blob: 477136deb0d01a2fb5993d20699c776e0103a228 [file] [log] [blame]
_______________________
BUILDING THE FSQA SUITE
_______________________
Ubuntu or Debian
----------------
1. Make sure that package list is up-to-date and install all necessary packages:
$ sudo apt-get update
$ sudo apt-get install acl attr automake bc dbench dump e2fsprogs fio gawk \
gcc git indent libacl1-dev libaio-dev libcap-dev libgdbm-dev libtool \
libtool-bin liburing-dev libuuid1 lvm2 make psmisc python3 quota sed \
uuid-dev uuid-runtime xfsprogs linux-headers-$(uname -r) sqlite3 \
libgdbm-compat-dev
2. Install packages for the filesystem(s) being tested:
$ sudo apt-get install exfatprogs f2fs-tools ocfs2-tools udftools xfsdump \
xfslibs-dev
Fedora
------
1. Install all necessary packages from standard repository:
$ sudo yum install acl attr automake bc dbench dump e2fsprogs fio gawk gcc \
gdbm-devel git indent kernel-devel libacl-devel libaio-devel \
libcap-devel libtool liburing-devel libuuid-devel lvm2 make psmisc \
python3 quota sed sqlite udftools xfsprogs
2. Install packages for the filesystem(s) being tested:
$ sudo yum install btrfs-progs exfatprogs f2fs-tools ocfs2-tools xfsdump \
xfsprogs-devel
RHEL or CentOS
--------------
1. Enable EPEL repository:
- see https://docs.fedoraproject.org/en-US/epel/#How_can_I_use_these_extra_packages.3F
2. Install all necessary packages which are available from standard repository
and EPEL:
$ sudo yum install acl attr automake bc dbench dump e2fsprogs fio gawk gcc \
gdbm-devel git indent kernel-devel libacl-devel libaio-devel \
libcap-devel libtool libuuid-devel lvm2 make psmisc python3 quota sed \
sqlite udftools xfsprogs
Or, EPEL packages could be compiled from sources, see:
- https://dbench.samba.org/web/download.html
- https://www.gnu.org/software/indent/
3. Build and install 'liburing':
- see https://github.com/axboe/liburing.
4. Install packages for the filesystem(s) being tested:
For XFS install:
$ sudo yum install xfsdump xfsprogs-devel
For exfat install:
$ sudo yum install exfatprogs
For f2fs build and install:
- see https://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs-tools.git/about/
For ocfs2 build and install:
- see https://github.com/markfasheh/ocfs2-tools
SUSE Linux Enterprise or openSUSE
---------------------------------
1. Install all necessary packages from standard repositories:
$ sudo zypper install acct automake bc dbench duperemove dump fio gcc git \
indent libacl-devel libaio-devel libattr-devel libcap libcap-devel \
libtool liburing-devel libuuid-devel lvm2 make quota sqlite3 xfsprogs
2. Install packages for the filesystem(s) being tested:
For btrfs install:
$ sudo zypper install btrfsprogs libbtrfs-devel
For XFS install:
$ sudo zypper install xfsdump xfsprogs-devel
Build and install test, libs and utils
--------------------------------------
$ git clone git://git.kernel.org/pub/scm/fs/xfs/xfstests-dev.git
$ cd xfstests-dev
$ make
$ sudo make install
Setup Environment
-----------------
1. Compile XFS/EXT4/BTRFS/etc. into your kernel or load as module. For example,
for XFS, enable XFS_FS in your kernel configuration, or compile it as a
module and load it with 'sudo modprobe xfs'. Most of the distributions will
have these filesystems already in the kernel/as module.
2. Create TEST device:
- format as the filesystem type you wish to test.
- should be at least 10GB in size.
- optionally populate with destroyable data.
- device contents may be destroyed.
3. (optional) Create SCRATCH device.
- many tests depend on the SCRATCH device existing.
- not need to be formatted.
- should be at least 10GB in size.
- must be different to TEST device.
- device contents will be destroyed.
4. (optional) Create SCRATCH device pool.
- needed for BTRFS testing
- specifies 3 or more independent SCRATCH devices via the SCRATCH_DEV_POOL
variable e.g SCRATCH_DEV_POOL="/dev/sda /dev/sdb /dev/sdc"
- device contents will be destroyed.
- SCRATCH device should be left unset, it will be overridden
by the SCRATCH_DEV_POOL implementation.
5. Copy local.config.example to local.config and edit as needed. The TEST_DEV
and TEST_DIR are required.
6. (optional) Create fsgqa test users and groups:
$ sudo useradd -m fsgqa
$ sudo useradd 123456-fsgqa
$ sudo useradd fsgqa2
$ sudo groupadd fsgqa
The "123456-fsgqa" user creation step can be safely skipped if your system
doesn't support names starting with digits, only a handful of tests require
it.
