Skip to content

Zero 2 W loses Wi-Fi on WPA2/WPA3 transition AP after wpasupplicant 2:2.10-24+rpt1: SAE now advertised, rpi-brcmfmac.conf misses SAE_EXT (bit 25) #7634

Open

Description

@DK5EN

Describe the bug

Since wpasupplicant 2:2.10-24+rpt1 (archive.raspberrypi.com, trixie, 2026-09-17) a Raspberry Pi Zero 2 W loses Wi-Fi permanently on a WPA2/WPA3 transition-mode access point. The link drops at the moment the package upgrade restarts wpa_supplicant.service and does not come back after a reboot. On a headless box this is a full loss of access.

Mechanism (all three parts verified, details below):

  1. 2:2.10-24+rpt1 carries one patch, 0020-dbus-Fix-reporting-of-the-SAE-KeyMgmt.patch (upstream 26c7f1bc, Debian #1147955). wpa_supplicant now lists sae in its D-Bus KeyMgmt capabilities when the driver sets NL80211_FEATURE_SAE.
  2. brcmfmac in rpi-6.18.y sets NL80211_FEATURE_SAE when BRCMF_FEAT_SAE_EXT is enabled (cfg80211.c, brcmf_setup_wiphy). The shipped /usr/lib/modprobe.d/rpi-brcmfmac.conf (feature_disable=0x282000, firmware-nonfree 348a967) clears FWSUP (bit 13), SAE (bit 19) and DUMP_OBSS (bit 21), but not SAE_EXT, which is bit 25 in the 6.18 driver's BRCMF_FEAT_LIST. So the driver still advertises SAE: wpa_cli get_capability auth_alg prints OPEN SHARED LEAP SAE on the Zero 2 W.
  3. NetworkManager 1.52.1 (trixie) appends SAE to key_mgmt for every wpa-psk profile in station mode as soon as the supplicant reports SAE, PMF and BIP (nm-supplicant-config.c). wpa_supplicant 2.10 prefers SAE over PSK when both are offered, the BCM43430/1 firmware (7.45.96.s1, 2023-06-14) does not complete the external-auth SAE handshake, and NetworkManager never falls back to WPA2. The upstream NetworkManager fix that makes pmf=disable opt out in station mode (b00c6749, 2025-06-05) is only in NetworkManager 1.56+, so a trixie Pi has no configuration-level escape.

Before the upgrade, the old supplicant only reported sae from capa.key_mgmt, which brcmfmac never populates, so NetworkManager stayed on WPA-PSK WPA-PSK-SHA256 FT-PSK and everything worked. This is the same failure shape as #4976 (closed), where the community workaround brcmfmac.feature_disable=0x82000 was the fix, and the Zero 2 W data point that was missing in #6130 / #7610.

Steps to reproduce the behaviour

  1. Raspberry Pi Zero 2 W, Raspberry Pi OS trixie (Lite), kernel 6.18.50+rpt-rpi-v8, NetworkManager 1.52.1-1+rpt4, firmware-brcm80211 1:20260519-1~bpo13+1+rpt1, connected via NetworkManager (key-mgmt=wpa-psk) to an AP in WPA2/WPA3 transition mode (Netgear Orbi here; nmcli dev wifi list shows RSN-FLAGS pair_ccmp group_ccmp psk sae).
  2. apt-get upgrade to wpasupplicant 2:2.10-24+rpt1.
  3. The postinst restarts wpa_supplicant.service; the station never re-associates. Reboot: same.
  4. Roll back to Debian's wpasupplicant 2:2.10-24 and hold it: NetworkManager logs Config: added 'key_mgmt' value 'WPA-PSK WPA-PSK-SHA256 FT-PSK' and the association completes within a second.

Control: a second Zero 2 W on the same SSID running bookworm (wpasupplicant 2:2.10-12+deb12u3, NetworkManager 1.42.4) associates with key_mgmt=WPA2-PSK, pmf=1.

Device (s)

Raspberry Pi Zero 2 W

System

Raspberry Pi Zero 2 W Rev 1.0
Debian GNU/Linux 13 (trixie), 6.18.50+rpt-rpi-v8
brcmfmac: Firmware: BCM43430/1 wl0: Jun 14 2023 07:27:45 version 7.45.96.s1 (gf031a129) FWID 01-70bd2af7 es7
wpasupplicant 2:2.10-24+rpt1 (broken) / 2:2.10-24 (works)
network-manager 1.52.1-1+rpt4
firmware-brcm80211 1:20260519-1~bpo13+1+rpt1, /usr/lib/modprobe.d/rpi-brcmfmac.conf = options brcmfmac roamoff=1 feature_disable=0x282000

Logs

Last journal lines before the link died (shipped to a peer via systemd-journal-upload, the box itself has volatile logs):

22:26:53 sudo: COMMAND=/usr/bin/bash /tmp/bootstrap/mcapp.sh --dev --tag v2.0.10-dev.1   (runs apt-get upgrade)
22:27:42 /usr/sbin/wpa_supplicant replaced (inode ctime), /var/lib/dpkg/info/wpasupplicant.list 22:27:43
22:27:47 systemd[1]: Stopping wpa_supplicant.service - WPA supplicant...
         (nothing after this; NetworkManager's /var/lib/NetworkManager/timestamps kept the profile at 22:27:48 across the next two boots)

With the 2:2.10-24 binary, same boot config:

NetworkManager[535]: device (wlan0): state change: need-auth -> prepare
NetworkManager[535]: Config: added 'key_mgmt' value 'WPA-PSK WPA-PSK-SHA256 FT-PSK'
wpa_supplicant[536]: wlan0: Associated with 5a:af:97:2e:2b:8b
wpa_supplicant[536]: wlan0: CTRL-EVENT-CONNECTED - Connection to 5a:af:97:2e:2b:8b completed

Capabilities on the Zero 2 W (2:2.10-24, i.e. what the driver exposes):

$ sudo wpa_cli get_capability key_mgmt
NONE IEEE8021X WPA-EAP WPA-PSK WPA-EAP-SUITE-B OWE DPP FT-PSK FT-EAP
$ sudo wpa_cli get_capability auth_alg
OPEN SHARED LEAP SAE

The +rpt1 patch makes the second line leak into the first over D-Bus.

Additional context

Suggested fixes, any one of which breaks the chain:

  • firmware-nonfree: extend rpi-brcmfmac.conf to also clear SAE_EXT, i.e. feature_disable=0x2282000 on 6.18 kernels, at least for the 43430 family. Note the mask is a raw bit index into BRCMF_FEAT_LIST and shifts when the enum grows; SAE_EXT did not exist when 0x282000 was chosen.
  • kernel: do not set NL80211_FEATURE_SAE for chips/firmware where external SAE authentication is known not to work (43430/1 with 7.45.96.s1).
  • wpa packaging: hold the D-Bus SAE-reporting patch back until the driver side no longer advertises SAE on those chips, or ship it together with the modprobe change.
  • network-manager (trixie): backport b00c6749 so wifi-sec.pmf=disable opts a wpa-psk profile out of SAE transition mode in station mode. That would at least give a configuration-level workaround.

Current workaround on the affected box: apt-mark hold wpasupplicant at 2:2.10-24, plus an apt pin. Alternatively brcmfmac.feature_disable=0x2282000 on the kernel command line, per #4976, which I have not tested on this box.

