blob: 5118b1943afc111f64e0182607a5ca171d43bbff [file] [log] [blame]
Linux Test Project HOWTO
10 October 2000
Nate Straz
Abstract
This document explains some of the more in depth topics of
the Linux Test Project and related testing issues. It does
not cover basic installation procedures. See the INSTALL
and README files in the tarball for that information.
1 Preface
This document was written to help bring the community up
to speed on the ins and outs of the Linux Test Project.
1.1 Copyright
Copyright (c) 2000 by SGI, Inc.
Please freely copy and distribute (sell or give away) this
document in any format. It's requested that corrections
and/or comments be fowarded to the document maintainer.
You may create a derivative work and distribute it provided
that you:
* Send your derivative work (in the most suitable format
such as sgml) to the LDP (Linux Documentation Project)
or the like for posting on the Internet. If not the LDP,
then let the LDP know where it is available.
* License the derivative work with this same license or use
GPL. Include a copyright notice and at least a pointer
to the license used.
* Give due credit to previous authors and major contributors.
If you're considering making a derived work other than a
translation, it's requested that you discuss your plans
with the current maintainer.
1.2 Disclaimer
Use the information in this document at your own risk. I
disavow any potential liability for the contents of this
document. Use of the concepts, examples, and/or other content
of this document is entirely at your own risk.
All copyrights are owned by their owners, unless specifically
noted otherwise. Use of a term in this document should
not be regarded as affecting the validity of any trademark
or service mark.
Naming of particular products or brands should not be seen
as endorsements.
You are strongly recommended to take a backup of your system
before major installation and backups at regular intervals.
2 Introduction
2.1 What is the Linux Test Project?
The Linux Test Project (LTP) is an effort to create a set
of tools and tests to verify the functionality and stability
of the Linux kernel. We hope this will support Linux development
by making unit testing more complete and minimizing user
impact by building a barrier to keep bugs from making it
to the user.
2.2 What is wrong with the current testing model?
The Linux development community utilizes two important (some
out argue most important) testing techniques in its normal
operations: Design and Code Inspections. The intent of LTP
is to support this by giving developers an ever growing
set of tools to help identify any operational problems in
their code that may be missed by human review. One of the
toughest categories of problems to catch with inspection
is that of interaction of features. With a continuously
improving set of tests and tools, developers can get an
indication of whether their changes may have broken some
other functionality.
There is no such thing as a perfect test base. It is only
useful it if keeps up with new and changing functionality,
and if it actually gets used.
2.3 Are you doing benchmarking?
Not at this time. We are more interested in functional,
regression, and stress testing the Linux kernel. Benchmarking
may be workable to compare the performance among kernel
versions.
2.4 Are you doing standards testing?
No, we are leaving that to the Linux Standards Base (LSB).
See the Linux Standards Base [http://www.linuxbase.org/||web site]
for more information.
3 Structure
The basic building block of the test project is a test case
that consists of a single action and a verification that
the action worked. The result of the test case is usually
restricted to PASS/FAIL.
A test program is a runnable program that contains one or
more test cases. Test programs often understand command
line options which alter their behavior. The options could
determine the amount of memory tested, the location of temporary
files, the type of network packet used, or any other useful
parameter.
Test tags are used to pair a unique identifier with a test
program and a set of command line options. Test tags are
the basis for test suites.
4 Writing Tests
Writing a test case is a lot easier than most people think.
Any code that you write to examine how a part of the kernel
works can be adapted into a test case. All that is needed
is a way to report the result of the action to the rest
of the world. There are several ways of doing this, some
more involved than others.
4.1 Exit Style Tests
Probably the simplest way of reporting the results of a test
case is the exit status of your program. If your test program
encounters unexpected or incorrect results, exit the test
program with a non-zero exit status, i.e. exit(1). Conversely,
if your program completes as expected, return a zero exit
status, i.e. exit(0). Any test driver should be able to
handle this type of error reporting. If a test program has
multiple test cases you won't know which test case failed,
but you will know the program that failed.
4.2 Formatted Output Tests
The next easiest way of reporting the results is to write
the results of each test case to standard output. This allows
for the testing results to be more understandable to both
the tester and the analysis tools. When the results are
written in a standard way, tools can be used to analyze
the results.
5 Testing Tools
The Linux Test Project has not yet decided on a "final" test
harness. We have provided a simple solution with pan to
make due until a complete solution has been found/created
that compliments the Linux kernel development process.
