Btrfs: fix relocation races

The recent commit to get rid of our trans_mutex introduced
some races with block group relocation.  The problem is that relocation
needs to do some record keeping about each root, and it was relying
on the transaction mutex to coordinate things in subtle ways.

This fix adds a mutex just for the relocation code and makes sure
it doesn't have a big impact on normal operations.  The race is
really fixed in btrfs_record_root_in_trans, which is where we
step back and wait for the relocation code to finish accounting
setup.

Signed-off-by: Chris Mason <chris.mason@oracle.com>

4 files changed, 105 insertions, 13 deletions

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 8490ee063709..a2c91a102b72 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -967,6 +967,12 @@ struct btrfs_fs_info {
 	struct srcu_struct subvol_srcu;
 
 	spinlock_t trans_lock;
+	/*
+	 * the reloc mutex goes with the trans lock, it is taken
+	 * during commit to protect us from the relocation code
+	 */
+	struct mutex reloc_mutex;
+
 	struct list_head trans_list;
 	struct list_head hashers;
 	struct list_head dead_roots;
@@ -1172,6 +1178,14 @@ struct btrfs_root {
 	u32 type;
 
 	u64 highest_objectid;
+
+	/* btrfs_record_root_in_trans is a multi-step process,
+	 * and it can race with the balancing code.   But the
+	 * race is very small, and only the first time the root
+	 * is added to each transaction.  So in_trans_setup
+	 * is used to tell us when more checks are required
+	 */
+	unsigned long in_trans_setup;
 	int ref_cows;
 	int track_dirty;
 	int in_radix;
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index 20c111b3fa0d..0b2b4b759136 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -1620,6 +1620,7 @@ struct btrfs_root *open_ctree(struct super_block *sb,
 	spin_lock_init(&fs_info->fs_roots_radix_lock);
 	spin_lock_init(&fs_info->delayed_iput_lock);
 	spin_lock_init(&fs_info->defrag_inodes_lock);
+	mutex_init(&fs_info->reloc_mutex);
 
 	init_completion(&fs_info->kobj_unregister);
 	fs_info->tree_root = tree_root;
diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c
index f25b10a22a0a..086b1e6b8614 100644
--- a/fs/btrfs/relocation.c
+++ b/fs/btrfs/relocation.c
@@ -1368,7 +1368,7 @@ int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
 	int ret;
 
 	if (!root->reloc_root)
-		return 0;
+		goto out;
 
 	reloc_root = root->reloc_root;
 	root_item = &reloc_root->root_item;
@@ -1390,6 +1390,8 @@ int btrfs_update_reloc_root(struct btrfs_trans_handle *trans,
 	ret = btrfs_update_root(trans, root->fs_info->tree_root,
 				&reloc_root->root_key, root_item);
 	BUG_ON(ret);
+
+out:
 	return 0;
 }
 
@@ -2142,10 +2144,11 @@ int prepare_to_merge(struct reloc_control *rc, int err)
 	u64 num_bytes = 0;
 	int ret;
 
-	spin_lock(&root->fs_info->trans_lock);
+	mutex_lock(&root->fs_info->reloc_mutex);
 	rc->merging_rsv_size += root->nodesize * (BTRFS_MAX_LEVEL - 1) * 2;
 	rc->merging_rsv_size += rc->nodes_relocated * 2;
-	spin_unlock(&root->fs_info->trans_lock);
+	mutex_unlock(&root->fs_info->reloc_mutex);
+
 again:
 	if (!err) {
 		num_bytes = rc->merging_rsv_size;
@@ -2214,9 +2217,16 @@ int merge_reloc_roots(struct reloc_control *rc)
 	int ret;
 again:
 	root = rc->extent_root;
-	spin_lock(&root->fs_info->trans_lock);
+
+	/*
+	 * this serializes us with btrfs_record_root_in_transaction,
+	 * we have to make sure nobody is in the middle of
+	 * adding their roots to the list while we are
+	 * doing this splice
+	 */
+	mutex_lock(&root->fs_info->reloc_mutex);
 	list_splice_init(&rc->reloc_roots, &reloc_roots);
-	spin_unlock(&root->fs_info->trans_lock);
+	mutex_unlock(&root->fs_info->reloc_mutex);
 
 	while (!list_empty(&reloc_roots)) {
 		found = 1;
@@ -3590,17 +3600,19 @@ next:
 static void set_reloc_control(struct reloc_control *rc)
 {
 	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
-	spin_lock(&fs_info->trans_lock);
+
+	mutex_lock(&fs_info->reloc_mutex);
 	fs_info->reloc_ctl = rc;
-	spin_unlock(&fs_info->trans_lock);
+	mutex_unlock(&fs_info->reloc_mutex);
 }
 
 static void unset_reloc_control(struct reloc_control *rc)
 {
 	struct btrfs_fs_info *fs_info = rc->extent_root->fs_info;
-	spin_lock(&fs_info->trans_lock);
+
+	mutex_lock(&fs_info->reloc_mutex);
 	fs_info->reloc_ctl = NULL;
-	spin_unlock(&fs_info->trans_lock);
+	mutex_unlock(&fs_info->reloc_mutex);
 }
 
 static int check_extent_flags(u64 flags)
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index 2b3590b9fe98..833996a0c628 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -126,28 +126,85 @@ static noinline int join_transaction(struct btrfs_root *root, int nofail)
  * to make sure the old root from before we joined the transaction is deleted
  * when the transaction commits
  */
-int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
+static int record_root_in_trans(struct btrfs_trans_handle *trans,
 			       struct btrfs_root *root)
 {
 	if (root->ref_cows && root->last_trans < trans->transid) {
 		WARN_ON(root == root->fs_info->extent_root);
 		WARN_ON(root->commit_root != root->node);
 
