Re: [PATCH] block: fix residual byte count handling

[Mike Christie <michaelc@cs.wisc.edu>]

FUJITA Tomonori wrote:
> sum(sg) == rq->data_len is already broken; sg sends such requests
> (though it would be nice if it doesn't).
> 

Actually, I think I was half wrong on that when you asked about 
scsi_debug. The scatterlist that sg.c uses is never seen by the block 
layer or scsi layer. It is just used as a container to hold segments. 
sg.c and st.c use their scatterlist to manage their preallocated 
pages/segments. When they pass it to scsi_execute_async, that function 
will create a request struct and add bios for the pages.

In 2.6.24 and below, sg.c will send a scatterlist length that does not 
match the IO length, and scsi_execute_async will goof up and send a 
rq->data_len that does not match the sum of the bios. That is what I was 
trying to fix in 2.6.24, but the patch got messed up. In 2.6.25-rc2 and 
above that is fixed and scsi_execute_async will catch sg.c doing this 
and set rq->data_len and the bio lengths correctly.

So hopefully that helps any fixes you might have planned.