Related: #4976 (identical symptom, Zero 2 W, closed with the feature_disable workaround), #6130, #7610 (SAE capability discussion, 2026-09-10, where a Zero 2 W data point was missing), #7528 (the issue the Debian bug references).

Activity

  1. Vandroiy78 commented on Sep 19, 2026

    @Vandroiy78

    I can reproduce this on my Raspberry Pi 4 2GB running Debian 13 'moode'.
    Linux moode 6.18.50+rpt-rpi-v8 #1 SMP PREEMPT Debian 1:6.18.50-1+rpt1 (2026-09-11) aarch64 GNU/Linux

    After downgrading:
    $ journalctl -b -u NetworkManager | grep key_mgmt
    Sep 19 07:57:35 moode NetworkManager[577]: [1789797455.3377] Config: added 'key_mgmt' value 'WPA-PSK WPA-PSK-SHA256 FT-PSK'

  2. 1p0d commented on Sep 19, 2026

    @1p0d

    Also affected: Raspberry Pi 4 Model B Rev 1.4 (BCM4345/6, i.e. CYW43455) on an AVM FRITZ!Box in WPA2/WPA3 mode (RSN-FLAGS pair_ccmp group_ccmp psk sae).

    • Kernel 6.18.50+rpt-rpi-v8, NetworkManager 1.52.1-1+rpt4, firmware-brcm80211 1:20260519-1~bpo13+1+rpt1
    • brcmfmac: Firmware: BCM4345/6 wl0: Aug 29 2023 01:47:08 version 7.45.265 (28bca26 CY) FWID 01-b677b91b

    The failure looked the same as in the report. unattended-upgrades installed 2:2.10-24+rpt1 at 06:37:29, and NetworkManager's timestamps file last recorded the profile as active at 06:37:32. It never reassociated, and reboots didn't help.

    I reproduced the failure deliberately by reloading brcmfmac with the shipped feature_disable=0x282000 and running wpa_supplicant with -d. On this box the firmware is not what fails. The SAE handshake never starts, because wpa_supplicant has no password:

    NetworkManager: Config: added 'key_mgmt' value 'WPA-PSK WPA-PSK-SHA256 FT-PSK SAE FT-SAE'
    wpa_supplicant: wlan0: WPA: AP key_mgmt 0x402 network profile key_mgmt 0xd42; available key_mgmt 0x402
    wpa_supplicant: wlan0: RSN: using KEY_MGMT SAE
    wpa_supplicant: wlan0: Trying to associate with <bssid> (SSID=<ssid> freq=5200 MHz)
    wpa_supplicant: nl80211: External auth action: 0, AKM: 0x8ac0f00
    wpa_supplicant: SAE: No password available
    wpa_supplicant: SAE: Failed to build SAE commit
    wpa_supplicant: wlan0: nl80211: External auth status: 1
    kernel: ieee80211 phy1: brcmf_cyw_external_auth: External authentication failed: status=1
    wpa_supplicant: wlan0: CTRL-EVENT-DISCONNECTED bssid=<bssid> reason=3 locally_generated=1
    

    This repeats every ~12.6 s, alternating between the 5 GHz and 2.4 GHz BSS, and never falls back to PSK.

    The profile was provisioned by Raspberry Pi Imager, which stores the Wi-Fi credential as the 64-hex-digit PSK (the PBKDF2 output) rather than the passphrase. network-config on the boot partition, the netplan YAML and NetworkManager's 802-11-wireless-security.psk all hold 64 hex characters. WPA2-PSK can use that key directly; SAE needs the passphrase. NetworkManager 1.52.1 still appends SAE FT-SAE, and wpa_supplicant then picks SAE, which can never succeed.

    With the actual passphrase, SAE works on the 43455. I tested with the same driver config, so SAE was advertised, using in-memory NetworkManager profiles:

    • key-mgmt=sae, pmf=required: connects, key_mgmt=SAE, sae_group=19
    • key-mgmt=wpa-psk with the passphrase (NetworkManager appends SAE as above): connects, key_mgmt=SAE

    The wpa-psk case:

    NetworkManager: Config: added 'key_mgmt' value 'WPA-PSK WPA-PSK-SHA256 FT-PSK SAE FT-SAE'
    wpa_supplicant: wlan0: RSN: using KEY_MGMT SAE
    wpa_supplicant: nl80211: External auth action: 0, AKM: 0x8ac0f00
    wpa_supplicant: SAE: Selecting supported ECC group 19
    wpa_supplicant: SAE: Derive PWE from PT
    wpa_supplicant: SAE: peer-send-confirm 1
    wpa_supplicant: wlan0: nl80211: External auth status: 0
    wpa_supplicant: wlan0: WPA: using KEY_MGMT SAE and proto 2
    wpa_supplicant: wlan0: WPA: Key negotiation completed with <bssid> [PTK=CCMP GTK=CCMP]
    wpa_supplicant: wlan0: CTRL-EVENT-CONNECTED - Connection to <bssid> completed [id=0 id_str=]
    

    So the chain here is: the driver advertises SAE, NetworkManager adds SAE to a wpa-psk profile, and that profile only holds a raw PSK. Any Imager-provisioned Pi whose driver advertises SAE should hit this on a transition-mode AP, whatever the chip.

    For the Zero 2 W it may be worth checking whether the profile also holds a 64-hex PSK (sudo nmcli -s -g 802-11-wireless-security.psk connection show <profile> | tr -d '\n' | wc -c prints 64) and whether a failing attempt logs SAE: No password available. If so, the 43430 firmware may be fine too.

    That suggests one more place to break the chain: NetworkManager, or wpa_supplicant's key-management selection, could leave SAE out when the profile holds only a raw 64-hex PSK, since SAE can never succeed with one. I haven't checked whether newer NetworkManager or hostap releases already do this.

    Workaround in use here, keeping 2:2.10-24+rpt1: a same-named /etc/modprobe.d/rpi-brcmfmac.conf that shadows the /usr/lib/modprobe.d copy. The kernel command line stays untouched, and brcmfmac isn't in the initramfs under MODULES=dep, so no update-initramfs is needed:

    options brcmfmac roamoff=1 feature_disable=0x2282000
    

    With it, auth_alg reports OPEN SHARED LEAP, NetworkManager configures key_mgmt 'WPA-PSK WPA-PSK-SHA256 FT-PSK', and the connection comes up as key_mgmt=WPA2-PSK. Bit 25 = SAE_EXT matches BRCMF_FEAT_LIST in rpi-6.18.y's feature.h. Alternatively, at least on the 43455, storing the passphrase instead of the hash makes WPA3 work without touching the driver.

  3. ValdikSS commented on Sep 19, 2026

    @ValdikSS
    Contributor

    Well, we have many bugs in different places here.

    (not a bug of rpi, just for info) wpa_cli reports different information over its cli interface and dbus interface, which is used by NetworkManager. This is a known issue. In this case, wpa_cli reports SAE in auth_alg, but dbus interface in KeyMtmg. Just make sure to 'read it properly'. And the key_mgmt is filled from the template, not a real data from the fw.

    1. A bug of Raspberry Pi Imager & NetworkManager

    wpa_supplicant: SAE: No password available
    wpa_supplicant: SAE: Failed to build SAE commit
    The profile was provisioned by Raspberry Pi Imager, which stores the Wi-Fi credential as the 64-hex-digit PSK (the PBKDF2 output) rather than the passphrase.

