Connectathon test suite README
What is Connectathon test suite
Connectathon test suite - Linux NFS
The Connectathon tests run on top of an NFS mount, and therefore test the behavior of a real (kernel) NFS client against a server. They can also be used for testing filesystems other than NFS.
#
# @(#)README 1.18 2003/12/30 Connectathon Testsuite
#
NFS and Connectathon are trademarks of Sun Microsystems, Inc.
Introduction to the Connectathon NFS Testsuites
These directories contain programs that can be used to test an implementation of the NFS Protocol. The tests run on a UNIX client and test server and client functions. (See READWIN.txt for information about running on DOS and Windows.) The tests are divided into three groups:
basic - basic file system operations tests
general - general file system tests
special - tests that poke certain common problem areas
lock - tests that exercise network locking
This README is divided into five sections. The first section is the introduction, which you are reading now. That is followed by a description of what you have to do before you run the testsuites on your machine. Then comes a description of how the testsuites are run in general followed by a description of how they are used at Connectathon. The last section describes what each test does in detail.
This testsuite should run on both BSD and System V based systems. The System V Release 3 port of the Connectathon Testsuite is provided courtesy of the Lachman Technology, Incorporated, 1901 N. Naper Blvd., Naperville, IL. 60563.
Preparing to run the Testsuites
To prepare to run the testsuites on your machine, change directories
to the highest level testsuite directory (it should be the same one
that contains this README file), edit tests.init according to the
platform you are on, and type make
to compile the test programs. If
you are not sure you are in the correct directory, type ls -CF
and
you should see the following files and directories:
Makefile basic/ lock/ tests.h
README domount.c runtests tests.init
READWIN.txt general/ server tools/
Testitems getopt.c special/ unixdos.h
The server
script uses getopt
. A source file of a public-domain
version of getopt
is included in the directory. The Makefile will
compile it for you.
The tests are configured according to parameters found in the script,
tests.init. It contains various definitions for commands and
parameters used by the various Makefiles and shell scripts. This file
should be checked and then perhaps modified to correctly match your
system. In particular, the values of MOUNTCMD
, UMOUNTCMD
, PATH
,
CFLAGS
, and LIBS
should be checked and set correctly. There are
several sets of suggested values which may be used as possible starting
places.
Two special targets are included in the Makefiles: copy and dist. The
command, make copy DESTDIR="path"
, where path
is the absolute name
of a directory, will cause the compiled tests to be copied to path
.
The command, make dist DESTDIR="path"
, where path
is the absolute
name of a directory, will copy the test sources to path
. DESTDIR
must be specified on the make command line when making either of these
targets.
Modifications may be required so the programs compile on your machine. If that is so, we would like to know what they are so that we can incorporate them into our distribution.
When defaults are used, the test programs expect the directory, /server, to exist on the server. The test driver will use the directory /mnt/'server_name' on the client, creating it first if necessary (where 'server_name' is the name of the server you are testing against). These defaults can be overridden at run time. Directions for doing this are contained in the next section.
How to run the Testsuites
There are two ways to run the tests: use the server shell script or mount, run the tests yourself, and unmount. We recommend you use the server script to run the tests.
The server script:
The server script executes one or more of the test sets depending on what option is given (see below). It is set up to mount, run tests using the runtests program, and unmount. It will attempt to unmount anything mounted on the mount point before attempting to mount the server file system. If a test fails, the run is aborted and the file system is left mounted to assist in troubleshooting the failure.
The server script uses the domount program to mount and unmount the test file systems. Since mount can only be executed by root, domount must have root permission. The Makefile will attempt to setuid the domount program to root. The server script can be run as a nonprivileged user. Alternately, you may login as root before you run server.
server [-a|-b|-g|-s|-l] [-f|-t] [-n] [-o mnt_options] [-p server_path] [-m mntpoint] [-N numpasses] server_name
-a|-b|-g|-s|-l - will be passed on to the runtests scripts. This argument
is optional. The default is read from the initialization
file, tests.init. The variable, TEST, contains this
argument.
