When a station roams between bands on the same VLAN, the vlan_remove
event from the old band can arrive after the vlan_add event from the
new band, causing the VLAN to be incorrectly deleted whilst still in use.
Add reference counting that tracks how many interfaces are using each
VLAN ID. Only create VLAN devices on the first reference and only
remove them when the last reference is dropped.
On startup, initialise refcounts from current station state to handle
daemon restarts correctly.
Signed-off-by: John Crispin <john@phrozen.org>
676e155 dhcp_inject: fix a syntax issue in dhcp_inject
e594c44 captive.uam.mac-auth: the default was wrong
Signed-off-by: John Crispin <john@phrozen.org>
LAN switch exposes single eth1 instead of per-port interfaces for RAP750W-311A.
Using VLAN to separate LAN to each physical port for RAP750W-311A.
Fixes: WIFI-15163
Signed-off-by: jackcybertan <jack.tsai@cybertan.com.tw>
make openwrt/profiles relative so moving or renaming the repo directory won't break the symlink
Signed-off-by: Shubham Vishwakarma <shubhamvis98@fossfrog.in>
Changelog:
- Fix MAC address assignment for Ethernet ports
- Fix Ethernet port configuration (was not working in the current DTS
because GPIOs 23 and 24, used for LEDs, are RGMII2 pins)
- Add package kmod-7915-firmware to enable Wi-Fi
Signed-off-by: Shubham Vishwakarma <shubhamvis98@fossfrog.in>
Set max_ap_assoc at wiphy init instead of vif init for mt7915.
Hard code max_ap_assoc to 128 for EAP112 in wifi_max_user.uc
Fixes WIFI-15027
Signed-off-by: Arif Alam <arif.alam@netexperience.com>
100c045 Fix cloud cannot show association list when WDS-AP in state.uc
168f6a4 dhcp_inject: Support multiple upstream
Signed-off-by: John Crispin <john@phrozen.org>
Introduce a blocklist mechanism to avoid retrying failed discovery
methods within the same discovery cycle. Each time a method fails
validation, it is added to the blacklist. The blacklist is cleared
once the device transitions to ONLINE or after all discovery methods
have been attempted.
This prevents repeated attempts of failing methods and ensures the
discovery process progresses more reliably.
Signed-off-by: Marek Kwaczynski <marek@shasta.cloud>
In cases where gateway.json exists, the discovery method may be unset.
Writing an empty value to discovery.state.json is not useful, so
avoid updating the file in this case.
Signed-off-by: Marek Kwaczynski <marek@shasta.cloud>
Always obtain EST certificates before starting the discovery process.
This ensures certificates are already available from the EST server, since
the FQDN may be provided via DHCP discovery or another discovery method,
and requires valid certificates to proceed.
Fixes: WIFI-15123
Signed-off-by: Marek Kwaczynski <marek@shasta.cloud>
Commit 2e4972e9ad ("ipq53xx: Add KERNEL_IPQ_MEM_PROFILE for IPQ53XX")
introduced KERNEL_IPQ_MEM_PROFILE but didn't set it for all IPQ53xx
boards, causing them to boot with incorrect RAM size settings.
This adds CONFIG_KERNEL_IPQ_MEM_PROFILE=0 to the affected board profiles:
- cig_wf189, cig_wf189h, cig_wf189w, cig_wf672
- edgecore_eap105
- sonicfi_rap7110c-341x
Signed-off-by: John Crispin <john@phrozen.org>
Reverse byte order in non-OUI part of MAC address to prevent overlap
when MBSSID is enabled. Swaps bytes 3 and 5 and masks lower nibble
of byte 5 before applying index XOR.
Signed-off-by: John Crispin <john@phrozen.org>
Ensure that LEDs are configured to be ON before attempting to change their state.
Previously, if the LED was configured to be OFF, it would still enter a double-blink
state when the cloud connection was lost, and then switch to solid ON upon
reconnection—ignoring the configured OFF state.
This update changes that behavior:
- If LEDs are configured OFF, they will remain OFF even during cloud
disconnection (no double-blink).
- After temporary state changes (e.g., during factory reset), the LED will
return to its configured state (either OFF or ON).
Signed-off-by: Paul White <paul@shasta.cloud>
The MIB loop cnt variable was defined as a static variable in the function that
implements the loop, however this function can be called for more than one switch
on some platforms. This results in a race condition that leads to memory
corruption and kernel crashes.
The fix moves the loop cnt variable to the passed in switch handle, this way
there is one per switch chip. Thix fix was identified by looking at newer
versions of the qca-ssdk software package from QCA.
Signed-off-by: Paul White <paul@shasta.cloud>
A scenario was seen where UCI config was not flushed to disk before
an AP power-cycle after uci-defaults was completed. Since these
scripts are deleted after being ran once, there is no way to recover
without a factory reset.
Adding this sync operation proved to help avoid this situation from
happening
Signed-off-by: Paul White <paul@shasta.cloud>