+		/*
+		 * see below for in_trans_setup usage rules
+		 * we have the reloc mutex held now, so there
+		 * is only one writer in this function
+		 */
+		root->in_trans_setup = 1;
+
+		/* make sure readers find in_trans_setup before
+		 * they find our root->last_trans update
+		 */
+		smp_wmb();
+
 		spin_lock(&root->fs_info->fs_roots_radix_lock);
 		if (root->last_trans == trans->transid) {
 			spin_unlock(&root->fs_info->fs_roots_radix_lock);
 			return 0;
 		}
-		root->last_trans = trans->transid;
 		radix_tree_tag_set(&root->fs_info->fs_roots_radix,
 			   (unsigned long)root->root_key.objectid,
 			   BTRFS_ROOT_TRANS_TAG);
 		spin_unlock(&root->fs_info->fs_roots_radix_lock);
+		root->last_trans = trans->transid;
+
+		/* this is pretty tricky.  We don't want to
+		 * take the relocation lock in btrfs_record_root_in_trans
+		 * unless we're really doing the first setup for this root in
+		 * this transaction.
+		 *
+		 * Normally we'd use root->last_trans as a flag to decide
+		 * if we want to take the expensive mutex.
+		 *
+		 * But, we have to set root->last_trans before we
+		 * init the relocation root, otherwise, we trip over warnings
+		 * in ctree.c.  The solution used here is to flag ourselves
+		 * with root->in_trans_setup.  When this is 1, we're still
+		 * fixing up the reloc trees and everyone must wait.
+		 *
+		 * When this is zero, they can trust root->last_trans and fly
+		 * through btrfs_record_root_in_trans without having to take the
+		 * lock.  smp_wmb() makes sure that all the writes above are
+		 * done before we pop in the zero below
+		 */
 		btrfs_init_reloc_root(trans, root);
+		smp_wmb();
+		root->in_trans_setup = 0;
 	}
 	return 0;
 }
 
+
+int btrfs_record_root_in_trans(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root)
+{
+	if (!root->ref_cows)
+		return 0;
+
+	/*
+	 * see record_root_in_trans for comments about in_trans_setup usage
+	 * and barriers
+	 */
+	smp_rmb();
+	if (root->last_trans == trans->transid &&
+	    !root->in_trans_setup)
+		return 0;
+
+	mutex_lock(&root->fs_info->reloc_mutex);
+	record_root_in_trans(trans, root);
+	mutex_unlock(&root->fs_info->reloc_mutex);
+
+	return 0;
+}
+
 /* wait for commit against the current transaction to become unblocked
  * when this is done, it is safe to start a new transaction, but the current
  * transaction might not be fully on disk.
@@ -882,7 +939,7 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	parent = dget_parent(dentry);
 	parent_inode = parent->d_inode;
 	parent_root = BTRFS_I(parent_inode)->root;
-	btrfs_record_root_in_trans(trans, parent_root);
+	record_root_in_trans(trans, parent_root);
 
 	/*
 	 * insert the directory item
@@ -900,7 +957,7 @@ static noinline int create_pending_snapshot(struct btrfs_trans_handle *trans,
 	ret = btrfs_update_inode(trans, parent_root, parent_inode);
 	BUG_ON(ret);
 
-	btrfs_record_root_in_trans(trans, root);
+	record_root_in_trans(trans, root);
 	btrfs_set_root_last_snapshot(&root->root_item, trans->transid);
 	memcpy(new_root_item, &root->root_item, sizeof(*new_root_item));
 	btrfs_check_and_init_root_item(new_root_item);
@@ -1247,6 +1304,13 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	} while (atomic_read(&cur_trans->num_writers) > 1 ||
 		 (should_grow && cur_trans->num_joined != joined));
 
+	/*
+	 * the reloc mutex makes sure that we stop
+	 * the balancing code from coming in and moving
+	 * extents around in the middle of the commit
+	 */
+	mutex_lock(&root->fs_info->reloc_mutex);
+
 	ret = create_pending_snapshots(trans, root->fs_info);
 	BUG_ON(ret);
 
@@ -1312,6 +1376,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 	root->fs_info->running_transaction = NULL;
 	root->fs_info->trans_no_join = 0;
 	spin_unlock(&root->fs_info->trans_lock);
+	mutex_unlock(&root->fs_info->reloc_mutex);
 
 	wake_up(&root->fs_info->transaction_wait);