This argument selects which tests to run:
-a run basic, general, special, and lock tests
-b run basic tests only
-g run general tests only
-s run special tests only
-l run lock tests only
-f|-t - will be passed on to the runtests scripts. This argument
is optional. The default is read from the initialization
file, tests.init. The variable, TESTARG, contains this
argument.
This argument selects how the basic tests are to be run:
-f a quick functionality test
-t extended test mode with timings
-n - Don't perform the mkdir and rmdir operations to create
and destroy the test directory.
-o mnt_options - will be passed on to the mount command. This argument is
optional. The default is read from the initialization
file, tests.init. The variable, MNTOPTIONS, contains this
argument.
-p server_path - specifies a directory on the server to mount. This
argument is optional. The default is read from the
initialization file, tests.init. The variable, SERVPATH,
contains this argument.
-m mntpoint - specifies a mount point on your client. This argument is
optional. The default is read from the initialization
file, tests.init. The variable, MNTPOINT, contains this
argument.
-N numpasses - will be passed to the runtests script. This argument
is optional. It specifies the number of times to run
through the tests.
server_name - the server you want to exercise. This is the only
required argument.
The test programs create a sub-directory in the mntpoint directory with the name, 'hostname'.test, (where 'hostname' is the name of the machine on which you run the tests). This name can not be overridden if you use the server script although it can be if you use runtests directly.
Example: (the client machine is eddie)
eddie% server -o hard,intr,rw slartibartfarst
Start tests on path /mnt.slartibartfast/eddie.test [y/n]? y
<output from tests>
:
:
All tests completed
eddie%
See the script for more details.
Run tests yourself:
There is a runtest script in the highest level directory (the master runtests) which uses tests.init to set up the test environment and then executes the runtest scripts in the basic, general, and/or special sub-directories.
runtests [-a|-b|-g|-s|-l] [-f|-n|-t] [-N numpasses] [test-directory]
-a - Run the basic, general, special, and lock tests. This
is the default.
-b - Run the basic tests.
-g - Run the general tests.
-s - Run the special tests.
-l - Run the lock tests.
-f - Set parameters for a quick functional test. It
applies only to basic tests.
-n - Suppress directory operations (mkdir and rmdir) on the
test-directory. See descriptions of basic tests for
more details.
-t - Run full-length test with running time statistics. It
only applies to basic tests. This is the default mode
for the basic tests.
-N numpasses - Run the tests "numpasses" times.
test-directory - The name of test directory that the test programs
create on the client. runtests executes the basic
tests in place and they work on the test directory.
The general tests are copied over to the test
directory and executed there. When the -n flag is
used, the test directory is assumed to already exist.
The default test-directory is
/mnt.'servername'/'hostname'.test (where 'servername'
is the name of the server being tested, and
'hostname' is the name of the machine on which you
are running the tests). There are three ways to
override the default test directory name. One it to
put the test_directory on the command line. Another
way is to set the environment variable, NFSTESTDIR,
equal to the directory name. The command line method
overrides setting the environment variable. The
third way can only be used for the tests in the basic
sub-directory. There you can set the TESTDIR
variable in tests.h. The command line and
environment variable both override this method.
Running the tests without mounting your NFS server on /mnt will run the tests locally (if /mnt is local disk). We recommend that you do this once to make sure the testsuites run properly before you use them to test NFS.
The runtests in the sub-directories, basic, general, and special, may be invoked with the same arguments as the master runtests if you wish to run each suite separately.
How to run the Testsuites at Connectathon
The tests should be run in the following order: basic, general, and special. The basic tests should be passed completely before other tests are attempted.
The NFS Test Suite should be run in three phases:
Phase 1 - Run test programs locally.
Phase 2 - Run the tests against a Sun. Run them on your machine using the Sun as the server and then run them on the Sun using your machine as the server.
Phase 3 - NxN Testing. Run the tests on your machine using every other machine as a server, one at a time. After the tests are successfully completed using a particular server, log that with the electronic board software provided. Check the electronic board to make sure that the tests run successfully on every other machine that uses your machine as a server.
