Re: 5.41 - Authentication Failed with kernel 4.2.0

Hi Johan,

quoted

quoted

quoted

quoted

ACL Data RX: Handle 70 flags 0x02 dlen 11                    [hci0] 28.348460

    SMP: Pairing Response (0x02) len 6
      IO capability: NoInputNoOutput (0x03)
      OOB data: Authentication data not present (0x00)
      Authentication requirement: Bonding, No MITM, Legacy, No Keypresses (0x01)
      Max encryption key size: 16
      Initiator key distribution: EncKey Sign (0x05)
      Responder key distribution: EncKey IdKey Sign (0x07)
< ACL Data TX: Handle 70 flags 0x00 dlen 21                    [hci0] 28.348510
    SMP: Pairing Confirm (0x03) len 16
      Confim value: 12bd199ba4739a017395109cfa4621b3

quoted

ACL Data RX: Handle 70 flags 0x02 dlen 21                    [hci0] 28.488508

    SMP: Pairing Confirm (0x03) len 16
      Confim value: 41f5f7a6212356d0eefa268f7994844a
< ACL Data TX: Handle 70 flags 0x00 dlen 21                    [hci0] 28.488545
    SMP: Pairing Random (0x04) len 16
      Random value: 4e8d532caf34ab845ff243acf6a0a681

quoted

ACL Data RX: Handle 70 flags 0x02 dlen 6                     [hci0] 28.628456

    SMP: Pairing Failed (0x05) len 1
      Reason: Confirm value failed (0x04)
< HCI Command: Disconnect (0x01|0x0006) plen 3                 [hci0] 28.628506
      Handle: 70
      Reason: Authentication Failure (0x05)
@ Authentication Failed: 5C:31:3E:5E:54:5E (1) status 0x05

That "Confirm value failed" response still looks like something strange
is happening on the remote side - maybe a bug there. Unfortunately I
don't really have any other ideas except going for raw bisecting here if
you have a kernel version this did work with.

The inputs to the function that generates the confirm value are the
local and remote public keys and the random number that was sent over
the air, so there's not much that can go wrong there - also nothing I
can remember that would have changed on the Linux side regarding this
for a long time. This of course assumes that the remote device is using
this error code correctly and doesn't send it for some other condition.

I bet that the peripheral side wrongly clears some RFU bits. Which
leads to wrong values for confirm values.

Johannes, could you try to experiment and see whether this is the issue?
You can temporarily disable Secure Connections support with the command
"btmgmt sc off"

I really wonder if we should start using HCI_MON_SYSTEM_NOTE / send_monitor_note to actually send these funky behaviors to the monitor channel once we detect it. At least that way, we would know why we trigger certain PDUs.

Or maybe we are doing something specific for SMP with a debugfs switch that allows us to enable to send SMP debug / tigers via monitor channel. An alternative would be to just be able to push the mgmt traffic via monitor channel so we actually have correct timing and can analyze these things together.

Regards

Marcel