Home network is bullying

[m k]

Happened so that one wired W10(laptop) said can't find the server.
Then another wired W10(desktop) said the same.

Then testing and desktop couldn't ping the router.
In between is two switches and inside a wall cablings, central 16 port switch has also few bad ports.

Testing continues and directly to router wired Linux laptop can operate just fine.
Then third directly wired W10(desktop) can't access the router.

Testing continues and first desktop is connected directly and is getting DHCP addresses but can't ping.
Then W10 laptop is connected through bigger switch connected wireless AP and operates just fine.
Third W10 desktop has wireless also and connect using the same AP just fine.

So at this point the situation is that 2/3 W10 machines are connected wirelessly just fine.
Between AP and router is still that 16 port switch and wired connections to same physical output port are not working.
Wireshark of the third machine can't see its NIC.

Verdict, Windows update has messed Ethernet interface but left WLAN interface intact.
Resetting, cleaning and whatnot of that last W10 machine but no cheese.

Then heavy stuff, Windows reinstall of first desktop.
Still no wired connection, can ping though, like the other desktop occasionally.
More of that kind, 10Mb hub and old Vista laptop for Wiresharking what is actually happening inside the cable of that third machine.

Then watta moment, Vista laptop can't access internet through the hub.

Router factory reset and everything is fine again.
Minus wasted time and anger.

The question.
Can the router see differences between following connections?
Computer 1 - switch port A to X - router port Y
Computer 2 - switch - switch port B to X - router port Y
Computer 3 - wireless AP - switch port C to X - router port Y
Computer L - router port Z

[nightfire]

Yes. The router operates also with the MAC adresses bound to every interface, that the switch also has to cache in its buffer. Moresome, on a computer you have the arp cache that also plays some role in name resolution and discovery.
It is not unusual that some stuff (especially in those small devices) sometimes gets a bit of a hiccup after a few weeks of 24/7 operation and a reboot will fix this- There are some brands of routers (that also do other fancy stuff in addition to rounting) where a reboot once a month is common knowledge and recommendation alongst the community...

[m k]

Yes, my bad, I was living at the IP level.
Surely there must be a physical address also somewhere.

But this ADSL + 4G hybrid box has then been pretty decent since initially rebooting wasn't in my mind at all.
Anyway, nasty thing.

Few more things, time and configuration.
There usually is a situation where something is better happen sooner than later.
Home router is also most likely pretty capable of doing its thing with factory settings.

In my case the router was not in factory settings because I once had two and both had same network addresses.
Some static addresses are also present but difficult ones are long gone already.

The hybrid box has its default DHCP range 3-253 but it's also using /16 prefix.
So all static stuff can also be far away but it's not really a home network anymore if machine count is many dozens, yes, IoT is changing that.
Now I changed the range starting from 100 but I guess changing those statics starting from 50 or 60 is better.

One can also think that changing factory address space farer from the start of the class is some kind of a protection against malware, like the case where malware was scanning only few from the beginning of the 3rd octet.
It's obsolete news when you read it.

Then a Windows thing.
Setting an IP address manually is quite a bit away from what it used to be.

[MIS42N]

Yes. The router operates also with the MAC adresses bound to every interface, that the switch also has to cache in its buffer. Moresome, on a computer you have the arp cache that also plays some role in name resolution and discovery.
It is not unusual that some stuff (especially in those small devices) sometimes gets a bit of a hiccup after a few weeks of 24/7 operation and a reboot will fix this- There are some brands of routers (that also do other fancy stuff in addition to rounting) where a reboot once a month is common knowledge and recommendation alongst the community...

The arp cache is loaded with the IP and MAC addresses of devices that are discovered on the local network. If a machine 1 tries to contact machine 2 for the first time, and the IP mask says it is on the same network, then machine 1 sends a broadcast packet saying 'are you there'. Machine 2 replies 'yes'. At the lowest level, all communication is via interface addresses (on ethernet and wifi, the MAC address). Both machines will then store each other's IP and MAC address in their arp caches. All communication then takes place machine to machine using MAC addresses.

