PfSense on a Riverbed Steelhead

stephenw10

Ok, so unless you can see any jumpers that might set them then, as I said two years ago, you will have to start poking at GPIOs to try to find what controls them. You might find a clue in the RIOS boot logs or filesystem.

Steve

Okijames

Regarding GPIOs, rather than poking around thi maybe this can be of use. Using ipmitool under Linux, these commands are said to enable inpath nics. Never tried it myself.

ipmitool -v raw 0x3e 0x20 0x80 0x7f 0x00 0x00
ipmitool -v raw 0x3e 0x20 0x80 0x7f 0x01 0x00

I think the above alters some i2c registers to do the trick. Try it under Linux first. If it works, try a FreeBSD method for setting such registers.

I took a look at the, again Linux based, approach used by rbMode on github. That guy's script refers to SMBUS and I can't figure how how SMBUS addresses relate to the ipmitool addresses.

Okijames

Quick note: Riverbed used Silicom bypass nics in their larger appliances. Silicom has FreeBSD drivers with source code on their web site. The source code might provide insight into methods for enabling your bypass nics. URL for Silicom 2port bypass nic here....

https://www.silicom-usa.com/pr/server-adapters/networking-bypass-adapters/gigabit-ethernet-bypass-networking-server-adapters/pe2g2bpi80-bypass-card/

stephenw10

Hmm, yeah that's interesting. It could almost certainly be done just using shell script then. Just need to figure out the smbus addresses......

ipmitool is already installed though and takes the same input as Linux so should would if that data in correct. Though it still requires an IPMI device of some kind in raw mode.
Anyone able to test it?

pauloalb

@Okijames said in PfSense on a Riverbed Steelhead:

ipmitool -v raw 0x3e 0x20 0x80 0x7f 0x01 0x00

Just tried it on my riverbed 250 and unfortunetly it did not work. Error is:

Could not open device at /dev/ipmi0 or /dev/ipmi/0 or /dev/ipmidev/0: No such file or directory

There is a python script that i have tested to work but only on linux, Debian in my case. Pfsense's FreeBSD no luck so far.
Maybe the script can help : Script

stephenw10

What happened when you ran rbmode in pfSense? What error?

You probably have to load the smbus drivers for that to work:

kldload ichsmb
kldload smb

Steve

pauloalb

@stephenw10 said in PfSense on a Riverbed Steelhead:

kldload smb

Hello,

Thank you stephenw10 for your help.
First the script imports smbus module but cant find it. I had to install pip and then smbus2 and change the script. That error went away.
I executed the kldload commands with "success".

After that rbmode gave this error:

[2.3.5-RELEASE][root@pfSense.localdomain]/usr/local/sbin: rbmode u
Setting mode to universal

Traceback (most recent call last):
File "/usr/local/sbin/rbmode", line 78, in <module>
main()
File "/usr/local/sbin/rbmode", line 70, in main
setMode(args.mode)
File "/usr/local/sbin/rbmode", line 38, in setMode
s = smbus2.SMBus(0)
File "/usr/local/lib/python2.7/site-packages/smbus2/smbus2.py", line 279, in init
self.open(bus)
File "/usr/local/lib/python2.7/site-packages/smbus2/smbus2.py", line 308, in open
self.fd = os.open(filepath, os.O_RDWR)
OSError: [Errno 2] No such file or directory: '/dev/i2c-0'

Indeed there is no /dev/i2x-0 but there is a /dev/smb0 so i replaced the reference in smbus2.py from i2c to smb but then another error occurred after re-executing rbmode:

IOError: [Errno 25] Inappropriate ioctl for device

Note: I got this script to work on Debian flavour of linux with success.

My knowledge of python is very very low as is FreeBSD so i am stuck at the moment...

