Besides being undefined behaviour, signed integer overflow can cause
unexpected comparison results. In the case of network_rank_compare(),
a connected network with rank INT_MAX would cause newly inserted
networks with negative rank to be inserted earlier in the ordered
network list. This is reflected in the GetOrderedMethods() DBus method
as can be seen in the following iwctl output:
[iwd]# station wlan0 get-networks
Network name Security Signal
----------------------------------------------------
BEOLAN 8021x **** }
BeoBlue psk *** } all unknown,
UI_Test_Network psk *** } hence assigned
deneb_2G psk *** } negative rank
BEOGUEST open **** }
> titan psk ****
Linksys05274_5GHz_dmt psk ****
Lyngby-4G-4 5GHz psk ****
If an application has a bug and hangs on SIGTERM this causes
test-runner to hang as well. This is obviously an issue with
the application in question, but test-runner should have a way
of continuing onto the next test rather than hanging.
Instead we can use WNOHANG and a sleep to allow applications
some amount of time to exit, and if they haven't use SIGKILL
instead as well as print an error. Similar to how
wait_for_socket works. The timeout is hard coded to 2 seconds
(100ms sleep + 20 iterations).
Previously iwmon was running per-test, which would jumble any subtests
together into the same log file making it hard to parse. Now create
a separate directory for each subtest and put the monitor log and
pcap there.
Using mac80211_hwsim can sometimes result in out of order messages
coming from the kernel. Since mac80211_hwsim immediately sends out
frames and the kernel keeps command responses in a separate queue,
bad scheduling can result in these messages being out of order.
In some cases we receive Auth/Assoc frames before the response to
our original CMD_CONNECT. This causes autotests to fail randomly,
some more often than others.
To fix this we can introduce a small delay into hwsim. Just a 1ms
delay makes the random failures disappear in the tests. This delay
is also makes hwsim more realistic since actual hardware will always
introduce some kind of delay when sending or receiving frames.
When running test-runner as non-root the environment variables
SUDO_GID/SUDO_UID were unset, causing atoi to segfault. This replaces
atoi with strtol, and checks the existance of SUDO_GID/SUDO_UID
before trying to turn it into an integer. This patch also allows
the uid/gid to be read from the user if running as non-root.
Note: running as non-root does require the users permissions to be
setup properly. Directories and files are created when running with
logging, so if the user running test-runner does not have these
permissions the creation of these files will fail.
The configuration value of iwd_config_dir was defaulting to /etc/iwd
which, in the context of test-runner, is probably not the best idea.
The system may have a main.conf file in /etc/iwd which could cause
tests to fail or behave unexpectedly.
In addition all tests which use iwd_config_dir set it to /tmp anyways.
Because of this, the new default value will be /tmp and no tests will
even need to bother setting this.
The configuration value itself is not being removed because it may be
useful to set arbitrary paths (e.g. /etc/iwd) for example when using
the shell functionality.
This key is special in hostapd, and was being treated as a normal hostapd
config file. This special radius config file needs to be kept unpaired from
any interfaces so now its passed in as a separate argument and appended to
the end of the hostapd execute command.
Tests which use a standalone RADIUS server may crash due to
the wiphy array not taking into account the 'radius_server'
key which is skipped during setup.
The goto was jumping to a label which freed the wiphy list which
had not yet been initialized. This also fixes another similar issue
if chdir fails (in this case tmpfs_extra_stuff would get freed
before being allocated).
Some test cases require (at least with recent hostapd versions) a
stand alone radius server. This is done using driver=none in the
hostapd config file. For this use case hostapd does not need any
radio since its not doing anything wireless related.
Now inside the hw.conf file, under the HOSTAPD group, you can
specify a config file as the value to 'radius_server' key. This
config file will be used without any associated radio when hostapd
is started.
These two processes are executed per-test so they can be passed the
test name and have the logs stored only in tests that actually need
them rather than at the top level.
After changing the valgrind log to --log-fd, it appears that
the log file just gets appended, where before it was overwritten.
This makes test print out all previous tests valgrind results.
Now after printing out the valgrind info we can erase the file
for the next test.
This was a mistake in the original implementation. Since test-runner
is always run from the tools directory using --log with no directory
specified would result in many log files being put into the iwd tree.
Instead we can make --log take a required argument so its more
obvious where all the logs are going to go.
localtime indexes month starting at zero so adding 1 gives us a folder
name with the correct month.
The year is also set as 'years since 1900', so we need to add 1900 to
the year to get the actual year.
Historically if you wanted to see output from a python test you needed
to specify -v pytests. This was also the case if IWD was started from
python.
Nearly every time I run test-runner I would specify "-v iwd,pytests"
only to get the IWD output on these specific tests.
Instead we can special case 'python3' (previously 'pytests') inside
execute_program so that turning on verbosity for 'iwd' also turns it
on for the python tests.
After the logging changes verbose IWD with valgrind did not show any IWD
output. This commit fixes this by checking the verbosity against log_name
rather than argv[0] since log_name has a special case for valgrind/iwd.
The valgrind logic in start_iwd was refactored to only use the --log-fd
option rather than using --log-file in addition to --log-fd.
All the processes verbose output is just written to stdout, and very
hard to parse after the test runs. Having test-runner write the
processes output to separate log files is much nicer to view after
the test runs.
A read/write file system is created which is separate and isolated
from the current FS mount (which mounts the whole host file system).
Doing this requires the user to create a folder somewhere to be used
as the mount point. This folder is where all the log files will end
up after test-runner runs.
Since test-runner must be run as root, care was taken to keep the
log files owned by the user which runs test-runner. The UID/GID
are passed to the VM, and any log files created are chown'ed back
to the user who ran test-runner.
execute_program was refactored, and both verbose and log arguments
were removed. The verbose argument was not really required since
we can do check_verbosity(argv[0]) internally. The 'log' flag was
only used along with --shell, and now the user can simply use
--log instead of dumping /tmp/iwd.log.
You can use this feature by specifying --log[=path] to test-runner.
If no path is provided the current directory will be used. Using
the --log flag will result in all processes being run with the
--verbose flag.
A new folder will be created under the --log path. The folder will
be of the name "run-<year>-<month>-<dat>-<PID>". Under this folder
you will find any global process logs. These are processes that
are only run once for ALL tests (hwsim, ifconfig, dbus etc.). There
will also be folders for specific tests that were run. Inside these
test folders will be logs of processes that are run per-test (iwd,
hostapd, python etc.).
Coverity reported this as a leak, but the test queue is actually
getting freed later and does not need to be freed locally in add_path
This basically reverts c0863e5bc6
The -U parameter only allowed for a list of unit tests to be run.
Most of the time for sanity checking you want to run all the unit
tests so this has been changed to take an optional argument.
Now, the -U flag (by itself) will run all unit tests. Running a
single or list of unit tests can still be achieved by:
--unit-tests=test-eapol,test-crypto
This makes every full test run consistent. The test list was being
stored as a hashmap, which has no been changed to a queue so we can
insert each test sorted.
Specifically, this defines the behavior when --shell is used when no
specific test is specified. In this case test-runner will assume the
'shell' test/sandbox should be used as the test environment as
running all autotests with --shell is not useful or feasable.
