netifd_add_dynamic(): missing duplicate check causes endless restart loop

[Sm00shed]

This issue is related to openwrt/openwrt#21422
(openwrt/openwrt#21422)
where the same symptom was reported and worked around, but the root cause was not identified.
I can confirm this from personal experience – I am running PPPoE with IPv6 and configured wan6 with option device @wan. The result was an endless restart loop. Renaming wan6 to wan_6 resolved the symptom but not the underlying cause.
Root cause (confirmed – PPPoE + IPv6):
netifd_add_dynamic() in ubus.c does not check whether an interface with the same parent_ifname and proto_handler already exists. When wan6 is configured in UCI, dhcpv6.sh calls add_dynamic("wan_6") – netifd creates this without checking for the conflict → endless loop.
Potentially affected scenarios (code-based, not personally verified):

PPPoE + IPv6 (confirmed): dhcpv6.script creates wan_6, conflicts with UCI wan6
DS-Lite: dhcpv6.script creates wan6_4, may conflict with UCI wan6_4
MAP-E / lw4o6: dhcpv6.script creates wan6_4, may conflict with UCI wan6_4
464XLAT: dhcpv6.script creates wan6_4, may conflict with UCI wan6_4
6rd: dhcp.script creates wan_6, may conflict with UCI wan6

These are personal observations from reading the source code – not a professional assessment. Corrections are welcome.

[Sm00shed]

Another manifestation of this bug: OpenVPN with IPv6 PD via PPPoE
https://forum.openwrt.org/t/openvpn-ipv6-via-pppoe/248151/4
Same symptoms - wan6 in UCI, but wan_6 triggers in hotplug logs.
Additional note: The OpenWrt wiki script at https://openwrt.org/docs/guide-user/services/vpn/openvpn/server_ip6prefix
contains a workaround with:
bash/sbin/ifup wan6
sleep 30
This actually triggers the wan6/wan_6 bug and the sleep 30 is pointless. Once netifd is fixed, both lines should be removed from the wiki and the script condition should check for the correct interface name.

[Sm00shed]

GDB analysis (live system, attached to running netifd via gdbserver --attach $(pidof netifd)):
$3 = 0x7f85293ac0 "add_dynamic"
interface_alloc (name=0x7f85293af4 "wan_6", dynamic=true)
$4 = 0x7f85293af4 "wan_6"
Value returned is $5 = (struct interface *) 0x7f852c1a00
Return value is a valid pointer, not NULL. wan_6 is created successfully even though wan6 already exists in UCI.
netifd_ubus_add_interface (iface=0x7f852c1a00)
$7 = 0x7f852c1fe0 "wan_6"
$8 = 0
$x0 = 0 = UBUS_STATUS_OK
proto_ext_notify -> $1 = "wan_6"
proto_ext_update_link -> $3 = "wan6"
$x0 = 1 = UBUS_STATUS_NOT_FOUND
wan_6 cannot claim its device because wan6 already holds it. Infinite restart loop confirmed.

[Sm00shed]

@nbd would you have time to look at this? GDB analysis confirms the root cause.