btrfs: track data relocation with simple quota

Relocation data allocations are quite tricky for simple quotas. The
basic data relocation sequence is (ignoring details that aren't relevant
to this fix):

- create a fake relocation data fs root
- create a fake relocation inode in that root
- for each data extent:
  - preallocate a data extent on behalf of the fake inode
  - copy over the data
- for each extent
  - swap the refs so that the original file extent now refers to the new
    extent item
- drop the fake root, dropping its refs on the old extents, which lets
  us delete them.

Done naively, this results in storing an extent item in the extent tree
whose owner_ref points at the relocation data root and a no-op squota
recording, since the reloc root is not a legit fstree. So far, that's
OK. The problem comes when you do the swap, and leave an extent item
owned by this bogus root as the real permanent extents of the file. If
the file then drops that ref, we free it and no-op account that against
the fake relocation root. Essentially, this means that relocation is
simple quota "extent laundering", since we re-own the extents into a
fake root.

Simple quotas very intentionally doesn't have a mechanism for
transferring ownership of extents, as that is exactly the complicated
thing we are trying to avoid with the new design. Further, it cannot be
correctly done in this case, since at the time you create the new
"real" refs, there is no way to know which was the original owner before
relocation unless we track it.

Therefore, it makes more sense to trick the preallocation to handle
relocation as a special case and note the proper owner ref from the
beginning. That way, we never write out an extent item without the
correct owner ref that it will eventually have.

This could be done by wiring a special root parameter all the way
through the allocation code path, but to avoid that special case
touching all the code, take advantage of the serial nature of relocation
to store the src root on the relocation root object. Then when we finish
the prealloc, if it happens to be this case, prepare the delayed ref
appropriately.

We must also add logic to handle relocating adjacent extents with
different owning roots. Those cannot be preallocated together in a
cluster as it would lose the separate ownership information.

This is obviously a smelly bit of code, but I think it is the best
solution to the problem, given the relocation implementation.

Signed-off-by: Boris Burkov <boris@bur.io>
Signed-off-by: David Sterba <dsterba@suse.com>

3 files changed, 59 insertions, 6 deletions

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 0c059f20533d..2758fbae7e39 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -298,6 +298,9 @@ struct btrfs_root {
 	/* Used only by log trees, when logging csum items */
 	struct extent_io_tree log_csum_range;
 
+	/* Used in simple quotas, track root during relocation. */
+	u64 relocation_src_root;
+
 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
 	u64 alloc_bytenr;
 #endif
diff --git a/fs/btrfs/extent-tree.c b/fs/btrfs/extent-tree.c
index 3ad8aec22b8d..c8e5b4715b49 100644
--- a/fs/btrfs/extent-tree.c
+++ b/fs/btrfs/extent-tree.c
@@ -59,7 +59,7 @@ static void __run_delayed_extent_op(struct btrfs_delayed_extent_op *extent_op,
 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 				      u64 parent, u64 root_objectid,
 				      u64 flags, u64 owner, u64 offset,
-				      struct btrfs_key *ins, int ref_mod);
+				      struct btrfs_key *ins, int ref_mod, u64 oref_root);
 static int alloc_reserved_tree_block(struct btrfs_trans_handle *trans,
 				     struct btrfs_delayed_ref_node *node,
 				     struct btrfs_delayed_extent_op *extent_op);
@@ -1579,7 +1579,7 @@ static int run_delayed_data_ref(struct btrfs_trans_handle *trans,
 		ret = alloc_reserved_file_extent(trans, parent, ref->root,
 						 flags, ref->objectid,
 						 ref->offset, &key,
-						 node->ref_mod);
+						 node->ref_mod, href->owning_root);
 		if (!ret)
 			ret = btrfs_record_squota_delta(trans->fs_info, &delta);
 		else
@@ -4744,7 +4744,7 @@ static int alloc_reserved_extent(struct btrfs_trans_handle *trans, u64 bytenr,
 static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 				      u64 parent, u64 root_objectid,
 				      u64 flags, u64 owner, u64 offset,
-				      struct btrfs_key *ins, int ref_mod)
+				      struct btrfs_key *ins, int ref_mod, u64 oref_root)
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
 	struct btrfs_root *extent_root;
@@ -4791,7 +4791,7 @@ static int alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 	if (simple_quota) {
 		btrfs_set_extent_inline_ref_type(leaf, iref, BTRFS_EXTENT_OWNER_REF_KEY);
 		oref = (struct btrfs_extent_owner_ref *)(&iref->offset);
-		btrfs_set_extent_owner_ref_root_id(leaf, oref, root_objectid);
+		btrfs_set_extent_owner_ref_root_id(leaf, oref, oref_root);
 		iref = (struct btrfs_extent_inline_ref *)(oref + 1);
 	}
 	btrfs_set_extent_inline_ref_type(leaf, iref, type);
@@ -4902,6 +4902,9 @@ int btrfs_alloc_reserved_file_extent(struct btrfs_trans_handle *trans,
 
