Message reg 0: log operation to perform
(e.g., L4_VCON_WRITE_OP)
• Message reg 1: number of characters
• Message reg 2...: characters to write


Writing Output: The Code

#include <l4/re/env.h>
#include <l4/sys/ipc.h>

[..]

l4re_env_t *env = l4re_env(); // get environment
l4_msg_regs_t *mr = l4_utcb_mr(); // get msg regs

mr->mr[0] = L4_VCON_WRITE_OP;
mr->mr[1] = 7; // 'hellon' = 6 chars + 0 char
memcpy(&mr->mr[2], “hellon”, 7);

l4_msgtag_t tag, ret;
tag = l4_msgtag(L4_PROTO_LOG, 4, /* 4 msg words /
0, L4_IPC_NEVER);

ret = l4_ipc_send(env->log, l4_utcb(), tag); // System Call!

if (l4_msgtag_has_error(ret)) {
/* error handling */
}

Writing Output: The Code

#include <l4/re/env.h>
#include <l4/sys/ipc.h>

[..]

l4re_env_t *env = l4re_env(); // get environment
l4_msg_regs_t mr = l4_utcb_mr(); // get msg regs
In real code, please just do
mr->mr[0] = L4_VCON_WRITE_OP;
mr->mr[1] = 7; // 'hellon' = 6 chars + 0 char
memcpy(&mr->mr[2], “hellon”, 7);
puts(“hello”);
l4_msgtag_t tag, ret;
tag = l4_msgtag(L4_PROTO_LOG, 4, /* 4 msg words /
0, 0, L4_IPC_NEVER);

ret = l4_ipc_send(env->log, l4_utcb(), tag); // System Call!

if (l4_msgtag_has_error(ret)) {
/* error handling */
}

Multithreading

• Fiasco.OC allows multithreading
– Many threads sharing the same address space
– Spread across multiple physical CPUs

• Classical Problem: critical sections

global: int i = 0;

Thread 1 Thread 2

for (unsigned j = 0; j < 10; for (unsigned j = 0; j < 10;
++j) ++j)
i += 1; i += 1;

• The result is rarely i == 20!

Synchronization

for (unsigned j = 0; j < 10; ++j)

i += 1; Critical Section

• Critical Sections need to be protected
– Disable interrupts → infeasible for user space
– Spinning → burns CPU / energy / time quanta

• What we want: blocking lock
– Thread tests flag: critical section free yes/no
– waits (sleeping) until section is free


Expected behavior

Thread1 leaves
critical section

Thread 1

Thread 2
Threads try to Thread2 leaves
enter critical critical section
section time


Synchronization - pthreads

• L4Re provides libpthread, so we can simply use
pthread_mutex operations:

pthread_mutex_t mtx = PTHREAD_MUTEX_INITIALIZER;

[..]

for (unsigned j = 0; j < 10; ++j) {
pthread_mutex_lock(&mtx);
i += 1;
pthread_mutex_unlock(&mtx);
}

• Fiasco.OC's IPC primitives allow for another solution,
though.


Synchronization: Serializer Thread

• IPC operations are synchronous by default:
– Sender and receiver both need to be in an IPC system call
• There's a combination of sending and receiving a
message: l4_ipc_call().
• This allows synchronization using a serializer thread:

Thread 1
Blocking Done
call

Serializer
Blocking Reply
call

Thread 2

TU Dresden, 2012-07-18 Microkernels - Intro time

Downloading and Compiling

• Fiasco.OC and L4Re are available from
http://os.inf.tu-dresden.de/L4Re

• There are download and build instructions.
– We will use the 32bit versions for this course
→ simply leave all configuration settings at their defaults
– Note, you have to do 2 separate builds: one for
Fiasco.OC and one for the L4Re.
– GCC-4.7 did not work for me at the moment.


L4Re directory structure

• src/l4
• Important subdirectories: pkg/, conf/
• pkg/contains all applications (each in its own
package)
– Packages have subdirs again:
• server/ → the application program
• lib/ → library to be used by clients
• include/ → header files shared between
server and clients


Running Fiasco.OC/L4Re

• We'll use QEMU to run our setups.
• L4Re's build system has QEMU support
integrated, which is configured through files
in src/l4/conf:
– modules.lst → contains multiboot setup info,
similar to a GRUB menu.lst
– Makeconf.boot → contains overall settings
(where to search for binaries, qemu, ...)


modules.lst
Have this once in your
modaddr 0x01100000 modules.lst file.

Each entry has a name roottask is the initial task
to boot. --init rom/hello asks
entry hello it to load the hello binary
roottask moe --init=rom/hello from the ROM file system
module l4re
module hello

modules are additional
files. They are loaded into
memory and can then be
accessed through the ROM
file system under the name
rom/<filename>.


Makeconf.boot

• Start from the example in src/l4/conf
(rename it to Makeconf.boot)

• At least set:
– MODULE_SEARCH_PATH (have it include the
path to your Fiasco.OC build directory)


Booting QEMU

• Go to L4Re build directory

• Run “make qemu”
– Select 'hello' entry from the dialog
• If there's no dialog, you need to install the
'dialog' package.
• You can also circument the dialog:
make qemu E=<entry>
where entry is the name of a modules.lst
entry.


Assignments

• Download and compile Fiasco.OC and L4Re.

• Run the hello world example in QEMU.

• Modify the hello world example (it is in
l4/pkg/hello/server/src):
– Replace the puts() call with a manual system
call to the log object.
– You can use the example code from these
slides.


Further Reading

• P. Brinch-Hansen: The Nucleus of a Multiprogramming
System
http://brinch-hansen.net/papers/1970a.pdf
Microkernels were invented in 1969!

• J. Liedtke: On microkernel construction
http://os.inf.tu-dresden.de/papers_ps/jochen/Mikern.ps
Shaping the ideas found in L4 microkernels.

• D. Engler et al.: Exokernel – An operating system
architecture for application-level resource management
http://pdos.csail.mit.edu/6.828/2008/readings/engler95exokernel.pdf
Taking user-level policy implementation to the extreme.


Introduction to Microkernels

Recommended

More Related Content

What's hot (20)

Similar to Introduction to Microkernels (20)

More from Vasily Sartakov (20)

Recently uploaded (20)

Introduction to Microkernels