[PATCH v2.1 09/10] btrfs: qgroup: Commit transaction in advance to reduce early EDQUOT

[Qu Wenruo <wqu@suse.com>]

Unlike previous method to try commit transaction inside
qgroup_reserve(), this time we will try to commit transaction using
fs_info->transaction_kthread to avoid nested transaction and no need to
worry about lock context.

Since it's an asynchronous function call and we won't wait transaction
commit, unlike previous method, we must call it before we hit qgroup
limit.

So this patch will use the ratio and size of qgroup meta_pertrans
reservation as indicator to check if we should trigger a transaction
commitment.
(meta_prealloc won't be cleaned in transaction commitment, it's useless
anyway)

Signed-off-by: Qu Wenruo <wqu@suse.com>
---
v2.1
  Allow transaction_kthread to commit without waiting for commit
  interval.
  Previous version it doesn't really work due to the commit interval
  limit.

  Now btrfs can handle small file creation much better. Although
  btrfs/139 still hit EDQUOT earlier than expected.
---
 fs/btrfs/ctree.h       |  6 ++++++
 fs/btrfs/disk-io.c     |  1 +
 fs/btrfs/qgroup.c      | 43 +++++++++++++++++++++++++++++++++++++++++--
 fs/btrfs/transaction.c |  1 +
 fs/btrfs/transaction.h | 14 ++++++++++++++
 5 files changed, 63 insertions(+), 2 deletions(-)

diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h
index 0c58f92c2d44..141a36571bd7 100644
--- a/fs/btrfs/ctree.h
+++ b/fs/btrfs/ctree.h
@@ -724,6 +724,12 @@ struct btrfs_delayed_root;
  */
 #define BTRFS_FS_EXCL_OP			16
 
+/*
+ * To info transaction_kthread we need an immediate commit so it doesn't
+ * need to wait for commit_interval
+ */
+#define BTRFS_FS_NEED_ASYNC_COMMIT		17
+
 struct btrfs_fs_info {
 	u8 fsid[BTRFS_FSID_SIZE];
 	u8 chunk_tree_uuid[BTRFS_UUID_SIZE];
diff --git a/fs/btrfs/disk-io.c b/fs/btrfs/disk-io.c
index d8eaadac4330..a33e1da195ec 100644
--- a/fs/btrfs/disk-io.c
+++ b/fs/btrfs/disk-io.c
@@ -1787,6 +1787,7 @@ static int transaction_kthread(void *arg)
 
 		now = get_seconds();
 		if (cur->state < TRANS_STATE_BLOCKED &&
+		    !test_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags) &&
 		    (now < cur->start_time ||
 		     now - cur->start_time < fs_info->commit_interval)) {
 			spin_unlock(&fs_info->trans_lock);
diff --git a/fs/btrfs/qgroup.c b/fs/btrfs/qgroup.c
index 6d5b476feb08..57d8fd605655 100644
--- a/fs/btrfs/qgroup.c
+++ b/fs/btrfs/qgroup.c
@@ -24,6 +24,7 @@
 #include <linux/slab.h>
 #include <linux/workqueue.h>
 #include <linux/btrfs.h>
+#include <linux/sizes.h>
 
 #include "ctree.h"
 #include "transaction.h"
@@ -2395,8 +2396,21 @@ int btrfs_qgroup_inherit(struct btrfs_trans_handle *trans,
 	return ret;
 }
 
-static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
+/*
+ * Two limits to commit transaction in advance.
+ *
+ * For RATIO, it will be 1/RATIO of the remaining limit
+ * (excluding data and prealloc meta) as threshold.
+ * For SIZE, it will be in byte unit as threshold.
+ */
+#define QGROUP_PERTRANS_RATIO		32
+#define QGROUP_PERTRANS_SIZE		SZ_32M
+static bool qgroup_check_limits(struct btrfs_fs_info *fs_info,
+				const struct btrfs_qgroup *qg, u64 num_bytes)
 {
+	u64 limit;
+	u64 threshold;
+
 	if ((qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_RFER) &&
 	    qgroup_rsv_total(qg) + (s64)qg->rfer + num_bytes > qg->max_rfer)
 		return false;
@@ -2405,6 +2419,31 @@ static bool qgroup_check_limits(const struct btrfs_qgroup *qg, u64 num_bytes)
 	    qgroup_rsv_total(qg) + (s64)qg->excl + num_bytes > qg->max_excl)
 		return false;
 
