Re: [rt2x00-users] [PATCH RFC] rt2500usb: disable broken HW encryption by default

[Ondrej Zary]

On Tuesday 23 March 2010, Ivo Van Doorn wrote:
> On Tue, Mar 23, 2010 at 4:09 PM, Ondrej Zary <linux@xxxxxxxxxxxxxxxxxxxx> 
wrote:
> > On Tuesday 23 March 2010, Ivo Van Doorn wrote:
> >> On Tue, Mar 23, 2010 at 10:27 AM, Ondrej Zary
> >>
> >> <linux@xxxxxxxxxxxxxxxxxxxx> wrote:
> >> > On Monday 22 March 2010, Ivo Van Doorn wrote:
> >> >> >> But I though it was mentioned that disabling HW crypto didn't
> >> >> >> solve the issue due to a second bug in a later kernel?
> >> >> >
> >> >> > That was a false positive. Probably because the device was not
> >> >> > unplugged between the tests (and looks like the driver does not
> >> >> > initialize the chip completely). It's not reliable, it sometimes
> >> >> > stops working after reboot.
> >> >>
> >> >> Ah well that at least simplifies the problem. I'll have to retest
> >> >> rt2500usb soon to see why the HW crypto failed. I am sure I had it
> >> >> working for WEP, WPA and WPA2
> >> >> before I submitted the patch.
> >> >
> >> > So let's try to fix it instead of disabling.
> >> >
> >> > First, the unrealiability (keeping HW encryption disabled). With the
> >> > driver loaded but not doing anything more, the register dumps are same
> >> > for both working and non-working case (dump-init.txt).
> >> >
> >> > dump-good-connected.txt is a dump after successful association and
> >> > DHCP dump-bad-attempt.txt is a dump after successful association
> >> > during non-working DHCP attempt
> >> > dump-bad-after.txt is a dump after DHCP timed out
> >>
> >> With association working, but DHCP failing it most likely means that
> >> somehow the frame was malformatted.
> >> The code for HW crypto alters the frame (alters IV/EIV/ICV data etc).
> >> And that is commonly the source of
> >> problems, because what has to be done depends heavily on the encryption
> >> type.
> >>
> >> So could you verify which of the encryption types (WEP,WPA,WPA2) is
> >> failing or working? That would give a starting
> >> position on which bytes might be corrupted.
> >
> > I was testing only with WPA2 before. I did some more testing today. The
> > results:
> >
> > No encryption - works always
> > WEP - sometimes works, sometimes not - same with and without HW
> > encryption WPA - sometimes works, sometimes not - same with and without
> > HW encryption WPA2 - never works with HW encryption
> >     - sometimes works, sometimes not without HW encryptionn
> >
> > So it seems that there are two problems:
> >  1. random problems with any encryption
> >  2. WPA2 is broken with HW encryption
> >
> > When the "random" problem appears, this appears in dmesg:
> > wlan1: authenticate with 00:13:d4:0f:f3:17 (try 1)
> > wlan1: authenticated
> > wlan1: associate with 00:13:d4:0f:f3:17 (try 1)
> > wlan1: RX AssocResp from 00:13:d4:0f:f3:17 (capab=0x411 status=0 aid=2)
> > wlan1: associated
> > phy0 -> rt2500usb_set_device_state: Error - Device failed to enter state
> > 3 (-16). phy0 -> rt2500usb_set_device_state: Error - Device failed to
> > enter state 3 (-16). No probe response from AP 00:13:d4:0f:f3:17 after
> > 500ms, disconnecting.
> >
> > Disabling call to rt2500usb_set_state() in rt2500usb_set_device_state()
> > seems to fix problem 1. After this change, WEP and WPA work always
> > regardless of HW encryption (and WPA2 works always without it).
>
> Ok, lets focus on the HW crypto for now.
>
> If WPA2 fails, then the WPA2 specific code for IV/EIV/ICV isn't
> working. Most likely
> the device either doesn't send all data back to the driver (while the
> driver does expect it)
> or it adds additional data which the driver doesn't expect. (See
> rt2x00crypto.c for the frame
> manipulation during HW crypto).
>
> Could you check the full packet data of a DHCP request with and
> without HW crypto?
> That might give a clue about what the hardware is sending for extra data.

How do I access the full packets? I tried using another machine with "iwconfig 
wlan0 mode monitor" and tcpdump. It captured many packets and I'm unable to 
identify the interesting ones.

-- 
Ondrej Zary
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/