Test Descriptions
System and library calls that are used by the testsuites are included in parentheses. Look at the source if you are interested in how time statistics are recorded since that is not included in this description.
BASIC TESTS:
Many of the programs listed below have optional calling parameters that can be used to override existing parameters. These are not used at this time so they are not described.
test1: File and Directory Creation Test
This program creates the test directory (mkdir) on the client and
changes directories (chdir) to it, unless the -n flag is used in which
case it simply changes directories to the test directory. Then it
builds a directory tree N levels deep, where each directory (including
the test directory) has M files and P directories (creat, close, chdir,
and mkdir). For the -f option, N = 2, M = 2, and P = 2 so a total of
six files and six directories are created. For other options, N = 5,
M = 5, and P = 2. The files that are created are given names that
begin with file.
and directories with names that begin with dir.
.
test2: File and directory removal test
This program changes directory to the test directory (chdir and/or mkdir) and removes the directory tree (unlink, chdir, and rmdir) that was just created by test1. The number of levels, files, and directories, and the name prefixes, are the same as in test1.
This routine will not remove a file or directory that was not created by test1 and will fail if it finds one. It determines this by looking at the prefix on the name of the object it's trying to remove.
test3: Lookups across mount point
This program changes directory to the test directory (chdir and/or mkdir) and gets the file status of the working directory (getwd or getcwd and stat). For the -f option, the getwd or getcwd is done once. For other options, 250 getcwds or getcwds are done.
test4: setattr, getattr, and lookup
This program changes directory to the test directory (chdir and/or mkdir) and creates ten files (creat). Then the permissions are changed (chmod) and the file status is retrieved (stat) for each file. For the -f option, one chmod and stat on each file is done. For other options, 50 getcwds or getcwds and stats on each file are done.
test4a: getattr, and lookup
This test exists but is not called as part of the testsuite. You can edit runtests in the basic directory so this test is called.
This program changes directory to the test directory (chdir and/or mkdir) and creates ten files (creat). Then the file status is retrieved (stat) for each file. For the -f option, the stat is done once per file. For other options, 50 stats are done per file.
test5: read and write
This program changes directory to the test directory (chdir and/or mkdir) and then:
- Creates a file (creat)
- Gets status of file (fstat)
- Checks size of file
- Writes 1048576 bytes into the file (write) in 8192 byte buffers.
- Closes file (close)
- Gets status of file (stat)
- Checks the size of the file
For the -f option, the file is created and written once. For other options, file is created and written 10 times.
Then the file is opened (open) and read (read) in 8192 byte buffers. It's contents are compared with what was written. The file is then closed (close).
Then the file is then re-opened (open) and re-read (read) before it is removed (unlink). For the -f option, this sequence is done once. For other options, this sequence is done 10 times.
test5a: write
This test exists but is not called as part of the testsuite. You can edit runtests in the basic directory so this test is called.
This program changes directory to the test directory (chdir and/or mkdir) and then:
- Creates a file (creat)
- Gets status of file (fstat)
- Checks size of file
- Writes 1048576 bytes into the file (write) in 8192 byte buffers.
- Closes file (close)
- Gets status of file (stat)
- Checks the size of the file
For the -f option, the file is created and written once. For other options, file is created and written 10 times.
test5b: read
This test exists but is not called as part of the testsuite. You can edit runtests in the basic directory so this test is called.
The file created in test5a is opened (open) and read (read) in 8192 byte buffers. It's contents are compared with what was written. The file is then closed (close) and removed (unlink).
For the -f option, the file is opened and read once. For other options, file is created and written 10 times.
test6: readdir
This program changes directory to the test directory (chdir and/or mkdir) and creates 200 files (creat). The current directory is opened (opendir), the beginning is found (rewinddir), and the directory is read (readdir) in a loop until the end is found. Errors flagged are:
- No entry for
.
- No entry for
..
- Duplicate entry
- Filename that doesn't begin with
file.
- The suffix of the filename is out of range
- An entry is returned for an unlinked file. (This error can only be found when the test is run with an option other than -f. For other options the rewinddir/readdir loop is done 200 times and a file is unlinked each time).