If the mask says the IP address is NOT on the same network, machine 1 will send to the router. Unless the only communication is local, every machine needs to know the router IP address. If hard coding the IP address in a machine, also need to hard code the router and get the network mask right. If using DHCP, DHCP provides the router address and mask as part of negotiations.

Windows machines, if configured for DHCP but can't contact the DHCP server will often load a default address, which usually means it can't talk to anything. The solution is to find out why DHCP is not working and fix it. Putting in static addresses is a recipe for later head scratching. If you want a fixed IP address, most DHCP servers can associate a particular MAC address with an IP address. Use static addresses sparingly.

Switches are sandwiched between machines and have to learn MAC addresses. When machine 1 sends any message to the switch, it records the MAC address as being on that port. If the message is a broadcast (the 'are you there' message) the switch will send it out all ports. But if it is a specific MAC address it will know which port to send out, so only 2 ports are active. As far as I know, switches don't know IP addresses.

Unplugging from one port and into another should be detected by the device with the ports, but it's not bulletproof. The MAC address to port mapping has to be relearned, that means software, and that means may not work.

Which means in any extended network, things can get messy. The shotgun solution to solving a problem is to turn everything off, then turn on the DHCP server (or servers). DHCP servers should always have static addresses so don't rely on any other equipment. Then turn on anything connected to the DHCP server (which if it is your wifi router usually means everything else). If there are wireless APs they are turned on before the machines that rely on them. Don't forget the printers.

I have two DHCP servers on the same network but with different address ranges. Unless both are unreachable (almost never), I don't get the 'Windows booted and decided on a different network' problem.

IPv6 fixes some of this, and sometimes things work because the target is reachable by IPv6 but not IPv4. This can lead to machines not connecting to the Internet because the destination needs IPv4. Although IPv6 is definitely the way to go, it sometimes makes things 'interesting'.

[PKTKS]

(..)
The question.
Can the router see differences between following connections?
Computer 1 - switch port A to X - router port Y
Computer 2 - switch - switch port B to X - router port Y
Computer 3 - wireless AP - switch port C to X - router port Y
Computer L - router port Z

Well what do you expect from these W10 crappy pile of junk ?

They crafted 3rd drivers and stuff to run their extortion algorithm....
Clean implementations of a sane TCP/IP stack can readily flush the ARP cache..

Recently I have wiped out some W10 machines totally crippled with
bad TCP/IP stacks and  routing issues... 

FLUSH YOUR ARP CACHE:

Code: [Select]

>arp -v -n
Address                  HWtype  HWaddress           Flags Mask            Iface
192.168.10.1            ether   x8:ax:f7:xb:x3:xf   CM                    bonded_IF
192.168.10.2                     (incomplete)                                  bonded_IF
192.168.10.5            ether   x4:xc:x3:x8:x3:x7   CM                   bonded_IF
192.168.10.7                     (incomplete)                                  bonded_IF
.
.
.
.

to make sure your broadcasts are properly done..
in particular trying to resolve those (incomplete) arps

Needless to say  to get rid of these W10 junk..

Paul

 

[JohnnyMalaria]

Seems that Windows 10 is getting blamed for the router acting up. All these problems including on a Vista machine (i.e., not Windows 10) that miraculous sort themselves out after rebooting the router???

[PKTKS]

Seems that Windows 10 is getting blamed for the router acting up. All these problems including on a Vista machine (i.e., not Windows 10) that miraculous sort themselves out after rebooting the router???

Not to my knowledge... in the past 10m or so the pile of problems
on the WXX machines related to bad routing issues and totally
crippled "updates" ...

made me so angry to the point of wiping them totally out..

After one of those "upgrades"  some machines just gone for a walk
and never booted again

Replaced them for dummy users with clean MINT replacement..