    • Raspberry Pi Imager should include the password, or configure connection to WPA2-PSK (limit the WPA standard).
    • NetworkManager/wpa_supplicant should check what kind of password data is available to it. SAE does not use pre-shared key (PSK).
    1. If the firmware exposes SAE, it should work. If it doesn't, the firmware should not expose it.

    This patch in wpa_supplicant has just fixed what NetworkManager expects to see for SAE, that's all. If the firmware is broken for some reason, Raspberry Pi should use the firmware that does not expose SAE support. Or better, fix the firmware.

    1. NetworkManager connection configuration file

    In NetworkManager configuration file, key-mgmt=wpa-psk means WPA2 & WPA3 Personal, and key-mgmt=sae means WPA3-Personal Only. I don't see any way to force WPA2 only. Disabling PMF, if it were available in Trixie version, is not a good solution either (you'd rather use it with WPA2).

    See https://networkmanager.dev/docs/api/latest/settings-802-11-wireless-security.html

    Could you debug why SAE does not work on 2W instead? You'd need to run wpa_supplicant with debug_level debug, not sure how to do that from NetworkManager from the start, but running wpa_cli log_level debug and then trying to connect to the network should do the trick, check journalctl.

  4. gkoch02 commented on Sep 19, 2026

    @gkoch02

    Another data point, plus confirmation of the 0x2282000 workaround on the chip in the title.

    Confirmed on Zero 2 W / BCM43430/1 with 2:2.10-24+rpt1 still installed

    Same failure shape as the report: Wi-Fi died the moment apt-get upgrade restarted wpa_supplicant.service, and never returned across reboots.

    • Raspberry Pi Zero 2 W Rev 1.0, Raspberry Pi OS trixie, kernel 6.18.50+rpt-rpi-v8, NetworkManager 1.52.1
    • brcmfmac: Firmware: BCM43430/1 wl0: Jun 14 2023 07:27:45 version 7.45.96.s1 (gf031a129) FWID 01-70bd2af7 es7
    • AP in WPA2/WPA3 transition mode — every BSSID advertises psk sae (NM's AP dump shows R:0588, i.e. PAIR_CCMP|GROUP_CCMP|KEY_MGMT_PSK|KEY_MGMT_SAE)

    Failing, with the stock feature_disable=0x282000 and +rpt1:

    NetworkManager: Config: added 'key_mgmt' value 'WPA-PSK WPA-PSK-SHA256 FT-PSK SAE FT-SAE'
    NetworkManager: Config: added 'ieee80211w' value '0'
    wpa_supplicant: FT: Invalid key management type (1024)
    wpa_supplicant: wlan0: Authentication with <bssid> timed out.
    wpa_supplicant: wlan0: CTRL-EVENT-DISCONNECTED bssid=<bssid> reason=3 locally_generated=1
    wpa_supplicant: wlan0: CTRL-EVENT-SSID-TEMP-DISABLED id=0 ssid="<ssid>" auth_failures=1 duration=10 reason=CONN_FAILED
    

    1024 = WPA_KEY_MGMT_SAE. This repeated indefinitely at ~10 s per attempt. Note ieee80211w=0: setting NM's pmf=1 on the profile did not stop NM appending SAE FT-SAE, consistent with the 1.56+ note in the report.

    After adding SAE_EXT to the mask (brcmfmac.feature_disable=0x2282000; I used the kernel command line, the /etc/modprobe.d shadow works equally), with 2:2.10-24+rpt1 still installed:

    auth_alg: OPEN SHARED LEAP                    (was OPEN SHARED LEAP SAE)
    NetworkManager: supported features: AP? FT+ SAE- BIP+
    NetworkManager: Config: added 'key_mgmt' value 'WPA-PSK WPA-PSK-SHA256 FT-PSK'
    

    Associates in about a second, stable since. I deliberately reinstalled +rpt1 after applying the mask, with no pin or downgrade in place, to confirm the mask alone is sufficient on this chip.

    The mask value is kernel-specific — 0x2282000 is wrong on 6.12

    BRCMF_FEAT_LIST was reordered between kernels. From feature.h:

    kernel bit 22 bit 25
    rpi-6.12.y SAE_EXT PMKID_V3
    rpi-6.18.y SCAN_V2 SAE_EXT

    So on bookworm/6.12 the equivalent is 0x682000; copying 0x2282000 there disables PMKID_V3 and leaves SAE_EXT enabled. Conversely a 6.12-derived 0x682000 becomes wrong after a trixie upgrade, where bit 22 is SCAN_V2. Worth flagging, since anyone pre-empting this on bookworm (or following #4976's older 0x82000) will copy a literal. For a kernel-independent guard on boxes I did not want to touch the driver on, I pinned the origin for this package instead:

    Package: wpasupplicant
    Pin: release o=Raspberry Pi Foundation
    Pin-Priority: -1
    

    Re @1p0d's question about the 43430 and SAE: No password available

    Partial answer only. The profile here is Imager-provisioned and does hold the 64-hex PSK rather than a passphrase (both the netplan YAML and NM's 802-11-wireless-security.psk are 64 hex characters), so the same precondition applies. But at default supplicant verbosity the 43430 logged Authentication with <bssid> timed out — a 10 s auth timeout — rather than SAE: No password available / SAE: Failed to build SAE commit and an immediate External auth status: 1. I did not run with -d before applying the fix, so I cannot say whether the 43430 also fails at the "no password" stage or genuinely stalls in external auth. Happy to reproduce with -d if that data point would help settle whether the 43430 firmware is implicated at all.

    Minor, but it made this much harder to diagnose

    Pi OS ships /usr/lib/systemd/journald.conf.d/40-rpi-volatile-storage.conf with Storage=volatile, and a drop-in overrides journald.conf. On a headless Wi-Fi-only box this failure therefore leaves no logs at all to read afterwards — the box boots fine, sshd starts, and nothing is recorded. An /etc/systemd/journald.conf.d/ drop-in with Storage=persistent is needed before the failure to capture anything. Might be worth a line in any eventual workaround note.

  5. 429pm commented on Sep 20, 2026

    @429pm

    Have just worked through a similar problem but on a Pi4.

    Hardware: Raspberry Pi 4
    OS: Raspberry Pi OS Trixie, 64-bit
    Network: NetworkManager + Netplan
    AP: TP-Link Archer C80
    Band: 2.4 GHz, channel 11 / 2462 MHz
    SSID: AnS.HQ

    Fresh Trixie installation initially had:
    wpasupplicant 2:2.10-24

    With the Archer configured as WPA3-Personal + WPA2-PSK[AES], the first apt upgrade reached:
    Setting up wpasupplicant (2:2.10-24+rpt1) ... and the Pi disappeared from the network.

    Reflashed the card and upgraded all packages except wpasupplicant. The Pi remained connected.

    Installed 2:2.10-24+rpt1 manually while the Archer was configured WPA2-PSK[AES] only. Upgrade completed successfully; approximately 14 pings from independent window were lost during the supplicant restart, then connectivity recovered.

    Rebooted successfully with 2:2.10-24+rpt1 and WPA2.

    • Changed the Archer back to WPA3-Personal + WPA2-PSK[AES]. Connectivity immediately failed.

