- Add a dependency array to section template
- Lift dependency handling from AbstractValue to Node
- Split resolve_depends() into a function handling dependencies of a
node itself (resolve_node_depends()), which now is used for all nodes,
and the existing recursive descent through the node tree
- When any node's dependencies are not satisfied, all descendents are
reset recursively, meaning that an option in a section that is hidden
will always be ignored, as if the option's own dependencies were
unsatisfied
Fixes: #1349
- If luasrcdiet fails for a Lua script, this is a bug that should be
noticed rather than including the unminified script
- Use .tmp suffix for temporary file rather than .o
- Replace chmod + stat + mv with simpler cp + rm to overwrite file while
preserving file mode
gluonutil_get_interface_lower() recursively determines the lower of an
interface until the base interface is found.
gluonutil_get_interface_type() tries do distinguish wired, wireless and
tunnel interfaces. For tunnel interfaces, TUN/TAP, L2TPETH and Wireguard
are supported at the moment.
- Rename obtain_if_addr() to get_linklocal_address()
- Pass buffer of size INET6_ADDRSTRLEN instead of the oversized NI_MAXHOST
- Check if an address is link-local before converting to a string
- Replace an incorrect use of strncmp() with strcmp()
- Return status to caller
- Streamline control flow
While we're at it, the function handle_neighbour(), which is one of the
callers of get_linklocal_address() is slightly cleaned up as well.
This switches hostapd variant used for SAE and OWE from hostapd-openssl
to hostapd-wolfssl.
The bug shich one broke the wolfssl implementation was resolved upstream
with commit 631c437a91c2 ("hostapd: backport wolfssl bignum fixes").
This particular commit also got backported to OpenWrt 19.07.
This allows for multiple mesh legs out of one box. Useful for offloaders
and networks using vlan for separate mesh transport.
Custom mesh interfaces in a VXLAN domain are otherwise firewalled.
This fixes#1479
This adds support for the beacon interval to be set on a per-band base.
This has the potential to reduce the amount of airtime used up for
sending beacon frames.
Separate wireless-related helper methods from the util module to a
new wireless module. This keeps them separated, as the amount of
wireless helpers increased in the past, justifying a separate module.
xt_CT was added to kmod-ipt-conntrack a while back. Update our
GLUON_SPECIALIZE_KERNEL rules accordingly.
When building xt_CT into the kernel, one of the symbols IP_NF_RAW and
IP6_NF_RAW must be enabled as well, even though there is no runtime
dependency when building as modules. The kernel grows by less than 1KB
even when both IP_NF_RAW and NETFILTER_XT_TARGET_CT are enabled, making
this change a (very slight) net win in both flash and RAM usage.
As a partial fix to #496, do not touch the MAC address of the WAN
interface when using VXLANs (as only the MAC address of the VXLAN
interface matters to batman-adv).
On PoE-powered devices, we usually want to use WAN for the setup mode.
As all outdoor devices are PoE-powered, we can reuse this function. This
switches the setup mode interface for new installations / after config
reset on a few devices that were missing in this list before.
The 'preserve' flag can be used to mark a peer so it is not removed or
modified on upgrades. In addition, groups containing preserved peers are
not removed.
Fixes: #557
The netdev() lookup is confusing to use: whenever a interface does not
exist during boot (for example VLAN) or when the address is overridden
from board.json (which is not obvious at all), it will yield either no
address, or a different address than expected.
To avoid this confusion, using board.json-based interface() is
preferable. This converts all uses of netdev() to the corresponding
lan/wan lookups, except for the final fallback for eth0.
- Replace misnamed, closure-returning sysfs() to a reusable read() function
- Rename eth() to netdev(), pass full interface name
- Rename board() to interface()
- Split reuable get_netdev_addr() out of netdev()
gsub() returns the number of matches as its second return value. This
was unintendedly passed through by the util functions trim() and
node_id(). It can be presumed that this had no effect in practice, but
it can lead to surprising output when passing values to print() for
debugging.
Allows reconfigurtion of remote syslog from within site.conf.
Conflicts with the gluon-web-logging package as user made changes
will be overwritten, because this package will reconfigure the syslog
destination on every upgrade.
Resolves#1845
Use the value of the `name` site.conf field as label (it was
accidentally unused before).
