[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Is JFFS a full featured filesystem?



On Wed, Aug 02, 2000 at 03:20:04PM +0100, David Woodhouse wrote:
> alex@xxxxxxx.se said:
> >  Exactly how did you plan to work out the dirty ranges? Compare with
> > on-flash contents? 
> 
> Apparently it's easy when using generic_file_write, because prepare_write 
> will give you the exact range. 

Both prepare_write and commit_write, so we have full control over what
happens.  The arguments to {prepare,commit}_write are offsets within the
actual page (so <= PAGE_SIZE), but page->index holds the higher bits of
the logical offset.

> When doing writable mmap() we have to do something cleverer - basically yes,
> comparing with the on-flash contents. We can either do that by going and
> reading the flash nodes again on writepage(), or we can keep a copy of the 
> clean page in RAM before it's dirtied. 

Or we can just make writable mmap()s fail - applications should handle that
case, and it seems rather easy to fuck up and corrupt an fs by not handling
it correctly.

> To start with, I'm inclined just to accept the hit of the 4Kb writes, and 
> let the GC combine nodes later as necessary. Comparing with old contents 
> can come later. This is only going to be a problem with writable mmap(), 
> which isn't supported at the moment _anyway_. The normal write() case is 

I don't see where a writable mmap() fails in 2.2 - in 2.4 it's obvious.

	Philipp