btrfs: introduce delayed_refs_rsv

Traditionally we've had voodoo in btrfs to account for the space that
delayed refs may take up by having a global_block_rsv.  This works most
of the time, except when it doesn't.  We've had issues reported and seen
in production where sometimes the global reserve is exhausted during
transaction commit before we can run all of our delayed refs, resulting
in an aborted transaction.  Because of this voodoo we have equally
dubious flushing semantics around throttling delayed refs which we often
get wrong.

So instead give them their own block_rsv.  This way we can always know
exactly how much outstanding space we need for delayed refs.  This
allows us to make sure we are constantly filling that reservation up
with space, and allows us to put more precise pressure on the enospc
system.  Instead of doing math to see if its a good time to throttle,
the normal enospc code will be invoked if we have a lot of delayed refs
pending, and they will be run via the normal flushing mechanism.

For now the delayed_refs_rsv will hold the reservations for the delayed
refs, the block group updates, and deleting csums.  We could have a
separate rsv for the block group updates, but the csum deletion stuff is
still handled via the delayed_refs so that will stay there.

Historical background:

The global reserve has grown to cover everything we don't reserve space
explicitly for, and we've grown a lot of weird ad-hoc heuristics to know
if we're running short on space and when it's time to force a commit.  A
failure rate of 20-40 file systems when we run hundreds of thousands of
them isn't super high, but cleaning up this code will make things less
ugly and more predictible.

Thus the delayed refs rsv.  We always know how many delayed refs we have
outstanding, and although running them generates more we can use the
global reserve for that spill over, which fits better into it's desired
use than a full blown reservation.  This first approach is to simply
take how many times we're reserving space for and multiply that by 2 in
order to save enough space for the delayed refs that could be generated.
This is a niave approach and will probably evolve, but for now it works.
Signed-off-by: Josef Bacik <jbacik@fb.com>
Reviewed-by: David Sterba <dsterba@suse.com> # high-level review
[ added background notes from the cover letter ]
Signed-off-by: David Sterba <dsterba@suse.com>

fs/btrfs/ctree.h

@@ -468,6 +468,7 @@ enum {
 	BTRFS_BLOCK_RSV_TRANS,
 	BTRFS_BLOCK_RSV_CHUNK,
 	BTRFS_BLOCK_RSV_DELOPS,
+	BTRFS_BLOCK_RSV_DELREFS,
 	BTRFS_BLOCK_RSV_EMPTY,
 	BTRFS_BLOCK_RSV_TEMP,
 };
@@ -831,6 +832,8 @@ struct btrfs_fs_info {
 	struct btrfs_block_rsv chunk_block_rsv;
 	/* block reservation for delayed operations */
 	struct btrfs_block_rsv delayed_block_rsv;
+	/* block reservation for delayed refs */
+	struct btrfs_block_rsv delayed_refs_rsv;
 
 	struct btrfs_block_rsv empty_block_rsv;
 
@@ -2816,6 +2819,13 @@ int btrfs_cond_migrate_bytes(struct btrfs_fs_info *fs_info,
 void btrfs_block_rsv_release(struct btrfs_fs_info *fs_info,
 			     struct btrfs_block_rsv *block_rsv,
 			     u64 num_bytes);
+void btrfs_delayed_refs_rsv_release(struct btrfs_fs_info *fs_info, int nr);
+void btrfs_update_delayed_refs_rsv(struct btrfs_trans_handle *trans);
+int btrfs_delayed_refs_rsv_refill(struct btrfs_fs_info *fs_info,
+				  enum btrfs_reserve_flush_enum flush);
+void btrfs_migrate_to_delayed_refs_rsv(struct btrfs_fs_info *fs_info,
+				       struct btrfs_block_rsv *src,
+				       u64 num_bytes);
 int btrfs_inc_block_group_ro(struct btrfs_block_group_cache *cache);
 void btrfs_dec_block_group_ro(struct btrfs_block_group_cache *cache);
 void btrfs_put_block_group_cache(struct btrfs_fs_info *info);

fs/btrfs/delayed-ref.c

@@ -473,12 +473,14 @@ static int insert_delayed_ref(struct btrfs_trans_handle *trans,
  * helper function to update the accounting in the head ref
  * existing and update must have the same bytenr
  */
-static noinline void
-update_existing_head_ref(struct btrfs_delayed_ref_root *delayed_refs,
+static noinline void update_existing_head_ref(struct btrfs_trans_handle *trans,
 			 struct btrfs_delayed_ref_head *existing,
 			 struct btrfs_delayed_ref_head *update,
 			 int *old_ref_mod_ret)
 {
+	struct btrfs_delayed_ref_root *delayed_refs =
+		&trans->transaction->delayed_refs;
+	struct btrfs_fs_info *fs_info = trans->fs_info;
 	int old_ref_mod;
 