Our site.conf currently doesn't define a specific order for the branch
entries. To avoid changing branch orders, sort entries by this label.
Fixes: #1961
Register to 'reset' event on form element and make call to 'update' function
delayed in 'data-update' handler to allow the form values to update beforehand.
When using a form's 'reset' button, form field visibility was not updated.
This could lead to situations where a checkbox had to be toggled again
twice to display the detail text inputs. (Example taken from private
wifi package)
This adds a helper method, which determines if the current platform
supports WPA3 or not.
WPA3 is supported if
- the device is not in the featureset category "tiny"
- the WiFi driver supports 802.11w management frame protection
The gluon-wireless-encryption package selects a WPA3 supporting
hostapd package as a dependency and stores the information, which
encryption method is supported to the device.
This package adds support for SAE on 802.11s mesh connections.
Enabling this package will require all 802.11s mesh connections
to be encrypted using the SAE key agreement scheme. The security
of SAE relies upon the authentication through a shared secret.
In the context of public mesh networks a shared secret is an
obvious oxymoron. Still this functionality provides an improvement
over unencrypted mesh connections in that it protects against a
passive attacker who did not observe the key agreement. In addition
Management Frame Protection (802.11w) gets automatically enabled on
mesh interfaces to prevent protocol-level deauthentication attacks.
If `wifi.mesh.sae` is enabled a shared secret will automatically be
derived from the `prefix6` variable. This is as secure as it gets
for a public mesh network.
For *private* mesh networks `wifi.mesh.sae_passphrase` should be
set to your shared secret.
Fixes#1636
Remove a lot of redundant code by switching to a match table listing
the targets and boards for each candidate for the primary MAC interface.
In addition, we add some flexiblity by allow to switch out the sysfs file
data source for the MAC address.
This reverts commits
- caf2dd037b.
- 07ebac6a49
- 55eff45f96
I accidentally pushed these commits as I had them lying around on a
dirty checkout I did testing on.
In addition this PR contains:
- split of gluon-respondd provider into multiple source files
- minor additional cleanups in gluon-mesh-babel respondd provider
(untested, as the babel respondd provider already doesn't compile prior
to these changes...)
many AVM devices do not have RESET/WPS buttons. So use the otherwise unused DECT/PHONE button to boot the device into setup mode.
This patch allows to enter the setup-mode by pressing the phone button
(often labeled as DECT) in addition to WPS and reset button.
This patch is necessary to allow supporting boards without a WPS and reset
button (e.g. AVM FRITZ!Box 7312).
With this commit, the status-led is set to be the "led-running"
device-tree alias for targets which do not implement the get_status_led
method in /etc/diag.sh.
This reverts commit 9b1eb40fe7.
With the batman-adv v2019.2 upgrade reverted (c1a7733956), the batman-adv
multicast-to-multi-unicast feature is not available yet. Without that it is
going to be very unlikely of the batman-adv multicast optimizations to
take effect. E.g. some outdated nodes would disable it.
To avoid confusion and diversion with a few communities having it enabled
and most implicitly deactivated, just deactivate it for all for now
until batman-adv is updated to v2019.2 or greater again.
mac_to_ip() calculates an ipv6 address from a mac address according to
RFC 4291. For wireguard we have to use specially crafted addresses that
must be unique. This allows calculating such unique mac-based addresses
by allowing to optionally specifying the bytes to be inserted into the
address.
The new routing_algo site.conf value BATMAN_IV_LEGACY is introduced. With
these changes, the routing_algo setting becomes mandatory.
Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue>
We cannot add the same file (here: /lib/gluon/mesh-batman-adv/compat) to
two, installed packages. Therefore, instead of determining the compat
version number from this file, infer it from the batman-adv release
version number instead.
Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue>
This is a fix for the broken ingress traffic-shaping in gluon v2018.2.2
and possibly earlier.
For ingress traffic shaping the kernel option NET_ACT_POLICE is needed.
Before this patch there was no dependency to this. Neither in
gluon_core, gluon-mesh-vpn-core nor in the package.
This patch adds this dependency.
[Matthias Schiffer: move dependency from GLUON_CORE to gluon-mesh-vpn-core]
Fixes#1790
package/gluon-web-network/luasrc/lib/gluon/config-mode/model/admin/network.lua:122:16: (W431) shadowing upvalue f on line 19
Fixes: bab4af01e ("gluon-web-network: improve PoE GPIO name translation
handling")
The gluon-authorized-keys is usually installed to use SSH keys to
authenticate a user against the device. To make this useful, it is also
required to disable passwordless SSH access to the device.
