netifd, branch master

config: accept 'true' for the interface disabled option

2026-07-17T17:59:12Z

Every boolean option converted through uci_to_blob accepts both '1' and 'true', but the interface disabled check compared against '1' only, so 'option disabled true' was silently ignored and the interface brought up anyway. Signed-off-by: Felix Fietkau

config: keep vlan config entries until interfaces are parsed

2026-07-17T17:58:37Z

config_init_vlans() drained and freed the config_vlans list before config_init_interfaces() ran, so config_bridge_has_vlans() always returned false for bridges defined inline in interface sections and the vlan_filtering/__has_vlans fixup no longer triggered for them, a regression from the pre list based code which scanned the uci sections directly. Iterate the list without freeing and release the entries after interface parsing. Signed-off-by: Felix Fietkau

config: match the device when overriding vlan config entries

2026-07-17T17:58:14Z

The override loop in config_insert_vlan_entry matched entries by name only. The names inserted by config_procd_bridge_cb are the vid keys of the per bridge vlans table, which are only unique within one bridge, so a vlan entry for one bridge silently deleted a same named entry belonging to a different bridge. Require a matching device name as well; the procd data override for uci sections keeps working since it refers to the same device. Signed-off-by: Felix Fietkau

system-linux: validate FMR ealen and offset before use

2026-07-17T17:58:00Z

Both values were parsed as u32 and passed to nla_put_u8 without a range check, silently truncating out of range values to their low 8 bits instead of rejecting the configuration. Fail tunnel setup with -EINVAL for values that do not fit, matching the encap limit validation. Signed-off-by: Felix Fietkau

system-linux: fix byte order of multicast GRE tunnel keys

2026-07-17T17:57:46Z

The fallback key for keyless multicast GRE tunnels was taken directly from the network byte order remote address and then passed through htonl() again when adding it to the netlink message, swapping the key on little endian systems. Convert the address to host byte order when deriving the key, matching the byte order of user supplied keys and the key emitted by iproute2. Signed-off-by: Felix Fietkau

extdev: cancel pending retry timers on device free

2026-07-17T17:57:38Z

Bridge devices arm two retry timers: the member enable retry and the dependency retry, both rearmed periodically while a member keeps failing or a dependency is absent. Freeing a device with such a timer still pending left a dangling uloop timeout that fires on freed memory, for example when a config reload removes a bridge whose dependency never appeared. Signed-off-by: Felix Fietkau

extdev: clear the bridge empty state when removed from the config

2026-07-17T17:57:04Z

The empty flag was set sticky on reload, and force_active, set by the empty path of __bridge_config_init, was otherwise only cleared when a present member is removed. Removing or disabling the empty option therefore kept the bridge force activated and present with zero members and kept extdev_bridge_remove_member from marking the device absent until netifd was restarted. Track the option transition and revert both; the hotplug prepare path sets force_active independently and is left alone. Signed-off-by: Felix Fietkau

bridge: ignore stale member check timer events

2026-07-17T17:56:42Z

The vlan membership check timer armed by a link-up event was not cancelled when the bridge was brought down, so a pending timer could fire after bridge_set_down, re-create the kernel bridge via bridge_enable_member and enslave the port while the owning interface is down. The check is only meaningful for a member that is still enslaved; bail out if the member is no longer active. Signed-off-by: Felix Fietkau

bridge: drop force_active when the bridge_empty option is removed

2026-07-17T17:55:29Z

force_active, set by bridge_config_init for empty bridges, is otherwise only cleared when a present member is removed, so removing the bridge_empty option from a memberless bridge left it force activated and present until restart, the same defect just fixed in extdev. Revert it on the option transition in bridge_apply_settings; the hotplug prepare path sets it independently and is left alone. Signed-off-by: Felix Fietkau

system-linux: check the nlmsg_parse result in cb_rtnl_event

2026-07-17T17:55:19Z

A truncated link message made nlmsg_parse fail before it initialised the attribute table, so the subsequent lookups read indeterminate stack contents. Signed-off-by: Felix Fietkau