Daniel Wagner [Wed, 30 Aug 2023 09:20:18 +0000 (11:20 +0200)]
nvme/{033,034,035,036,37}: drop port handle between passthru target setup and cleanup
The passthru nvmet setup and cleanup helpers are using the port as
handle to track resources.
Instead returning the port from the setup call, we figure out in the
cleanup code which resources have been allocated. This avoids passing
around awkward handles.
The commit enabled the shellcheck warning SC2119 based on the
understanding that bash would pass arguments list "$@" from caller to
callee when functions are called without arguments. However, it was not
correct then merit of SC2119 is not so important. On the other hand,
SC2119 reports false positive warning when bash functions take optional
arguments and they are called without arguments.
Per discussion on the list [1], we concluded that the merit of optional
arguments is larger than that of SC2119. Hence, disable SC2119 again.
Shin'ichiro Kawasaki [Thu, 17 Aug 2023 05:21:30 +0000 (14:21 +0900)]
nvme: remove "udevadm settle" after _nvme_connect_subsys
The previous commit introduced "udevadm settle" command at the end of
_nvme_connect_subsys. Then the command is no longer required after
calling _nvme_connect_subsys in test cases.
Reviewed-by: Daniel Wagner <dwagner@suse.de> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Shin'ichiro Kawasaki [Thu, 10 Aug 2023 11:17:51 +0000 (20:17 +0900)]
nvme/rc: fix nvme device readiness check after _nvme_connect_subsys
The helper function _nvme_connect_subsys() creates a nvme device. It may
take some time after the function call until the device gets ready for
I/O. So it is expected that the test cases call _find_nvme_dev() after
_nvme_connect_subsys() before I/O. _find_nvme_dev() returns the path of
the created device, and it also waits for uuid and wwid sysfs attributes
of the created device get ready. This wait works as the wait for the
device I/O readiness.
However, this wait by _find_nvme_dev() has two problems. The first
problem is missing call of _find_nvme_dev(). The test case nvme/047
calls _nvme_connect_subsys() twice, but _find_nvme_dev() is called only
for the first _nvme_connect_subsys() call. This causes too early I/O to
the device with tcp transport [1]. Fix this by moving the wait for the
device readiness from _find_nvme_dev() to _nvme_connect_subsys(). Also
add --no-wait option to _nvme_connect_subsys(). It allows to skip the
wait in _nvmet_passthru_target_connect() which has its own wait for
device readiness.
The second problem is wrong paths for the sysfs attributes. The paths
do not include namespace index, so the check for the attributes always
fail. Still _find_nvme_dev() does 1 second wait and allows the device
get ready for I/O in most cases, but this is not intended behavior.
Fix this by checking sysfs paths with the namespace index. Get list of
namespace indices for the sub-system and do the check for all indices.
On top of the checks for sysfs attributes, add 'udevadm settle' and a
check for the created device file. These ensures that the create device
is ready for I/O.
Daniel Wagner [Fri, 11 Aug 2023 09:36:13 +0000 (11:36 +0200)]
nvme: Add explicitly host to allow_host list
Only allow to connect to our setup target with the correct hostnqn.
Thus we have to explicitly add the test hostnqn to the test subsysnqn
allow_host list.
In nvme/030 the genctr is updated as side effect by writing the
attr_allow_any_host variable. Though writting to attr_allow_any_host is
not allowed after switching to the allow_host list. We can achieve the
same effect by adding and removing the host to the allow_host list.
Signed-off-by: Daniel Wagner <dwagner@suse.de> Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Daniel Wagner [Fri, 11 Aug 2023 09:36:07 +0000 (11:36 +0200)]
nvme/{041,042,043,044,045,048}: Remove local variable hostnqn and hostid
Commit acc408477e85 ("nvme/{041,042,043,044,045,048}: Use default
hostnqn and hostid") switched the test over to use the default
hostnqn. It missed the change to remove the local variable and
use the def_hostnqn/def_hostid directly.
Signed-off-by: Daniel Wagner <dwagner@suse.de> Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Daniel Wagner [Fri, 11 Aug 2023 09:36:02 +0000 (11:36 +0200)]
nvme/{003,004,005,013,046,049}: Group all variables declarations
Group all variable declarations together at the beginning of the
function. Many of the nvme tests do this but not all. Thus make these
tests use the same style. This allows to spot the odd balls in the
refactoring of the tests.