The directory is then closed (closedir) and the files that were created are removed (unlink).
test7: link and rename
This program changes directory to the test directory (chdir and/or mkdir) and creates ten files. For each of these files, the file is renamed (rename) and file statistics are retrieved (stat) for both the new and old names. Errors that are flagged are:
- Old file still exists
- New file doesn't exist (can't stat)
- The new file's number of links doesn't equal one
Then an attempt is made to link the new file to it's old name (link) and file stats are again retrieved (stat). An error is flagged if:
- Can't link
- Stats on new file can't be retrieved after link
- The new file's number of links doesn't equal two
- Stats on old file can't be retrieved after link
- The old file's number of links doesn't equal two
Then the new file is removed (unlink) and file stats are retrieved for the old file (stat). An error is flagged if:
- Stats on old file can't be retrieved after unlink
- The old file's number of links doesn't equal one
For the -f option, the rename/link/unlink loop is done once for each file. For other options, the rename/link/unlink loop is done 10 times for each file.
Any files that remain at the end of the test are removed (unlink).
test7a: rename
This test exists but is not called as part of the testsuite. You can edit runtests in the basic directory so this test is called.
This program changes directory to the test directory (chdir and/or mkdir) and creates ten files. For each of these files, the file is renamed (rename) and file statistics are retrieved (stat) for both the new and old names. Errors that are flagged are:
- Old file still exists
- New file doesn't exist (can't stat)
- The new file's number of links doesn't equal one
The file is then renamed back to its original name and the same tests are applied.
For the -f option, the rename/rename loop is done once for each file. For other options, the rename/rename loop is done 10 times for each file.
Any files that remain at the end of the test are removed (unlink).
test7b: link
This test exists but is not called as part of the testsuite. You can edit runtests in the basic directory so this test is called.
This program changes directory to the test directory (chdir and/or mkdir) and creates ten files. A link (link) is done for each of these files and file stats are retrieved for the old and new files (stat). An error is flagged if:
- Can't link
- Stats on either file can't be retrieved after link
- The either file's number of links doesn't equal two
This is followed by an unlink (unlink) of the new file. An error is flagged if:
- Stats on the old file can't be retrieved after unlink
- The old file's number of links doesn't equal one
For the -f option, the link/unlink loop is done once for each file. For other options, the link/unlink loop is done 10 times for each file.
Any files that remain at the end of the test are removed (unlink).
test8: symlink and readlink
NOTE: Not all operating systems support symlink and readlink. If the errno, EOPNOTSUPP, is returned during test8, the test will be counted as passing. For clients not supporting S_IFLNK, the test will not be attempted.
This program changes directory to the test directory (chdir and/or mkdir) and makes 10 symlinks (symlink). It reads (readlink), and gets statistics for (lstat) each, and then removes them (unlink). Errors flagged are:
- Unsupported function
- Can't get statistics (lstat failed)
- The mode in the stats is not symlink
- The value of the symlink is incorrect (returned from readlink)
- The linkname is wrong
- The unlink failed
For the -f option, the symlink/readlink/unlink loop is done for each symlink. For other options, the symlink/readlink/unlink loop is done 20 times for each symlink.
test9: statfs
This program changes directory to the test directory (chdir and/or mkdir) and gets the file system status on the current directory (statfs). For the -f option, the statfs is done once. For other options, the statfs is done 1500 times.
GENERAL: General tests to look at server loading.
Runs a small compile, tbl, nroff, a large compile, four simultaneous large compiles, and make.
SPECIAL: Information specific to the special tests
The special directory is set up to test special problems that have come
up in the past. These tests are meant to be advisory, things to watch
out for. It is not required that you pass
these tests but we
strongly suggest that you do.
The tests try to:
check for proper open/unlink operation
check for proper open/rename operation
check for proper open/chmod 0 operation
check for lost reply on non-idempotent requests
test exclusive create
test negative seek
test rename
LOCK:
The lock directory contains a test program which can be used to test the kernel file and record locking facilities. This is done to test the network lock manager.
The test program contains 13 sets of locking tests. They test basic locking functionality.