Usually the folks get used to MINT very fast..
and drop W10 totally..

Paul

PS.  You sum that shitty OS  with the lack of a proper "MIDIA"  on recent installations..
and the maintenance of infra structure  based on "Internet" recovery
has just hit the unsustainable point..
 

[PKTKS]

e.g...    as pathetic as it may be...

An OS that refuses to properly boot after some upgrades..

.. and when it does...    it insists  in doing Internet based  "recovery"..

with a totally crippled TCP/IP stack that can not  do proper routing
or find a gateway...

Is sounds  ridiculous,..  but that occurrence  repeated some dozen times..

Enough.  Wiped that out.

Paul

[m k]

Seems that Windows 10 is getting blamed for the router acting up. All these problems including on a Vista machine (i.e., not Windows 10) that miraculous sort themselves out after rebooting the router???

Windows is getting much blame because it's on the table and actively what ever, that is many times just that.
Then obviously it's among the first ones also when it shouldn't.

My case went to wrong direction mostly because of untrusted mechanics of connections and earlier central switch hickups.

Once I also had a situation where a machine had manual IP settings and was pinging the router.
Answer was that differently(can be APIPA) addressed machine can't reach the router.
Then checking IP through W10 setup and all but DNS server setting were empty, same through the old way.

Originally settings were done old way,
since W10 setup said that can't save, something must be changed,
when IPv6 was off and IPv4 had valid values in all boxes.

I think the main problem MS has with Windows is the constant need of new automations.

Linux then has a different problem, graphical interfaces, like dropping system-config-samba.

[madires]

It is not unusual that some stuff (especially in those small devices) sometimes gets a bit of a hiccup after a few weeks of 24/7 operation and a reboot will fix this- There are some brands of routers (that also do other fancy stuff in addition to rounting) where a reboot once a month is common knowledge and recommendation alongst the community...

Yep, that's a common problem with the cheap plastic boxes. In many cases it seems to be caused by power supply issues, like a too cheap wall wart or bad electrolytics in the buck converters. In other cases by poorly written software. From my experience the plastic boxes can run quite stable for long periods when powered by a good quality PSU.

[Syntax Error]

Just a question about what you said earlier...

The hybrid box has its default DHCP range 3-253 but it's also using /16 prefix.

Do you mean your subnet is? 255.255.0.0

This yields a LAN IP range from 192.168.0.0 to 192.168.255.255

In theory, the subnet should be 255.255.255.0 or /24
Otherwise the DHCP server or W10 might just freak out.

[m k]

It's not that straight forward, class C private is 192.168.0.0/16.

But anyway, I've got a new network range and it's inuse, 192.168.1.0/24.
Others are 176.16.0.0/16 and 10.0.0.0/16, not exactly kosher, in a sense, with the first one.
The box is Fast5370AIR-01.

Earlier only class C alternative was 192.168.0.0/16 and I'm quite sure it was also so yesterday.
The router is now 192.168.1.1 and not full Sherman but close that the first client after change was 192.168.1.100 with mask 255.255.0.0.

I've also heard that around here was an online banking attack few days ago.
Maybe ISP did some indirectly forced resets, and yes, it's a bit tin foil.
(just wondering those agronyms, like GCHQ)

[madires]

From RFC 1918:
- 10.0.0.0/8 (10.0.0.0 - 10.255.255.255)
- 172.16.0.0/12 (172.16.0.0 - 172.31.255.255)
- 192.168.0.0/16 (192.168.0.0 - 192.168.255.255)

And a class C network is a /24 (netmask 255.255.255.0). So you can create 256 class C networks inside 192.168.0.0/16:
- 192.168.0.0/24
- 192.168.1.0/24
...
- 192.168.255.0/24

[MIS42N]

You probably know all this, but just in case it is useful:

The mask on an IP address determines if a machine tries a direct connection or hands the problem off to a router. It will try the direct connection if the machine it wants to contact has the same address when masked as its own. So 10.14.23.56/16 will try to contact 10.14.27.44/16 directly, but 10.14.23.56/24 to 10.14.27.44/24 will hand it off to its designated router, even if they are hanging off the same switch. The important thing is all machines on one side of a router and the router port(s) should have the same masked address otherwise the router gets hammered unnecessarily. If there's 400 machines, then make the mask /23 (255.255.254.0), there's nothing that says it has to be a multiple of 8 bits.

