In order to first integrate frame-xchg some refactoring needed to
be done. First it is useful to allow queueing frames up rather than
requiring the module (p2p, anqp etc) to wait for the last frame to
finish. This can be aided by radio management but frame-xchg needed
some refactoring as well.
First was getting rid of this fx pointer re-use. It looks like this
was done to save a bit of memory but things get pretty complex
needed to check if the pointer is stale or has been reset. Instead
of this we now just allocate a new pointer each frame-xchg. This
allows for the module to queue multiple requests as well as removes
the complexity of needed to check if the fx pointer is stale.
Next was adding the ability to track frame-xchgs by ID. If a module
can queue up multiple requests it also needs to be able to cancel
them individually vs per-wdev. This comes free with the wiphy work
queue since it returns an ID which can be given directly to the
caller.
Then radio management was simply piped in by adding the
insert/done APIs.
These APIs will handle fairness and order in any operations which
radios can only do sequentially (offchannel, scanning, connection etc.).
Both scan and frame-xchg are complex modules (especially scanning)
which is why the radio management APIs were implemented generic enough
where the changes to both modules will be minimal. Any module that
requires this kind of work can push a work item into the radio
management work queue (wiphy_radio_work_insert) and when the work
is ready to be started radio management will call back into the module.
Once the work is completed (and this may be some time later e.g. in
scan results or a frame watch) the module can signal back that the
work is finished (wiphy_radio_work_done). Wiphy will then pop the
queue and continue with the next work item.
A concept of priority was added in order to allow important offchannel
operations (e.g. ANQP) to take priority over other work items. The
priority is an integer, where lower values are of a higher priority.
The concept of priority cleanly solves a lot of the complexity that
was added in order to support ANQP queries (suspending scanning and
waiting for ANQP to finish before connecting).
Instead ANQP queries can be queued at a higher priority than scanning
which removes the need for suspending scans. In addition we can treat
connections as radio management work and insert them at a lower
priority than ANQP, but higher than scanning. This forces the
connection to wait for ANQP without having to track any state.
Many tests force a reauth after the initial connection. When the tests
were written there was no way of ensuring the reauth completed except
waiting (IWD.wait()). Now we can wait for hostapd events in the tests,
which is faster and more reliable than busy waiting.
This test was not reliably passing. Busy waiting is not really reliable,
but in this specific case its really the only option as the blacklist
must expire based on time.
When roaming, iwd tries to scan a limited number of frequencies to keep
the roaming latency down. Ideally the frequency list would come in from
a neighbor report, but if neighbor reports are not supported, we fall
back to our internal database for known frequencies of this network.
iwd tries to keep the number of scans down to a bare minimum, which
means that we might miss APs that are in range. This could happen
because the user might have moved physically and our frequency list is
no longer up to date, or if the AP frequencies have been reconfigured.
If a limited scan fails to find any good roaming candidates, re-attempt
a full scan right away.
If the roam failed and we are no longer connected, station_disassociated
is called which ends up calling station_roam_state_clear. Thus
resetting the variables is not needed. Reflow the logic to make this a
bit more explicit.
If the roam attempt fails, do not reset this to false. Generally this
is set by the fact that we lost beacon and to not attempt neighbor
reports, etc. This hint should be preserved across roam attempts.
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).
frame_xchg_startv was using sizeof(mmpdu) to check the minimum length
for a frame. Instead mmpdu_header_len should be used since this checks
fc.order and returns either 24 or 28 bytes, not 28 bytes always.
This change adds the requirement that the first iovec in the array
must contain at least the first 2 bytes (mmpdu_fc) of the header.