By default, mandatory locking is not tested. Mandatory locking is generally not supported on NFS files.
MISC:
'Testitems' is a list of NFS functionality that can be used for reference.
Programs in 'tools' are provided for your use as you see fit. Please feel free to add to this (or any other) directory! If you do, please make sure that Mike Kupfer mike.kupfer@sun.com gets a copy so we can add it to the master test distribution.
The code in this tree was checked August 1998 for Y2000 problems. None were found.
See READWIN.txt for information about running the tests under DOS or Windows.
Changes
Changes for 2004 include the following:
Fix lock/tlock.c to be consistent about when to use stdarg and when to use varargs; reported by Samuel Sha sam@austin.ibm.com.
Change
make all
so that the variousruntests
scripts have the execute bit set; reported by Erik Deumens deumens@qtp.ufl.edu.Removed some lint; from James Peach jpeach@sgi.com.
Irix 6.5.19 support from James Peach jpeach@sgi.com.
The
server
script now exports MNTOPTIONS, so that options that are added toserver
can be detected by the rest of the suite. From Chuck Lever Charles.Lever@netapp.com.The tests now correctly check for errors returns from mmap(). From David Robinson david.robinson@sun.com.
MacOS X support from Mike Mackovitch macko@apple.com.
tests.init now includes a CC= line for Linux, in case your distribution doesn't include
cc
. Reported by Rodney Brown rodney@lehman.com.Changes for AIX, from Erik Deumens deumens@qtp.ufl.edu.
Changes for the latest Tru64 Unix, from Eric Werme werme@hp.com.
The general tests should be more robust in the face of errors from make(1). Based on comments from Chuck Lever Charles.Lever@netapp.com and a patch from Mike Mackovitch macko@apple.com.
The
make lint
target for the basic tests now includes subr.c.-
Improvements to special/bigfile2:
- error messages now print the complete low-order word (from Mike Mackovitch macko@apple.com.
- the test file is opened with O_SYNC, so that problems are detected right away.
Fix to special/op_chmod so that it uses CHMOD_NONE instead of 0. From Pascal Schmidt der.eremit@email.de.
Changes for 2003 include the following:
HPUX fixes from Brian Love blove@rlmsoftware.com and Brian McEntire brianm@fsg1.nws.noaa.gov.
AIX support, based on patches from saul@exanet.com.
gcc command-line options for building 64-bit binaries, from Sergey Klyushin sergey.klyushin@hummingbird.com.
The messages from the server script are now a little clearer about leaving the server mounted after a test failure. Thanks to Vincent McIntyre Vince.McIntyre@atnf.csiro.au for the suggestion.
The locking tests should now work with NFS Version 4 and servers that enforce mandatory locking. Thanks to Bill Baker bill.baker@sun.com for the test12 fix.
The general tests have been fixed to use the
stat
program that comes with the tests, instead of any systemstat
program.
Changes for 2002 include the following:
The special tests do a better job of recognizing when NFS version 2 was specified (based on a patch from Jay Weber jweber@mail.thatnet.net).
Compilation and runtime fixes for *BSD systems, based on patches from Marty Johnson martyj@traakan.com.
The default local mount point was changed from /mnt.'server_name' to /mnt/'server_name'. This is so that if the server dies or hangs, it is less likely to cause operational problems on the client.
The
server
script will try to usemkdir -p
if it's available.The general and special tests do a better job of checking for errors during initialization.
The bigfile tests have been moved to the end of the special tests because they can take so long to run.
Fixed the definition of signal handlers for Tru64 UNIX.
Updated Linux configuration information from Jay Weber jweber@mail.thatnet.net.
Changes for 2001 include the following:
Added a
-N numpasses
option to the top-levelserver
andruntests
script.Updated HPUX compilation flags for the benefit of the special/bigfile2 test (from Anand Paladugu paladugu_anand@emc.com).
Minor portability fixes to special/bigfile2.c.
The basic tests no longer assume that
.
is in $PATH.The basic and special tests should be easier to build under Windows (from Rick Hopkins rhopkins@ssc-corp.com).