    • Changed the Archer back to WPA2-PSK[AES]. After several connection attempts, connectivity recovered.

    Note: nmcli -f active,ssid,security dev wifi reports AnS.HQ as WPA2 when in the WPA2-only configuration.

    I also have two Zero 2 Ws that have now started responding again since the C80 router has been switched to WPA2-PSK[AES]. They were also absent from AnS.HQ after the upgrade to wpasupplicant 2:2.10-24+rpt1.

  6. ValdikSS commented on Sep 20, 2026

    @ValdikSS
    Contributor

    Could you debug why SAE does not work on 2W instead? You'd need to run wpa_supplicant with debug_level debug, not sure how to do that from NetworkManager from the start, but running wpa_cli log_level debug and then trying to connect to the network should do the trick, check journalctl.

  7. pelwell commented on Sep 20, 2026

    @pelwell
    Contributor

    Could you debug why SAE does not work on 2W instead?

    That's a very different issue. 43436P, 43436S and 43456 are all Synaptics parts, and the Synaptics firmware handles SAE offload differently. There's a ticket open with them about SAE on 43456, and hopefully if we get that working we might be able to backport those changes to other parts, but I don't think it will be possible without new firmwares.

  8. ValdikSS commented on Sep 20, 2026

    @ValdikSS
    Contributor

    Well, my favor to ask applies to Rpi4 as well, which has 43455. It should be working, right?

  9. Knubbel commented on Sep 20, 2026

    @Knubbel

    Data point: Raspberry Pi 4B (BCM4345/6) — same regression, headless WLAN-only box, recovery required physical access + SD image restore

    I have seen a Pi 4B mentioned further down in this thread — here is the detailed evidence from our affected Pi 4B box. Same package, same mechanism, different chip family than the Zero 2 W report above.

    System

    Raspberry Pi 4 Model B Rev 1.5
    Debian GNU/Linux 13 (trixie)
    kernel 6.18.39+rpt-rpi-v8 (at incident) / 6.18.50+rpt-rpi-v8 (after kernel upgrade, workaround still fine)
    brcmfmac: Firmware: BCM4345/6 wl0: Aug 29 2023 01:47:08 version 7.45.265 (28bca26 CY) FWID 01-b677b91b
    wpasupplicant 2:2.10-24+rpt1 (broken) / 2:2.10-24 (works)
    network-manager 1.52.1-1+rpt4
    /usr/lib/modprobe.d/rpi-brcmfmac.conf = options brcmfmac roamoff=1 feature_disable=0x282000
    AP: consumer mesh system, SSID in WPA2/WPA3 transition mode, 5 GHz (channel 116)
    

    Note the modprobe mask is identical to the Zero 2 W report (0x282000) — the shipped Pi OS image uses the same mask for the 4345/6 family, and it has the same gap: SAE_EXT (bit 25) is not cleared.

    Impact on this box (why this is worse than a transient disconnect)

    This Pi is a headless home-automation + energy-management server (ioBroker; it feeds live meter data to a battery-storage system) and it is deployed WLAN-only — no ethernet at the installation site, no fallback console.

    When the upgrade restarted wpa_supplicant.service, the link dropped and never came back. SSH gone, device gone from the network, reboot did not help. Recovery sequence we had to go through:

    1. physically relocate the device and attach LAN + peripherals,
    2. restore the last SD-card image (created 2026-09-14 by our scheduled backup),
    3. repair FAT boot-partition damage that came with the restore,
    4. the image rolled the box back ~5 days — most of the application-level history
      data was only recoverable afterwards (and partially) via a separate app-internal
      backup; the last hours before the outage are permanently gone.

    So an unattended apt upgrade escalated into an on-site intervention with data loss — because of this package regression.

    Timeline

    • 2026-09-17: wpasupplicant 2:2.10-24+rpt1 published on archive.raspberrypi.com (trixie).
    • 2026-09-17…19: an upgrade on the production box pulled the package → WLAN collapsed mid-transaction (from the operator's view the upgrade "aborted" — actually the SSH session died when the supplicant restarted). The live journal was lost because the box had to be re-imaged on 2026-09-20 to regain access.
    • 2026-09-20: controlled re-run of the upgrade (detached via systemd-run, LAN attached for access) — failure reproduced exactly, with full logging. Evidence below is from this reproduction.

    Evidence

    NetworkManager key_mgmt before/after the single package change (same boot config, same AP):

    # working, wpasupplicant 2:2.10-24 (Debian):
    NetworkManager: Config: added 'key_mgmt' value 'WPA-PSK WPA-PSK-SHA256 FT-PSK'
    wpa_supplicant: wlan0: CTRL-EVENT-CONNECTED - Connection to <BSSID> completed
    
    # broken, wpasupplicant 2:2.10-24+rpt1:
    NetworkManager: device (wlan0): Couldn't initialize supplicant interface: Name owner lost
    NetworkManager: Config: added 'key_mgmt' value 'WPA-PSK WPA-PSK-SHA256 FT-PSK SAE FT-SAE'
    kernel: ieee80211 phy0: brcmf_cyw_external_auth: External authentication failed: status=1
    wpa_supplicant: wlan0: Authentication with <BSSID> timed out.
    wpa_supplicant: wlan0: CTRL-EVENT-SSID-TEMP-DISABLED id=0 ssid="<home-5GHz-AP>" auth_failures=1 duration=10 reason=CONN_FAILED
    

    The external_auth failed: status=1 errors then repeat on every re-association attempt (dozens within minutes), with the SSID being temp-disabled in a loop. Permanent, survives reboot — identical to the Zero 2 W description.

    Driver-level capability mismatch on this Pi 4B box (BCM4345/6, firmware 7.45.265, currently on Debian 2:2.10-24 — this is what the driver exposes, independent of the supplicant version):

    $ sudo wpa_cli -i wlan0 get_capability key_mgmt
    NONE IEEE8021X WPA-EAP WPA-PSK WPA-EAP-SUITE-B OWE DPP FT-PSK FT-EAP
    $ sudo wpa_cli -i wlan0 get_capability auth_alg
    OPEN SHARED LEAP SAE
    

    Exactly the same shape as reported for the Zero 2 W: auth_alg advertises SAE, key_mgmt does not — and the +rpt1 D-Bus patch makes the second line leak into NetworkManager's key_mgmt calculation. So the mechanism is not limited to the 43430/1 family; the 4345/6 on Pi 4B is affected the same way, with the same shipped feature_disable=0x282000 mask that misses SAE_EXT.

    pmf=disable does NOT provide an escape on trixie — confirmed

    Before downgrading, we tried the configuration-level workaround on the NetworkManager profile:

    802-11-wireless-security.pmf = 1 (disable)
    → NetworkManager 1.52.1-1+rpt4 still logged:
      Config: added 'key_mgmt' value 'WPA-PSK WPA-PSK-SHA256 FT-PSK SAE FT-SAE'
    → authentication still failed with the same external_auth errors
    

    This confirms the analysis in the issue: the upstream NM fix (b00c6749) that makes pmf=disable opt out of SAE in station mode is not in trixie's NetworkManager 1.52.1. On a trixie Pi there is currently no configuration-level opt-out — package rollback or a kernel/firmware-side change are the only levers.