Signed-off-by: Daniel Wagner <dwagner@suse.de> Reviewed-by: Hannes Reinecke <hare@suse.de> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Shin'ichiro Kawasaki [Fri, 4 Aug 2023 12:20:07 +0000 (21:20 +0900)]
block/004: reset zones of TEST_DEV before fio operation
When test target is a zoned block device with max_active_zones limit
larger than max_open_zones, fio write operation may fail depending on
zone conditions. To avoid the failure, reset zones of the device before
the fio run.
CONTRIBUTING, README: recommend patch post for contributions
There are voices that blktests changes through GitHub pull requests are
not visible to relevant block sub-system kernel developers who
communicate through linux-block mailing list. Some GitHub pull requests
in the past needed additional discussion in the linux-block mailing list
again to confirm that the changes were good for the kernel developers.
To reduce the repeated discussion in the mailing list and GitHub,
clarify that contribution by patch post to linux-block is preferred to
GitHub pull request. Still GitHub pull requests are open mainly for
quick, minor fixes.
The knowledge required to implement blktests test cases are documented
in './new' script which generates test case script template. This is
handy when we implement new test cases. However, it is difficult to
refer the documentation when we do not implement new test cases.
To help to refer the documentation, add a pointer to it and describe
what it documents.
It was guided for test statements in blktests scripts to use double
square brackets [[ ]] form instead of single square brackets [ ] form.
However, a number of patch contributors use [ ]. It is not productive to
replace them with [[ ]] and discouraging contributions. To avoid those
drawbacks, allow both forms and still keep [[ ]] as the preferred form.
While at it, fix a typo.
Theodore Ts'o [Thu, 3 Aug 2023 21:20:52 +0000 (17:20 -0400)]
src/Makefile: fix static linking of miniublk
When using static linking, the libraries need to be placed after the
.o or .c files so they are searched. Otherwise, the build will fail:
cc -O2 -Wall -Wshadow -DHAVE_LINUX_BLKZONED_H -D_GNU_SOURCE -lpthread -luring -o miniublk miniublk.c
/bin/ld: /tmp/ccfjiUvb.o: in function `ublk_ctrl_init':
miniublk.c:(.text+0xaeb): undefined reference to `io_uring_queue_init_params'
/bin/ld: /tmp/ccfjiUvb.o: in function `ublk_queue_deinit':
miniublk.c:(.text+0xb73): undefined reference to `io_uring_unregister_ring_fd'
/bin/ld: miniublk.c:(.text+0xb85): undefined reference to `io_uring_unregister_files'
/bin/ld: /tmp/ccfjiUvb.o: in function `ublk_io_handler_fn':
miniublk.c:(.text+0xd62): undefined reference to `io_uring_queue_init_params'
/bin/ld: miniublk.c:(.text+0xd77): undefined reference to `io_uring_register_ring_fd'
/bin/ld: miniublk.c:(.text+0xd8c): undefined reference to `io_uring_register_files'
/bin/ld: miniublk.c:(.text+0xe94): undefined reference to `io_uring_submit_and_wait_timeout'
/bin/ld: /tmp/ccfjiUvb.o: in function `__ublk_ctrl_cmd':
miniublk.c:(.text+0x14a6): undefined reference to `io_uring_submit'
/bin/ld: miniublk.c:(.text+0x1533): undefined reference to `__io_uring_get_cqe'
collect2: error: ld returned 1 exit status
make: *** [Makefile:61: miniublk] Error 1
Bart Van Assche [Mon, 12 Jun 2023 19:52:28 +0000 (19:52 +0000)]
common/scsi_debug: Use _unload_module() instead of modprobe -r
It may happen that unloading the scsi_debug kernel module fails a few
times before it finally succeeds. Fix this by using _unload_module
instead of modprobe -r.
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
The test cases block/004, zbd/005 and zbd/006 do write to zoned block
devices which require mq-deadline scheduler. When TEST_DEV is a bio
based device-mapper, mq-deadline scheduler shall be set to destination
devices. For this purpose, call _test_dev_set_scheduler instead of
_test_dev_queue_set.