This new dependency is only required when the user doesn't have
gluon-setup-mode enabled already.
Fixes: #1777
Reported-by: yanosz <github@yanosz.net>
Fixes: a753fa79e3 ("gluon-authorized-keys: add keys from site.conf")
Signed-off-by: Sven Eckelmann <sven@narfation.org>
This MR includs only the VPN MODE of the hoodselector whitch simply set
hoods base on their geopositions.
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
check_site.lua: fix language syntax
muss -> must
rage -> range
at lease -> at least
coordiantes -> coordinates
realaise -> realised
gluon-hoodselector: fix language syntax in hoodselector
can not -> can't
routers -> router's
continure -> continue
to next -> to the next
TMP -> temporary
for current -> for the current
continure -> continue
with next -> with the next
thier -> there
provides -> provide
possition -> position
therfore -> therefore
gluon-hoodselector: fix language syntax in util.lua
realaise -> realised
gluon-hoodselector: fix language syntax and use autoupdate lock mechanism.
gluon-hoodselector: fix spelling/grammar
gluon-hoodselector: automatically set SECTION and CATEGORY for Gluon packages
gluon-hoodselector-add-VPN-MODE: add micrond & libjson-c dependency
gluon-hoodselector-add-VPN-MODE: check running hoodselector before loading lua
gluon-hoodselector-add-VPN-MODE: remove nixio dependency from hoodselector util
Revert "gluon-hoodselector-add-VPN-MODE: check running hoodselector before loading lua"
This reverts commit 535b0a1b2fb73e563bf6a44b568a796440bd307f.
add luaposix and luabitop to pakage dependency
sbin/hoodselector: remove nixio requiemend
sbin/hoodselector: load hoods only if necessary
gluon-hoodselector: use VPN abstraction layer. the hoodselectore does
not need to know about all individual VPN protocols.
gluon-hoodselector: Makefile add gluon-mesh-vpn-core as dependency
gluon-hoodselector: apply changes of mesh vpn lib
gluon-hoodselector: remove outdated comments
package/gluon-hoodselector: check_site.lua rm domain seed check thus its already checked by gluon-core
package/gluon-hoodselector: util.lua code cleanup and refactoring
package/gluon-hoodselector: hoodselector code cleanup and refactoring
gluon-hoodselector: util.lua, use taps instead of spaces. Use posix.unistd.access instead of io.open
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
gluon-hoodselector: hoodselector, use taps instead of spaces.
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
gluon-hoodselector: check_site.lua: replace hood with domain
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
gluon-hoodselector: drop VPN mode and rename hood to domain. Furthermore implement geolocator mode as neorayder way
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
package/gluon-hoodselector: rm duplicated print output
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
package/gluon-hoodselector util: fix wrong function signature
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
small typo fixes
small typo fixes
Update util.lua
processes are really restarted now. new (old) problem: nodes will not forget their former ipv6-addresses. watchdog could here with that.
gluon-hoodselector util.lua: replace i iterator with _
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
Update util.lua
now polygons with holes are recognized correctly. also a mix of nested polygons and boxes should be possible as shapes[]
package/gluon-hoodselector: hoodselector use gluon-reload for daemon restarts/reloads
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
package/gluon-hoodselector: util.lua use math-polygon lib and rm restart_services function. Rectengles will be converted into polygons now
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
package/gluon-hoodselector: Makefile rewrite description update depends list
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
package/gluon-hoodselector: check_site.lua reduce complexity
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
package/gluon-hoodselector: use : for gluon_version Val
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
package/gluon-hoodselector: fix if equal syntax
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
luasrc/usr/lib/lua/hoodselector/util.lua: check_site.lua simplify checksite script and fix if logic
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
package/gluon-hoodselector: set space after comma, rm unnecessary error handling
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
package/gluon-hoodselector: use only brackes on require function no mixup
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
package/gluon-hoodselector: check_site.lua rm unuse variables and fix non std global function
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
package/gluon-hoodselector: util.lua rm unuse include
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
package/gluon-hoodselector: rm comment return nil in function get_geolocation()
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
package/gluon-hoodselector: Makefile refactor pkg description
Signed-off-by: Jan-Tarek Butt <tarek@ring0.de>
We now keep the VPN enable state, bandwidth limit enable and actual limits
in the core config to avoid having to recover "user intent" from different
config files when the used VPN packages change.
