> Here's a port of this to an upstream kernel: untested - the reordering
> of the dec_pending looks suspicious - io now referenced after it's
> freed?
Right, I think now everything is done in the right order. Though I had
to add another field to target_io.
> Please send patches against upstream kernels where possible, and
> remember to add 'Signed-off-by' lines to everything!
>
>Alasdair
I reworked the patch against the 2.6.17 kernel and did a quick test run
with a simple target. In the upstream kernel the bioset stuff has only
to be modified to be local to each mapped device instead of one for all.
Stefan
Signed-off-by: Stefan Bader <shbader@xxxxxxx>
diff -Nurp linux-2.6.17/drivers/md/dm.c linux-2.6.17-new/drivers/md/dm.c
--- linux-2.6.17/drivers/md/dm.c 2006-06-19 14:01:40.000000000 +0200
+++ linux-2.6.17-new/drivers/md/dm.c 2006-08-15 16:21:30.000000000 +0200
@@ -44,6 +44,7 @@ struct dm_io {
struct target_io {
struct dm_io *io;
struct dm_target *ti;
+ struct bio_set *bs;
union map_info info;
};
@@ -92,6 +93,11 @@ struct mapped_device {
*/
mempool_t *io_pool;
mempool_t *tio_pool;
+ /*
+ * Bio objects are allocated from this private pool instead of the
+ * global one.
+ */
+ struct bio_set *bio_pool;
/*
* Event handling.
@@ -113,16 +119,10 @@ struct mapped_device {
static kmem_cache_t *_io_cache;
static kmem_cache_t *_tio_cache;
-static struct bio_set *dm_set;
-
static int __init local_init(void)
{
int r;
- dm_set = bioset_create(16, 16, 4);
- if (!dm_set)
- return -ENOMEM;
-
/* allocate a slab for the dm_ios */
_io_cache = kmem_cache_create("dm_io",
sizeof(struct dm_io), 0, 0, NULL, NULL);
@@ -156,8 +156,6 @@ static void local_exit(void)
kmem_cache_destroy(_tio_cache);
kmem_cache_destroy(_io_cache);
- bioset_free(dm_set);
-
if (unregister_blkdev(_major, _name) < 0)
DMERR("devfs_unregister_blkdev failed");
@@ -386,7 +384,7 @@ static int clone_endio(struct bio *bio,
{
int r = 0;
struct target_io *tio = bio->bi_private;
- struct dm_io *io = tio->io;
+ struct mapped_device *md = tio->io->md;
dm_endio_fn endio = tio->ti->type->end_io;
if (bio->bi_size)
@@ -405,9 +403,13 @@ static int clone_endio(struct bio *bio,
return 1;
}
- free_tio(io->md, tio);
- dec_pending(io, error);
+ dec_pending(tio->io, error);
+ /*
+ * First release bio because destructor needs valid reference to
+ * target_io to find the correct bio_set.
+ */
bio_put(bio);
+ free_tio(md, tio);
return r;
}
@@ -465,10 +467,14 @@ static void __map_bio(struct dm_target *
else if (r < 0) {
/* error the io and bail out */
- struct dm_io *io = tio->io;
- free_tio(tio->io->md, tio);
- dec_pending(io, r);
+ struct mapped_device *md = tio->io->md;
+ dec_pending(tio->io, r);
+ /*
+ * First release bio because destructor needs valid reference
+ * to target_io to find the correct bio_set.
+ */
bio_put(clone);
+ free_tio(md, tio);
}
}
@@ -484,7 +490,11 @@ struct clone_info {
static void dm_bio_destructor(struct bio *bio)
{
- bio_free(bio, dm_set);
+ struct target_io *tio = bio->bi_private;
+
+ BUG_ON(!tio);
+
+ bio_free(bio, tio->bs);
}
/*
@@ -492,12 +502,12 @@ static void dm_bio_destructor(struct bio
*/
static struct bio *split_bvec(struct bio *bio, sector_t sector,
unsigned short idx, unsigned int offset,
- unsigned int len)
+ unsigned int len, struct bio_set *bs)
{
struct bio *clone;
struct bio_vec *bv = bio->bi_io_vec + idx;
- clone = bio_alloc_bioset(GFP_NOIO, 1, dm_set);
+ clone = bio_alloc_bioset(GFP_NOIO, 1, bs);
clone->bi_destructor = dm_bio_destructor;
*clone->bi_io_vec = *bv;
@@ -517,11 +527,13 @@ static struct bio *split_bvec(struct bio
*/
static struct bio *clone_bio(struct bio *bio, sector_t sector,
unsigned short idx, unsigned short bv_count,
- unsigned int len)
+ unsigned int len, struct bio_set *bs)
{
struct bio *clone;
- clone = bio_clone(bio, GFP_NOIO);
+ clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs);
+ __bio_clone(clone, bio);
+ clone->bi_destructor = dm_bio_destructor;
clone->bi_sector = sector;
clone->bi_idx = idx;
clone->bi_vcnt = idx + bv_count;
@@ -544,6 +556,7 @@ static void __clone_and_map(struct clone
tio = alloc_tio(ci->md);
tio->io = ci->io;
tio->ti = ti;
+ tio->bs = ci->md->bio_pool;
memset(&tio->info, 0, sizeof(tio->info));
if (ci->sector_count <= max) {
@@ -552,7 +565,8 @@ static void __clone_and_map(struct clone
* the remaining io with a single clone.
*/
clone = clone_bio(bio, ci->sector, ci->idx,
- bio->bi_vcnt - ci->idx, ci->sector_count);
+ bio->bi_vcnt - ci->idx, ci->sector_count,
+ tio->bs);
__map_bio(ti, clone, tio);
ci->sector_count = 0;
@@ -575,7 +589,8 @@ static void __clone_and_map(struct clone
len += bv_len;
}
- clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len);
+ clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len,
+ tio->bs);
__map_bio(ti, clone, tio);
ci->sector += len;
@@ -598,13 +613,15 @@ static void __clone_and_map(struct clone
tio = alloc_tio(ci->md);
tio->io = ci->io;
tio->ti = ti;
+ tio->bs = ci->md->bio_pool;
memset(&tio->info, 0, sizeof(tio->info));
}
len = min(remaining, max);
clone = split_bvec(bio, ci->sector, ci->idx,
- bv->bv_offset + offset, len);
+ bv->bv_offset + offset, len,
+ tio->bs);
__map_bio(ti, clone, tio);
@@ -871,6 +888,10 @@ static struct mapped_device *alloc_dev(u
if (!md->tio_pool)
goto bad3;
+ md->bio_pool = bioset_create(16, 16, 4);
+ if (!md->bio_pool)
+ goto bad_no_bioset;
+
md->disk = alloc_disk(1);
if (!md->disk)
goto bad4;
@@ -891,6 +912,8 @@ static struct mapped_device *alloc_dev(u
return md;
bad4:
+ bioset_free(md->bio_pool);
+ bad_no_bioset:
mempool_destroy(md->tio_pool);
bad3:
mempool_destroy(md->io_pool);
@@ -912,6 +935,7 @@ static void free_dev(struct mapped_devic
}
mempool_destroy(md->tio_pool);
mempool_destroy(md->io_pool);
+ bioset_free(md->bio_pool);
del_gendisk(md->disk);
free_minor(minor);
put_disk(md->disk);
--
dm-devel mailing list
dm-devel@xxxxxxxxxx
https://www.redhat.com/mailman/listinfo/dm-devel