The test case block/005 requires TEST_DEV to have the queue/scheduler
sysfs attribute. However, kernel no longer provides the attribute since
version 6.5. Add the helper function _require_test_dev_sysfs to check
the requirement.
common/{rc,dm}: introduce functions to set scheduler of dm destinations
Since version v6.5, kernel no longer provides scheduler sysfs attribute
of bio based device-mapper. If TEST_DEV is a bio based device-mapper and
the test case requires specific scheduler, it is required to modify
sysfs attributes of destination devices instead of TEST_DEV.
To set scheduler to the sysfs attribute of destination devices, add the
helper function _dm_destination_dev_set_scheduler. It saves the original
scheduler value in the associative array SYSFS_QUEUE_SAVED so that it is
restored at each test case end. Also add _test_dev_set_scheduler which
sets scheduler regardless whether TEST_DEV is a bio based device-mapper
or not.
Current implementation saves sysfs attributes under queue/ directory
in the associative array TEST_DEV_QUEUE_SAVED using attribute file names
as keys. The saved attributes are restored after each test case run.
When TEST_DEV is a device-mapper, this attribute restore does not cover
attributes of device-mapper destination devices. As a preparation to
cover the destination devices, use path of the attributes as keys
instead of file names. Also rename the associative array
TEST_DEV_QUEUE_SAVED to SYSFS_QUEUE_SAVED.
common/ublk: avoid modprobe failure for built-in ublk_drv
When ublk_drv driver is not a loadable module but a built-in module,
modprobe for the driver fails in _init_ublk. This results in unexpected
test case skips with the message "requires ublk_drv".
To not skip the test cases with built-in ublk_drv, call modprobe only
when the driver is loadable and its module file exists. Also, do not set
SKIP_REASONS to handle modprobe failure as test case failure.
Fixes: 840ccf1fc33e ("block/033: add test to cover gendisk leak") Reviewed-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Daniel Wagner [Mon, 3 Jul 2023 09:16:47 +0000 (11:16 +0200)]
nvme/{041,042,043,044,045,048}: Use default hostnqn and hostid
The host might have enabled the udev/systemd auto connect feature.
This disturbs the blktests for the fc transport. nvme-cli is able
to distinguish between the different invocations via the --context
option.
Instead creating random generated IDs or reuse the hostnqn, it's safer
to always use the same hostnqn to reduce the risk that the matching
logic doesn't work. For this purpose, drop references to
/etc/nvme/hostid and /etc/nvme/hostnqn.
Signed-off-by: Daniel Wagner <dwagner@suse.de>
[Shin'ichiro: resolved conflict and noted /etc/nvme/* in commit message] Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
When the host has enabled the udev/systemd autoconnect services for the
fc transport it interacts with blktests and make tests break.
nvme-cli learned to ignore connects attemps when using the --context
command line option paired with a volatile configuration. Thus we can
mark all the resources created by blktests and avoid any interaction
with the systemd autoconnect scripts.
Only enabled this for the fc transport.
Signed-off-by: Daniel Wagner <dwagner@suse.de> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Daniel Wagner [Mon, 3 Jul 2023 09:16:45 +0000 (11:16 +0200)]
nvme/048: Check for queue count check directly
The test monitored the state changes live -> resetting -> connecting ->
live, to figure out the queue count change was successful.
The fc transport is reconnecting very fast and the state transitions
are not observed by the current approach.
So instead trying to monitor the state changes, let's just wait for the
live state and the correct queue number.
As queue count is depending on the number of online CPUs we explicitly
use 1 and 2 for the max_queue count. This means the queue_count value
needs to reach either 2 or 3 (admin queue included).
Signed-off-by: Daniel Wagner <dwagner@suse.de> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Alan Adamson [Mon, 3 Jul 2023 18:25:49 +0000 (11:25 -0700)]
nvme/043: use a valid hostkey
The test does not generate a hostkey and uses a NULL when setting
up and connecting to a target. The test passes, but a valid hotkey
should be used. This patch generates a valid hostkey and uses it
when setting up the target.