Fixes#1736
Several fixes and enhancements related to multicast were added upstream
in batman-adv. So let's give the batman-adv multicast optimizations
another go.
Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue>
The is_outdoor function is placed inside the gluon.platform module, not
the platform_info module. Currently, the outdoor-mode wizard component
and the upgrade script fail due to nil-value calls.
adds a section to the wizard for outdoor capable devices
that informs the user of of the regulatory situation and
allows a quick toggle of the outdoor mode.
Add the `wifi5.outdoor_chanlist` site configuration that
allows specifying an outdoor channel range that can be
switched to for regulatory compliance.
Upon enabling the outdoor option the device will
- configure the `outdoor_chanlist` on all 5 GHz radios
- which may enable DFS/TPC, based on the regulatory domain
- disable ibss/mesh on the 5 GHz radio, as DFS *will*
break mesh connections
- allow for htmode reconfiguration on 5 GHz radios
The outdoor option can be toggled from
- Advanced Settings
- W-LAN
- Outdoor Installation
The `preserve_channel` flag overrules the outdoor channel
selection.
The batctl v2013.4 build was removed from the batman-adv-legacy package
as the current, upstream batctl releases work with batman-adv-legacy,
too.
As a replacement we need to add the upstream batctl dependency to
gluon-mesh-batman-adv-14 to have a batctl available again here.
Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue>
The commit a080049735 ("gluon-status-page-mesh-batman-adv: Retrieve TQ of
neighbors with non-best direct link") removed the check whether a neighbor
has the BATADV_ATTR_FLAG_BEST set. But consumers may still want to filter
out or mark neighbors which don't have this flag set. To assist with such a
feature, enhance the neighbor object with an extra boolean "best" attribute
which stores whether the BATADV_ATTR_FLAG_BEST was found or not.
Reported-by: Vincent Wiemann <webmaster@codefetch.de>
The commit ee63ed42fe ("gluon-mesh-batman-adv: List neighbors with
non-best direct link") removed the check whether a neighbor has the
BATADV_ATTR_FLAG_BEST set. But consumers may still want to filter out or
mark neighbors which don't have this flag set. To assist with such a
feature, enhance the neighbor object with an extra boolean "best" attribute
which stores whether the BATADV_ATTR_FLAG_BEST was found or not.
Reported-by: Vincent Wiemann <webmaster@codefetch.de>
Links between two direct neighbors are not always the best route between
these devices. The flag BATADV_ATTR_FLAG_BEST would not be set for these
originator entries and the respondd module would just ignore this entry.
If these neighbors are not accepted and returned to the status page then
some of the neighbor entries will show a name, (acceptable) signal strength
and mac address but no TQ value.
Fixes: 28668c8c52 ("gluon-status-page: API")
Links between two direct neighbors are not always the best route between
these devices. The flag BATADV_ATTR_FLAG_BEST would not be set for these
originator entries and the respondd module would just ignore this entry.
This causes missing links in meshviewer and similar tools. And when the
link quality is nearly equal and but fluctuates slightly, these links will
from time to time appear and disappear on the map.
Fixes: 2e0e24a992 ("announce neighbours using alfred/gluon-announce")
The device is broken until the next release. The LEDs are currently not
working (fixed in current OpenWRT master).
Also give a brief explanation about the BROKEN status being dependent on
the WiFi chip used and not the SoC family in general.
When two domains alias the same name (or one aliases another), display a
meaningful error message like:
Failed to alias domain 'foo' as 'bar', name already taken by
domain 'baz'.
The amount of local wifi clients is currently counted by two different
ways:
* asking the kernel wifi layer for the number of of clients on 2.4GHz and
5GHz band
* asking batman-adv for the number of non-timed out entries in the local
translation table with WiFi flag
The number of wifi24+wifi5 and the number of TT wifi client counts are
reported via respondd to various consumers. The ffrgb meshviewer is
displaying these values as:
* 2,4 GHz: wifi24
* 5 GHz: wifi5
* other: (TT local wifi+non-wifi clients) - (wifi24 + wifi5)
But the local translation table is holding entries much longer than the
wifi layer. It can therefore easily happen that a wifi client disappears in
the kernel wifi layer and batman-adv still has the entry stored in the
local TT.
