[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [dm-devel] [PATCH 1/2] block: export __make_request



On Wed, 2011-09-14 at 20:17 +0300, Boaz Harrosh wrote:
> On 9/14/2011 12:19 AM, Christoph Hellwig wrote:
> > On Mon, Sep 12, 2011 at 08:04:46PM +0200, Jens Axboe wrote:
> >>> I really hate naming things different from the method they are
> >>> implementing.  I've tried to figure out what the point of the
> >>> old blk_make_request is - why would we not go through
> >>> generic_make_request for this?
> >>>
> >>> Boaz, any idea?
> >>
> >> I tend to agree, we could rename the existing blk_make_request(). It
> >> could be blk_make_request_from_bio() or something like that, since
> >> that's what it does.
> >
> > It should at very least be renamed.  But I still can't figure out what
> > it is for exactly.
> >
> > There are three users:
> >
> >   (1) virtio_blk::virtblk_get_id():
> > 	This looks like it really should just use blk_rq_map_kern.
> >   (2) osd_initiator::_make_request():
> > 	This one looks like it should just use the same scheme as
> > 	sg_io(), as it's doing the same thing.
> 
> Good god what sg_io? That broken pointr+length from user-mode that sg.c
> and bsg.c are using? no can do it's not user-mode pointers, and it's not
> pointer+length it's pages pointers of a bio. The only other structure
> that could carry the same information is struct sg, but we work very 
> hard to get rid of this contraption. (scsi_execute_async or something 
> that it was)
> 
> blk_make_request() was made to be the parallel of __make_request, to be
> used from filesystem level users. But with two differences.
> 1. Mainly support for none-FS BLOCK_PC requests
> 2. Also support chained bios. (was added later)
> 
> >   (3) target_core_pscsi::__pscsi_map_SG():
> > 	Same as (2).
> >
> 
> There is no better suitable structure in current Kernel to carry a list
> of pages, with optional offset and length, then bio struct. Given a bio
> at hand. how do you make a block request out of it? (If it's not an 
> FS_PC type IO?)
> 
> As I remember target_core had their own pages-linked-list structure, and
> how do you make a request out of that? again best at hand is bio.
> 

FYI, for v3.1-rc code, target core has been converted to use native
struct scatterlist for everything, and the legacy struct se_mem that
existed before the advent of modern >= .24 SGL linking, et al. has been
completely removed.

--nab



[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]