 	BUG_ON(existing->is_data != update->is_data);
@@ -536,10 +538,18 @@ update_existing_head_ref(struct btrfs_delayed_ref_root *delayed_refs,
 	 * versa we need to make sure to adjust pending_csums accordingly.
 	 */
 	if (existing->is_data) {
-		if (existing->total_ref_mod >= 0 && old_ref_mod < 0)
+		u64 csum_leaves =
+			btrfs_csum_bytes_to_leaves(fs_info,
+						   existing->num_bytes);
+
+		if (existing->total_ref_mod >= 0 && old_ref_mod < 0) {
 			delayed_refs->pending_csums -= existing->num_bytes;
-		if (existing->total_ref_mod < 0 && old_ref_mod >= 0)
+			btrfs_delayed_refs_rsv_release(fs_info, csum_leaves);
+		}
+		if (existing->total_ref_mod < 0 && old_ref_mod >= 0) {
 			delayed_refs->pending_csums += existing->num_bytes;
+			trans->delayed_ref_updates += csum_leaves;
+		}
 	}
 	spin_unlock(&existing->lock);
 }
@@ -645,7 +655,7 @@ add_delayed_ref_head(struct btrfs_trans_handle *trans,
 			&& head_ref->qgroup_reserved
 			&& existing->qgroup_ref_root
 			&& existing->qgroup_reserved);
-		update_existing_head_ref(delayed_refs, existing, head_ref,
+		update_existing_head_ref(trans, existing, head_ref,
 					 old_ref_mod);
 		/*
 		 * we've updated the existing ref, free the newly
@@ -656,8 +666,12 @@ add_delayed_ref_head(struct btrfs_trans_handle *trans,
 	} else {
 		if (old_ref_mod)
 			*old_ref_mod = 0;
-		if (head_ref->is_data && head_ref->ref_mod < 0)
+		if (head_ref->is_data && head_ref->ref_mod < 0) {
 			delayed_refs->pending_csums += head_ref->num_bytes;
+			trans->delayed_ref_updates +=
+				btrfs_csum_bytes_to_leaves(trans->fs_info,
+							   head_ref->num_bytes);
+		}
 		delayed_refs->num_heads++;
 		delayed_refs->num_heads_ready++;
 		atomic_inc(&delayed_refs->num_entries);
@@ -793,6 +807,12 @@ int btrfs_add_delayed_tree_ref(struct btrfs_trans_handle *trans,
 	ret = insert_delayed_ref(trans, delayed_refs, head_ref, &ref->node);
 	spin_unlock(&delayed_refs->lock);
 
+	/*
+	 * Need to update the delayed_refs_rsv with any changes we may have
+	 * made.
+	 */
+	btrfs_update_delayed_refs_rsv(trans);
+
 	trace_add_delayed_tree_ref(fs_info, &ref->node, ref,
 				   action == BTRFS_ADD_DELAYED_EXTENT ?
 				   BTRFS_ADD_DELAYED_REF : action);
@@ -874,6 +894,12 @@ int btrfs_add_delayed_data_ref(struct btrfs_trans_handle *trans,
 	ret = insert_delayed_ref(trans, delayed_refs, head_ref, &ref->node);
 	spin_unlock(&delayed_refs->lock);
 