It is not necessary, but is desirable that the private network addresses be used for private networks. It is possible to set up a network using the same IP address range as Yahoo, but you'd never be able to send or receive to Yahoo. For preference I use 10.x.x.x whenever I'm setting up a network, just because it has huge headroom. But any of the others are fine.

If multiple networks are needed then it's necessary to start managing the router tables in the routers. The default router action is 'send it to the Internet', so routers have to be told that local traffic is local traffic. And routers on the Internet will not route private addresses.

DHCP is a good way of standardising the router address, the DNS addresses, the correct mask.

DNS is sometimes a problem. Some routers support DNS with a caching server. Nice for reducing traffic to the Internet but sometimes dodgy in implementation. If multiple DNS addreses can be configured in DHCP then one local and and one (or more) Internet based ones is good.

[m k]

The hybrid box in question,
it has a dropdown list of LAN/DHCP ranges and that's it.
After selection all settings are free but when I tried to change that original /16 mask the box refuced.
One must wonder what kind of a programmer has been involved.

A bit more from IPv4 addresses.
The class is
A when MSb is 0,
B when MSb is 1 and next is 0,
C when 2 MSbs are 1 and next is 0,
D when 3 MSbs are 1 and next is 0 and
E when 4 MSbs are 1.

Class A, B and C defines *default* subnet masks(/8 /16 /24), witch are not a must in any way, but max width of course.

Nowadays it's not unusual that ISPs use semi private addressigns where the whole segment is under one private space.
Around here a business customer side is many times class A private /24.
Quite nice for global privacy thingies, in case you trust your ISP.

So if Windows is stuck in /24 class C private network it's clearly not very up to date.

What kind of a static public IPv4 address one gets from ISP.
My guess is that what ever it is the mask is /30.

But there's more.
https://www.ucl.ac.uk/isd/services/get-connected/wired-networks/use-private-rfc1918-address-space-ucl-internal-network

"In practise addresses in 192.168.0.0/21 (ie. 192.168.0.0 through 192.168.7.255 inclusive) will never be allocated."

"Note: there is an exception to this. The School of Pharmacy were using 192.168.6.0/23 so this prefix is now routed on the backbone."

It seems that one better stay quite low within the class.
And maybe that earlier malware was from London metro area.

[Monkeh]

But there's more.
https://www.ucl.ac.uk/isd/services/get-connected/wired-networks/use-private-rfc1918-address-space-ucl-internal-network

"In practise addresses in 192.168.0.0/21 (ie. 192.168.0.0 through 192.168.7.255 inclusive) will never be allocated."

"Note: there is an exception to this. The School of Pharmacy were using 192.168.6.0/23 so this prefix is now routed on the backbone."

It seems that one better stay quite low within the class.
And maybe that earlier malware was from London metro area.

That is entirely to do with their internal network and has nothing to do with you.

[madires]

The hybrid box in question,
it has a dropdown list of LAN/DHCP ranges and that's it.
After selection all settings are free but when I tried to change that original /16 mask the box refuced.
One must wonder what kind of a programmer has been involved.

Which vendor and model?

A bit more from IPv4 addresses.
The class is

Classful addressing is totally outdated. We're using CIDR for more than 25 years! Please see https://en.wikipedia.org/wiki/Classless_Inter-Domain_Routing.

What kind of a static public IPv4 address one gets from ISP.
My guess is that what ever it is the mask is /30.

