vtpm manager NVM problem

[Erdem Bayer <ebayer@bayer.gen.tr>]

Hi

I am sending my previous mail about saving the state of vtpm with some 
new questions and information.

I am using xen 3.1.3 with these patches applied:

http://lists.xensource.com/archives/html/xen-devel/2008-02/msg01092.html
http://lists.xensource.com/archives/html/xense-devel/2007-04/msg00005.html

When I start the vtpm_manager I get the following output.

INFO[VTPM]: Starting VTPM.
INFO[TCS]: Constructing new TCS:
INFO[TCS]: Calling TCS_OpenContext:
INFO[VTSP]: OIAP.
INFO[VTSP]: Loading Key into TPM.
INFO[VTSP]: Unbinding 256 bytes of data.
INFO[VTPM]: Loaded saved state (dmis = 1).
INFO[VTSP]: Loading Key into TPM.
INFO[VTPM]: Creating new DMI instance 0 attached.
INFO[TCS]: Calling TCS_OpenContext:
INFO[VTPM]: [Backend Listener]: Backend Listener waiting for messages.
INFO[VTPM]: [VTPM Listener]: VTPM Listener waiting for messages.
INFO[VTPM]: [Hotplug Listener]: Hotplug Listener waiting for messages.
INFO[VTPM]: Re-attaching DMI instance 1.
INFO[TCS]: Calling TCS_OpenContext:
INFO[VTPM]: Launching DMI on PID = 4407
INFO[VTSP]: Binding 16 bytes of data.
INFO[VTPM]: Saved 256 bytes of E(symkey) + 656 bytes of E(data)
INFO[VTPM]: Saved VTPM Manager state (status = 0, dmis = 1)
INFO[VTPM]: [Hotplug Listener]: Hotplug Listener waiting for messages.

When I start a domain with the option vtpm = [ 'instance=1, backend=0' ] 
vtpm_manager on dom0 correctly starts a new vtpmd process with the 
following options:

vtpmd clear pvm 1

I can accomplish all tpm operations on this vtpm from domU. I can see 
the instance is recorded to vtpm database correctly:

cat /etc/xen/vtpm.db

#Database for VM to vTPM association
#1st column: domain name
#2nd column: TPM instance number
pardus-client 1

However, vtpm_manager complains about loading NVM.

TPMD[1]: tpmd.c:126: Info: Initializing tpm state: clear, type: pvm, id: 1

TPMD[1]: tpm/tpm_cmd_handler.c:4143: Debug: tpm_emulator_init()
TPMD[1]: tpm/tpm_startup.c:30: Info: TPM_Init()
TPMD[1]: tpm/tpm_testing.c:242: Info: TPM_SelfTestFull()
TPMD[1]: tpm/tpm_testing.c:42: Debug: tpm_test_prng()
TPMD[1]: tpm/tpm_testing.c:70: Debug: Monobit: 9880
TPMD[1]: tpm/tpm_testing.c:71: Debug: Poker:   19.3
TPMD[1]: tpm/tpm_testing.c:72: Debug: run_1:   2413, 2466
TPMD[1]: tpm/tpm_testing.c:73: Debug: run_2:   1205, 1219
TPMD[1]: tpm/tpm_testing.c:74: Debug: run_3:   658, 627
TPMD[1]: tpm/tpm_testing.c:75: Debug: run_4:   316, 320
TPMD[1]: tpm/tpm_testing.c:76: Debug: run_5:   179, 148
TPMD[1]: tpm/tpm_testing.c:77: Debug: run_6+:  166, 156
TPMD[1]: tpm/tpm_testing.c:78: Debug: run_34:  0
TPMD[1]: tpm/tpm_testing.c:112: Debug: tpm_test_sha1()
TPMD[1]: tpm/tpm_testing.c:156: Debug: tpm_test_hmac()
TPMD[1]: tpm/tpm_testing.c:183: Debug: tpm_test_rsa_EK()
TPMD[1]: tpm/tpm_testing.c:185: Debug: rsa_generate_key()
TPMD[1]: tpm/tpm_testing.c:190: Debug: testing endorsement key
TPMD[1]: tpm/tpm_testing.c:196: Debug: rsa_sign(RSA_SSA_PKCS1_SHA1)
TPMD[1]: tpm/tpm_testing.c:199: Debug: rsa_verify(RSA_SSA_PKCS1_SHA1)
TPMD[1]: tpm/tpm_testing.c:202: Debug: rsa_sign(RSA_SSA_PKCS1_DER)
TPMD[1]: tpm/tpm_testing.c:205: Debug: rsa_verify(RSA_SSA_PKCS1_DER)
TPMD[1]: tpm/tpm_testing.c:209: Debug: rsa_encrypt(RSA_ES_PKCSV15)
TPMD[1]: tpm/tpm_testing.c:213: Debug: rsa_decrypt(RSA_ES_PKCSV15)
TPMD[1]: tpm/tpm_testing.c:217: Debug: verify plain text
TPMD[1]: tpm/tpm_testing.c:220: Debug: rsa_encrypt(RSA_ES_OAEP_SHA1)
TPMD[1]: tpm/tpm_testing.c:224: Debug: rsa_decrypt(RSA_ES_OAEP_SHA1)
TPMD[1]: tpm/tpm_testing.c:228: Debug: verify plain text
TPMD[1]: tpm/tpm_testing.c:260: Info: Self-Test succeeded
TPMD[1]: tpm/tpm_startup.c:45: Info: TPM_Startup(1)
Loading NVM.
        Sending LoadNVM command
ERROR[VTPM]: Failed to load NVM
.INFO[VTPM]: [VTPM Listener]: VTPM Listener waiting for messages.
        Reading LoadNVM header

