I forgot to send this one.
For Jens: Device mapper dm-io stores some additional information in the
bio vector past the vector end, so the test for zero-data barrier
bio->bi_io_vec != NULL cannot be used. It would mistakenly treat zero
barriers as data-barrier if they have zero data and non-NULL vector.
For Alasdair: this must be applied before all the raid1-barrier patches.
If you apply raid1 patches without this, it will compile, but it will
behave strangely. Any other target-barrier patches can be applied without
this.
Mikulas
---
Change definition of bio_has_data (and bio_empty_barrier that uses it).
With the following dm-io patch, it may happen that we create a null
barrier bio that has non-zero bi_io_vec but zero bi_size and bi_vcnt.
(dm-io uses data past the vector end to store its own bookkeeping
information, so the vector must be non-null).
This patch changes the logic so that bio with zero length and non-zero
vector pointer is recognized as an empty-barrier bio. Without this patch,
such bio would be mistreated at a number of locations in the bio stack as
well as in device mapper.
I grepped for all users of bio_has_data and didn't find problem with any
of them. I also grepped for other code locations that may mistakenly use
bio->bi_io_vec != NULL to test for an empty barrier and there were not any
other.
Signed-off-by: Mikulas Patocka <mpatocka@xxxxxxxxxx>
---
include/linux/bio.h | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
Index: linux-2.6.30-rc2-devel/include/linux/bio.h
===================================================================
--- linux-2.6.30-rc2-devel.orig/include/linux/bio.h 2009-04-24 05:28:39.000000000 +0200
+++ linux-2.6.30-rc2-devel/include/linux/bio.h 2009-04-24 05:29:15.000000000 +0200
@@ -504,7 +504,7 @@ static inline char *__bio_kmap_irq(struc
*/
static inline int bio_has_data(struct bio *bio)
{
- return bio && bio->bi_io_vec != NULL;
+ return bio && bio->bi_size != 0;
}
/*
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
