2008/043 [Phase 1 of OSS for Solaris]

[Freeman Liu]

Garrett D'Amore wrote:
> Freeman Liu wrote:
>> Garrett D'Amore wrote:
>>> Freeman Liu wrote:
>>>> Gary Winiger wrote:
>>>>>> Yes. Visibility and stability of /dev/dsp will be promoted in the 
>>>>>> following phases.
>>>>>>
>>>>>>  
>>>>>>> When you say that only the root user can access /dev/dsp, do you 
>>>>>>> mean
>>>>>>> to say that the underlying device file is owned by root and has
>>>>>>> permission 600?  Or does the device's driver make explicit 
>>>>>>> permission
>>>>>>> checks?
>>>>>>>  
>>>>>>>
>>>>>>>       
>>>>>> It is owned by root and has permission 600.
>>>>>>     
>>>>>
>>>>>     What suid 0 programs are required that are not present in S10?
>>>>>   
>>>> I am not sure if  I understand you correct. I guess you think that 
>>>> since the 600 permission, there must
>>>> be some suid 0 applications to use it. Actually, there is no such 
>>>> applications, this file will only be
>>>> used by sadasupport. We make the permission 600 to avoid normal 
>>>> users to open it.
>>>> Correct me if I misunderstand you.
>>>
>>> If its only used internally, there was a way (IIRC) to create a 
>>> "ddi-internal-pathname" so that only in-kernel components could see 
>>> the node.  I forgot the actual call, but IIRC the keyboard/mouse 
>>> code on SPARC does this.
>>>
>>> That might be a preferable solution, or at least worth investigation.
>> Actually, we have thought about that solution. The main reason that 
>> we take the current approach is that /dev/dsp will be promoted to 
>> public interface in the following phases and this approach will make 
>> the migration smooth with only trivial change.
>
> From my perspective, if the /dev/dsp interface is private, and has no 
> consumers in userland today, then it is very very wrong to publish it.
>
> Its not hard to change the call to ddi_create_internal_pathname (or 
> whatever) into ddi_create_minor_node() in the future, when you are 
> ready to make it public.
>
> Otherwise, if the interface is stable enough that you think it is 
> ready to be made public *today*, then I'd like to see that done *now*, 
> rather than putting the node out there, hoping you won't change it, 
> and if you do, that nobody uses it.  (In other words, either commit to 
> the public interface, or don't expose it.  Pick one.)
After all these discussion, I feel the rule here and I accept your 
suggestion.

Best regards
  Freeman
>
>    -- Garrett
>>
>> Best regards
>>  Freeman
>>>
>>>    -- Garrett
>>>>
>>>> Best regards
>>>>  Freeman
>>>>> Gary..
>>>>>   
>>>>
>>>
>>
>