Messages

A router is not a switch.

Conceptually, OPT1 and LAN are two different interfaces. So it all depends on how you configure them. What you probably want is to bridge both physical ports to one logical interface, not have to different interfaces with different subnets.

This is explained here.

If that is something different than you actually want, you should probably read this first.

[special]

I suggest you to read:

https://forum.opnsense.org/index.php?topic=42985.0, especially point 3

and

https://forum.opnsense.org/index.php?topic=44159.0

if you have not done so already.

On a side note: "VLAN 1" is special - on some switching platforms, it denotes the "untagged" VLAN, on some, it doesn't. So be sure that your switch, your Proxmox installation and OpnSense are in sync about what "VLAN 1" means in terms of bridge view, logical and physical interface views, respectively. Also check that port 5 has PVID 20; making it an untagged member of VLAN 20 only controls egress tagging.

You will find that you have chosen an environment with the maximum amount of potential complications. If you do not have an advanced level of networking knowledge, you will have a steep learning curve ahead - and I do not mean that condescending, but as an unbiased early warning, only considering the fact that this was your first posting here. What I mean by that is that I doubt that there is a step-by-step guide and an LLM will probably be of little use, either, especially if it is a free or local variety with limited capabilities.

Yes, actually your impression is correct. The limited rate after the duplicated rules obviously was limited to 9 log entries per second, because I originally had 9 rules and imposed an 1/1 limit on them. I just want to see indications as to what is happening in the logs. I still can look at the counters if I want to see how many requests come in.

I cannot say for sure how many requests per second were there at the time of attack and I also think many may have been dropped because the storage could not keep up. It is a 1 Gbps connection, but with just port probing, that may still amount to a few thousand requests per second, each of which has a significant log impact. I cannot really say what has happened, because obviously the filterlog process was out for several hours (there is also a gap into log file history on one day), such that not even remote logging shows anything. As a matter-of-fact IDK if the CPU load was only a side effect.

Logging for default deny is off here.

It seems to me the rule should be applied regardless if the log rate limit has been reached. Obviously, without logging.

While that would be desirable, pf does not work that way today, so I made it work.

My question is actually three-fold:

1. Can anyone think of a cleverer way to make it work without using duplicated rules?
2. Would there be a way to have the desired behavior for block rules (i.e. "rate-limit logging without affecting rule matching"), probably as a feature request for OpnSense?
3. Should there be a more prominent web UI warning about this "unexpected" behaviour for block rules?

And no, I do not want to use syncookies for functional reasons - also they would not fix the problem: state table exhaustion is not the issue, it is the mere logging that is throttling the firewall.

AOC brauchen pro Port 1-2,5W, Passive DACs maximal 0,1W. Wo soll das sparen?

Die Vorteile liegen bei der maximalen Länge der Leitung, Störfestigkeit und galvanischer Entkopplung, wie korrekt bemerkt mit wesentliche höheren Kosten.

Und wenn Du gute NICs und moderne Switches hast, wären selbst längere 10 GbE-Leitungen nicht mehr so viel schlechter bei 1 Watt pro Port.

Realistischerweise also: Im Serverraum DAC für kurze Verbindungen, wenn lange Leitung benötigt bis ca. 20m: 10GbE oder 2.5GbE, wenn längere Leitungen und Verkabelung "alt": optische Kabel.

Recently, I had a problem when some of my friends told me that out voice chat was choppy. When I investigated, I found my OpnSense at 100% CPU. That turned out to be due to 7 WAN block rules for various blocklists and an ongoing DDOS attack.

I had logging enabled for those rules and the log was constantly being written to. I could not even reliably use the web UI.

This made me think about how to rate-limit the logs. Apparently, with 26.7.2, there is now max-pkg-rate from pf available.
So, now, you can create two block rules, one with logging enabled, quick block and "Max packet rate [packets]" = 1 and "Max packet rate [seconds]" = 1 to have a rule that matches at most once per second and logs and another quick block rule without logging and without rate limits immediately following that.

By using that, you still have (rate-limited) logging, but without the risk for a totally overwhelmed OpnSense.

Alas, you need two rules, because the first one will not match any more when the rate limit has been exceeded. Would there be a better way of having such a thing (i.e. eat the cake and still have it)?

P.S.: Allowing rate-limiting on block rules is a potential security hole, IMHO: When you use that without thinking, the block will not be applied when the rate is too high, i.e. block rules can be circumvented by just hitting more frequently at the target....
I guess the original idea was to do this for pass rules only and rate-limiting those on the side.

I believe that, in some respects, they have come very far, but I think we are talking about two different things. I certainly expect an AI to outperform me at individual tasks — otherwise it would not be much of a tool. What I do not expect it to do is decide what my goals should be or which goals are worth pursuing.

Asking clarifying questions, decomposing a problem or comparing solution strategies neither requires nor indicates that kind of agency. Those are extremely useful capabilities, but they still operate within a goal and context supplied by somebody else.

I tend to regard current AIs as collections of remarkably powerful, but uneven, cognitive capabilities rather than thinking entities pursuing goals of their own. If they ever became better than us not merely at solving problems, but at deciding which problems ought to be solved and why, then we would indeed have a rather different problem.

I also do like the "OPNsense Science Fiction and Philosophy" wording, and maybe robotics really is further behind than "thinking", so that might be an alternative. There are still everyday tasks that even the most advanced robots struggle to perform reliably, while a six-year-old does them without thinking — such as picking up a key, orienting it correctly and opening an unfamiliar lock.

That is an interesting and difficult question - I am grateful I do not have to ponder that out of sheer necessity.

What I see now is that LLMs have some specific advantages over humans. For example, the more capable models can keep a much larger amount of information in their working context at once - something I have found much less useful with smaller local or free models.. They can therefore do tasks "in memory" for which I would need pencil and paper, or several intermediate steps. They can also access information on the Internet much faster than I can.

