While removing a file with dedup extents, we could have a great number of delayed refs pending to process, and these refs refer to droping a ref of the extent, which is of BTRFS_DROP_DELAYED_REF type.
But in order to prevent an extent's ref count from going down to zero when there still are pending delayed refs, we first select those "adding a ref" ones, which is of BTRFS_ADD_DELAYED_REF type. So in removing case, all of our delayed refs are of BTRFS_DROP_DELAYED_REF type, but we have to walk all the refs issued to the extent to find any BTRFS_ADD_DELAYED_REF types and end up there is no such thing, and then start over again to find BTRFS_DROP_DELAYED_REF. This is really unnecessary, we can improve this by tracking how many BTRFS_ADD_DELAYED_REF refs we have and search by the right type. Signed-off-by: Liu Bo <bo.li....@oracle.com> --- fs/btrfs/delayed-ref.c | 10 ++++++++++ fs/btrfs/delayed-ref.h | 3 +++ fs/btrfs/extent-tree.c | 17 ++++++++++++++++- 3 files changed, 29 insertions(+), 1 deletion(-) diff --git a/fs/btrfs/delayed-ref.c b/fs/btrfs/delayed-ref.c index b0d5d79..9596649 100644 --- a/fs/btrfs/delayed-ref.c +++ b/fs/btrfs/delayed-ref.c @@ -543,6 +543,10 @@ update_existing_head_ref(struct btrfs_delayed_ref_node *existing, * update the reference mod on the head to reflect this new operation */ existing->ref_mod += update->ref_mod; + + WARN_ON_ONCE(update->ref_mod > 1); + if (update->ref_mod == 1) + existing_ref->add_cnt++; } /* @@ -604,6 +608,12 @@ static noinline void add_delayed_ref_head(struct btrfs_fs_info *fs_info, head_ref->must_insert_reserved = must_insert_reserved; head_ref->is_data = is_data; + /* track added ref, more comments in select_delayed_ref() */ + if (count_mod == 1) + head_ref->add_cnt = 1; + else + head_ref->add_cnt = 0; + INIT_LIST_HEAD(&head_ref->cluster); mutex_init(&head_ref->mutex); diff --git a/fs/btrfs/delayed-ref.h b/fs/btrfs/delayed-ref.h index 70b962c..9377b27 100644 --- a/fs/btrfs/delayed-ref.h +++ b/fs/btrfs/delayed-ref.h @@ -84,6 +84,9 @@ struct btrfs_delayed_ref_head { struct list_head cluster; struct btrfs_delayed_extent_op *extent_op; + + int add_cnt; + /* * when a new extent is allocated, it is just reserved in memory * The actual extent isn't inserted into the extent allocation tree diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c index 009980c..a6fb5fa 100644 --- a/fs/btrfs/extent-tree.c +++ b/fs/btrfs/extent-tree.c @@ -2287,6 +2287,16 @@ select_delayed_ref(struct btrfs_delayed_ref_head *head) struct rb_node *node; struct btrfs_delayed_ref_node *ref; int action = BTRFS_ADD_DELAYED_REF; + + /* + * track the count of BTRFS_ADD_DELAYED_REF, + * in the case that there's no BTRFS_ADD_DELAYED_REF while there're a + * a great number of BTRFS_DROP_DELAYED_REF, + * it'll waste time on searching BTRFS_ADD_DELAYED_REF, usually this + * happens with dedup enabled. + */ + if (head->add_cnt == 0) + action = BTRFS_DROP_DELAYED_REF; again: /* * select delayed ref of type BTRFS_ADD_DELAYED_REF first. @@ -2301,8 +2311,11 @@ again: rb_node); if (ref->bytenr != head->node.bytenr) break; - if (ref->action == action) + if (ref->action == action) { + if (action == BTRFS_ADD_DELAYED_REF) + head->add_cnt--; return ref; + } node = rb_prev(node); } if (action == BTRFS_ADD_DELAYED_REF) { @@ -2378,6 +2391,8 @@ static noinline int run_clustered_refs(struct btrfs_trans_handle *trans, * there are still refs with lower seq numbers in the * process of being added. Don't run this ref yet. */ + if (ref->action == BTRFS_ADD_DELAYED_REF) + locked_ref->add_cnt++; list_del_init(&locked_ref->cluster); btrfs_delayed_ref_unlock(locked_ref); locked_ref = NULL; -- 1.8.2.1 -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html