This really shouldn't be a problem since all current users of
frame-xchg put the entire header (or entire frame) into the first
iovec in the array.
explicit_bzero is used in src/p2p.c since commit
1675c765a3 but src/missing.h is not
included, as a result build with uclibc fails on:
/home/naourr/work/instance-0/output-1/per-package/iwd/host/opt/ext-toolchain/bin/../lib/gcc/mips64el-buildroot-linux-uclibc/5.5.0/../../../../mips64el-buildroot-linux-uclibc/bin/ld: src/p2p.o: in function `p2p_connection_reset':
p2p.c:(.text+0x2cf4): undefined reference to `explicit_bzero'
/home/naourr/work/instance-0/output-1/per-package/iwd/host/opt/ext-toolchain/bin/../lib/gcc/mips64el-buildroot-linux-uclibc/5.5.0/../../../../mips64el-buildroot-linux-uclibc/bin/ld: p2p.c:(.text+0x2cfc): undefined reference to `explicit_bzero'
This logic was using l_hashmap_insert, which supports duplicates. Since
some entries were inserted multiple times, they ended up being printed
multiple times. Fix that by introducing a macro that uses
l_hashmap_replace instead.
Right now, if the connection fails, then network always thinks that the
password should be re-asked. Loosen this to only do so if the
connection failed at least in the handshake phase. If the connection
failed due to Association / Authentication timeout, it is likely that
something is wrong with the AP and it can't respond.
Using the new station ANQP watch network can delay the connection
request until after ANQP has finished. Since station may be
autoconnecting we must also add a check in network_autoconnect
which prevents it from autoconnecting if we have a pending Connect
request.
This is to allow network to watch for ANQP activity in order to
fix the race condition between scanning finishing and ANQP finishing.
Without this it is possible for a DBus Connect() to come in before
ANQP has completed and causing the network to return NotConfigured,
when its actually in the process of obtaining all the network info.
The watch was made globally in station due to network not having
a station object until each individual network is created. Adding a
watch during network creation would result in many watchers as well
as a lot of removal/addition as networks are found and lost.
Change signature of network_connect_new_hidden_network to take
reference to the caller's l_dbus_message struct. This allows to
set the caller's l_dbus_message struct to NULL after replying in
the case of a failure.
==201== at 0x467C15: l_dbus_message_unref (dbus-message.c:412)
==201== by 0x412A51: station_hidden_network_scan_results (station.c:2504)
==201== by 0x41EAEA: scan_finished (scan.c:1505)
==201== by 0x41EC10: get_scan_done (scan.c:1535)
==201== by 0x462592: destroy_request (genl.c:673)
==201== by 0x462987: process_unicast (genl.c:988)
==201== by 0x462987: received_data (genl.c:1087)
==201== by 0x45F5A2: io_callback (io.c:126)
==201== by 0x45E8FD: l_main_iterate (main.c:474)
==201== by 0x45E9BB: l_main_run (main.c:521)
==201== by 0x45EBCA: l_main_run_with_signal (main.c:643)
==201== by 0x403B15: main (main.c:512)
Introduce hidden_pending to keep reference to the dbus message object
while we wait for the scan results to be returned while trying to
connect to a hidden network. This simplifies the logic by separating it
into two independent logical units: scanning, connecting and eliminates
a possibility of a memory leak in the case when Network.Connect being
initiated while Station.ConnectHiddenNetwork is in progress.
If a connection is initiated (via dbus) while a quick scan is in
progress, the quick scan will be aborted. In this case,
station_quick_scan_results will always transition to the
AUTOCONNECT_FULL state regardless of whether it should or not.
Fix this by making sure that we only enter AUTOCONNECT_FULL if we're
still in the AUTOCONNECT_QUICK state.
Reported-by: Alvin Šipraga <alsi@bang-olufsen.dk>
If start_scan_next_request() is called while a scan request
(NL80211_CMD_TRIGGER_SCAN) is still running, the same scan request will
be sent again. Add a check in the function to avoid sending a request if
one is already in progress. For consistency, check also that scan
results are not being requested (NL80211_CMD_GET_SCAN), before trying to
send the next scan request. Finally, remove similar checks at
start_next_scan_request() callsites to simplify the code.
