Ban by subnet in http component (with working code)

Development
1

I’ve been plagued by hackers trying to break my Home Assistant by attacking the file structure.
HA typically warns about these with log messages like

2026-04-10 22:51:45.572 WARNING (MainThread) [homeassistant.http.security_filter] Filtered a potential harmful request from 216.167.35.4 to: /cgi-bin/../../../../../../../../../../bin/sh
2026-04-10 22:51:45.783 WARNING (MainThread) [homeassistant.http.security_filter] Filtered a potential harmful request from 216.167.35.4 to: /cgi-bin/%2e%2e/%2e%2e/%2e%2e/%2e%2e/%2e%2e/%2e%2e/%2e%2e/bin/sh
2026-04-10 22:51:57.557 WARNING (MainThread) [homeassistant.http.security_filter] Filtered a request from 216.167.35.4 with a potential harmful query string: /index.php
2026-04-10 22:51:57.780 WARNING (MainThread) [homeassistant.http.security_filter] Filtered a request from 216.167.35.4 with a potential harmful query string: /index.php

I added each offender to ip_bans.yaml but they just increment their ip address to get around the ban.

To help with this I’ve forked and modified the http component to be able to ban by subnet (a.b.c.d/n etc) and this has eased the problem considerably as you can exclude whole parts of the internet with one line.
For example 123.0.0.0/8 excludes 123.[0-255].[0-255].[0-255]

Home Assistant handles this with the core http component. I have developed this into a custom component and it can be found here

Would anybody find this useful?

To test my forked http component:
Make sure your HA version is at least 2026.4.1
Copy the whole http folder into the custom_components config folder.

To ban networks, add the following to configuration.yaml in the http section

http:
  ....
  ip_ban_enabled: True # optional as defaults to True
  banned_networks: # Make sure the format is right, e.g. /16 must end in 0.0/16
    - 111.7.0.0/16
    - 45.0.0.0/8
    - 179.43.0.0/16
    - 123.160.0.0/14
    - 195.178.0.0/16
  log_banned_networks: True # optional as defaults to True
  notify_banned_networks: True # optional as defaults to True

Make sure you never exclude local addresses (10.x, 172.x, 192.x)

If you enable logging it will report as info messages so if the default logger setting is as above this you will need to add a logs: line to the logger: in configuration.yaml like this

logger:
  default: critical
  logs:
    custom_components.http: info

Risks:
I strongly suggest you install the samba app/add-on so you can access the config folder without using the ha file editor app/addon in case something breaks and you are locked out of ha. You should also have ssh access through a client like putty so you can restart if needed.

Views please
[edited 11/4/26 to match current Github location and configuration options]

7 Likes
Add an IP Range to IP Bans List?
[Security] Improvement of IPBan [HTTP Integration]
Add ip_ban_whitelist for HTTP Integration
Upgrade from 2026.3.4 to 2026.4 endup with "Unable to connect to Home Assistant."
2

Yeah, been getting daily attempts lately. It would be great if this option to ban entire subnets were made part of core functionality because having to fork a component means you’d have to take responsibility for keeping it up to date.

Not sure why this option hasn’t made it into core. Just add checks that you cannot ban internal IP ranges so that you don’t get locked out and it should be safe for general use.

3

Good point - I’ll look into this. I think I can get it from get_supervisor_ip in homeassistant.helpers.hassio

If I get traction here I’ll look into tidying up the configuration.
ATM it’s in network_bans.yaml but should be in configuration.yaml as it’s only accessed at load time.

I know how to generate schemas for configuration.yaml but I don’t know how to retrieve the configuration settings from the hass (HomeAssistant) object - any pointers?

Again, if it gets traction I’ll then look to a pull request so it’s incorporated into the ha core.

1 Like
4

Can’t help with pointers beyond reiterating the obvious - throw “not allowed” error on any subnet which starts in 10.0.0.0/8-32, 172.16.0.0/12-32 or 192.168.0.0/24-32. Hell, I’d simply throw an error if the subnet you’re covering might potentially cover anything starting in 10, 172 or 192 out of a sheer abundance of caution.

I seem to recall this was already proposed as a PR and rejected by the almighty powers, but I have no concrete links to back it up. Maybe with the internal IP safeguards in place, this might be more acceptable.

If you do get a PR raised, please link it in this thread. There are plenty of people who would back this request, both from a regular user perspective as well as from the people who created the whitelist option PR 3 times unsuccessfully.

5

Please note the changes in my first post as the configuration has changed.
The banned networks are now in configuration.yaml to be more conststent.
You no longer need the extra file banned_networks.yaml

6

Installed and tested.