Okijames

FWIWI tried compiling the Silicom drivers on a FreeBSD 10.3 (to match pfSense 2.3.5) and turns out the bpmod.ko kernel module (note this is pre-built, NOT actually available as source code in the driver package) won't work, it complains about not finding a Silicom card, as expected I guess...

stephenw10

With those smb modules loaded try running: smbmsg -p

That should return all the available smbus devices available. But it also might hang the bus or even the whole box so be ready to power cycle it. It will probably list some things though. Hopefully including 0x24.

The smbus python tools appear to be more like raw i2c tools. The actual output command in rbmode appears to write to a device at 0x24 with subaddress (or smbus command) 0x55 and whatever the output of that checksum function is. My python sucks!

Steve

You can probably do that easily enough with smbmsg.

Okijames

root@:~ # kldload ichsmb
ichsmb0: <Intel 631xESB/6321ESB (ESB2) SMBus controller> port 0x540-0x55f irq 19 at device 31.3 on pci0
smbus0: <System Management Bus> on ichsmb0
root@:~ # kldload smb
smb0: <SMBus generic I/O> on smbus0
root@:~ # smbmsg -p
Probing for devices on /dev/smb0:
Device @0x10: w
Device @0x48: rw
Device @0x5c: rw
Device @0x60: rw
Device @0x64: rw
Device @0x66: rw
ichsmb0: device timeout, status=0x41
ichsmb0: device timeout, status=0x41
ichsmb0: device timeout, status=0x41
ichsmb0: device timeout, status=0x41

Okijames

That's what I get, and then the "device timeout" messages continue forever. Rerunning smbmsg -p shows nothing but the timeout messages, so pretty much need to reboot after running once. No x24 addresses of course! curses...

stephenw10

Hmm. Have you also seen that script work in Linux?

I could imagine the device might come up at a different address in FreeBSD perhaps...

Okijames

@pauloalb Since you got the script to work under Linux...

What were the values for "calcCheckSums" in this bit of the script?

s.write_block_data(0x24, 0x55, calcChecksums(cmd_0))
time.sleep(0.1)
s.write_block_data(0x24, 0x55, calcChecksums(cmd_1))

This would give us clues as to what we should give smbmsg to send the same values under BSD. Changing the 0x24 address location per what we see from using "smbmsg -p" .

Okijames

Per some docs...

write_block_data(i2c_addr, register, data, force=None)
Write a block of byte data to a given register.
Parameters
• i2c_addr (int) – i2c address
• register (int) – Start register
• data (list) – List of bytes
• force (Boolean) –

So I'm thinking the smbmsg equivalent could be something like this per my smbus scan...

smbmsg -s 0x48 -c 0x55 -o 3 0x?? 0x?? 0x??

With the ?? pieces being the values the rbmod script calculates in the above bit of python.

stephenw10

With some extra print calls and no actual smbus writes:

steve@steve-MMLP7AP-00 ~/Documents $ sudo ./rbmode.py u
[sudo] password for steve:      
Setting mode to universal
[3, None]
[3, 1]
[3, 252, 1, 254, 102, 153]
[3, 0]
[3, 252, 0, 255, 102, 153]

So I expect the required smbmsg command to be something like:

smbmsg -s 0x24 -c 0x55 -o 6 0x03 0xfc 0x01 0xfe 0x66 0x99
smbmsg -s 0x24 -c 0x55 -o 6 0x03 0xfc 0x00 0xff 0x66 0x99

BUT I'm unsure how the checksum applies. i2c doesn't have a checksum but the script calculates the checksum from the data because smbus does. So if smbmsg does the checksum for us it might be:

smbmsg -s 0x24 -c 0x55 -o 2 0x03 0x01
smbmsg -s 0x24 -c 0x55 -o 2 0x03 0x00

That all depends on 0x24 being the real device address....