This also fixes a crash that occurs if the following conditions are met:
- the duplicated request is the only request in the scan request
queue, and
- both scan requests fail with an error not EBUSY.
In this case, the first callback to scan_request_triggered() will delete
the request from the scan request queue. The second callback will find
an empty queue and consequently pass a NULL scan_request pointer to
scan_request_failed(), causing a segmentation fault.
If scanning is suspended, have scan_common() queue its scan request
rather than issuing it immediately. This respects the assumption that
scans are not requested while sc->suspended is true.
This bug is caused by the following behavior:
1. Start a frame-xchg, wait for callback
2. From callback start a new frame-xchg, same prefix.
The new frame-xchg request will detect that there is a duplicate watch,
which is correct behavior. It will then remove this duplicate from the
watchlist. The issue here is that we are in the watchlist notify loop
from the original xchg. This causes that loop to read from the now
freed watchlist item, causing an invalid read.
Instead of freeing the item immediately, check if the notify loop is in
progress and only set 'id' to zero and 'stale_items' to true. This will
allow the notify loop to finish, then the watchlist code will prune out
any stale items. If not in the notify loop the item can be freed as it
was before.
Don't match the default group's (group_id 0) wdev_id against the
provided wdev_id because the default group can be used on all wdevs and
its wdev_id is 0. Also match individual item's wdev_id in the group to
make up for this although it normally wouldn't matter.
802.11ai mandates that the RSN element is included during authentication
for FILS. This previously was happening by chance since supplicant_ie
was being included with CMD_AUTHENTICATE. This included more than just
the RSNE so that was removed in an earlier commit. Now FILS builds the
RSNE itself and includes this with CMD_AUTHENTICATE.
build_cmd_ft_authenticate and build_cmd_authenticate were virtually
identical. These have been unified into a single builder.
We were also incorrectly including ATTR_IE to every authenticate
command, which violates the spec for certain protocols, This was
removed and any auth protocols will now add any IEs that they require.
In certain cases the autoconnect portion of each subtest was connecting
to the network so fast that the check for obj.scanning was never successful
since IWD was already connected (and in turn not scanning). Since the
autoconnect path will wait for the device to be connected there really isn't
a reason to wait for any scanning conditions. The normal connect path does
need to wait for scanning though, and for this we can now use the new
scan_if_needed parameter to get_ordered_networks.
There is a very common block of code inside many autotests
which goes something like:
device.scan()
condition = 'obj.scanning'
wd.wait_for_object_condition(device, condition)
condition = 'not obj.scanning'
wd.wait_for_object_condition(device, condition)
network = device.get_ordered_network('an-ssid')
When you see the same pattern in nearly all the tests this shows
we need a helper. Basic autotests which merely check that a
connection succeeded should not need to write the same code again
and again. This code ends up being copy-pasted which can lead to
bugs.
There is also a code pattern which attempts to get ordered
networks, and if this fails it scans and tries again. This, while
not optimal, does prevent unneeded scanning by first checking if
any networks already exist.
This patch solves both the code reuse issue as well as the recovery
if get_ordered_network(s) fails. A new optional parameter was
added to get_ordered_network(s) which is False by default. If True
get_ordered_network(s) will perform a scan if the initial call
yields no networks. Tests will now be able to simply call
get_ordered_network(s) without performing a scan before hand.
These values were meant only to force IWD's BSS preference but
since the RSSI's were so low in some cases this caused a roam
immediately after connecting. This patch changes the RSSI values
to prevent a roam from happening.
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.
In this situation the kernel is sending a low RSSI event which netdev
picks up, but since we set netdev->connected so early the event is
forwarded to station before IWD has fully connected. Station then
tries to get a neighbor report, which may fail and cause a known
frequency scan. If this is a new network the frequency scan tries to
get any known frequencies in network_info which will be unset and
cause a segfault.
This can be avoided by only sending RSSI events when netdev->operational
is set rather than netdev->connected.
