Re: bio_chain: proposed solution for bio_alloc failure and large IO simplification

[Andrew Morton <akpm@zip.com.au>]

Jens Axboe wrote:
> 
> ...
> > What I did, and what I'd suggest as a convention is:
> >
> > During BIO assembly, bi_vcnt indicates the maximum number of
> > bvecs which the BIO can hold. And bi_idx indexes the next-free
> > bvec within the BIO.
> 
> Hmm I don't like that too much. For reference, bi_vcnt from the block
> layer is the number of bio_vecs in the bio. And bi_idx is the index into
> the 'current' bio_vec. To tie that in with the above, how about just
> changing bi_max to be a real number. Internal bio can still find the
> pool from that, and private bios can just fill it out.

But then bi_max is the _actual_ size of the BIO, and not the
size which the caller requested.

umm, err, actually, that suits me just fine ;)  We could leave
bi_size as-is and just implement

	unsigned bio_nr_bvecs(struct bio *bio);

But that may not work for privately allocated BIOs.  "bios not
coming from bio_alloc()"?

What _are_ these private BIOs, anyway?  Is any in-kernel code
constructing them at present?

-