Steve

Okijames

Considering this comment was in the rbmode script all along...

universal

port 0: 06-03-fc-01-fe-66-99 -> 0x03, 0x01

port 1: 06-03-fc-00-ff-66-99 -> 0x03, 0x00

Lines up precisely with your converted output and possible smbmsg commands...

smbmsg -s 0x24 -c 0x55 -o 6 0x03 0xfc 0x01 0xfe 0x66 0x99
smbmsg -s 0x24 -c 0x55 -o 6 0x03 0xfc 0x00 0xff 0x66 0x99

I guess the info was staring us in the face this whole time :)

Now will it work? smbmsg syntax is cryptic, but will give it a go.

Thanks!

Okijames

Huge thanks to Steve, his print output of rbmode filled in the final pieces!

First a bit of a caveat: I did this with FreeBSD 12.1. I used a full install of a current FreeBSD version to improve chances of stuff working, have access to extra tools, drivers, etc. that are not in pfSense, let alone and old version of pfSense.

One thing to note BTW is 12.1 installed from a USB key inserted in the front panel directly to an internal SATA without issue. FreeBSD 10.3 and pfSense 2.3.5 based on same were unable to install directly to an internal SATA.

So anyway...

Success! Instant relay click and nic link lights with the first smbmsg command below! LAN0_0 (em0) and WAN0_0 (em1) both working!

Note they nics did require an "ifconfig emX up" before they could be used but that's to be expected.

smbmsg -s 0x48 -c 0x55 -o 6 0x03 0xfc 0x01 0xfe 0x66 0x99
smbmsg -s 0x48 -c 0x55 -o 6 0x03 0xfc 0x00 0xff 0x66 0x99

The second command doesn't seem to do anything but does not return an error. I rebooted, which does reset the relays, and tried the second command alone, still doesn't seem to do anything, good or bad.

So next steps would entail getting the following on pfSense, if it's not already there...

-ichsmb.ko
-smb.ko
-smbmsg

Then alter boot-up config to load the kernel modules and issue the smbmsg commands automatically.

I won't be doing any of the pfSense bits cause I don't actually want to run a 32bit box. It just always bugged me that half the nics wouldn't work without the original software.

Anybody want to buy one? :)

I think I have 2 250s and 2 550s. The 550s for sure (because that's been my box for all this hackery) have a Xeon Sossaman and 4GB ECC RAM. IIRC the 250s have a Celeron and 2GB ECC RAM. No USB DOMs, no HDDs, no software.

stephenw10

The modules are there in pfSense, just not loaded. To run that you would need to add to /boot/loader.conf.local:

ichsmb_load=yes
smb_load=yes

Then run that smbmsg with a shellcmd at boot. Pretty easy.
Of course not if that is 32bit hardware though.

I assume the second line is for larger models that have two pairs of ports.

The fact that code was already in the comments there was confusing. The way it's written looked like they got the i2c output from original OS maybe. The extracted the actual commands from it. But then recreate the same string by calculating the checksum... odd.

Steve

pauloalb

Special thanks to @stephenw10 and @okijames and everyone that helped.

I am on my lunch break so i couldn't test this much but i call this a SUCCESS! Both nics do light up and show activity.

I am running this on a Rivebed Steelhead 250 with pfSense 2.3.5 running from a flash drive plugged in the front of the device with no hdd and no subonModule internal flash plugged in.

The exact sequence that worked for me, gathering all the bits from previous messages is this:

kldload ichsmb
kldload smb
ifconfig em0 up
ifconfig em1 up
smbmsg -s 0x48 -c 0x55 -o 6 0x03 0xfc 0x01 0xfe 0x66 0x99
smbmsg -s 0x48 -c 0x55 -o 6 0x03 0xfc 0x00 0xff 0x66 0x99

Thanks again for the fantastic comunity support and engagement from everyone

pauloalb

stephenw10

Nice!

Interesting that it's apparently at a different address in FreeBSD/pfSense.

I wouldn't expect you to have to ifconfig up the interface assuming it is assigned and enabled. Usually in that sort of setup the relays disconnect the physical ports but the OS can always see the NIC chip.

Steve