2010-07-15 15:45:55

by Christof Schmitt

Subject: [patch 1/1] block: Apply segment size and boundary limits to integrity data

From: Christof Schmitt <[email protected]>

Apply the conditions used in __blk_recalc_rq_segments also to
integrity data: Adhere to the maximum segment size and the segment
boundary set by the driver. Without this change, a driver would
receive integrity data blocks that do not adhere to the limits set for
the request queue.

Signed-off-by: Christof Schmitt <[email protected]>
---
block/blk-integrity.c | 18 ++++++++++++++++--
1 file changed, 16 insertions(+), 2 deletions(-)

--- a/block/blk-integrity.c
+++ b/block/blk-integrity.c
@@ -41,15 +41,22 @@ int blk_rq_count_integrity_sg(struct req
{
struct bio_vec *iv, *ivprv;
struct req_iterator iter;
- unsigned int segments;
+ unsigned int segments, seg_size;

ivprv = NULL;
segments = 0;
+ seg_size = 0;

rq_for_each_integrity_segment(iv, rq, iter) {

- if (!ivprv || !BIOVEC_PHYS_MERGEABLE(ivprv, iv))
+ if (!ivprv ||
+ !BIOVEC_PHYS_MERGEABLE(ivprv, iv) ||
+ seg_size + iv->bv_len > queue_max_segment_size(rq->q) ||
+ !BIOVEC_SEG_BOUNDARY(rq->q, ivprv, iv)) {
segments++;
+ seg_size = iv->bv_len;
+ } else
+ seg_size += iv->bv_len;

ivprv = iv;
}
@@ -81,9 +88,16 @@ int blk_rq_map_integrity_sg(struct reque
rq_for_each_integrity_segment(iv, rq, iter) {

if (ivprv) {
+ if (sg->length + iv->bv_len
+ > queue_max_segment_size(rq->q))
+ goto new_segment;
+
if (!BIOVEC_PHYS_MERGEABLE(ivprv, iv))
goto new_segment;

+ if (!BIOVEC_SEG_BOUNDARY(rq->q, ivprv, iv))
+ goto new_segment;
+
sg->length += iv->bv_len;
} else {
new_segment:

2010-07-15 15:53:32

by Jens Axboe

Subject: Re: [patch 1/1] block: Apply segment size and boundary limits to integrity data

On 07/15/2010 09:34 AM, Christof Schmitt wrote:
> From: Christof Schmitt <[email protected]>
>
> Apply the conditions used in __blk_recalc_rq_segments also to
> integrity data: Adhere to the maximum segment size and the segment
> boundary set by the driver. Without this change, a driver would
> receive integrity data blocks that do not adhere to the limits set for
> the request queue.

Thanks. The level of duplication there is nasty, particularly for such
intricate code as this that has had bugs in the past in the core.

--
Jens Axboe