[libvirt] [PATCH] phyp: ssh authentication with pub keys fixed

[Matthias Bolte]

2009/11/10 Eduardo Otubo <otubo at linux.vnet.ibm.com>:
> Matthias Bolte wrote:
>>
>> 2009/11/9 Eduardo Otubo <otubo at linux.vnet.ibm.com>:
>>>
>>> Matthias Bolte wrote:
>>>>>
>>>>> diff --git a/src/phyp/phyp_driver.c b/src/phyp/phyp_driver.c
>>>>> index a92046a..f96d2d6 100644
>>>>> --- a/src/phyp/phyp_driver.c
>>>>> +++ b/src/phyp/phyp_driver.c
>>>>
>>>> [...]
>>>>>
>>>>> @@ -282,10 +297,8 @@ openSSHSession(virConnectPtr conn,
>>>>> virConnectAuthPtr
>>>>> auth,
>>>>>    /* Trying authentication by pubkey */
>>>>>    while ((rc =
>>>>>            libssh2_userauth_publickey_fromfile(session, username,
>>>>
>>>> You assign conn->uri->user to username and use it without checking for
>>>> NULL. You should either check conn->uri->user for NULL in phypOpen(),
>>>> as you do it for conn->uri->server and conn->uri->path, and return
>>>> VIR_DRV_OPEN_ERROR if its NULL or request a username via the auth
>>>> callback if conn->uri->user is NULL.
>>>
>>> Ok.
>>>
>>>>> -                                                "/home/user/"
>>>>> -                                                ".ssh/id_rsa.pub",
>>>>> -                                                "/home/user/"
>>>>> -                                                ".ssh/id_rsa",
>>>>> +                                                pubkey,
>>>>> +                                                pvtkey,
>>>>>                                                password)) ==
>>>>
>>>> The password (actually the passphrase) is NULL at this point. Is this
>>>> really working?
>>>
>>> Talking with libssh2 guys, this feature is not exactly working well, they
>>> said that it is possible to pass a random passphrase (or even NULL) that
>>> it
>>> will authenticate using pub and pvt keys. So, I assumed this as a
>>> hardcoded
>>> NULL just until they fix this function.
>>
>> Hm, okay. May be you should add a comment about this.
>>
>>>>>           LIBSSH2_ERROR_EAGAIN) ;
>>>>>    if (rc) {
>>>>
>>>> So you fallback to username/password authentication if keyfile
>>>> authentication failed (rc != 0). According to the
>>>> libssh2_userauth_publickey_fromfile manpage it may return this error
>>>> codes:
>>>>
>>>> LIBSSH2_ERROR_ALLOC - An internal memory allocation call failed.
>>>> LIBSSH2_ERROR_SOCKET_SEND - Unable to send data on socket.
>>>> LIBSSH2_ERROR_SOCKET_TIMEOUT
>>>> LIBSSH2_ERROR_PUBLICKEY_UNRECOGNIZED - The username/public key
>>>> combination was invalid.
>>>> LIBSSH2_ERROR_PUBLICKEY_UNVERIFIED - The username/public key
>>>> combination was invalid, or the signature for the supplied public key
>>>> was invalid.
>>>
>>> Appearently, going further the man pages and tracing all the function
>>> return
>>> points, I figured out that this function may also return
>>> LIBSSH2_ERROR_SOCKET_NONE or LIBSSH2_ERROR_NONE for many reasons. As far
>>> as
>>> I understand, LIBSSH2_ERROR_NONE is for a succesful pubkey
>>> authentication,
>>> and LIBSSH2_ERROR_SOCKET_NONE is for a non succesful. Adjusted all values
>>> for this if construction.
>>>
>>>> IMHO its not useful to fallback to username/password authentication
>>>> for the first three possible errors, only if a keyfile related error
>>>> occurs like the last two.
>>>
>>> In this case I explicit check for errors (LIBSSH2_ERROR_ALLOC,
>>> LIBSSH2_ERROR_SOCKET_SEND and LIBSSH2_ERROR_SOCKET_TIMEOUT) before
>>> fallback.
>>>
>>>> I wonder which error code will be returned if one or both keyfiles
>>>> don't exist. Maybe you should check if both keyfiles exist before
>>>> calling libssh2_userauth_publickey_fromfile() and fallback to
>>>> username/password authentication if one or both are missing.
>>>
>>> Ok. I am stating files now.
>>>
>>>>> @@ -341,15 +354,22 @@ openSSHSession(virConnectPtr conn,
>>>>> virConnectAuthPtr auth,
>>>>>            goto disconnect;
>>>>>        } else
>>>>>            goto exit;
>>>>> +    } else {
>>>>> +        goto exit;
>>>>>    }
>>>>>  disconnect:
>>>>>    libssh2_session_disconnect(session, "Disconnecting...");
>>>>>    libssh2_session_free(session);
>>>>>  err:
>>>>> +    VIR_FREE(userhome);
>>>>> +    VIR_FREE(pubkey);
>>>>> +    VIR_FREE(pvtkey);
>>>>>    VIR_FREE(password);
>>>>>    return NULL;
>>>>>
>>>>>  exit:
>>>>> +    VIR_FREE(userhome);
>>>>
>>>> VIR_FREE(pubkey) is missing here, it's there in the first version of
>>>> this
>>>> patch.
>>>
>>> Ok.
>>>
>>>
>>> Thanks again :)
>>> []'s
>>
>>
>>> diff --git a/src/phyp/phyp_driver.c b/src/phyp/phyp_driver.c
>>> index a92046a..94581b2 100644
>>> --- a/src/phyp/phyp_driver.c
>>> +++ b/src/phyp/phyp_driver.c
>>
>> [...]
>>>
>>> @@ -280,15 +302,19 @@ openSSHSession(virConnectPtr conn,
>>> virConnectAuthPtr auth,
>>>     }
>>>
>>>     /* Trying authentication by pubkey */
>>> +    if (stat(pvtkey, &pvt_stat) || stat(pubkey, &pub_stat))
>>
>> You could have used access(pvtkey, R_OK) instead, but stat() is okay.
>>
>> Don't you want to try username/password authentication in case of
>> missing keyfiles? Instead you goto err.
>>
>>> +        goto err;
>>> +
>>>     while ((rc =
>>>             libssh2_userauth_publickey_fromfile(session, username,
>>> -                                                "/home/user/"
>>> -                                                ".ssh/id_rsa.pub",
>>> -                                                "/home/user/"
>>> -                                                ".ssh/id_rsa",
>>> -                                                password)) ==
>>> +                                                pubkey,
>>> +                                                pvtkey,
>>> +                                                NULL)) ==
>>>            LIBSSH2_ERROR_EAGAIN) ;
>>> -    if (rc) {
>>> +
>>> +    if (rc == LIBSSH2_ERROR_NONE
>>
>> Didn't you say LIBSSH2_ERROR_NONE would be returned in case of
>> successful authentication? I think you wanted to check for
>> LIBSSH2_ERROR_SOCKET_NONE here, didn't you?
>>
>>> +        || rc == LIBSSH2_ERROR_PUBLICKEY_UNRECOGNIZED
>>> +        || rc == LIBSSH2_ERROR_PUBLICKEY_UNVERIFIED) {
>>>         int i;
>>>         int hasPassphrase = 0;
>>>
>>
>> Keep it up, you'll get this patch right :-)
>>
>> Matthias
>
> Hope this is the last patch I send on this topic! :)
> All fixed.
>

It is :-)

There is one small issue left. GCC doesn't like the variable
declarations after the keyboard_interactive, so I moved them to the
beginning of the function.

ACK, committed and pushed. Thanks.

Matthias