On Thu, Jun 3, 2021 at 6:52 PM Jeff Layton <jlayton@xxxxxxxxxx> wrote:
>
> They both say that the snap_rwsem must be held for write, but I don't
> see any real reason for it, and it's not currently always called that
> way.
>
> The lookup is just walking the rbtree, so holding it for read should be
> fine there. The "get" is bumping the refcount and (possibly) removing
> it from the empty list. I see no need to hold the snap_rwsem for write
> for that.
>
> Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxx>
> ---
> fs/ceph/snap.c | 8 ++++----
> 1 file changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/fs/ceph/snap.c b/fs/ceph/snap.c
> index bc6c33d485e6..f8cac2abab3f 100644
> --- a/fs/ceph/snap.c
> +++ b/fs/ceph/snap.c
> @@ -60,12 +60,12 @@
> /*
> * increase ref count for the realm
> *
> - * caller must hold snap_rwsem for write.
> + * caller must hold snap_rwsem.
> */
> void ceph_get_snap_realm(struct ceph_mds_client *mdsc,
> struct ceph_snap_realm *realm)
> {
> - lockdep_assert_held_write(&mdsc->snap_rwsem);
> + lockdep_assert_held(&mdsc->snap_rwsem);
>
> dout("get_realm %p %d -> %d\n", realm,
> atomic_read(&realm->nref), atomic_read(&realm->nref)+1);
> @@ -139,7 +139,7 @@ static struct ceph_snap_realm *ceph_create_snap_realm(
> /*
> * lookup the realm rooted at @ino.
> *
> - * caller must hold snap_rwsem for write.
> + * caller must hold snap_rwsem.
> */
> static struct ceph_snap_realm *__lookup_snap_realm(struct ceph_mds_client *mdsc,
> u64 ino)
> @@ -147,7 +147,7 @@ static struct ceph_snap_realm *__lookup_snap_realm(struct ceph_mds_client *mdsc,
> struct rb_node *n = mdsc->snap_realms.rb_node;
> struct ceph_snap_realm *r;
>
> - lockdep_assert_held_write(&mdsc->snap_rwsem);
> + lockdep_assert_held(&mdsc->snap_rwsem);
>
> while (n) {
> r = rb_entry(n, struct ceph_snap_realm, node);
> --
> 2.31.1
>
Ah, since you are relaxing the requirement some of those lockdep
asserts from the previous patch aren't actually redundant. This seems
fine to me: lookup definitely doesn't need the write lock and allowing
concurrent gets should be OK. The write lock made some sense back when
get was an atomic_read followed by atomic_inc but not now.
Reviewed-by: Ilya Dryomov <idryomov@xxxxxxxxx>
Thanks,
Ilya