    Workaround in production (verified)

    sudo apt-get install -y --allow-downgrades wpasupplicant=2:2.10-24
    sudo apt-mark hold wpasupplicant
    • NetworkManager immediately logs key_mgmt without SAE again; association completes within ~1 second.
    • Verified across multiple reboots, and — useful to know — also across a kernel upgrade 6.18.39 → 6.18.50 on the same day: kernel 6.18.50 + Debian 2:2.10-24 + same AP → connects fine (because without the D-Bus patch NM never learns the driver would do SAE).
    • Security note: Debian 2:2.10-24 and the RPT rebuild +rpt1 carry the same CVE state; the delta in +rpt1 is only the SAE D-Bus reporting patch (upstream 26c7f1bc / Debian #1147955), so the hold does not fall behind on security fixes.
    • We did not test brcmfmac.feature_disable=0x2282000 on the kernel command line — downgrade + hold was sufficient for us. (If the mask approach is chosen as the real fix: the same 0x282000 is shipped on Pi 4 images, so the corrected mask needs to go out for the 4345/6 family as well, not only 43430/1.)

    Suggested-fix scope

    With respect to the suggestions in the opening post: our data point extends the affected set from "43430/1 with 7.45.96.s1" to 4345/6 with 7.45.265 (Pi 4B) — same failure, same missing SAE_EXT bit in the shipped mask. A chip-allowlist approach in the kernel would need to cover both families; a corrected rpi-brcmfmac.conf mask would fix both in one place. From an operator's perspective, shipping the modprobe change together with (or before) any wpa packaging change that enables SAE reporting would prevent this class of stranding — for headless WLAN-only devices this package update is a brick-level event, not a brief blip.

  10. pwnsen commented on Sep 21, 2026

    @pwnsen

    Same problem here – sharing in case it helps. The workaround from this issue works on my Zero 2 W.

    Disclosure: I ran into the same failure and I am not a Pi/Wi-Fi/driver expert. The diagnosis and this write-up were produced by an AI assistant (Claude Fable 5.1, running in Claude Code). It worked read-only on my SD card first and then on the live system. I reviewed it, and all command output below is from my own device. Please treat the interpretation with appropriate caution. The measurements are real.

    brcmfmac.feature_disable=0x2282000 on the kernel command line (marked as untested in the original report) fixes it here, with 2:2.10-24+rpt1 still installed.

    Setup

    • Raspberry Pi Zero 2 W, Raspberry Pi OS Lite trixie 32-bit (armhf), image 2026-09-15, kernel 6.18.50+rpt-rpi-v7
    • BCM43430/1 wl0: Jun 14 2023 07:27:45 version 7.45.96.s1 (gf031a129) FWID 01-70bd2af7 es7
    • network-manager 1.52.1-1+rpt4, firmware-brcm80211 1:20260519-1~bpo13+1+rpt1, /usr/lib/modprobe.d/rpi-brcmfmac.conf = options brcmfmac roamoff=1 feature_disable=0x282000
    • AP: AVM FRITZ!Box mesh, "WPA2 + WPA3" transition mode
    • Wi-Fi provisioned by Raspberry Pi Imager 2.0.x (cloud-init -> netplan -> NetworkManager), key-management: psk with the 64-hex PSK the Imager writes

    Failure: apt full-upgrade on 2026-09-20 pulled wpasupplicant 2:2.10-24 -> 2:2.10-24+rpt1. The link dropped in the very second the postinst restarted the service (/var/lib/NetworkManager/timestamps frozen at that second, /etc/resolv.conf emptied by NM in the same second) and did not come back across 4 reboots.

    Fix: appended brcmfmac.feature_disable=0x2282000 to /boot/firmware/cmdline.txt (edited on another machine, the box was unreachable). Nothing else changed.

    After reboot:

    $ grep -o "brcmfmac[^ ]*" /proc/cmdline
    brcmfmac.feature_disable=0x2282000
    $ sudo wpa_cli -i wlan0 get_capability auth_alg
    OPEN SHARED LEAP
    $ sudo wpa_cli -i wlan0 status | grep -E "key_mgmt|pmf|wpa_state"
    key_mgmt=WPA2-PSK
    pmf=1
    wpa_state=COMPLETED
    NetworkManager: Config: added 'key_mgmt' value 'WPA-PSK WPA-PSK-SHA256 FT-PSK'
    NetworkManager: device (wlan0): Activation: successful, device activated.
    

    Compare with the original report: there get_capability auth_alg prints OPEN SHARED LEAP SAE. With bit 25 masked, SAE is gone and NetworkManager no longer appends SAE to key_mgmt.

    Two additional data points:

    1. Both firmware blobs a Zero 2 W can end up with advertise extsae in their feature string (brcmfmac43436s-sdio.bin 7.45.96.s1 and brcmfmac43436-sdio.bin 9.88.4.77), so both board variants should be affected.
    2. Possible aggravating factor for Imager-provisioned systems: the Imager stores the PSK as the 64-hex hash, not the passphrase. NetworkManager 1.52.1 appends SAE to key_mgmt regardless of the PSK format, but SAE cannot be performed without the passphrase. If that reading is right, a firmware fix alone would not be sufficient on such systems.

    Note for anyone recovering a headless box: systemd.run= on the kernel command line had no effect on the Pi OS image here, because /etc/systemd/system/default.target takes precedence over the generator's default.target symlink. Editing cmdline.txt on the FAT partition from another computer was the quickest recovery.

  11. mrman486 commented on Sep 21, 2026

    @mrman486
    • Changed the Archer back to WPA3-Personal + WPA2-PSK[AES]. Connectivity immediately failed.
    • Changed the Archer back to WPA2-PSK[AES]. After several connection attempts, connectivity recovered.

    Same issues with Pi3

    Changed the router settings as above removing the WPA3 and the Pi3 connects again. No other changes were made to the Pi3 just the update over the weekend of wpasupplicant

    I went into nmtui and checked what the password was and it was different to what it should be. I cleared it, saved the correct one, rebooted and now WPA3 is working along with WPA2.

    My build was from the imager_2.0.11.1 Rasp Pi OS lite x64

  12. added a commit that references this issue on Sep 21, 2026
    76782d5
  13. Omer8990 commented on Sep 21, 2026

    @Omer8990

    Same bug on two boards that are not Zero 2 W. Both BCM4345/6 (43455), fw 7.45.265, kernel 6.18.50+rpt, fresh Trixie images from today, WPA2/WPA3 mixed AP, NetworkManager 1.52.1.

    Pi 4 (4GB): Wi-Fi dropped during apt full-upgrade at the exact second dpkg.log shows wpasupplicant 2:2.10-24+rpt1 being configured (13:31:51), never came back, also not after reboot.
    Pi 5 (8GB): identical symptom, I reflashed before understanding it.

    With the new supplicant left installed, adding brcmfmac.feature_disable=0x2282000 to cmdline.txt fixes it: wpa_cli get_capability key_mgmt no longer lists SAE, NM associates WPA2-PSK, stable since. Checked the bits against rpi-6.18.y feature.h: 13=FWSUP, 19=SAE, 21=DUMP_OBSS, 25=SAE_EXT, so the shipped 0x282000 predates SAE_EXT.

    PR with the one-line change to debian/rpi-brcmfmac.conf: RPi-Distro/firmware-nonfree#61

  14. oxivanisher commented on Sep 21, 2026

    @oxivanisher

    I don't know if this helps in any way, and I did not debug it further, but on one Pi5 (no dektop env.) the update went trough without any problems on another Pi5 (with a desktop env) I had to downgrade wpasupplicant to get online again.