Yi Zhang [Fri, 30 Jun 2023 12:00:28 +0000 (20:00 +0800)]
zbd/009: skip have_good_mkfs_btrfs when mkfs.btrfs not avaiable
When mkfs.btrfs is not available, have_good_mkfs_btrfs in the test case
zbd/009 reports a bash script error. Skip have_good_mkfs_btrfs to avoid
the error.
tests/zbd/009: line 24: mkfs.btrfs: command not found
zbd/009 (test gap zone support with BTRFS) [not run]
driver btrfs is not available
mkfs.btrfs is not available
Shin'ichiro Kawasaki [Wed, 28 Jun 2023 12:43:43 +0000 (21:43 +0900)]
nvme/rc: specify hostnqn and hostid to nvme discover and connect
After the kernel commit ae8bd606e09b ("nvme-fabrics: prevent overriding
of existing host"), 'nvme discover' and 'nvme connect' commands will
fail in case hostid and hostnqn don't maintain 1:1 mapping in the
system. This caused failure of many test cases in the nvme group with
kernel messages "nvme_fabrics: found same hostid XXX but different
hostnqn YYY".
To avoid the failures, specify valid hostnqn and hostid to the nvme
commands always. Prepare def_hostnqn and def_hostid even when
/etc/nvme/hostnqn or /etc/nvme/hostid is not available. Using these
values, add --hostnqn and --hostid options to the nvme commands in
_nvme_discover() and _nvme_connect_subsys().
Nitesh Shetty [Wed, 14 Jun 2023 17:06:42 +0000 (22:36 +0530)]
block/034: Test memory is released by null-blk driver with memory_backed=1
This tests memory leak, by loading/unloading nullblk driver.
Steps:
1. Load nullblk driver with memory_backed=1
2. "dd" of 50M
3. Unload null-blk driver
We do it for 5 iterations to avoid any noise.
Shin'ichiro Kawasaki [Fri, 26 May 2023 04:58:43 +0000 (13:58 +0900)]
block/011: recover test target NVME device capacity
The test case runs fio while disabling and enabling PCI device of the
test target block device. When the block device is a NVME PCI device,
the test triggers NVME controller reset. When an error happens during
the reset, NVME PCI driver marks zero capacity for the device. This
zero capacity device causes failures of the following test cases.
To avoid the failures by zero device capacity, check the capacity at the
test end. If it is zero, remove the device and rescan PCI bus to detect
the device again, and regain the correct capacity.
Shin'ichiro Kawasaki [Fri, 26 May 2023 04:58:42 +0000 (13:58 +0900)]
block/011: skip when mounted block devices are affected
The test case disables PCI device of the test target block device. When
the PCI device has other block devices mounted, those block devices are
disabled also. If the mounted device is the system disk, the test screws
up the system. To avoid such dangerous operation, check if the target
PCI device has mounted block devices. In that case, skip the test.
Daniel Wagner [Tue, 6 Jun 2023 07:54:43 +0000 (09:54 +0200)]
nvmeof-mp: Remove nvmeof-mp tests
The nvmeof-mp are designed to test a NVMe multipathing setup using
dm_multipath, which was a configuration RHEL shipped. In the meantime,
RHEL switched also to NVMe native multipathing which is tested with the
nvme tests.
The nvme tests cover all the use cases which are testes by nvmeof-mp
including the transport failure, see nvme/040.
During LSFMM 2023 conference it was decided it is better to drop these
tests and reduce the maintenance burden on blktests.
Cc: John Meneghini <jmeneghi@redhat.com> Signed-off-by: Daniel Wagner <dwagner@suse.de> Reviewed-by: John Meneghini <jmeneghi@redhat.com> Reviewed-by: Johannes Thumshirn <johannes.thumshirn@wdc.com> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Yang Xu [Wed, 31 May 2023 01:07:01 +0000 (09:07 +0800)]
nvme/{016,017}: use _check_genctr instead of _filter_discovery
Since commit 328943e3 ("Update tests for discovery log page changes"),
blktests also include the discovery subsystem itself. But it
will lead these cases fails on older nvme-cli system.
To avoid this, like nvme/002, use _check_genctr to check instead of
comparing many discovery Log Entry output.
Signed-off-by: Yang Xu <xuyang2018.jy@fujitsu.com> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Ziyang Zhang [Wed, 24 May 2023 08:55:41 +0000 (16:55 +0800)]
tests: Add ublk tests
It is very important to test ublk crash handling since the userspace
part is not reliable. Especially we should test removing device, killing
ublk daemons and user recovery feature.
Add five new tests for ublk to cover these cases.
Signed-off-by: Ziyang Zhang <ZiyangZhang@linux.alibaba.com> Reviewed-by: Ming Lei <ming.lei@redhat.com> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Daniel Wagner [Thu, 11 May 2023 14:09:53 +0000 (16:09 +0200)]
nvme/{016,017}: Make the number iterations configurable
Some tests hard code high values of iterations. This makes them run
relatively long compared to the other tests. Introduce a new environment
variable nvme_num_iter to allow tune the runtime.
Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> Signed-off-by: Daniel Wagner <dwagner@suse.de> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Daniel Wagner [Thu, 11 May 2023 14:09:51 +0000 (16:09 +0200)]
nvme/{045,047}: Calculate IO size for random fio jobs
_nvme_calc_run_io_size() returns the jobs size for _run_fio_rand_io()
function. The jobs size is the size per job, thus we have to divide
through the number of CPUs.
_xfs_run_fio_verify_io() is replaced with _run_fio_rand_io() because the
former has a minimum nvme_img_size of 350M. Both tests nvme/{045,047}
just want some IO to verify that the path is working. Thus reduce the
min nmve_img_size requirement switch to _run_fio_rand_io()
Signed-off-by: Daniel Wagner <dwagner@suse.de> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Daniel Wagner [Thu, 11 May 2023 14:09:50 +0000 (16:09 +0200)]
nvme: Add minimal test image size requirement
Some tests need a minimal test image size to work correctly. Thus add a
helper to check the size and update these tests accordingly.
The image minimum is 4M because some of the test have hard coded values.
All tests which use the xfs fio verification job have a minimum
requirement of 350M impossed by the xfs filesystem.
Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> Signed-off-by: Daniel Wagner <dwagner@suse.de> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Daniel Wagner [Thu, 11 May 2023 14:09:49 +0000 (16:09 +0200)]
nvme: Make test image size configurable
The reduce the overall runtime of the testsuite by making the default
size of the test image small. For verification jobs, the default can be
overwriten via the newly introduced nvme_img_size environment variable.
Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> Signed-off-by: Daniel Wagner <dwagner@suse.de> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Daniel Wagner [Thu, 11 May 2023 14:09:46 +0000 (16:09 +0200)]
common/xfs: Limit fio size job to fit into xfs fs
The usable capacity of the filesystem is less than the raw
partition/device size due to the additional meta/log data.
Ensure that the job size for fio is not exceeding the limits.
Because we have hard coded the path where we mount the filesystem
and don't want to expose this, we just update max size inside
_xfs_run_fio_verify_io(). No need to leak this into the caller.
Daniel Wagner [Thu, 11 May 2023 14:09:45 +0000 (16:09 +0200)]
common/xfs: Make size argument optional for _xfs_run_fio_verify_io
Make the size argument optional by reading the filesystem info. The
caller doesn't have to guess (or calculate) how big the max IO size.
The log data structure of XFS is reducing the capacity.
Daniel Wagner [Thu, 11 May 2023 14:09:43 +0000 (16:09 +0200)]
nvme/rc: Auto convert test device size info
Introduce a convert_to_mb() helper which converts the size argument
to MBytes and use in test device require function. This makes it
possible to use user input strings in future.
Reviewed-by: Hannes Reinecke <hare@suse.de> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> Signed-off-by: Daniel Wagner <dwagner@suse.de> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
block/{001,002,027}: allow to run with built-in scsi_debug and sd_mod
To allow the test cases run with build-in scsi_debug, replace
'_have_module scsi_debug' with _have_scsi_debug, and replace
_init_scsi_debug with _configure_scsi_debug.
Also to allow block/001 run with built-in sd_mod, replace
'_have_module sd_mod' with '_have_kernel_option BLK_DEV_SD'. When sd_mod
driver is built-in, /sys/module/sd_mod directory is not created. Then
_have_driver() can not detect availability of the driver. Instead, refer
the kernel config to check availability of the driver.
scsi/{004,005}: allow to run with built-in scsi_debug
To allow the test case run with build-in scsi_debug, replace
'_have_module scsi_debug' with _have_scsi_debug, and replace
_init_scsi_debug with _configure_scsi_debug.
To set up scsi_debug devices with built-in scsi_debug module, introduce
a new helper function _configure_scsi_debug. It works in similar manner
as _init_scsi_debug which sets up scsi_debug devices with loadable
scsi_debug module.
_configure_scsi_debug takes parameters of scsi_debug devices in format
of 'key=value' as its arguments. It calls another new helper function
_scsi_debug_key_path to find sysfs files corresponding to the keys in
/sys/module/scsi_debug/parameters or /sys/bus/pseudo/drivers/scsi_debug.
When the file is found, write the value to the file.
Before setting the parameters through sysfs files, save current values
of scsi_debug parameters in ORIG_SCSI_DEBUG_PARAMS. Use the saved values
to restore parameters in _exit_scsi_debug. Do this value restore not
only for the parameters modified in _configure_scsi_debug but also for
the parameters modified by test cases.
Among the parameters, 'add_host' has special meaning to add new hosts.
Then handle it separately so that it is set at last in
_configure_scsi_debug, and restored at first in _exit_scsi_debug.
Also record the hosts which exist before _configure_scsi_debug in the
array ORIG_SCSI_DEBUG_HOSTS. Those hosts should not be used for testing,
then do not add them to SCSI_DEBUG_HOSTS.
common/scsi_debug: factor out _setup_scsi_debug_vars
As a preparation to introduce a new helper function to configure
scsi_debug device with built-in scsi_debug module, factor out a part
of _init_scsi_debug to a new function _setup_scsi_debug_vars.
As a preparation to adapt test cases to built-in scsi_debug module, re-
define the _have_scsi_debug function. It checks that the scsi_debug
module is built as a loadable module. Modify it to check that the
scsi_debug module is available as built-in module or loadable module.
Also replace all _have_scsi_debug calls in test cases with
"_have_module scsi_debug" so that the change of _have_scsi_debug do not
affect the test cases. Following commits will modify them to call
_have_scsi_debug, only for test cases ready to run with built-in
scsi_debug.
common/rc: skip module file check if modules path does not exist
When all of the kernel modules are built-in, /lib/modules/*/kernel path
may not exist. In this case, check for the path results in failure. Skip
the check when the path does not exist.
When kernel version numbers have postfix letters, _have_fio_ver fail to
parse the version. For example, uname -r returns "6.3.0+", it handles
"0+" as a number and fails to parse. Fix it by dropping all letters
other than numbers or period.
Akinobu Mita [Sun, 16 Apr 2023 04:37:44 +0000 (13:37 +0900)]
block/{014,015,030}: don't require modular null_blk for fault-injection
This blktests change changes null_blk fault-injection settings to be
configured via configfs instead of module parameters.
This allows null_blk fault-injection tests to run even if the null_blk is
built-in the kernel and not built as a module.
If the null_blk does not yet support configuring fault-injection via
configfs, fall back to set up with module parameter.
nvme/039: avoid failure by error message rate limit
The test case nvme/039 tests that expected error messages are printed
for errors injected to the nvme driver. However, the test case fails by
chance when previous test cases generate many error messages. In this
case, the kernel function pr_err_ratelimited() may suppress the error
messages that the test case expects. Also, it may print messages that
the test case does not expect, such as "blk_print_req_error: xxxx
callbacks suppressed".
To avoid the failure, make two improvements for the test case. Firstly,
wait DEFAULT_RATE_LIMIT seconds at the beginning of the test to ensure
the expected error messages are not suppressed. Secondly, exclude the
unexpected message for the error message check. Introduce a helper
function last_dmesg() for the second improvement.
User can communicate to NVMe char device (/dev/ngXnY) using the
uring-passthrough interface. This test exercises some of these
communication pathways, using the 'io_uring_cmd' ioengine of fio.
Daniel Wagner [Wed, 29 Mar 2023 09:02:02 +0000 (11:02 +0200)]
nvme/047: Test different queue types for fabrics transports
Test if the fabrics transports are handling different queues types
correctly.
Currently, only TCP and RDMA from the fabric transport family support
this feature, so limit the test to these two.