+	/*
+	 * Need to update the delayed_refs_rsv with any changes we may have
+	 * made.
+	 */
+	btrfs_update_delayed_refs_rsv(trans);
+
 	trace_add_delayed_data_ref(trans->fs_info, &ref->node, ref,
 				   action == BTRFS_ADD_DELAYED_EXTENT ?
 				   BTRFS_ADD_DELAYED_REF : action);
@@ -910,6 +936,12 @@ int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info,
 			     NULL, NULL, NULL);
 
 	spin_unlock(&delayed_refs->lock);
+
+	/*
+	 * Need to update the delayed_refs_rsv with any changes we may have
+	 * made.
+	 */
+	btrfs_update_delayed_refs_rsv(trans);
 	return 0;
 }
 

fs/btrfs/disk-io.c

@@ -2678,6 +2678,9 @@ int open_ctree(struct super_block *sb,
 	btrfs_init_block_rsv(&fs_info->empty_block_rsv, BTRFS_BLOCK_RSV_EMPTY);
 	btrfs_init_block_rsv(&fs_info->delayed_block_rsv,
 			     BTRFS_BLOCK_RSV_DELOPS);
+	btrfs_init_block_rsv(&fs_info->delayed_refs_rsv,
+			     BTRFS_BLOCK_RSV_DELREFS);
+
 	atomic_set(&fs_info->async_delalloc_pages, 0);
 	atomic_set(&fs_info->defrag_running, 0);
 	atomic_set(&fs_info->qgroup_op_seq, 0);
@@ -4446,6 +4449,7 @@ void btrfs_cleanup_dirty_bgs(struct btrfs_transaction *cur_trans,
 
 		spin_unlock(&cur_trans->dirty_bgs_lock);
 		btrfs_put_block_group(cache);
+		btrfs_delayed_refs_rsv_release(fs_info, 1);
 		spin_lock(&cur_trans->dirty_bgs_lock);
 	}
 	spin_unlock(&cur_trans->dirty_bgs_lock);

fs/btrfs/transaction.c

@@ -454,7 +454,7 @@ start_transaction(struct btrfs_root *root, unsigned int num_items,
 		  bool enforce_qgroups)
 {
 	struct btrfs_fs_info *fs_info = root->fs_info;
-
+	struct btrfs_block_rsv *delayed_refs_rsv = &fs_info->delayed_refs_rsv;
 	struct btrfs_trans_handle *h;
 	struct btrfs_transaction *cur_trans;
 	u64 num_bytes = 0;
@@ -483,13 +483,28 @@ start_transaction(struct btrfs_root *root, unsigned int num_items,
 	 * the appropriate flushing if need be.
 	 */
 	if (num_items && root != fs_info->chunk_root) {
+		struct btrfs_block_rsv *rsv = &fs_info->trans_block_rsv;
+		u64 delayed_refs_bytes = 0;
+
 		qgroup_reserved = num_items * fs_info->nodesize;
 		ret = btrfs_qgroup_reserve_meta_pertrans(root, qgroup_reserved,
 				enforce_qgroups);
 		if (ret)
 			return ERR_PTR(ret);
 
+		/*
+		 * We want to reserve all the bytes we may need all at once, so
+		 * we only do 1 enospc flushing cycle per transaction start.  We
+		 * accomplish this by simply assuming we'll do 2 x num_items
+		 * worth of delayed refs updates in this trans handle, and
+		 * refill that amount for whatever is missing in the reserve.
+		 */
 		num_bytes = btrfs_calc_trans_metadata_size(fs_info, num_items);
+		if (delayed_refs_rsv->full == 0) {
+			delayed_refs_bytes = num_bytes;
+			num_bytes <<= 1;
+		}
+
 		/*
 		 * Do the reservation for the relocation root creation
 		 */
@@ -498,8 +513,24 @@ start_transaction(struct btrfs_root *root, unsigned int num_items,
 			reloc_reserved = true;
 		}
 
-		ret = btrfs_block_rsv_add(root, &fs_info->trans_block_rsv,
-					  num_bytes, flush);
+		ret = btrfs_block_rsv_add(root, rsv, num_bytes, flush);
+		if (ret)
+			goto reserve_fail;
+		if (delayed_refs_bytes) {
+			btrfs_migrate_to_delayed_refs_rsv(fs_info, rsv,
+							  delayed_refs_bytes);
+			num_bytes -= delayed_refs_bytes;
+		}
+	} else if (num_items == 0 && flush == BTRFS_RESERVE_FLUSH_ALL &&
+		   !delayed_refs_rsv->full) {
+		/*
+		 * Some people call with btrfs_start_transaction(root, 0)
+		 * because they can be throttled, but have some other mechanism
+		 * for reserving space.  We still want these guys to refill the
+		 * delayed block_rsv so just add 1 items worth of reservation
+		 * here.
+		 */
+		ret = btrfs_delayed_refs_rsv_refill(fs_info, flush);
 		if (ret)
 			goto reserve_fail;
 	}