  15. 36 remaining items

  16. jmozmoz commented on Oct 2, 2026

    @jmozmoz

    I am not really sure what I am talking about here, but: The mask should be 0x2282000, not 0x82000. So you have to add another 2 to the beginning of the one which is used by default (0x282000)

    The Wi-Fi connection failed immediately and the Zero 2 W became unreachable over SSH. I also tested the brcmfmac.feature_disable=0x82000 workaround before installing rpt3. I added: brcmfmac.feature_disable=0x82000

    to /boot/firmware/cmdline.txt, rebooted, and confirmed it was active in /proc/cmdline. With that parameter active and wpasupplicant 2:2.10-24 still installed, Wi-Fi remained fully functional. I then upgraded only

  17. Linus-007 commented on Oct 2, 2026

    @Linus-007

    Follow-up: I retested using the corrected mask brcmfmac.feature_disable=0x2282000.
    I confirmed after reboot that the running kernel had:
    brcmfmac.feature_disable=0x2282000

    active in /proc/cmdline.
    Wi-Fi was working normally with wpasupplicant 2:2.10-24.
    I then upgraded only:
    wpasupplicant 2:2.10-24 -> 2:2.10-24+rpt3

    The Wi-Fi connection failed immediately during package installation, before any reboot.
    So on this Raspberry Pi Zero 2 W, the corrected 0x2282000 workaround does not prevent the regression with 2:2.10-24+rpt3.

  18. jmozmoz commented on Oct 2, 2026

    @jmozmoz

    The Wi-Fi connection failed immediately during package installation, before any reboot. So on this Raspberry Pi Zero 2 W, the corrected 0x2282000 workaround does not prevent the regression with 2:2.10-24+rpt3.

    Have you tried to reboot after upgrading wpasupplicant? I observed the same behaviour: The connection fails during the package installation. But after a reboot, the connection to WiFi worked.

  19. jmozmoz commented on Oct 4, 2026

    @jmozmoz

    Update, the router lost the connection to the Pi Zero 2 W. In the log of the routher (Fritzbox) the following can be found:

    WLAN-Anmeldung ist gescheitert : Die erneute Anmeldung ist aufgrund aktiver "Unterstützung für geschützte Anmeldungen von WLAN-Geräten (PMF)" gescheitert. MAC-Adresse: 2C:CF:XX:XX:XX:XX.
    

    Translated: Wi-Fi authentication failed: Re-authentication failed due to active "Protected Management Frames (PMF)" support. MAC address: 2C:CF:XX:XX:XX:XX.

    Update: I explicitly added pmf=2 to [wifi-security] in the NetworkManger config. Let's see if this helps.

  20. jmozmoz commented on Oct 6, 2026

    @jmozmoz

    All this didn't help. Now I am back to wpasupplicant 2:2.10-24 and removed all settings for modprobe/pmf. Let's see, if this fixes the problem, the connection doesn't fail sometime due to pmf. At least connecting to Wifi with WPA2+WPA3 enabled in the FritzBox works.

    Update, the router lost the connection to the Pi Zero 2 W. In the log of the routher (Fritzbox) the following can be found:

    WLAN-Anmeldung ist gescheitert : Die erneute Anmeldung ist aufgrund aktiver "Unterstützung für geschützte Anmeldungen von WLAN-Geräten (PMF)" gescheitert. MAC-Adresse: 2C:CF:XX:XX:XX:XX.
    

    Translated: Wi-Fi authentication failed: Re-authentication failed due to active "Protected Management Frames (PMF)" support. MAC address: 2C:CF:XX:XX:XX:XX.

    Update: I explicitly added pmf=2 to [wifi-security] in the NetworkManger config. Let's see if this helps.

  21. carhensi commented on Oct 10, 2026

    @carhensi

    I have the same issue. Reproducible on wpasupplicant 2:2.10-24+rpt3 (Zero 2 W, BCM43430/1, kernel 6.18.50, NM 1.52.1). UniFi WPA2/WPA3 mixed mode. Verified the profile stores a real passphrase, not a 64-hex PSK.

    pmf=1 alone didn't help: auth timeouts, SAE still in NM's key_mgmt. Adding brcmfmac.feature_disable=0x2282000 and rebooting got it connected via WPA2. Same supplicant version.. I'd be happy to test the proper fix.

  22. carhensi commented on Oct 10, 2026

    @carhensi

    Still reproducible on wpasupplicant 2:2.10-24+rpt3 — and notably with a real passphrase, not a derived 64-hex PSK, so the +rpt3 derived-key fix doesn't cover this path.

    • Zero 2 W / BCM43430/1, kernel 6.18.50, NM 1.52.1, wpasupplicant 2:2.10-24+rpt3
    • AP: UniFi SSID in WPA2/WPA3 transition mode (iw scan → Authentication suites: PSK SAE)
    • Connection profile is key-mgmt=wpa-psk with a passphrase

    pmf=1 / ieee80211w=0 on the profile does not help. Only disabling SAE_EXT (bit 25) does — brcmfmac.feature_disable=0x2282000 → CTRL-EVENT-CONNECTED + DHCP immediately. Happy to run further diagnostics or test the proper fix on this hardware.

  23. Linus-007 commented on Oct 13, 2026

    @Linus-007

    I reviewed the relevant brcmfmac source path and the Raspberry Pi packaging around this issue. I think there is a concrete flaw worth testing rather than continuing to treat this only as a wpa_supplicant regression.

    The Raspberry Pi configuration currently relies on a hard-coded brcmfmac feature bitmap:

    feature_disable=0x282000
    

    That bitmap is tied to the bit positions generated by BRCMF_FEAT_LIST. Those positions have changed between kernel versions. In the current 6.18 tree, SAE_EXT is bit 25, so the existing mask does not disable it.

    The result is effectively:

    SAE      disabled
    SAE_EXT  still enabled
    

    brcmfmac therefore still reports external SAE capability upwards. Once the Raspberry Pi wpa_supplicant build started exposing SAE capability correctly over D-Bus, NetworkManager could legitimately select SAE on a WPA2/WPA3 transition network.

    That makes the hard-coded bitmap fairly fragile. Its meaning changes when the feature enumeration changes, without the configuration file itself changing.

    There also appears to be a possible place for a more durable fix in brcmfmac. The driver already has firmware-feature override infrastructure in feature.c, but it appears to trust the firmware sae_ext capability sufficiently to expose SAE without a BCM43430/1 firmware-specific exclusion.

    For the affected Zero 2 W firmware, I think it is worth considering whether BRCMF_FEAT_SAE_EXT should be suppressed by a proper chip/firmware quirk rather than by another global numeric feature_disable mask.

    That would avoid depending on enum bit positions and would prevent user space from being told that SAE external authentication is usable on hardware/firmware where it is not reliable.

    For reference, I also retested the corrected workaround:

    brcmfmac.feature_disable=0x2282000
    

    I confirmed that exact value was active in /proc/cmdline.

    With Debian:

    wpasupplicant 2:2.10-24
    

    Wi-Fi worked normally.

    I then upgraded only:

    wpasupplicant 2:2.10-24+rpt3
    

    and Wi-Fi failed immediately during package installation, before reboot.