Several people have said we should use such and such a test
harness. Until we find we need a large complex test harness,
we will apply the KISS concept.
5.1 Pan
pan is a simple test driver with the ability to keep track
of orphaned processes and capture test output. It works
by reading a list of test tags and command lines and runs
them. By default pan will select a command randomly from
the list of test tags, wait for it to finish. Through command
line options you can run through the entire list sequentially,
run n tests, keep n test running at all times, and buffer
test output. Pan can be nested to create very complex test
environments.
Pan uses an active file, also called a zoo file to keep track
of which tests are currently running. This file holds the
pid, tag, and a portion of the command line. When you start
pan it becomes a test tag in itself, thus it requires a
name for itself. Pan updates the active file to show which
test tags are currently running. When a test tag exits,
pan will overwrite the first character with a '#'. The active
file can be shared between multiple instances of pan so
you know which tests were running when the system crashes
by looking at one file.
A pan file contains a list of test tags for pan to run. The
format of a pan file is as follows:
testtag testprogram -o one -p two other command line options
# This is a comment. It is a good idea to describe the test
# tags in your pan file. Tests programs can have different
# behaviors depending on the command line options so it is
# helpful to describe what each test tag is meant to verify
or # provoke.
# Some more test cases
mm01 mmap001 -m 10000
# 40 Mb mmap() test.
# Creates a 10000 page mmap, touches all of the map, sync's
# it, and munmap()s it.
mm03 mmap001 -i 0 -I 1 -m 100
# repetitive mmapping test.
# Creates a one page map repetitively for one minute.
dup02 dup02
# Negative test for dup(2) with bad fd
kill09 kill09
# Basic test for kill(2)
fs-suite01 pan -e -a fs-suite01.zoo -n fs-suite01 -f runtest/fs
# run the entire set of file system tests
The test tags are simple identifiers, no spaces are allowed.
The test of the line is the program to run, which is done
using execvp(3). Lines starting with '#' are comments and
ignored by pan. It is a good practice to include descriptions
with your test tags so you can have a reminder what a certain
obscure test tag tries to do.
5.1.1 Examples
The most basic way to run pan is by passing the test program
and parameters on the command line. This will run the single
program once and wrap the output.
$ pan -a ltp.zoo -n tutor sleep 4
<<<test_start>>>
tag=cmdln stime=971450564
cmdline="sleep 4"
contacts=""
analysis=exit
initiation_status="ok"
<<<test_output>>>
<<<execution_status>>>
duration=103341903 termination_type=exited termination_id=0
corefile=no cutime=0 cstime=0
<<<test_end>>>
$ cat ltp.zoo
#9357,tutor,pan/pan -a ltp.zoo -n tutor sleep 4
#9358,cmdln,sleep 4
$
How it works
This example shows the two parameters that are always required
by pan, the active file and a test tag for pan. The "sleep
4" on the end of the command line is a test program and
parameters that pan should run. This test is given the tag
"cmdln." Pan will run one test randomly, which ends
up being cmdln since it is the only test that we told pan
about.
In the active file, ltp.zoo, pan writes the pid, test tag,
and part of the command line for the currently running tests.
The command lines are truncated so each line will fit on
an 80 column display. When a test tag finishes, pan will
place a '#' at the beginning of the line to mark it as available.
Here you can see that cmdln and tutor, the name we gave
pan, ran to completion. If the computer hangs, you can read
this file to see which test programs were running.
We have run one test once. Let's do something a little more
exciting. Let's run one test several times, at the same
time.
$ pan -a ltp.zoo -n tutor -x 3 -s 3 -O /tmp sleep 1
<<<test_start>>>
tag=cmdln stime=971465653
cmdline="sleep 1"
contacts=""
analysis=exit
initiation_status="ok"
<<<test_output>>>
<<<execution_status>>>
duration=103326814 termination_type=exited termination_id=0
corefile=no
cutime=1 cstime=0
<<<test_end>>>
<<<test_start>>>
tag=cmdln stime=971465653
cmdline="sleep 1"
contacts=""
analysis=exit
initiation_status="ok"
<<<test_output>>>
<<<execution_status>>>
duration=103326814 termination_type=exited termination_id=0
corefile=no
cutime=0 cstime=1
<<<test_end>>>
<<<test_start>>>
tag=cmdln stime=971465653
cmdline="sleep 1"
contacts=""
analysis=exit
initiation_status="ok"
<<<test_output>>>
<<<execution_status>>>
duration=103326814 termination_type=exited termination_id=0
corefile=no
cutime=0 cstime=0
<<<test_end>>>
How it works
In this example we run another fake test from the command
line, but we run it three times (-s 3) and keep three test
tags active at the same time (-x 3). The -O parameter is
a directory where temporary files can be created to buffer
the output of each test tag. You can see in the output that
cmdln ran three times. If the -O option were omitted, your
test output would be mixed, making it almost worthless.