I had the same experience: IP addresses from 145.148.10.0/24 testing loggi to HA daily and incrementing their IPs after being banned.

Banning the whole subnet helps.

I hope this will make its way into HA core.

7

Thanks for the support.

I’ve modified the security filter just now.
When logging unsafe requests the log didn’t show the offending source url, just the path.
It does now.

I also added the log message as a persistent notification because it’s easy to miss the attempted hacks if you don’t regularly read the logs.

A typical log entry/notification now looks like

2025-03-10 13:33:20.585 WARNING (MainThread) [custom_components.http.security_filter] Filtered a potential harmful request from 172.17.0.1 to: /cgi-bin/%2e%2e/%2e%2e/%2e%2e/%2e%2e/%2e%2e/%2e%2e/%2e%2e/bin/sh

Note this is a separate issue to network bans - it just makes it easier to identify hackers

I think I’m getting close to a pull request.
I’m away for a few days so I’ll do that at the end of the week unless something else crops up.

1 Like
8

One more thing you might want to look into before submitting the PR.

How are you handling overlapping network entries Eg. 1.2.3.4/32 & 1.2.3.0/24?

The current config checker will throw an error if it finds a duplicate IP. Do you need to guard against this scenario?

9

Not sure I do.
Why does it matter?
My “new” logic logic iterates each network given in the config and checks for if caller in network.
It will react to the first one found.

I could check at load time for every combination of network if a overlaps b but it seems overkill.
Am I missing something?

10

Logging the blocks has raised an interesting conundrum.
HA logged repeated attempts to probe with unsafe characters/potentially harmful requests from a user in network 45.156.128.0/24 so I blocked the complete range of the server.
It turns out that the server belongs to internet-census.org in the USA which describes itself as benignly scanning the internet for security weaknesses for benchmarks and yet the ICANN d/b shows the ip as an individual registrant.
I don’t like the thought of this probing in case they actually find a weakness in HA and crash it (or worse).
Should we block these?

11

My main concern is that currently ipbans throws a configuration error if you enter the same IP address twice. My comment was more of a heads up to ensure the same thing does not occur if you enter overlapping ranges.

If your code is only reacting to the first one found and handles the block gracefully without errors, then that’s even better.

I say treat it just like any other non-private IP and let the user decide whether they’re comfortable being probed.

12

Update
I want to add an optional description for each banned network in the config so the user can remind themselves in the persistent notification and log entry why they banned network if they want to.

Something like the current ban message followed by the (optional) description in quotes at the end…

2025-03-13 15:06:05.605 WARNING (MainThread) [custom_components.http.ban] Prevented access attempt from customer.sharktech.net (45.58.159.161) which is in banned network 45.0.0.0/8 “Banned whole subnet not just 45.58.128.0 - 45.58.191.255 because I like sledghammers!”

I plan to do the pr next week all being well.

13

PR created Improve http component to ban IP addresses by subnet #140661

It’s my first so I’m going to need help getting it through

1 Like
14

Left you a like on the PR, since that’s basically all I can contribute at this point.

Quick question though - any particular reason why the banned_networks need to go in the configuration.yaml file instead of the existing ip_bans.yaml file?

15

Thanks for the like.

For individual IP bans:
ip_bans.yaml is a dynamic file updated at runtime by the ban s/w (failed logins > threshold creates an entry).
It is only read at load load time.
New bans are written when they occur (ready for the next restart) and all the banned IPs are stored in the component.

For subnet bans:
These are not updated dynamically - only when the user adds them.
The banned networks list in configuration.yaml is loaded at boot time and is fixed until the user updates it and restarts HA - just like any other component.
I thought it best to keep static configuration away from dynamic component updates - which is possibly what the developer of the individual IP ban s/w thought hence creating ip_bans.yaml.

Long term if this pr gets adopted it would be good to have a button on the developer page to reload http so new subnet bans could be loaded without a restart…

16

Thanks, that clears it up!

17

I definitely need this feature <3 I’ve created a feature request exactly for this one, but didn’t thought to check if someone already developed something in another channel.
Thank you so much for your hard work and I really hope it’ll part of one of the next releases

18

The documentation PR is at Documentation update for HA PR #140661 (http banned networks) #38012

19

There’s a few comments on the PR now which basically say to remove persistent notifications and a worry that looking up ip addresses to see if they are in a banned network will be slow and affect every access.

Happy to remove persistent notifications and I pointed out looking up an ip address to see it it’s in an ip_network is quick (not iterative).

I think I can do all they need except prove the speed issue isn’t one…

2 Likes
20

Will tests that assert that the filtering times are within an “acceptable” threshold help ?