    So on this Zero 2 W, even explicitly disabling SAE_EXT with the corrected 6.18 mask does not make +rpt3 reliable.

    I do not think that proves the kernel capability advertisement is the only problem, but it does suggest two things may be present here:

    1. A fragile brcmfmac feature-disable mechanism that allowed SAE_EXT to remain exposed after the enum changed.
    2. An additional user-space or firmware interaction that still causes failure with +rpt3 even when the corrected mask is active.

    It may be worth testing a proper brcmfmac firmware/chip-specific SAE_EXT quirk instead of relying on the raw bitmap, then seeing whether the remaining +rpt3 failure persists.

  24. Linus-007 commented on Oct 13, 2026

    @Linus-007

    I looked deeper, and there is an important correction to our earlier conclusion.

    The most likely fix is still to disable SAE_EXT on the affected BCM43430/1 firmware, but our test did not actually prove that the corrected mask was effective. We only checked /proc/cmdline. That proves the parameter was present on the kernel command line; it does not prove what brcmfmac ultimately loaded as its effective feature_disable value, nor whether SAE was actually no longer advertised.

    The issue contains two strong data points on the same Zero 2 W / BCM43430/1 / 6.18 / +rpt3 combination:

    • jmozmoz changed the effective module configuration to feature_disable=0x2282000, rebooted, and SAE support became none advertised; WPA2 connection then worked.
    • carhensi reproduced the failure with a real plaintext passphrase and +rpt3; pmf=1 did nothing, but feature_disable=0x2282000 plus reboot immediately restored the connection using the same +rpt3 supplicant.

    That is much stronger evidence than I initially gave it credit for.

    The source confirms why.

    brcmfmac reads the firmware capability string, sees sae_ext, sets BRCMF_FEAT_SAE_EXT, and later applies settings->feature_disable by clearing those feature bits.

    If SAE_EXT survives, brcmfmac advertises SAE upwards and NetworkManager 1.52.1 can add SAE to a WPA2/WPA3 transition-mode profile.

    The actual defect is therefore this:

    firmware says sae_ext
            ↓
    brcmfmac enables BRCMF_FEAT_SAE_EXT
            ↓
    Pi OS tries to correct bad firmware capabilities with a raw numeric
    feature_disable bitmap
            ↓
    BRCMF_FEAT_LIST changes between kernels
            ↓
    the old bitmap stops disabling SAE_EXT
            ↓
    SAE is advertised when it should not be
    

    On 6.18 the shipped mask is:

    feature_disable=0x282000
    

    but SAE_EXT is now bit 25, so the effective mask required to disable it becomes:

    feature_disable=0x2282000
    

    This demonstrates why a raw numeric bitmap is a fragile long-term mechanism: the meaning of the mask changes when the enum changes.

    Proposed source-level fix

    I believe this should be fixed in brcmfmac, not in wpa_supplicant.

    For the affected firmware:

    BCM43430/1
    FWID 01-70bd2af7
    firmware 7.45.96.s1
    

    the driver should clear BRCMF_FEAT_SAE_EXT by feature name after firmware capability detection, rather than relying on Raspberry Pi OS to maintain a kernel-version-dependent numeric bitmap.

    Conceptually:

    if (drvr->bus_if->chip == BRCM_CC_43430_CHIP_ID &&
        !strcmp(drvr->fwver, "01-70bd2af7"))
            drvr->feat_flags &= ~BIT(BRCMF_FEAT_SAE_EXT);
    

    The existing brcmfmac source already contains chip-specific feature corrections after capability probing, so this would be consistent with the existing design.

    A cleaner implementation would be to extend the firmware override table so that it can explicitly disable features as well as enable them.

    For example:

    struct brcmf_feat_fwfeat {
            const char * const fwid;
            u32 enable_flags;
            u32 disable_flags;
    };
    

    with an entry such as:

    {
            "01-70bd2af7",
            0,
            BIT(BRCMF_FEAT_SAE_EXT)
    },
    

    and then:

    drv->feat_flags |= e->enable_flags;
    drv->feat_flags &= ~e->disable_flags;
    

    That would be kernel-version independent because it uses:

    BIT(BRCMF_FEAT_SAE_EXT)
    

    rather than a literal numeric mask such as:

    0x02000000
    

    If the enum moves again, the compiler calculates the correct bit automatically.

    Important correction to my own earlier test

    I previously reported that:

    brcmfmac.feature_disable=0x2282000
    

    did not prevent the regression with wpasupplicant 2:2.10-24+rpt3.

    That conclusion was too strong.

    Before installing +rpt3, I confirmed only that:

    brcmfmac.feature_disable=0x2282000
    

    appeared in /proc/cmdline.

    I did not verify the effective loaded module parameter or confirm that SAE had actually disappeared from the driver's advertised capabilities.

    The checks that should have been captured were:

    cat /sys/module/brcmfmac/parameters/feature_disable
    sudo wpa_cli get_capability auth_alg
    

    The decisive state would be an effective feature_disable value corresponding to 0x2282000 and no SAE capability being advertised.

    Because that was not verified, my earlier statement that the corrected mask itself failed is not conclusive.

    Based on the other controlled tests in this issue, the strongest supported fix is therefore:

    Blacklist BRCMF_FEAT_SAE_EXT for the affected BCM43430/1 firmware inside brcmfmac, rather than relying on a raw feature_disable numeric bitmap.

    That appears to address the underlying capability-advertisement defect while avoiding a kernel-version-dependent configuration workaround.

  25. wesleygas commented on Oct 31, 2026

    @wesleygas

    I can confirm this issue is still actively reproducible on a freshly flashed Pi Zero 2 W running Debian 13 (Trixie) with the latest packages.

    Looked at the handshake failure and wanted to share my findings and test results in case it helps

    Environment

    • Board: Raspberry Pi Zero 2 W (aarch64)
    • Wi-Fi Chip / Firmware: BCM43430/1 (brcmfmac43430-sdio.bin FWID 01-70bd2af7 es7)
    • Kernel: 6.18.50+rpt-rpi-v8 #1 SMP PREEMPT Debian 1:6.18.50-1+rpt1 (2026-09-11) aarch64
    • Packages:
      • wpasupplicant: 2:2.10-24+rpt3
      • network-manager: 1.52.1-1+rpt4
      • firmware-brcm80211: 1:20260519-1~bpo13+1+rpt1
    • AP Configuration: 2.4 GHz in WPA2/WPA3 Personal Transitional (mixed) mode.