* Using a pan file to run multiple tests
* Nesting pan
For more information on pan see the man page doc/man1/pan.1.
5.2 Scanner
scanner is a results analysis tool that understands the rts
style output which pan generates by default. It will produce
a table summarizing which tests passed and which failed.
5.3 The Quick-hitter Package
Many of the tests released use the Quick-hitter test package
to perform tasks like create and move to a temporary directory,
handle some common command line parameters, loop, run in
parallel, handle signals, and clean up.
There is an example test case, doc/examples/quickhit.c, which
shows how the quick-hitter package can be used. The file
is meant to be a supplement to the documentation, not a
working test case. Use any of the tests in tests/ as a template.
6 To Do
There are a lot of things that still need to be done to make
this a complete kernel testing system. The following sections
will discuss some of the to do items in detail.
6.1 Configuration Analysis
While the number of configuration options for the Linux kernel
is seen as a strength to developers and users alike, it
is a curse to testers. To create a powerful automated testing
system, we need to be able to determine what the configuration
on the booted box is and then determine which tests should
be run on that box.
The Linux kernel has hundreds of configuration options that
can be set to compile the kernel. There are more options
that can be set when you boot the kernel and while it is
running. There are also many patches that can be applied
to the kernel to add functionality or change behavior.
6.2 Result Comparison
A lot of testing will be done in the life of the Linux Test
Project. Keeping track of the results from all the testing
will require some infrastructure. It would be nice to take
that output from a test machine, feed it to a program and
receive a list of items that broke since the last run on
that machine, or were fixed, or work on another test machine
but not on this one.
7 Contact information and updates
URL: http://oss.sgi.com/projects/ltp/
email: owners-ltp@oss.sgi.com
mailing list: ltp@oss.sgi.com
list archive: http://oss.sgi.com/projects/ltp/mail-threaded/
Questions and comments should be sent to the LTP mailing
list at ltp@oss.sgi.com. To subscribe, send mail to majordomo@oss.sgi.com
with "subscribe ltp" in the body of the message.
The source is also available via CVS. See the web site for
a web interface and check out instructions.
8 Glossary
Test IEEE/ANSI([footnote] Kit, Edward, Software Testing in the Real World:
Improving the Process. P. 82. ACM Press, 1995.) :
(i) An activity in which a system or component is executed
under specified conditions, the results are observed or
record, and an evaluation is made of some aspect of the
system or component.
(ii) A set of one or more test cases.
Test Case A test assertion with a single result that
is being verified. This allows designations such as PASS
or FAIL to be applied to a single bit of functionality.
A single test case may be one of many test cases for
testing the complete functionality of a system.
IEEE/ANSI:
(i)A set of test inputs, execution conditions, and expected
results developed for a particular objective.
(ii) The smallest entity that is always executed as a unit,
from beginning to end.
Test Driver A program that handles the execution of
test programs. It is responsible for starting the test
programs, capturing their output, and recording their
results. Pan is an example of a test driver.
Test Framework A mechanism for organizing a group of
tests. Frameworks may have complex or very simple API's,
drivers and result logging mechanisms. Examples of frameworks
are TETware and DejaGnu.
Test Harness A Test harness is the mechanism that connects
a test program to a test framework. It may be a specification
of exit codes, or a set of libraries for formatting messages
and determining exit codes. In TETware, the tet_result()
API is the test harness.
Test Program A single invokable program. A test program
can contain one or more test cases. The test harness's
API allows for reporting/analysis of the individual test
cases.
Test Suite A collection of tests programs, assertions,
cases grouped together under a framework.
Test Tag An identifier that corresponds to a command
line which runs a test. The tag is a single word that
matches a test program with a set of command line arguments.