The ffrgb meshviewer would then show this count in the category "other".
This often results in confusions because "other" is usually for ethernet
clients. And nodes with a frequently disappearing larger group of clients
(near bus stations or larger intersections) often show most clients under
the group "other" even when this devices doesn't have a LAN ethernet port.
It is better for presentation to calculate the number of total wifi clients
by summing up wifi24 + wifi5. And getting the number of total clients (non
wifi + wifi) by adding the result of the previous calculation to the sum of
non-wifi client in the local batman-adv translation table.
Fixes: 89a9d8138c ("gluon-mesh-batman-adv-core: Announce client count by frequency")
Reported-by: Pascal Wettin <p.wettin@gmx.de>
Gluon has multiple ways to obtain unique MAC-addresses. They are either
provided by the WiFi driver or derived from the primary MAC-address.
Quoting the same file:
> It's necessary that the first 45 bits of the MAC address don't
> vary on a single hardware interface, since some chips are using
> a hardware MAC filter. (e.g 'rt305x')
This currently fails in case the rt35xx based chips mac address differs
from the primary MAC. In this case, the MAC address for the client0 radio
(vif 1) comes from the WiFi driver. As there is only a single
MAC-address provided by '/sys/class/ieee80211/phyX/addresses' but the
MAC-address for mesh 0 (vif 2) is derived from the Node-ID, resulting in
different first 45 bits. The WiFi won't come up altogether in this case.
This commit verifies at least 4 MAC-Addresses are provided by the WiFi
driver. If this is not the case, all MAC-addresses are derived from the
primary MAC. This way, affected radios are working correctly.
This commit distributes dualband radios evenly on 2.4 GHz and 5GHz with
2.4 GHz being prioritised higher than 5 GHz. This means in case a device
has only a single radio and this radio supports operation in both bands,
it will be set to 2.4 GHz.
Tested-by: Martin Weinelt <martin@darmstadt.freifunk.net>
Signed-off-by: David Bauer <mail@david-bauer.net>
Allow odhcp6c to fork the script to handle router
advertisments in 30 seconds intervals. This is the value
that was previously used in Gluon v2018.1 / LEDE 17.01.
The default value is 3 seconds and while it is RFC compliant
it can put alot of pressure on even moderately sized devices.
Signed-off-by: Martin Weinelt <martin@darmstadt.freifunk.net>
log-file /dev/stderr is broken for babeld as it eats log messages for debug log.
This commit gets rid of the option. This allows -d N to be used as babeld command
line option.
This patch makes use of the new feature in l3roamd to gracefully
add, remove and list the mesh interfaces that are currently in use. This
helps when changing mesh interfaces often - a characteristic of the
wireguard protocol implementation as in the previous behavior all local
clients are dropped when adjusting mesh interfaces.
gluon-wan is a sudo-like exec wrapper that switches the process group to
gluon-mesh-vpn, making it use the WAN dnsmasq rather than resolving over
the mesh.
Note that this only affects DNS at the moment. Processes running under
gluon-wan will still use the regular mesh IPv6 routing table, and not the
WAN routing table. This is not a problem for IPv4, as there is only one
IPv4 routing table.
Fixes#1575
Rename to respondd.c / respondd.so, gluon.mk expects these names. This way
we can remove the install code. The installed filename is changed to
gluon-mesh-babel.so, bringing it in line with out common naming scheme.
The redirect and dnat target are needed for gluon-alt-esc-client to
forward frames to the selected, alternative gateways.
Signed-off-by: Linus Lüssing <linus.luessing@c0d3.blue>
This adds support for the TP-Link TL-WR902Ac v1 travel router.
The device is marked as broken due to 64MB which might be insufficient
in certain environments.
Before 7827f89, mandatory hostname field in config mode was
pre-filled with the default hostname.
This commit adds the config_mode.hostname.prefill option for
controlling the default value.
this activates the package by default when using the batman feature
while still allowing to explicitly remove it like this:
GLUON_SITE_PACKAGES := \
-gluon-ebtables-limit-arp
gluon-config-mode-geo-location-osm extends the
gluon-config-mode-geo-location with a location picker based on
OpenStreetMaps.