7. (optional) If you wish to run the udf components of the suite install
mkudffs. Also download and build the Philips UDF Verification Software from
https://www.lscdweb.com/registered/udf_verifier.html, then copy the udf_test
binary to xfstests/src/.
8. (optional) To do io_uring related testing, please make sure below 3 things:
1) kernel is built with CONFIG_IO_URING=y
2) sysctl -w kernel.io_uring_disabled=0 (or set it to 2 to disable io_uring
testing dynamically if kernel supports)
3) install liburing development package contains liburing.h before building
fstests
For example, to run the tests with loopback partitions:
# xfs_io -f -c "falloc 0 10g" test.img
# xfs_io -f -c "falloc 0 10g" scratch.img
# mkfs.xfs test.img
# losetup /dev/loop0 ./test.img
# losetup /dev/loop1 ./scratch.img
# mkdir -p /mnt/test && mount /dev/loop0 /mnt/test
# mkdir -p /mnt/scratch
The config for the setup above is:
$ cat local.config
export TEST_DEV=/dev/loop0
export TEST_DIR=/mnt/test
export SCRATCH_DEV=/dev/loop1
export SCRATCH_MNT=/mnt/scratch
From this point you can run some basic tests, see 'USING THE FSQA SUITE' below.
Additional Setup
----------------
Some tests require additional configuration in your local.config. Add these
variables to a local.config and keep that file in your workarea. Or add a case
to the switch in common/config assigning these variables based on the hostname
of your test machine. Or use 'setenv' to set them.
Extra TEST device specifications:
- Set TEST_LOGDEV to "device for test-fs external log"
- Set TEST_RTDEV to "device for test-fs realtime data"
- If TEST_LOGDEV and/or TEST_RTDEV, these will always be used.
- Set FSTYP to "the filesystem you want to test", the filesystem type is
devised from the TEST_DEV device, but you may want to override it; if
unset, the default is 'xfs'
Extra SCRATCH device specifications:
- Set SCRATCH_LOGDEV to "device for scratch-fs external log"
- Set SCRATCH_RTDEV to "device for scratch-fs realtime data"
- If SCRATCH_LOGDEV and/or SCRATCH_RTDEV, the USE_EXTERNAL environment
Tape device specification for xfsdump testing:
- Set TAPE_DEV to "tape device for testing xfsdump".
- Set RMT_TAPE_DEV to "remote tape device for testing xfsdump"
variable set to "yes" will enable their use.
- Note that if testing xfsdump, make sure the tape devices have a tape which
can be overwritten.
Extra XFS specification:
- Set TEST_XFS_REPAIR_REBUILD=1 to have _check_xfs_filesystem run
xfs_repair -n to check the filesystem; xfs_repair to rebuild metadata
indexes; and xfs_repair -n (a third time) to check the results of the
rebuilding.
- Set FORCE_XFS_CHECK_PROG=yes to have _check_xfs_filesystem run xfs_check
to check the filesystem. As of August 2021, xfs_repair finds all
filesystem corruptions found by xfs_check, and more, which means that
xfs_check is no longer run by default.
- Set TEST_XFS_SCRUB_REBUILD=1 to have _check_xfs_filesystem run xfs_scrub in
"force_repair" mode to rebuild the filesystem; and xfs_repair -n to check
the results of the rebuilding.
- xfs_scrub, if present, will always check the test and scratch
filesystems if they are still online at the end of the test. It is no
longer necessary to set TEST_XFS_SCRUB.
Tools specification:
- dump:
- Set DUMP_CORRUPT_FS=1 to record metadata dumps of XFS, ext* or
btrfs filesystems if a filesystem check fails.
- Set DUMP_COMPRESSOR to a compression program to compress metadumps of
filesystems. This program must accept '-f' and the name of a file to
compress; and it must accept '-d -f -k' and the name of a file to
decompress. In other words, it must emulate gzip.
- dmesg:
- Set KEEP_DMESG=yes to keep dmesg log after test
- kmemleak:
- Set USE_KMEMLEAK=yes to scan for memory leaks in the kernel after every
test, if the kernel supports kmemleak.
- fsstress:
- Set FSSTRESS_AVOID and/or FSX_AVOID, which contain options added to
the end of fsstresss and fsx invocations, respectively, in case you wish
to exclude certain operational modes from these tests.
- core dumps:
- Set COREDUMP_COMPRESSOR to a compression program to compress crash dumps.
This program must accept '-f' and the name of a file to compress. In
other words, it must emulate gzip.
