Re: [PATCH] ramfs: pretend dirent sizes

[Jan Blunck <j.blunck@tu-harburg.de>]

Chris Wedgwood wrote:
> 
>>I'm using the i_size of directories in my patches.  When reading
>>from a union directory, I'm using the i_size to seek to the right
>>offset in the union stack.
> 
> 
> Ick.  That'a a bit of a hack.
> 

Don't think so:

1st dir:   [XXXXXXXXXXXXXXXXXXXXX]
       f_pos=0               f_pos=i_size(1st)


2nd dir:   [XXXXXXXXXXX|---------]
       f_pos=i_size(1st)     f_pos=i_size(1st+2nd)
                        ^
                        | f_pos=i_size(1st)+offset

Since these "arranged" values are also used as the offsets in the return 
dirent IMO it is quite clean.

> 
> Hence the value of 20 I guess --- assuming nothing will stack this
> high?
> 

Nope. This value is kind of traditional: tmpfs is using it 
(http://marc.theaimsgroup.com/?l=linux-kernel&m=103208296515378&w=2). I 
think a better value would be 1 (one) since this is also used as the 
dirent offset by dcache_readdir().

> 
> I personally would prefer that to be honest or some other way that
> doesn't change i_size.

The i_size of a directory isn't covered by the POSIX standard. IMO, it 
should be possible to seek in the range of i_size and a following 
readdir()  on the directory should succeed. But this isn't possible even 
not with real file systems like ext2.
But keeping the i_size bound to zero even if the directory contains 
entries does not make sense at all.

Jan