Patch to apply to the original version

[Averell7]

Hello @aetta I post here the five patches which were written by Claude, and the docx file of the full discussion with Claude.
The patches must be applied in the following order : warnings → graph-max-rate → graph-yscale → name-file → conf-path

bandwidthd-fix-warnings.patch
bandwidthd-graph-max-rate.patch
bandwidthd-graph-yscale.patch
bandwidthd-name-file.patch
bandwidthd-conf-path.patch

I think they can be used in a git command. On my box, I applied them by patch -p1 < [patch name]

Since we will let this repository public, a Readme should be added explaining the changes :

• To be able to compile, we had to add some parameters to the .configure command, because the original C code is old :
  ./configure --build=aarch64-unknown-linux-gnu CFLAGS="-Wall -g -O2 -fcommon -fgnu89-inline"
• There were still a lot of warnings, the first patch corrects this
• graph-max-rate allows to set a maximum to the graphs scale, to prevent a peak to make unreadable the standard traffic
• graph-yscale allows to use a non linear scale, to add readability with high spead connexions. Same problem as above, the two options can be used together (see screenshot below)
• name-file : the names to join to the IP address display can be read from a json file
• conf-path : adds /etc/bandwidth.d/bandwidthd.conf to the possible locations of the configuration file

[Averell7]

These parameters has been added to the config file :

# Vertical axis scale: "linear" (default) or "sqrt".
# sqrt lifts low/background traffic so spikes no longer flatten it.
graph_yscale "sqrt"

# Shared vertical ceiling in Kbit/s. 0 = auto-scale (old behaviour).
# Spikes above this clip at the top.
graph_max_rate 5000

# JSON map { "ip": "name", ... }. Names take precedence over reverse DNS.
name_file "/etc/unbound/ip_to_user.json"

[Averell7]

IMPORTANT NOTE

There was a bug in the original code because two fields of the cdf files were glued together. We separated them so that there is now 18 fields. The consequence is that this version will not be able to read the logs created by the original fork. The correction is present in the first diff of bandwidthd-graph-yscale.patch

[Averell7]

Discussion with Claude.ai (French)

Bandwidthd - Discussion with Claude.docx

[aetta]

Hi! I spent a good chunk of today testing this fork.

All 5 patches apply cleanly and it compiles without issues using the documented command.

Dependencies needed: libpng-dev, libgd-dev, libpcap-dev.

About the segfault I mentioned earlier: tracked it down with gdb to StoreIPDataInCDF in bandwidthd.c .

fopen for the cdf log file returns NULL if log_dir doesn't exist or isn't writable, and there's no NULL check before the fprintf.

Pre-existing bug, not from the patches.

Fix: add if (!cdf) return; after the fopen. Workaround for now: set log_dir explicitly in the config.

Once past that, graph_yscale sqrt works great with real traffic. Will test name_file tomorrow!

Thanks

[Averell7]

Fine ! You are ready to create the update.

Important thing : It is necessary to update the model /etc/idefix/models/bandwidthd.conf because the real conf file is generated by /usr/lib/idefix/write_network_conf.py from this model

[aetta]

Hi! Tested name_file today, works perfectly. The friendly name shows up in both the table and the graph titles.

A few things:

About the model: I saw that write_network_conf.py generates bandwidthd.conf from bandwidthd.model, which is currently empty on our boxes. Should we create a simple template with the new options (graph_yscale, graph_max_rate, name_file, log_dir) already included?

About name_file: it expects a simple {"ip": "name"} JSON. What's the best way to generate it? I am thinking of a small script that crosses the ARP table with idefix3.json to map IPs to usernames, run periodically by cron. Does that make sense or is there a better approach?

About aarch64: compiled and running fine on my test box (DietPi/Debian 12, aarch64). Do the other boxes use the same architecture?

[Averell7]

I saw that write_network_conf.py generates bandwidthd.conf from bandwidthd.model, which is currently empty on our boxes.

No, it is not empty, otherwise bandwidthd could not work ! You have it in /etc/idefix/models

####################################################
# Bandwidthd.conf
# version 1.0
#.
# Commented out options are here to provide
# documentation and represent defaults

# Subnets to collect statistics on...
subnet {lan_subnet}

# Device to listen on
dev "{lan_port}"

###################################################
# Options that don't usually get changed

#Put interface in promiscuous mode to score to traffic
#that may not be routing through the host machine.
#promiscuous true

promiscuous false

#Log data to cdf file htdocs/log.cdf
#output_cdf false

output_cdf true

#Read back the cdf file on startup
#recover_cdf false

recover_cdf true

#Set META REFRESH for static pages in seconds(default 150, use 0 to disable).

meta_refresh 150

#Set the static html output directory
htdocs_dir "/var/lib/bandwidthd/htdocs"

[Averell7]

Promiscuous must be set to true (at least in bridge mode).

[Averell7]

I am thinking of a small script that crosses the ARP table with idefix3.json to map IPs to usernames,

This code already exists and writes /etc/unbound/ip_to_user.json

[aetta]

@Averell7 ! All set! Ready to test!

[Averell7]

I added two patches.

1. Make visible the indications of 6h, 12h and 18h on the graph. Presently they are visible if there is no traffic but once there is traffic, they become invisible.
2. show the night time with a light gray background.

bandwidthd-sixhour-marks.patch
bandwidthd-night-shade.patch

Not critical, but good for readability.

[Averell7]

About the names : as long as we do not have our specific json file, use as default :
/etc/unbound/ip_to_user.json