Kernel/Modules related configuration:
- Set TEST_FS_MODULE_RELOAD=1 to unload the module and reload it between
test invocations. This assumes that the name of the module is the same
as FSTYP.
- Set MODPROBE_PATIENT_RM_TIMEOUT_SECONDS to specify the amount of time we
should try a patient module remove. The default is 50 seconds. Set this
to "forever" and we'll wait forever until the module is gone.
- Set KCONFIG_PATH to specify your preferred location of kernel config
file. The config is used by tests to check if kernel feature is enabled.
- Set REPORT_GCOV to a directory path to make lcov and genhtml generate
html reports from any gcov code coverage data collected by the kernel.
If REPORT_GCOV is set to 1, the report will be written to $REPORT_DIR/gcov/.
Test control:
- Set LOAD_FACTOR to a nonzero positive integer to increase the amount of
load applied to the system during a test by the specified multiple.
- Set TIME_FACTOR to a nonzero positive integer to increase the amount of
time that a test runs by the specified multiple.
- For tests that are a member of the "soak" group, setting SOAK_DURATION
allows the test runner to specify exactly how long the test should continue
running. This setting overrides TIME_FACTOR. Floating point numbers are
allowed, and the unit suffixes m(inutes), h(ours), d(ays), and w(eeks) are
supported.
Misc:
- If you wish to disable UDF verification test set the environment variable
DISABLE_UDF_TEST to 1.
- Set LOGWRITES_DEV to a block device to use for power fail testing.
- Set PERF_CONFIGNAME to a arbitrary string to be used for identifying
the test setup for running perf tests. This should be different for
each type of performance test you wish to run so that relevant results
are compared. For example 'spinningrust' for configurations that use
spinning disks and 'nvme' for tests using nvme drives.
- Set MIN_FSSIZE to specify the minimal size (bytes) of a filesystem we
can create. Setting this parameter will skip the tests creating a
filesystem less than MIN_FSSIZE.
- Set DIFF_LENGTH to "number of diff lines to print from a failed test",
by default 10, set to 0 to print the full diff
- set IDMAPPED_MOUNTS=true to run all tests on top of idmapped mounts. While
this option is supported for all filesystems currently only -overlay is
expected to run without issues. For other filesystems additional patches
and fixes to the test suite might be needed.
- Set REPORT_VARS_FILE to a file containing colon-separated name-value pairs
that will be recorded in the test section report. Names must be unique.
Whitespace surrounding the colon will be removed.
- set CANON_DEVS=yes to canonicalize device symlinks. This will let you
for example use something like TEST_DEV/dev/disk/by-id/nvme-* so the
device remains persistent between reboots. This is disabled by default.
______________________
USING THE FSQA SUITE
______________________
Running tests:
- cd xfstests
- By default the tests suite will run all the tests in the auto group. These
are the tests that are expected to function correctly as regression tests,
and it excludes tests that exercise conditions known to cause machine
failures (i.e. the "dangerous" tests).
- ./check '*/001' '*/002' '*/003'
- ./check '*/06?'
- Groups of tests maybe ran by: ./check -g [group(s)]
See the tests/*/group.list files after building xfstests to learn about
each test's group memberships.
- If you want to run all tests regardless of what group they are in
(including dangerous tests), use the "all" group: ./check -g all
- To randomize test order: ./check -r [test(s)]
- You can explicitly specify NFS/AFS/CIFS/OVERLAY, otherwise
the filesystem type will be autodetected from $TEST_DEV:
- for running nfs tests: ./check -nfs [test(s)]
- for running afs tests: ./check -afs [test(s)]
- for running cifs/smb3 tests: ./check -cifs [test(s)]
- for overlay tests: ./check -overlay [test(s)]
The TEST and SCRATCH partitions should be pre-formatted
with another base fs, where the overlay dirs will be created
The check script tests the return value of each script, and
compares the output against the expected output. If the output
is not as expected, a diff will be output and an .out.bad file
will be produced for the failing test.
Unexpected console messages, crashes and hangs may be considered
to be failures but are not necessarily detected by the QA system.
__________________________
ADDING TO THE FSQA SUITE
__________________________
Creating new tests scripts:
Use the "new" script.
Test script environment:
When developing a new test script keep the following things in
mind. All of the environment variables and shell procedures are
available to the script once the "common/preamble" file has been
sourced and the "_begin_fstest" function has been called.
1. The tests are run from an arbitrary directory. If you want to
do operations on an XFS filesystem (good idea, eh?), then do
one of the following:
(a) Create directories and files at will in the directory
$TEST_DIR ... this is within an XFS filesystem and world
writeable. You should cleanup when your test is done,
e.g. use a _cleanup shell procedure in the trap ... see
001 for an example. If you need to know, the $TEST_DIR
directory is within the filesystem on the block device
$TEST_DEV.