    Observed Behavior & Handshake Trace

    1. NetworkManager Failure:
      Connecting via nmcli dev wifi connect <SSID> fails after ~3 seconds with:
      device (wlan0): supplicant interface state: scanning -> associating
      device (wlan0): Activation: (wifi) association took too long
      device (wlan0): state change: config -> need-auth
      device (wlan0): Activation: failed for connection ... (reason 'no-secrets')
      
      
    2. Underlying wpa_supplicant Trace:
      NetworkManager injects SAE FT-SAE into key_mgmt:
      Config: added 'key_mgmt' value 'WPA-PSK WPA-PSK-SHA256 FT-PSK SAE FT-SAE'
      wpa_supplicant attempts Auth Type 4 (SAE) via nl80211:
      nl802111: sae_pwe=2
    • Auth Type 4
      nl80211: Connect request send successfully
      wlan0: Trying to associate with (SSID='...' freq=2462 MHz)
      wlan0: Authentication with timed out.
    1. Isolated Pure WPA2-PSK Test:
      When testing a standalone wpa_supplicant config against the identical transitional AP forcing pure key_mgmt=WPA-PSK:
      wlan0: State: ASSOCIATING -> ASSOCIATED
      wlan0: WPA: Key negotiation completed with [PTK=CCMP GTK=CCMP]
      wlan0: State: GROUP_HANDSHAKE -> COMPLETED
      wlan0: Radio work 'connect' done in 0.162845 seconds
      The password, radio, and WPA2 4-way handshake succeed in 0.16s. The failure is isolated entirely to the SAE external-auth exchange on this chip.
      ──────

    Still Open:

    1. rpi-brcmfmac.conf bitmask:
      The shipped /lib/modprobe.d/rpi-brcmfmac.conf in firmware-brcm80211 (1:20260519-1~bpo13+1+rpt1) still defines:
      options brcmfmac roamoff=1 feature_disable=0x282000
      Inspecting the kernel module feature table for 6.18.50+rpt-rpi-v8:
      • Bit 18 (0x40000): SAE
      • Bit 24 (0x1000000 / 25th bit 1-indexed): SAE_EXT
      Neither is covered by 0x282000. Masking 0x1000000 successfully clears the Device supports SAE with AUTHENTICATE command flag from iw phy0 info.
    2. D-Bus Capability Inconsistency:
      Even with the driver feature masked, wpa_supplicant (2:2.10-24+rpt3) continues to unconditionally advertise "sae" in its D-Bus KeyMgmt interface capabilities due to a static compile-time declaration
      (CONFIG_SAE=y) rather than gating on driver capabilities.
    3. NetworkManager Aggressive AKM Expansion:
      Because NetworkManager detects "sae" on the D-Bus interface, its supplicant generator in nm-supplicant-config.c forcibly expands wpa-psk profiles to include SAE FT-SAE. Neither proto rsn nor pmf 1 in NM
      prevents this expansion.
      • When we suppressed the "sae" string match in NM, NM passed WPA-PSK WPA-PSK-SHA256 FT-PSK, and the interface associated and acquired DHCP immediately.
  26. aidanxcan commented on Nov 7, 2026

    @aidanxcan

    I hit what looks like the same symptom on a Pi 4 (SAE completes, then status_code=16), but on Bookworm, and in my case neither the feature mask nor the stored PSK was involved. I found the cause and a workaround that keeps WPA3, so I'm posting it here in case it explains some of the 43455 reports.

    Setup:

    • Raspberry Pi 4 Model B (BCM4345/6), firmware 7.45.265 (01-b677b91b)
    • Raspberry Pi OS Bookworm, kernel 6.12.109+rpt-rpi-v8, wpa_supplicant 2.10, NetworkManager 1.42.4
    • NetworkManager profile key-mgmt=sae with the plaintext passphrase
    • UniFi APs on a WPA3-only SSID: AKM SAE, MFP required, SAE-H2E advertised (the AP's RSNXE in EAPOL-Key message 3 is f4 01 20)

    Every join failed right after a successful SAE exchange. With the supplicant at debug level, commit and confirm both completed with status 0 ("SAE completed - setting PMK"), and then:

    CTRL-EVENT-ASSOC-REJECT bssid=00:00:00:00:00:00 status_code=16
    

    That 16 doesn't come from the AP. brcmfmac reports every failed connect as status 16 (brcmf_bss_connect_done sets WLAN_STATUS_AUTH_TIMEOUT). I added logging to the driver's event handlers to see what the chip itself got back: E_AUTH status 0, then E_ASSOC status 1 reason 1. On an association failure that reason is the AP's 802.11 status code, so the AP was answering 1, "unspecified failure". Switching to firmware 7.45.286, or zeroing the PMKID in the driver (more on that below), didn't change it.

    hostapd only answers status 1 after SAE in one place, the downgrade check in check_assoc_ies(). When the AP runs sae_pwe=2, it refuses a station whose RSNXE claims SAE-H2E if the SAE exchange used hunting-and-pecking. That's what the Pi ends up doing here, for three reasons:

    1. Upstream wpa_supplicant 2.10 has no SaePwe D-Bus property. NetworkManager 1.42 tries to set SaePwe=2 (nm-supplicant-interface.c), the set fails, and sae_pwe stays at 0, so SAE always uses hunting-and-pecking. It can't have been 2: with 2 the supplicant would have picked H2E, because the AP's scan entry carried the H2E flag (the sae_pwe=1 test below only works if it does).
    2. On the external-auth path brcmfmac only forwards vendor elements from the host (brcmf_parse_vndr_ies), so the firmware builds the association request, and it seems to put an H2E-capable RSNXE in it no matter which method the host used. I didn't capture the frame over the air, so this part is inferred from the AP's answer and from the fix.
    3. The UniFi AP advertises H2E and enforces the check.

    Forcing sae_pwe=1 (H2E only) on the running supplicant got it on the first try:

    wpa_cli -i wlan0 set sae_pwe 1
    nmcli connection up <ssid>
    

    The chip reported E_ASSOC status 0 and the 4-way handshake completed. To make it stick across reboots under NetworkManager, I create the wlan0 supplicant interface from a config file before NetworkManager starts. NetworkManager adopts the interface that's already there:

    # /etc/wpa_supplicant/wlan0-h2e.conf
    sae_pwe=1
    update_config=0
    
    busctl call fi.w1.wpa_supplicant1 /fi/w1/wpa_supplicant1 fi.w1.wpa_supplicant1 \
      CreateInterface 'a{sv}' 3 Ifname s wlan0 Driver s nl80211 ConfigFile s /etc/wpa_supplicant/wlan0-h2e.conf
    

    That runs from a oneshot unit ordered Before=NetworkManager.service, and a device I have in production on this network now joins at boot that way. sae_pwe=2 should work too, since the supplicant then uses H2E whenever the AP advertises it, but I haven't tested that.

    wesleygas's trace above shows sae_pwe=2 reaching nl80211 on Trixie, so Trixie may not have the first gap and this might only affect Bookworm. Someone with a WPA3-only AP that does H2E and a Trixie Pi could settle that.

    One more thing I noticed in the driver while I was in there: in rpi-6.12.y, brcmf_cfg80211_external_auth() builds struct brcmf_auth_req_status_le on the stack without zeroing it and never fills pmkid, so the firmware gets stack contents as the PMKID. Mainline's brcmf_cyw_external_auth() zeroes the struct and copies params->pmkid. Fixing it didn't change the result for me, but it should probably be fixed anyway.

    Feel free to reach out to me here.

    Aidan Cannon (@aidanxcan)

  27. pelwell commented on Nov 16, 2026

    @pelwell
    Contributor

    Pull request #7762 adds three commits which together should allow both Zero 2 variants to connect to WPA3 APs. It should not need the brcmfmac feature_disable flags, but they should be harmless.

    You can install a trial build by running sudo rpi-update pulls/7762, but read the caveats first.

    N.B. This is intended to be run with wpasupplicant 2:2.10-24+rpt3, but install the trial kernel before updating wpasupplicant if you aren't running +rpt3 already.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    Wifi IssueAny issues related to wifi

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions