Now that parallel-vm.py is actually stopping VMs as soon as they are not
needed for retries, it is not really an unexpected exit to see a VM exit
while test cases remain in the queue as long as at least that many VMs
remain running. Get rid of confusing 'unexpected exit' status from the
UI in such cases.
Fixes: 4aaddecdd8 ("tests: Handle test retries through the same queue")
Signed-off-by: Jouni Malinen <j@w1.fi>
This removes the separate rerun step from the parallel-vm.py processing
and instead, simply requeues the failed test cases into the same queue
that is used for the initial run. This is simpler and more efficient
since reruns start as soon as any VM is ready for processing them
instead of having to wait for all VMs to complete the first round.
Furthermore, this allows VMs to be stopped sooner when no more test
cases remain and that is helpful especially with the time travel patches
that make the wait-for-next-test step in the VM use all available CPU.
Signed-off-by: Jouni Malinen <j@w1.fi>
When the serial ports are set into raw mode on stdio (fd:0,fd:1)
then Ctrl-C is sort of passed through, but not effective. Request
non-raw mode to avoid that and let us cancel test execution with
Ctrl-C properly (both in parallel-vm.py and vm-run.sh cases).
Note that this requires a currently out-of-tree patch, but so
does the virtual time. If the patch is not applied, the command
line argument is ignored.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
This speeds up test execution significantly by removing unnecessary
waiting for things to happen since the kernel log is allowed to jump
forward whenever there is nothing to do.
Signed-off-by: Jouni Malinen <jouni@codeaurora.org>
This allows UML builds to be used in running user mode without having to
run the full x86 kernel in virtual machine.
Signed-off-by: Jouni Malinen <j@w1.fi>
Print a list of full paths to log files from failed test cases both the
parallel.log and stdout so that they can be easily opened for analysis.
In addition, renumber the VM lines in the <timestamp>-parallel.log to
match the i+1 numbering used in the log directories and UI that is
tracking test execution.
Signed-off-by: Jouni Malinen <jouni@codeaurora.org>
This old interface has been obsoleted and should not have been used
since 2010, so remove testing for it in preparation to dropping the
interface completely from wpa_supplicant.
Signed-off-by: Lubomir Rintel <lkundrak@v3.sk>
This allows unexpected cases to terminate parallel-vm.py without being
hidden by the exception handler.
Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
The result of reading non blocked empty stream is different between
python2 and 3. The python2 sends "[Errno 11] Resource temporarily
unavailable" exception. The python3 could read "None" without
exception, so handle this "None" case as well.
Signed-off-by: Masashi Honma <masashi.honma@gmail.com>
In recent Debian versions, ebtables is an alias managed by
the alternatives(8) mechanism. This means /usr/sbin/ebtables
is a symlink to /etc/alternatives/ebtables, which in turn
links to the real binary.
As we mount a tmpfs over /etc, we cannot access this.
Fix this by bind-mounting the real /etc to /tmp/etc and
adding a symlink from /etc/alternatives to this.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
If telnetd is installed and --telnet <port> is passed on the
vm-run.sh command line, start a telnet server (directly connected
to bash, no login) inside the VM(s) to be able to look into them
when something is wrong. Use a user network in qemu with a single
host forward from the specified port for this, listening only on
'localhost'.
Please note that this provides unauthenticated access to the guest
system from anything that can open a TCP connection on the host system.
The guess system does have access to reading all files on the host that
the user account running kvm has access to (and even write access if the
default ROTAG ,readonly parameter is cleared). In other words, this
option should not be used on any multiuser systems where kvm is run
under user accounts that are not dedicated for testing purposes (i.e.,
do not have access to any files that should not be readable to
everyone).
This needs CONFIG_VIRTIO_NET=y in the guest kernel.
For parallel-vm.py, the --telnet argument specifies the base port
and each VM index (0, 1, ...) is added to it.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Recently, qemu/seabios grew an annoying console/terminal reset,
which also causes my terminal to be left in a state where long
lines don't work well and less gets confused because of this.
Suppress this by suppressing all output from qemu before a new
magic string printed from inside.sh.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
This is useful when running a test multiple times, looking at
log output etc. to not have to pick out the right directory
each and every time.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
parallel-vm.py has obsoleted this a long time ago and there is no need
to maintain two scripts for doing more or less the same thing.
