[Ksummit-2008-discuss] Suggested topic: possible new API's between file systems and the block layer

Ric Wheeler ricwheeler at gmail.com
Sat Jul 26 22:01:38 PDT 2008


Andrew Morton wrote:
> On Sat, 26 Jul 2008 17:12:09 -0400 David Woodhouse <dwmw2 at infradead.org> wrote:
>
>   
>>>>> For example, support for the T13 Trim command.  There will be others,
>>>>> but I surely don't know what they are.
>>>>>           
>>>> That's what they're calling the 'forget this sector' command?
>>>>         
>>> Think so.  It's the "we're not using this sector any more command".
>>>       
>> Yeah, I've been saying for years that we need that. I'm still
>> unconvinced that making flash pretend to be a hard drive and using a
>> 'normal' file system on it is the best way forward, but at least this
>> addresses the largest problem with the approach.
>>
>>     
>>>> I'm planning to look at that quite soon.
>>>>         
>>> Cool.  It will affect VFS, filesystems, block layer and device drivers.
>>>       
>> VFS?
>>     
>
> If we're talking about the same thing and if I'm understanding it
> correctly, the kernel need to tell the device "we're not using
> these sectors any more".  That means hooking into truncate.
>   
There are actually two similar proposals - one for SCSI and one for ATA 
devices. The SCSI version is aimed at higher end devices which want to 
allow  "over provisioned" storage, the low end ones are aimed at 
FLASH/SSD devices.

Hooking into the truncate code is the big missing gap, the schemes 
basically assume that any write to the device is good enough to indicate 
that things have been allocated (and actually used ;-))

ric




More information about the Ksummit-2008-discuss mailing list