 	BUG_ON(root_objectid == BTRFS_TREE_LOG_OBJECTID);
 
+	if (btrfs_is_data_reloc_root(root) && is_fstree(root->relocation_src_root))
+		owning_root = root->relocation_src_root;
+
 	btrfs_init_generic_ref(&generic_ref, BTRFS_ADD_DELAYED_EXTENT,
 			       ins->objectid, ins->offset, 0, owning_root);
 	btrfs_init_data_ref(&generic_ref, root_objectid, owner,
@@ -4957,7 +4960,7 @@ int btrfs_alloc_logged_file_extent(struct btrfs_trans_handle *trans,
 	spin_unlock(&space_info->lock);
 
 	ret = alloc_reserved_file_extent(trans, 0, root_objectid, 0, owner,
-					 offset, ins, 1);
+					 offset, ins, 1, root_objectid);
 	if (ret)
 		btrfs_pin_extent(trans, ins->objectid, ins->offset, 1);
 	ret = btrfs_record_squota_delta(fs_info, &delta);
diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c
index ce1db6152cc4..6e8e14d1aeaa 100644
--- a/fs/btrfs/relocation.c
+++ b/fs/btrfs/relocation.c
@@ -122,6 +122,7 @@ struct file_extent_cluster {
 	u64 end;
 	u64 boundary[MAX_EXTENTS];
 	unsigned int nr;
+	u64 owning_root;
 };
 
 struct reloc_control {
@@ -3166,6 +3167,7 @@ int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
 			 struct file_extent_cluster *cluster)
 {
 	int ret;
+	struct btrfs_root *root = BTRFS_I(inode)->root;
 
 	if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) {
 		ret = relocate_file_extent_cluster(inode, cluster);
@@ -3174,8 +3176,38 @@ int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key,
 		cluster->nr = 0;
 	}
 
-	if (!cluster->nr)
+	/*
+	 * Under simple quotas, we set root->relocation_src_root when we find
+	 * the extent. If adjacent extents have different owners, we can't merge
+	 * them while relocating. Handle this by storing the owning root that
+	 * started a cluster and if we see an extent from a different root break
+	 * cluster formation (just like the above case of non-adjacent extents).
+	 *
+	 * Without simple quotas, relocation_src_root is always 0, so we should
+	 * never see a mismatch, and it should have no effect on relocation
+	 * clusters.
+	 */
+	if (cluster->nr > 0 && cluster->owning_root != root->relocation_src_root) {
+		u64 tmp = root->relocation_src_root;
+
+		/*
+		 * root->relocation_src_root is the state that actually affects
+		 * the preallocation we do here, so set it to the root owning
+		 * the cluster we need to relocate.
+		 */
+		root->relocation_src_root = cluster->owning_root;
+		ret = relocate_file_extent_cluster(inode, cluster);
+		if (ret)
+			return ret;
+		cluster->nr = 0;
+		/* And reset it back for the current extent's owning root. */
+		root->relocation_src_root = tmp;
+	}
+
+	if (!cluster->nr) {
 		cluster->start = extent_key->objectid;
+		cluster->owning_root = root->relocation_src_root;
+	}
 	else
 		BUG_ON(cluster->nr >= MAX_EXTENTS);
 	cluster->end = extent_key->objectid + extent_key->offset - 1;
@@ -3705,6 +3737,21 @@ restart:
 				    struct btrfs_extent_item);
 		flags = btrfs_extent_flags(path->nodes[0], ei);
 
+		/*
+		 * If we are relocating a simple quota owned extent item, we
+		 * need to note the owner on the reloc data root so that when
+		 * we allocate the replacement item, we can attribute it to the
+		 * correct eventual owner (rather than the reloc data root).
+		 */
+		if (btrfs_qgroup_mode(fs_info) == BTRFS_QGROUP_MODE_SIMPLE) {
+			struct btrfs_root *root = BTRFS_I(rc->data_inode)->root;
+			u64 owning_root_id = btrfs_get_extent_owner_root(fs_info,
+								 path->nodes[0],
+								 path->slots[0]);
+
+			root->relocation_src_root = owning_root_id;
+		}
+
 		if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) {
 			ret = add_tree_block(rc, &key, path, &blocks);
 		} else if (rc->stage == UPDATE_DATA_PTRS &&