[LTP] [PATCH 1/2] ltp-aiodio/dio_sparse: Fix write in dio_sparse()

[Cyril Hrubis <chrubis@suse.cz>]

Hi!
> >> As the second case of kernel commit 9ecd10b says:
> >>     2. Direct writes starting from or beyong i_size (not inside i_size)
> >>        also could trigger block allocation and expose stale data.  For
> >>        example, consider a sparse file with i_size of 2k, and a write to
> >>        offset 2k or 3k into the file, with a filesystem block size of 4k.
> >>        (Thanks to Jeff Moyer for pointing this case out in his review.)
> > 
> > Ah, looking at the git log, the offset was just added recently, that's
> > why the code is confusing, previously the filesize was size of the
> > file...
> 
> Yes, it seems that the code get confusing after the offset was added,
> and yes, previously the filesize was the size of the file, but sorry,
> according to my understanding it still represents the size of the file now.

Ok we are mixing two things together. One is how much size the file
actually takes on a filesystem and second one is the file length.

> > 
> > So what about we rename the writesize to blocksize and filesize to
> > writesize so that it's clear what the function dio_sparse does.
> > 
> > int dio_sparse(char *filename, int align, int blocksize, int writesize, int offset)
> > 
> > Also shouldn't we truncate the file to offset + filesize rather than
> > just filesize? And then pass the size to the children as offset +
> > filesize as well?
> 
> In my humble opinion, there was no problem at all in the code before
> the offset was added, and look at the commit message of adding offset,
> the aim is to achieve the second case of kernel commit 9ecd10b above,
> make it to direct write from or beyond the end of the file.

The patch is correct, there is no doubt about that.

> So, I think the writesize, filesize and offset should be independent of
> each other, the "offset" here is not filesize's offset (i.e. filesize
> does not calculate with offset), but only to decide where we write start from.

Okay, but still the writesize should be rename to something as
block_size.

-- 
Cyril Hrubis
chrubis@suse.cz