Based-on-patch-by: Jan-Tarek Butt <tarek@ring0.de>
The new code is shorter and uses more readable variable names. It does not
depend on specifically named input fields anymore (allowing to use multiple
maps on the same page), and only uses well-defined interfaces to trigger
revalidation of input fields.
The Map model class allows to add OSM maps to gluon-web forms.
Let gluon-respondd expose "MemAvailable" from /proc/meminfo to allow for
a more realistic memory-usage estimation.
Information on MemAvailable can be found here:
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/
commit/?id=34e431b0ae398fc54ea69ff85ec700722c9da773
As already done with other config mode texts, the altitude field now has
default texts that are used when they are not set in the site i18n files.
The altitude-help text has been removed from site i18n; instead, the
geo-location-help text now overrides the whole section description
including the part that mentions the altitude.
This reverts commit b3d7011130.
with this change, DNS in batman-adv based networks is broken.
although the revert breaks babel based networks, this is not as big of a problem.
This device is a dual 5GHz device. It is recommended to manually change the
radio of the first device to the lower 5GHz channels and the second radio
to the upper 5GHz channels.
The commit b3762fc61c ("gluon-client-bridge: move IPv4 local subnet route
to br-client (#1312)") moves the IPv4 prefix from the local-port interface
to br-client. A client requesting an IPv4 connection to the IPv4 anycast
address of the node (the device running gluon) will create following
packets:
1. ARP packet from client to get the MAC of the mac address of the anycast
IPv4 address
2. ARP reply from node to client with the anycast MAC address for the IPv4
anycast address
3. IPv4 packet from client which requires reply (for example ICMP echo
request)
4. ARP request for the client MAC address for its IPv4 address in prefix4
(done with the mac address of br-client and transmitted over br-client)
5. IPv4 packet from node (transmitted over br-client with br-client MAC
address) as reply for the client IPv4 packet (for example ICMP echo
reply)
The step 4 and 5 are problematic here because packets use the node specific
MAC addresses from br-client instead of the anycast MAC address. The client
will receive the ARP packet with the node specific MAC address and change
their own neighbor IP (translation) table. This will for example break the
access to the status page to the connected device or the anycast DNS
forwarder implementation when the client roams to a different node.
This reverts commit b3762fc61c and adds an
upgrade code to remove local_node_route on on existing installations.
The commit b3762fc61c ("gluon-client-bridge: move IPv4 local subnet route
to br-client (#1312)") moves the IPv4 prefix from the local-port interface
to br-client. A client requesting an IPv4 connection to the IPv4 anycast
address of the node (the device running gluon) will create following
packets:
1. ARP packet from client to get the MAC of the mac address of the anycast
IPv4 address
2. ARP reply from node to client with the anycast MAC address for the IPv4
anycast address
3. IPv4 packet from client which requires reply (for example ICMP echo
request)
4. ARP request for the client MAC address for its IPv4 address in prefix4
(done with the mac address of br-client and transmitted over br-client)
5. IPv4 packet from node (transmitted over br-client with br-client MAC
address) as reply for the client IPv4 packet (for example ICMP echo
reply)
The step 4 is extremely problematic here. ARP replies with the anycast IPv4
address must not be submitted or received via bat0 - expecially not when it
contains an node specific MAC address as source. When it is still done then
the wrong MAC address is stored in the batadv DAT cache and ARP packet is
maybe even forwarded to clients. This latter is especially true for ARP
requests which are broadcast and will be flooded to the complete mesh.
Clients will see these ARP packets and change their own neighbor IP
(translation) table. They will then try to submit the packets for IPv4
anycast addresses to the complete wrong device in the mesh. This will for
example break the access to the status page to the connected device or the
anycast DNS forwarder implementation. Especially the latter causes extreme
latency when clients try to connect to server using a domain name or even
breaks the connection setup process completely. Both are caused by the
unanswered DNS requests which at first glance look like packet loss.
An node must therefore take care of:
* not transmitting ARP packets related to the anycast IPv4 address over
bat0
* drop ARP packets related to the anycast IPv4 when they are received on
bat0 from a still broken node
* don't accept ARP packets related to the anycast IPv4 replies on local
node when it comes from bat0
Fixes: b3762fc61c ("gluon-client-bridge: move IPv4 local subnet route to br-client (#1312)")