Usually you get a single dynamic IPv4 address (/32). Based on your country or ISP a static address could be also possible (still a /32). For business internet access you might get a small subnet. In conjunction with IPv6 many ISPs go for DS-Lite and you won't have a dedicated public IPv4 address anymore, i.e. each customer gets a private IPv4 address and those private addresses are NATted by the ISP.

[m k]

You can call it what you like but it's still the same 32bit address.

It's also very irrelevant how outdated that marking method is if it is required.
Another thing is if the meaning of an address has changed.

The box model is in reply #11, maker is Sagemcom.

That is entirely to do with their internal network and has nothing to do with you.

Nothing of this has had nothing to concretely do with me after that router reset.

Try to look it a bit wider, it was an example what can happen.
A person who is acting in the area of that backbone can't really trust many things concerning private IPv4 networks.
It's another thing how wide that kind of a practice is, and how well known even after excess information drumming.

"OT"
It seems that I may have read W10 manual IP settings by just moving my eyes.
Subnet mask box is now prefix length.

[madires]

It's quite strange that Sagemcom would do such BS.

[Monkeh]

It's quite strange that Sagemcom would do such BS.

Sagemcom make garbage boxes to be filled by ISP-molested firmware for a living. That they behave badly is a given.

[james_s]

I had some similar weird symptoms a while back that turned out to be a corrupted configuration on my router.

If you have a switch with bad ports it's probably prudent to replace that too. You don't really know what's going on under the hood, the bad ports might be the tip of the iceberg. It could be spewing out all sorts of garbage.

[m k]

My case has still another thing, the mobile mast.
It's so and so visible from behind the next house when the router is beside the wall, above the selective window and just below sheet metal roof side.

Then there is that DSL phone tap, it's not even near that router location

So there is a DIY cable.
An UTP stump where one end is RJ45 wall mount female and the other is last pair taken out RJ11 and normal 2 pairs RJ45.
That is then continuing to the wall socket by normal cable, maybe 15m, and then to the cross linkage panel, where a stump again split it and there the phone socket is the old 3-banana + RJ11 model.
Total length is probably less than 40m.

Final length to DSLAM is unknown.
Fibre is coming next summer.

The central switch is Digitus from close to last millenia, I guess.
Should have been out of the window a long ago but then again, working expressionlessly.
I've also moved all used ports to the other half of the switch.
Surely they are now far far away from possible problem area, even that I obviously don't have glue how those ports are internally connected.

[m k]

One more thing, a forgotten W7 laptop.

Once the situation was so that the central switch had issues pretty regularly.
It was finally sorted out so that one connection(that W7) made two switches hang, when simultanously mobile network was unable to support the area reasonably.

On its way the situation took couple central switch ports out.
Assumption is that the hybrid box is actually not very hybrid, it just uses DSL when mobile is not available and that DSL use through that DIY cable was too much for those switch ports.
It's also so that the hybrid box doesn't like to be without that DSL, or at least didn't when that /24 DHCP range was not available.

That W7 laptop then, also the one that rejected W10 upgrade,
it finally lost all it's network connections, including internal LAN and WLAN, plus USB connected LAN.
Windows of course was the initial accused one but Linux installation couldn't use internal stuff eighter, USB was ok.

And BTW, the wall has a double RJ45 socket.

[TomS_]

The question.
Can the router see differences between following connections?
Computer 1 - switch port A to X - router port Y
Computer 2 - switch - switch port B to X - router port Y
Computer 3 - wireless AP - switch port C to X - router port Y
Computer L - router port Z

The router, via the MAC address table of its switch, knows which port a particular MAC address lives on. But it doesnt know how many other devices are in between it and any given MAC address. The MAC address can be the first device encountered out of a switch port, or it can be many hops down the line.

The purpose of the MAC table is not to know how far away something is, only where it can be reached. There are other protocols that can be used to figure out how far away something is, but they are generally operating at a higher layer.