(b) mkfs a new XFS filesystem on $SCRATCH_DEV, and mount this
on $SCRATCH_MNT. Call the the _require_scratch function
on startup if you require use of the scratch partition.
_require_scratch does some checks on $SCRATCH_DEV &
$SCRATCH_MNT and makes sure they're unmounted. You should
cleanup when your test is done, and in particular unmount
$SCRATCH_MNT.
Tests can make use of $SCRATCH_LOGDEV and $SCRATCH_RTDEV
for testing external log and realtime volumes - however,
these tests need to simply "pass" (e.g. cat $seq.out; exit
- or default to an internal log) in the common case where
these variables are not set.
2. You can safely create temporary files that are not part of the
filesystem tests (e.g. to catch output, prepare lists of things
to do, etc.) in files named $tmp.<anything>. The standard test
script framework created by "new" will initialize $tmp and
cleanup on exit.
3. By default, tests are run as the same uid as the person
executing the control script "check" that runs the test scripts.
4. Some other useful shell procedures:
_get_fqdn - echo the host's fully qualified
domain name
_get_pids_by_name - one argument is a process name, and
return all of the matching pids on
standard output
_within_tolerance - fancy numerical "close enough is good
enough" filter for deterministic
output ... see comments in
common/filter for an explanation
_filter_date - turn ctime(3) format dates into the
string DATE for deterministic
output
_cat_passwd, - dump the content of the password
_cat_group or group file (both the local file
and the content of the NIS database
if it is likely to be present)
5. General recommendations, usage conventions, etc.:
- When the content of the password or group file is
required, get it using the _cat_passwd and _cat_group
functions, to ensure NIS information is included if NIS
is active.
- When calling getfacl in a test, pass the "-n" argument so
that numeric rather than symbolic identifiers are used in
the output.
- When creating a new test, it is possible to enter a custom name
for the file. Filenames are in form NNN-custom-name, where NNN
is automatically added by the ./new script as an unique ID,
and "custom-name" is the optional string entered into a prompt
in the ./new script. It can contain only alphanumeric characters
and dash. Note the "NNN-" part is added automatically.
6. Test group membership: Each test can be associated with any number
of groups for convenient selection of subsets of tests. Group names
must be human readable using only characters in the set [:alnum:_-].
Test authors associate a test with groups by passing the names of those
groups as arguments to the _begin_fstest function. While _begin_fstests
is a shell function that must be called at the start of a test to
initialise the test environment correctly, the the build infrastructure
also scans the test files for _begin_fstests invocations. It does this
to compile the group lists that are used to determine which tests to run
when `check` is executed. In other words, test files files must call
_begin_fstest with their intended groups or they will not be run.
However, because the build infrastructure also uses _begin_fstests as
a defined keyword, addition restrictions are placed on how it must be
formatted:
(a) It must be a single line with no multi-line continuations.
(b) group names should be separated by spaces and not other whitespace
(c) A '#' placed anywhere in the list, even in the middle of a group
name, will cause everything from the # to the end of the line to be
ignored.
For example, the code:
_begin_fstest auto quick subvol snapshot # metadata
associates the current test with the "auto", "quick", "subvol", and
"snapshot" groups. Because "metadata" is after the "#" comment
delimiter, it is ignored by the build infrastructure and so it will not
be associated with that group.
It is not necessary to specify the "all" group in the list because that
group is always computed at run time from the group lists.
Verified output:
Each test script has a name, e.g. 007, and an associated
verified output, e.g. 007.out.
It is important that the verified output is deterministic, and
part of the job of the test script is to filter the output to
make this so. Examples of the sort of things that need filtering:
- dates
- pids
- hostnames
- filesystem names
- timezones
- variable directory contents
- imprecise numbers, especially sizes and times
Pass/failure:
The script "check" may be used to run one or more tests.
Test number $seq is deemed to "pass" when:
(a) no "core" file is created,
(b) the file $seq.notrun is not created,
(c) the exit status is 0, and
(d) the output matches the verified output.
In the "not run" case (b), the $seq.notrun file should contain a
short one-line summary of why the test was not run. The standard
output is not checked, so this can be used for a more verbose
explanation and to provide feedback when the QA test is run
interactively.
To force a non-zero exit status use:
status=1
exit
Note that:
exit 1
won't have the desired effect because of the way the exit trap
works.
The recent pass/fail history is maintained in the file "check.log".
The elapsed time for the most recent pass for each test is kept
in "check.time".
The compare-failures script in tools/ may be used to compare failures
across multiple runs, given files containing stdout from those runs.
__________________
SUBMITTING PATCHES
__________________
Send patches to the fstests mailing list at fstests@vger.kernel.org.