Re: [PATCH] ramfs: pretend dirent sizes

[Peter Staubach <staubach@redhat.com>]

Jan Blunck wrote:

>>
>> Also, how is lseek + readdir supposed to work in general?
>
>
> This is how libc is reading directories (at least on arch s390x):
>
> getdents() != 0
> lseek() to d_off of last dirent
> getdents() != 0
> lseek() to d_off of last dirent
> getdents() == 0
> return
>
> Therefore I really need values that make sense for d_off. Therefore I 
> really need values that make (some kind of) sense for i_size.


Please be aware that not all filesystems treat directories as sequential
files, filled with with odd sized records.  Some directories are distributed
across the entire file system and the d_off fields really contain the
address on disk of the next entry in the directory.  The d_off field is 
really
just a cookie and is not meant to be interpreted as an offset into the
directory.  The lseek() system call is used, but only because that was a
handy way of positioning the pointer into the directory.  It is unfortunate
that the getdents(2) system call was not defined with a cookie as one of
its arguments to indicate where to start fetching entries from.  This would
have reduced the confusion about treating directories as files or not.

    Thanx...

       ps