Signed-off-by: Jouni Malinen <j@w1.fi>
The hwsim's start.sh script spawns hostapd process using "sudo".
Since sudo forks a child process, $! holds the pid of sudo itself.
Fix that by storing the PID of the child process instead.
Since in VM "sudo" is replaced with a dummy script, pass an additional
argument to run-all.sh and start.sh scripts to indicate that they are
running inside a VM.
This is needed to fix ap_config_reload and ap_config_reload_file test
cases on some platforms where sudo is apparently not relaying the
signals properly.
Signed-off-by: Andrei Otcheretianski <andrei.otcheretianski@intel.com>
mac80211_hwsim module typically dumps a lot of details into the kernel
message buffer. While it's probably okay in a dedicated VM, it's way too
chatty in other setups.
The kernel allows fine-tuning logging via the dynamic debugging
facility. Let's enable all logging locations in the mac80211_hwsim
module so that we don't loose debugging output when the kernel adopts
the dynamic debug mechanism for the driver.
Signed-off-by: Lubomir Rintel <lkundrak@v3.sk>
I find myself writing a version of this script every now and
then, but there's little point in that - just add one to the
tree so we can use it again.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
We capture the dmesg that contains everything, but if a test
causes a kernel crash we will miss all logging at higher levels
like debug. Change the printk level to catch all of that too.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
This kernel debugging option adds multiple seconds of extra latency to
interface removal operations. While this can be worked around by
increasing timeouts in number of test cases, there does not seem to be
any clean way of working around this for PMKSA cacheching test with
per-STA VIFs (e.g., pmksa_cache_preauth_vlan_used_per_sta_vif).
To avoid unnecessary test failures, remove CONFIG_DEBUG_KOBJECT_RELEASE
from the default config. If someone wants to test with this kernel debug
option, it can be enabled for custom kernel builds while understanding
that it can result in false failure reports and significantly extended
time needed to complete full testing run.
Signed-off-by: Jouni Malinen <j@w1.fi>
This is useful for now since the IPv6 support for proxyarp is not yet
included in the upstream kernel. This allows the IPv4 test cases to pass
with the current upstream kernel while allowing the IPv6 test cases to
report SKIP instead of FAIL.
Signed-off-by: Jouni Malinen <j@w1.fi>
This describes example steps on how to get the VM testing setup with
parallel VMs configured with Ubuntu Server 16.04.1.
Signed-off-by: Jouni Malinen <j@w1.fi>
This file is used only by hostapd_cli and wpa_cli and neither of those
are currently included in code coverage reporting. Avoid dropping the
coverage numbers by code that cannot be reached due to not being
included in the programs that are covered.
Signed-off-by: Jouni Malinen <j@w1.fi>
This reverts commit be9fe3d8af. While I
did manage to complete multiple test runs without failures, it looks
like this change increases full test run duration by about 30 seconds
when using seven VMs. The most visible reason for that seems to be in
"breaking" active scanning quite frequently with the Probe Response
frame coming out about 40 ms (or more) after the Probe Request frame
which is long enough for the station to already have left the channel.
Since this logging change is not critical, it is simplest to revert it
for now rather than make changes to huge number of test cases to allow
more scan attempts to be performed before timing out.
Signed-off-by: Jouni Malinen <j@w1.fi>
When running tests, make printk put all messages, including debug
messages, onto the serial console to go into the console file.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
This is useful for testing CRDA since it means you can use EPATH to
redirect the test scripts to a different crda binary.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
Now that vm-run.sh supports a long list of test cases without crashing
the VM kernel, there is no need to use the "parallel-vm.py -1 1 <tests>"
workaround. Print the re-run example commands with vm-run.sh instead. In
addition, add the --long argument if it was specified for the test run
to avoid skipping test cases in the re-run case.
Signed-off-by: Jouni Malinen <jouni@qca.qualcomm.com>
The script is currently limited by the maximum kernel command line
length and if that's exceeded the kernel panics at boot. Fix this by
writing the arguments to a file and reading it in the VM.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
If /tmp has a relatively small size limit, or multiple people run the
tests on the same machine, using the same output directory can easily
cause problems.
Make the test framework honor the new HWSIM_TEST_LOG_DIR environment
variable to make it easier to avoid those problems.
Signed-off-by: Johannes Berg <johannes.berg@intel.com>
