After adding some prints to the driver, it would seem the failure originates
in build_dir/target-mips_24kc_musl/linux-ath79_generic/backports-5.15.58-1/net/mac80211/cfg.c in ieee80211_add_key() :
During FT roaming, wpa_supplicant attempts to set the key before association. This used to be rejected, but as a side effect of my commit 66e67e41 ("mac80211: redesign auth/assoc") the key was accepted causing hardware crypto to not be used for it as the station isn't added to the driver yet. It would be possible to accept the key and then add it to the driver when the station has been added. However, this may run into issues with drivers using the state- based station adding if they accept the key only after association like it used to be. For now, revert to the behaviour from before the auth and assoc change.
Hence it would seem that worst-case may be to loose the hardware key decryption... I guess I'll just see what my router does with the ASSOC condition removed...
--- update 2 ---
Removing the WLAN_STA_ASSOC test indeed get's rid of the error message. Roaming seems to be a bit faster (roaming between AP 1 with this change, and an identical AP 2 without this change). Will keep an eye on this to see if there are negative events popping up in daily use.
Hope this makes it in stable release soon. I've tried 21.x, 22..x and Snaphot builds; different HW, FT over DS/Air, 802.11w v k on/off, .... nothing helped. FT/roaming just does not work.
FT is working here. It’s still ‘regression’ to previous openwrt release. Still seems to be good enough for my use.
I’ll keep investigating in parallel how to do it in accordance with the todo from the cfg.c file
-- update ---
Managed to create a 'release-equivalent' image @ 22.03, so able to use online repository for packages (not sure about kmods, as one may need the entire dependency chain...).
I've switched my Archer C7 v2 routers over to use this version, and would be happy to share if anyone is interested (also to receive suggestions on how to share ).
Thanks @fodiator, I'm definitely interested. Not only in a release equivalent image, but also in gaining some of the knowledge you have in order to get this far. Please share the TODO in the cfg.c file?
It seems to me, the challenge of OpenWRT is understanding where each and every component comes from. Why is this not already sorted in the upstream Linux components that OpenWRT relies on?
/*
* The ASSOC test makes sure the driver is ready to
* receive the key. When wpa_supplicant has roamed
* using FT, it attempts to set the key before
* association has completed, this rejects that attempt
* so it will set the key again after association.
*
* TODO: accept the key if we have a station entry and
* add it to the device after the station.
*/
I've included patch 100-allow_key_without_assoc.patch under
[buildroot]/openwrt/package/kernel/mac80211/patches/subsys
patch content
Used to remove FT key addition failed error on Archer C7 v2
Index: backports-5.15.58-1/net/mac80211/cfg.c
===================================================================
--- backports-5.15.58-1.orig/net/mac80211/cfg.c
+++ backports-5.15.58-1/net/mac80211/cfg.c
@@ -466,11 +466,15 @@ static int ieee80211_add_key(struct wiph
* TODO: accept the key if we have a station entry and
* add it to the device after the station.
*/
- if (!sta || !test_sta_flag(sta, WLAN_STA_ASSOC)) {
+ if (!sta) {
+ sdata_info(sdata, "mwv1 - no sta\n");
ieee80211_key_free_unused(key);
err = -ENOENT;
goto out_unlock;
}
+ if (!test_sta_flag(sta, WLAN_STA_ASSOC)) {
+ sdata_info(sdata, "mwv2 - no sta assoc\n");
+ }
}
switch (sdata->vif.type) {
This will automatically sourced in when (re)-building source.
Also, now have my ath10k driver ftrace enabled to increase my understanding on how to follow the TODO properly.
I'd be happy to supply a patch file, it's code-wise trivial .
Not sure if this will be accepted as a patch for main release, as it may have the side effect of non-hardware decryption.
For what it's worth, I am tracing the hostapd-mac80211 interplay. It's clear that there is provision for fail-retry settings (depending on ASSOC state), which I need to have a closer look at... Will update here once I understand this better.
---- update ----
the patch above removes the need for the retry cycle in my 802.1x + VLAN setup.
@fodiator understand that it might take longer to get this to a patch that will be accepted.
Meantime, how might we replicate what you have done to get it onto our own setups? Some pointers on where to get going would be helpful. Is it sufficient to follow the Developer Guide on the Wiki?
Any thoughts as to why this hasn't been corrected some time ago already?
@andybjackson building the patch should be fairly straightforward.
Assuming your build environment is set up as per Wiki, you can build the release equivalent as described here:
If you copy the patch content from above
and place it in [buildroot]/openwrt/package/kernel/mac80211/patches/subsys
and run make world you should get the sysupgrade.bin under [buildroot]/bin/targets/[path to your device]
You can confirm the patch is part of your build strings [buildroot]/[yourtarget]/[yourlinux]/backports-5.15.58-1/net/mac80211/mac80211.ko | grep mwv .
In the meantime I am still looking to improve the solution by adding the STA to the driver as per the note above. I have been able to get insight into the hostapd code. Hoping to have an implementation within a few weeks.
The reason this code was 'corrected' seems to be - according to a comment - about hardware decryption not being active when the key is added without being associated on some systems.
Does this happen with WPA2/3-PSK as well? Because I certainly have the issue with WPA2/3-EAP. Have currently turned 802.11r off in my production environment.
@fodiator Am now up to speed on building OpenWRT myself. How are you getting on with the patch. Can I help in any way, for example with testing?
PS Some advice for anyone also else getting up to speed with OpenWRT build following the openwrt.org Build System guidance in anticipation of @fodiator's patch:
Use the fastest computer you can - else the process is slow
git tag is a step that I don't yet fully understand. It displays a list of available build choices, but isn't clear how one might choose on. "q" gets you past the step and I use git checkout master rather than the release version so this seems redundant to me.
.