So I basically use them as tools that can be guided and directed by me - and much more efficiently and cheaply than a team of human assistants ever could.

For people doing intellectual work, I would argue that the important thing is to develop the skills needed to make the best use of these tools. That being said, it still takes a lot of expertise and problem-solving ability to direct them properly - at this point, they still need expert guidance and supervision.
My best guess would be that the gap between good and bad engineers will probably widen. There may be fewer people doing this kind of work, but those who remain in demand will be highly skilled and very productive.

So I would invest less in learning one particular new technology, and more in strengthening problem-solving skills: learning how to frame a problem, break it down, question assumptions, move between different levels of abstraction and determine which problem actually needs to be solved. Domain expertise certainly helps, but these skills transfer remarkably well between domains.

Quite often, the important question is not how to perform a task correctly, but whether it is the right task to perform in the first place. Those decisions should not be left to managers alone, and AI is certainly not suited to making them either.

That is something I have been practising for decades. Story of my life. Even without the AI part.

Nice find, but:

a. More interesting than the "no match: Go" line is the line immediately before it:

Remote protocol version 2.0, remote software version Go

So there really seems to be an SSH server responding — just apparently not OpenSSH. The presented server string seems to be "SSH-2.0-Go".

b. The odd part about that is that the server presents a known host key:

Host '192.168.11.1' is known and matches the RSA host key.

That can simply mean the user accepted that key on an earlier connection, however that host key was exactly the one that was stored for 192.168.11.1. So it would be useful to check whether this fingerprint actually matches the SSH host key of the OPNsense box.

c. I would compare both the banner and the host key directly:

Code Select Expand

# on OPNsense
nc 127.0.0.1 22
ssh-keygen -lf /conf/sshd/ssh_host_rsa_key.pub

# from the affected client
nc 192.168.11.1 22
ssh-keyscan -t rsa 192.168.11.1 | ssh-keygen -lf -

If localhost says OpenSSH... while the client gets SSH-2.0-Go, then something in the network path — e.g. NAT/redirect/proxying — is clearly intercepting the connection. The interesting question is: who answers here? I doubt that it is OpenSSH on OpnSense, so the version change does not matter.

Einfach mal die Suche nutzen, wie hier unter Punkt 16 beschrieben?

https://forum.opnsense.org/index.php?topic=21839

Oder sogar in der offiziellen Dokumentation:

https://docs.opnsense.org/manual/how-tos/pppoe_isp_setup.html

Ob das WAN-Interface auf einem VLAN aufsetzt, also die OpnSense selbst taggt oder nicht (also der Fraytek taggt) ändert nichts: WAN ist immer PPPoE bei einem solchen Setup.

From one retiree to another: On a side note, just because that is probably not too obvious for everybody (I learned this just now): Try the paid versions of any cloud LLM - they are order of magnitude better that just the free crap.

I found over the course of the last two years that the free stuff did not get better, but even seemed to get worse. This may have been the result of the companies trying to reduce non-paid efforts. I found that answers came too early and were imprecise to a level that they caused more harm than good. Switching to a paid version and selecting a better model immediately changed the game. In their current shape, cloud LLMs can really help speed up things. I could do most of that stuff myself, but it would take me much longer.

Could be depending on the type of key or SSH client used, especially if the key and/or client is very old and/or defective (there were compromised / blacklisted keys because of a bad RNG). The defaults for key security are ever increased with newer OpenSSH releases. That is, bad or insecure mechanisms are slowly being thrown out.

Try to create new SSH keys with a modern client, install the public key and try if that works.

Did you put the actual string "Hourly config backup to Git/Forgejo" into the cron job parameter for "remote backup" instead of a random delay?

Patrick is right. Don't bother with USB adapters on OpnSense, use APs.

As for the router idea: I think you are mixing up concepts here. If the router really acts as a router and does NAT, OpnSense will not be able to discern between devices behind that router, so no. If it can be repurposed as a switch, then in a sense, yes. However, all devices connected to the switch can interact without OpnSense even interfering.

Also, what you can do is limited once the devices have internet access, because traffic might happen on different ports and encrypted while you cannot see what's in it. Complete control is a myth - unless you block everything.

Rules per device is possible, but then again: how do you even identify a specific device? By IP? By MAC? Good luck with either.

That is perfectly possible, but see #24.

I have to make three corrections to formerly discussed points, though:

1. PCIe actually has its own mechanisms for data-integrity protection between the card and CPU/chipset: LCRC with link-level replay can detect and recover transmission errors, ECRC can provide additional end-to-end error detection, and AER can report PCIe errors.
This means that the observed data corruption may have occurred somewhere outside the portion protected by PCIe's link-level CRC/replay mechanism.

2. The X570S chipset was formerly believed to be different from the X570, yet there is no proof of that. On the contrary, ASRock said it was only fine-tuning that made passive cooling possible. Other manufacturers made claims to this as well.

3. In the meantime, my counter-proof to the PCIe initialisation on warm boots was contradicted - see: https://github.com/torvalds/linux/commit/ae1737e7339b513f8c2fc21b500a0fc215d155c3 and https://bugzilla.kernel.org/show_bug.cgi?id=220770

Maybe this was an implementation fault in either the Windows driver, defective NICs or bad board layout on Asrock's behalf - I cannot tell nor investigate, because I do not have the cards any more. All I can say is that the problems existed and were solved by switching to the other NIC - which also has the advantage of being able to fully use 10 GbE over my 17m CAT5 link, which could not reliably be achieved with either an X550 or X540 before.

On a side note: chalk those inaccuracies up to my discontent about having to investigate such problems for weeks when I just wanted to have more speed on my main workstation.