IPv6 via PPPoE often receives a prefix, not an address. In that case
mwan3_get_src_ip would not find an address and fall back to ::. However
in case of multiple IPv6 upstreams, this no longer means a ping (such as
done by uptime check) succeeds, for there are multiple addresses.
With this, mwan3 can effectively load-balance/fail-over on a PPPoE IPv6
prefix and a secondary (LTE) acquired /64 without immediately disabling
the IPv6 prefix interface as soon as the secondary IPv6 interface comes
up.
I think this change is also a fix to #26690
but I'm not 100% sure of that.
Signed-off-by: Fabian Groffen <grobian@bitzolder.nl>
`pq' yields with GNU sed:
sed: -e expression #1, char 15: extra characters after command
so, make q a separate command instead.
busybox accepts this too, and still returns only the first match (q
quits processing)
Signed-off-by: Fabian Groffen <grobian@bitzolder.nl>
I received an email from Aaron Goodman saying that he is no longer working
with mwan3 and asking me to remove him from the PKG_MAINTAINER list.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
`ubus.call(sprintf('network.interface.%s', ifname), 'status', {})`
returns null if the interface `ifname` doesn't exists (yet).
For pppoe interfaces using `option ipv6 auto`, a virtual interface suffixed `_6`
is automatically created once the connection is established,
but until then it doesn't exists.
Fixes: 6423781254 ("mwan3: reimplement rpcd plugin using ucode")
Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>
On my "test" router (5 wans, 2 tracking ips per wan), before any rework,
prometheus-node-exporter-lua mwan3 average scraping time was 1230ms
(scraping only the interfaces), after optimizing the shell version,
average time was down to 485ms, with ucode we are now at 41ms.
Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>
Using $(...) to call an internal function causes a fork().
Pass the name of the return variable as first parameter,
and write to it using 'export -n'.
Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>
pgrep is pretty slow as it needs to read all /proc/*/cmdline
or /proc/*/status, and it's called twice per wan.
Add STARTED and PID files to speedup mwan3_get_mwan3track_status().
Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>
This is a preparation for adding an utility function readfile() to read
directly to a variable. Using '$(cat ...)' causes a fork/execve which is
extremely slow just to read '~2' chars.
(commit message from Florian Eckert)
Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>
Just use 'job -p' to list the subprocesses currently running,
so we do not needed the global variables TRACK_PID and SLEEP_PID anymore
Signed-off-by: Etienne Champetier <champetier.etienne@gmail.com>
This redirects the debug output to stderr, allowing `mwan3 use` to be used in
scripts without polluting stdout.
Before:
mwan3 use wan curl -fsSL https://ifconfig.co/json | jq -er '.country_iso'
jq: parse error: Invalid numeric literal at line 1, column 8
curl: (23) Failure writing output to destination, passed 389 returned 0
After:
mwan3 use wan curl -fsSL https://ifconfig.co/json | jq -er '.country_iso'
Running 'curl -fsSL https://ifconfig.co/json' with DEVICE=eth2 SRCIP=192.168.0.1 FWMARK=0x3f00 FAMILY=ipv4
DE
Signed-off-by: Philipp Schmitt <philipp@schmitt.co>
This propagates the exit code of the command wrapped by `mwan3 use` and
allows for example to use `mwan3 use` in monitoring scripts.
Before change:
shell command:
mwan3 use wan false >/dev/null && echo ok || echo fail
result:
ok
After change:
shell command:
mwan3 use wan false >/dev/null && echo ok || echo fail
result:
fail
Signed-off-by: Philipp Schmitt <philipp@schmitt.co>
Forgot to bump the 'PGK_RELEASE' in the last change in 'mwan3'.
Fixes: 619629ce85 ("mwan3: close flock fd when starting mwan3.user scripts")
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
The awk expression in mwan3_delete_iface_rules splits the `ip rule list`
output by spaces, therefore $1 contains the trailing colon (e.g., "1:",
"1000:"). The < and > operators compare such values as strings instead
of numbers, producing unexpected results (for example, "1:" > "1000").
Change the field separator to ":" for correct number comparison, so that
the right rules are removed.
An example error message that may appear before the fix:
Error: argument "1:" is wrong: preference value is invalid
It happens because `substr($1,0,4)` selects short numbers along with
the colon. In other cases wrong rules may be removed, for example, if
there is rule 10051, then rule 1005 will be removed.