+	/*
+	 * Even if we passed the check, it's better to check if reservation
+	 * for meta_pertrans is pushing us near limit.
+	 * If there is too much pertrans reservation or it's near the limit,
+	 * let's try commit transaction to free some, using transaction_kthread
+	 */
+	if ((qg->lim_flags & (BTRFS_QGROUP_LIMIT_MAX_RFER |
+			      BTRFS_QGROUP_LIMIT_MAX_EXCL))) {
+		if (qg->lim_flags & BTRFS_QGROUP_LIMIT_MAX_EXCL)
+			limit = qg->max_excl;
+		else
+			limit = qg->max_rfer;
+		threshold = (limit - qg->rsv.values[BTRFS_QGROUP_RSV_DATA] -
+			    qg->rsv.values[BTRFS_QGROUP_RSV_META_PREALLOC]) /
+			    QGROUP_PERTRANS_RATIO;
+		threshold = min_t(u64, threshold, QGROUP_PERTRANS_SIZE);
+
+		/*
+		 * Use transaction_kthread to commit transaction, so we no
+		 * longer need to bother nested transaction nor lock context.
+		 */
+		if (qg->rsv.values[BTRFS_QGROUP_RSV_META_PERTRANS] > threshold)
+			btrfs_commit_transaction_locksafe(fs_info);
+	}
+
 	return true;
 }
 
@@ -2454,7 +2493,7 @@ static int qgroup_reserve(struct btrfs_root *root, u64 num_bytes, bool enforce,
 
 		qg = unode_aux_to_qgroup(unode);
 
-		if (enforce && !qgroup_check_limits(qg, num_bytes)) {
+		if (enforce && !qgroup_check_limits(fs_info, qg, num_bytes)) {
 			ret = -EDQUOT;
 			goto out;
 		}
diff --git a/fs/btrfs/transaction.c b/fs/btrfs/transaction.c
index ddae813c01dd..5e1c1d2295eb 100644
--- a/fs/btrfs/transaction.c
+++ b/fs/btrfs/transaction.c
@@ -2289,6 +2289,7 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans)
 	 */
 	cur_trans->state = TRANS_STATE_COMPLETED;
 	wake_up(&cur_trans->commit_wait);
+	clear_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags);
 
 	spin_lock(&fs_info->trans_lock);
 	list_del_init(&cur_trans->list);
diff --git a/fs/btrfs/transaction.h b/fs/btrfs/transaction.h
index c55e44560103..63f1dc7f9268 100644
--- a/fs/btrfs/transaction.h
+++ b/fs/btrfs/transaction.h
@@ -204,6 +204,20 @@ int btrfs_clean_one_deleted_snapshot(struct btrfs_root *root);
 int btrfs_commit_transaction(struct btrfs_trans_handle *trans);
 int btrfs_commit_transaction_async(struct btrfs_trans_handle *trans,
 				   int wait_for_unblock);
+
+/*
+ * Try to commit transaction asynchronously, so this is safe to call
+ * even holding a spinlock.
+ *
+ * It's done by informing transaction_kthread to commit transaction without
+ * waiting for commit interval.
+ */
+static inline void btrfs_commit_transaction_locksafe(
+		struct btrfs_fs_info *fs_info)
+{
+	set_bit(BTRFS_FS_NEED_ASYNC_COMMIT, &fs_info->flags);
+	wake_up_process(fs_info->transaction_kthread);
+}
 int btrfs_end_transaction_throttle(struct btrfs_trans_handle *trans);
 int btrfs_should_end_transaction(struct btrfs_trans_handle *trans);
 void btrfs_throttle(struct btrfs_fs_info *fs_info);
-- 
2.15.1