I have searched the code and lists for this NVM loading, and realized 
that the NVM for saving all vtpm instances resides on /var/vtpm 
directory. I have searched the lists for this information and found out 
that vtpm_manager creates the necessary directory structure when it 
starts running. However my vtpm_manager does not create any such directory.

/var/vtpm/:
toplam 20
drwxr-xr-x  4 root root 4096 Mar 26 16:40 .
drwxr-xr-x 17 root root 4096 Mar 26 16:35 ..
drwxr-xr-x  2 root root 4096 Mar 26 16:40 fifos
drwxr-xr-x  2 root root 4096 Mar 26 16:35 socks
-rw-------  1 root root 1532 Mar 27 15:54 VTPM

/var/vtpm/fifos:
toplam 8
drwxr-xr-x 2 root root 4096 Mar 26 16:40 .
drwxr-xr-x 4 root root 4096 Mar 26 16:40 ..
prw------- 1 root root    0 Mar 27 15:54 from_console.fifo
prw------- 1 root root    0 Mar 27 15:54 to_console.fifo
prw------- 1 root root    0 Mar 26 16:40 tpm_cmd_to_1.fifo
prw------- 1 root root    0 Mar 26 16:35 tpm_rsp_from_all.fifo
prw------- 1 root root    0 Mar 27 15:54 vtpm_cmd_from_all.fifo
prw------- 1 root root    0 Mar 27 15:54 vtpm_rsp_to_1.fifo

/var/vtpm/socks:
toplam 8
drwxr-xr-x 2 root root 4096 Mar 26 16:35 .
drwxr-xr-x 4 root root 4096 Mar 26 16:40 ..

When I restart or shutdown the domain and start again, vtpmd starts a 
new vtpm instance with clear option again, which I think is wrong. So 
all my previously created keys are lost on new instance, because 
previous SRK key is lost.

So the most important question follows: How do I save state of a vtpm 
across domU reboots?

I checked the code for this clear parameter, and my understanding is as 
follows:

vtpm is based on tpm_emulator and tpm_emulator have 3 states: 
deactivate, save, clear. Whenever I start a new domain, xen starts vtpm 
with clear parameter.

vtpm_create_instance() creates a new vtpm instance and determines what 
to do with it with the return value of vtpm_get_create_reason(), which 
returns the value of xenbus/resume. vtpm_create_instance() then sends a 
command to the tpm with a fifo about whether to resume or start a vtpm 
instance. When the command sent is start, vtpm just clears all the PCR's 
and keys on the existing vtpm instance.

Is this vtpm_resume something related to domain save/restore and 
suspend/resume therefore completely irrelevant to the subject? (like the 
backend driver restarted all frontend connections must be resumed) I 
assume this because I saw the code about netfront and blkfront driver 
codes, which includes this resume command sended with xenbus. But the 
tpm frontend  xenu driver does not include information abut this.

How do I save state of the vtpm across domU shutdowns?

Kind regards
Erdem Bayer