Signed-off-by: Maxim Mikityanskiy <maxtram95@gmail.com>
Previously the "use" command had the following shortcomings:
* a subprocess was created instead of replacing the shell process
* whitespace in arguments was not handled correctly
Implementation detail:
In shell context the `"$@"` expression should be used (instead of `$*`).
This allows the safe handling of arguments containing whitespace.
Closes: #20001
Signed-off-by: Lars Kruse <devel@sumpfralle.de>
This makes mwan3rtmon check if mwan3_get_routes returns a route
before removing it. This helps with IPv6 routes with source address
selector removal where multiple original routes are transformed to
the same mwan3 route if one of the source routes is removed while
the others are kept.
Signed-off-by: Jonas Lochmann <git@inkompetenz.org>
This updates mwan3 to use network_get_preferred_ipaddr6 instead of
network_get_ipaddr6 if possible to determine a source ip for the
connectivity checks. This avoids issues where the first ip address
that is returned from network_get_ipaddr6 does not work anymore while
the preferred one returned from network_get_preferred_ipaddr6 works.
Signed-off-by: Jonas Lochmann <git@inkompetenz.org>
The tracking and interface status was mixed up in the report. To fix
this, the interface status and the tracking status are now used
directly. The online, uptime and error information are appended to the
status line if needed. If certain routing tables and routing rules are
missing, the error number is also given.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
The tracker state is not shown via ubus. Only if the tracker was in
active state, then the boolean running was set or not. By adding the
tracking state to the ubus information we could also evaluate the state
of the tracker. To remain compatible, the runnig flag of the tracker is
not removed, which in fact displays the same information, but only if
the tracker is in state 'active' or not.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
The expression 'disabled' is more meaningful than 'not enabled' and can
therefore be better processed in the ubus output, since it is only one
word.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
The function 'get_mwan3_status' is reading the internal state from the
tracker via the status file. Do not use the state 'notracking' status
anymore. If the mwan3track is not running always return 'unknown'
and not 'notracking'. There is already an other function that evaluates
the external state of the tracker.
We have now the following states of the tracker:
internal (mwan3track):
- offline
- online
- diconnecting
- connecting
- disabled
- unknown
external (via pgrep and config):
- paused
- active
- down
- not enabled
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
During the renameing of mwan3_connected_v4 to mwan3_connected_ipv4 and
mwan3_connected_v6 to mwan3_connected_ipv6 the adjustment in the ubus
call was forgotten. This commit fixes this.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
Set the score value to the maximum value when the connected function is
called. The same happens with a disconnected event, the score value is
there set to zero.
Suggested-by: Anna Tikhomirova <vamp@vampik.ru>
Suggested-by: Maxim Mikityanskiy <maxtram95@gmail.com>
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
Refactoring the score handling, so that only one action could take place
during run. The behaviour should be more comprehensible, since several
score actions are not processed at the same time.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
Addition of routes to mwan3_connected ipset is broken. The ipset name was
changed from mwan3_connected_v4/6 to mwan3_connected_ipv4/6, but this
change was not reflected in mwan3rtmon.
Signed-off-by: Anna Tikhomirova <vamp@vampik.ru>
* Update commit message
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
Addition of iptables rules for mwan3 sticky rules is broken, resulting
in non-working sticky rules. The required parameters for the function
'mwan3_set_sticky_iptables' were passed in the wrong order.
Signed-off-by: Anna Tikhomirova <vamp@vampik.ru>
* Update commit message
* Quoting function arguments
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
This is a fix for the the following change:
3d824ea288
Before the change, it was only possible to execute a shell script. To
remove this restriction, a binary or other script language can now also
be used for 'mwan3.user'. Unfortunately, the old shell script was not
executable for older mwan3 version. During a sysupgrade with config transfer,
this 'mwan3.user' script could not be executed for newer mwan3 versions.
To fix this, the calling script checks whether the 'mwan3.user' is executable,
and if not, this executable bit is now set.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
Up to now on every interface down event a mwan3 disconnected event was
send. This is wrong because if the interface was never connected, then a
disconnected event should not get generated. This commit fixes this bug.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
The metrics and weight need to be the same. A 50% balanced would be
require member policies of the same metric and weight value.
Signed-off-by: Florian Eckert <fe@dev.tdt.de>
