Re: [PATCH 1/1] fs/omfs/inode.c: replace count*size kzalloc by kcalloc

[Bob Copeland <me@bobcopeland.com>]

On Wed, Jun 25, 2014 at 10:03:26PM +0200, Fabian Frederick wrote:
> > >     bitmap_size = DIV_ROUND_UP(sbi->s_num_blocks, 8);
> > >
> > Agreed - even though the FS data structures support 64-bit block
> > count, I've never seen an OMFS fs with more than about 2M blocks
> > (typical device had 20 gigs w/ 8k blocks).  So it would make
> > sense to bail in omfs_fill_super if that number is greater than
> > 2^31 or so.
> We could use unsigned int for bitmap instead of int or simply u64 ?

It doesn't really make sense to be a signed int, sure -- but even so
making it a u64 without at least including a sanity check is probably
not the way to go.

OMFS allocates space for the entire free-space bitmap in memory, rather
than loading its blocks on demand.  That's admittedly pretty dumb, but I
did it so that I could eventually support those FSes without a free-space
bitmap (I've never been asked for that feature, though, and didn't have
ReplayTV myself, so I don't believe that actually happened).

If s_num_blocks won't fit in a u32, well then that's a pretty huge chunk of
memory to allocate, and would represent a disk much bigger than the ones
that were available when this FS was used on a few devices.

(As for why the designers used u64 for all data structures, I guess just
optimism?)

-- 
Bob Copeland %% www.bobcopeland.com