We also issue some I/O to make sure that not just the plain connect
works. For this we have to use a file system which supports direct I/O
and hence we use a device backend.
Shin'ichiro Kawasaki [Wed, 1 Mar 2023 08:03:01 +0000 (17:03 +0900)]
src/Makefile: fix number sign handling in macro
GNU make version 4.3 introduced a backward-incompatible change. The
number sign '#' now should not have preceding backslash in a macro [1].
To make macros with number signs work regardless of make versions,
assign the number sign to a variable.
Ming Lei [Fri, 24 Feb 2023 12:45:02 +0000 (20:45 +0800)]
block/033: add test to cover gendisk leak
So far only sync ublk removal is supported, and the device's
last reference is dropped in gendisk's ->free_disk(), so it
can be used to test gendisk leak issue.
Signed-off-by: Ming Lei <ming.lei@redhat.com> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Ming Lei [Fri, 24 Feb 2023 12:45:01 +0000 (20:45 +0800)]
src: add mini ublk source code
Prepare for adding ublk related test:
1) ublk delete is sync removal, this way is convenient to
blkg/queue/disk instance leak issue
2) mini ublk has two builtin target(null, loop), and loop IO is
handled by io_uring, so we can use ublk to cover part of io_uring
workloads
3) not like loop/nbd, ublk won't pre-allocate/add disk, and always
add/delete disk dynamically, this way may cover disk plug & unplug
tests
4) ublk specific test given people starts to use it, so better to
let blktest cover ublk related tests
Add mini ublk source for test purpose only, which is easy to use:
./miniublk add -t {null|loop} [-q nr_queues] [-d depth] [-n dev_id]
default: nr_queues=2(max 4), depth=128(max 128), dev_id=-1(auto allocation)
-t loop -f backing_file
-t null
./miniublk del [-n dev_id] -a
-a delete all devices, -n delete specified device
./miniublk list [-n dev_id] -a
-a list all devices, -n list specified device, default -a
miniublk depends on liburing 2.2, adds HAVE_LIBURING for checking if
liburing 2.2 exists; also add HAVE_UBLK_HEADER for checking ublk kernel
UAPI header exits. If either of two dependencies can't be met, simply
ignore miniublk target.
Also v6.0 is the 1st linux kernel release with ublk.
Signed-off-by: Ming Lei <ming.lei@redhat.com> Reviewed-by: Chaitanya Kulkarni <kch@nvidia.com> Signed-off-by: Shin'ichiro Kawasaki <shinichiro.kawasaki@wdc.com>
Shin'ichiro Kawasaki [Mon, 13 Feb 2023 02:00:08 +0000 (11:00 +0900)]
check, common/rc: support normal user privilege
To run commands with normal user privilege, add a new config variable
NORMAL_USER and two helper functions _run_user and _require_normal_user.
The user name specified to NORMAL_USER is used to run the commands
specified to _run_user. The test cases which require NORMAL_USER shall
call _require_normal_user to ensure the NORMAL_USER is valid.
Ming Lei [Wed, 8 Feb 2023 01:02:35 +0000 (09:02 +0800)]
block/032: add test to cover umount one deleted disk
disk can be disappear any time because of error handling, when
it is usually being mounted. Make sure umount can be done successfully
after disk deleting is done from error handling.
Hannes Reinecke [Mon, 6 Feb 2023 14:51:42 +0000 (15:51 +0100)]
Update tests for discovery log page changes
With TP8013 the discovery log includes now a record for the
discovery subsystem itself, so we need to update the blktests
to be aware of the new discovery log page layout.
Bart Van Assche [Thu, 2 Feb 2023 18:20:39 +0000 (10:20 -0800)]
src/discontiguous-io.cpp: Fix the build
Fix the following build error:
g++ -O2 -std=c++11 -Wall -Wextra -Wshadow -Wno-sign-compare -Werror -DHAVE_LINUX_BLKZONED_H -o discontiguous-io discontiguous-io.cpp
discontiguous-io.cpp: In function ‘void dumphex(std::ostream&, const void*, size_t)’:
discontiguous-io.cpp:92:24: error: ‘uintptr_t’ was not declared in this scope
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
projects / users / sagi / blktests.git / log