bcachefs: Fix discard path journal flushing

The discard path is supposed to issue journal flushes when there's too
many buckets empty buckets that need a journal commit before they can be
written to again, but at some point this code seems to have been lost.

Bring it back with a new optimization to make sure we don't issue too
many journal flushes: the journal now tracks the sequence number of the
most recent flush in progress, which the discard path uses when deciding
which buckets need a journal flush.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

Author: Kent Overstreet

fs/bcachefs/alloc_background.c

@@ -1803,7 +1803,6 @@ struct discard_buckets_state {
 	u64		open;
 	u64		need_journal_commit;
 	u64		discarded;
-	u64		need_journal_commit_this_dev;
 };
 
 static int bch2_discard_one_bucket(struct btree_trans *trans,
@@ -1827,11 +1826,11 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 		goto out;
 	}
 
-	if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
-			c->journal.flushed_seq_ondisk,
-			pos.inode, pos.offset)) {
-		s->need_journal_commit++;
-		s->need_journal_commit_this_dev++;
+	u64 seq_ready = bch2_bucket_journal_seq_ready(&c->buckets_waiting_for_journal,
+						      pos.inode, pos.offset);
+	if (seq_ready > c->journal.flushed_seq_ondisk) {
+		if (seq_ready > c->journal.flushing_seq)
+			s->need_journal_commit++;
 		goto out;
 	}
 
@@ -1865,23 +1864,24 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 		discard_locked = true;
 	}
 
-	if (!bkey_eq(*discard_pos_done, iter.pos) &&
-	    ca->mi.discard && !c->opts.nochanges) {
-		/*
-		 * This works without any other locks because this is the only
-		 * thread that removes items from the need_discard tree
-		 */
-		bch2_trans_unlock_long(trans);
-		blkdev_issue_discard(ca->disk_sb.bdev,
-				     k.k->p.offset * ca->mi.bucket_size,
-				     ca->mi.bucket_size,
-				     GFP_KERNEL);
-		*discard_pos_done = iter.pos;
+	if (!bkey_eq(*discard_pos_done, iter.pos)) {
 		s->discarded++;
+		*discard_pos_done = iter.pos;
 
-		ret = bch2_trans_relock_notrace(trans);
-		if (ret)
-			goto out;
+		if (ca->mi.discard && !c->opts.nochanges) {
+			/*
+			 * This works without any other locks because this is the only
+			 * thread that removes items from the need_discard tree
+			 */
+			bch2_trans_unlock_long(trans);
+			blkdev_issue_discard(ca->disk_sb.bdev,
+					     k.k->p.offset * ca->mi.bucket_size,
+					     ca->mi.bucket_size,
+					     GFP_KERNEL);
+			ret = bch2_trans_relock_notrace(trans);
+			if (ret)
+				goto out;
+		}
 	}
 
 	SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
@@ -1929,6 +1929,9 @@ static void bch2_do_discards_work(struct work_struct *work)
 				   POS(ca->dev_idx, U64_MAX), 0, k,
 			bch2_discard_one_bucket(trans, ca, &iter, &discard_pos_done, &s, false)));
 
+	if (s.need_journal_commit > dev_buckets_available(ca, BCH_WATERMARK_normal))
+		bch2_journal_flush_async(&c->journal, NULL);
+
 	trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded,
 			      bch2_err_str(ret));
 
@@ -2024,7 +2027,7 @@ static void bch2_do_discards_fast_work(struct work_struct *work)
 			break;
 	}
 
-	trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded, bch2_err_str(ret));
+	trace_discard_buckets_fast(c, s.seen, s.open, s.need_journal_commit, s.discarded, bch2_err_str(ret));
 
 	bch2_trans_put(trans);
 	percpu_ref_put(&ca->io_ref);

fs/bcachefs/alloc_foreground.c

@@ -205,8 +205,12 @@ static inline bool may_alloc_bucket(struct bch_fs *c,
 		return false;
 	}
 
-	if (bch2_bucket_needs_journal_commit(&c->buckets_waiting_for_journal,
-			c->journal.flushed_seq_ondisk, bucket.inode, bucket.offset)) {
+	u64 journal_seq_ready =
+		bch2_bucket_journal_seq_ready(&c->buckets_waiting_for_journal,
+					      bucket.inode, bucket.offset);
+	if (journal_seq_ready > c->journal.flushed_seq_ondisk) {
+		if (journal_seq_ready > c->journal.flushing_seq)
+			s->need_journal_commit++;
 		s->skipped_need_journal_commit++;
 		return false;
 	}
@@ -570,7 +574,7 @@ static struct open_bucket *bch2_bucket_alloc_trans(struct btree_trans *trans,
 		? bch2_bucket_alloc_freelist(trans, ca, watermark, &s, cl)
 		: bch2_bucket_alloc_early(trans, ca, watermark, &s, cl);
 
-	if (s.skipped_need_journal_commit * 2 > avail)
+	if (s.need_journal_commit * 2 > avail)
 		bch2_journal_flush_async(&c->journal, NULL);
 
 	if (!ob && s.btree_bitmap != BTREE_BITMAP_ANY) {

fs/bcachefs/alloc_types.h

@@ -18,6 +18,7 @@ struct bucket_alloc_state {
 	u64	buckets_seen;
 	u64	skipped_open;
 	u64	skipped_need_journal_commit;
+	u64	need_journal_commit;
 	u64	skipped_nocow;
 	u64	skipped_nouse;
 	u64	skipped_mi_btree_bitmap;

fs/bcachefs/buckets_waiting_for_journal.c

@@ -22,23 +22,21 @@ static void bucket_table_init(struct buckets_waiting_for_journal_table *t, size_
 	memset(t->d, 0, sizeof(t->d[0]) << t->bits);
 }
 
-bool bch2_bucket_needs_journal_commit(struct buckets_waiting_for_journal *b,
-				      u64 flushed_seq,
-				      unsigned dev, u64 bucket)
+u64 bch2_bucket_journal_seq_ready(struct buckets_waiting_for_journal *b,
+				  unsigned dev, u64 bucket)
 {
 	struct buckets_waiting_for_journal_table *t;
 	u64 dev_bucket = (u64) dev << 56 | bucket;
-	bool ret = false;
-	unsigned i;
+	u64 ret = 0;
 
 	mutex_lock(&b->lock);
 	t = b->t;
 
-	for (i = 0; i < ARRAY_SIZE(t->hash_seeds); i++) {
+	for (unsigned i = 0; i < ARRAY_SIZE(t->hash_seeds); i++) {
 		struct bucket_hashed *h = bucket_hash(t, i, dev_bucket);
 
 		if (h->dev_bucket == dev_bucket) {
-			ret = h->journal_seq > flushed_seq;
+			ret = h->journal_seq;
 			break;
 		}
 	}

fs/bcachefs/buckets_waiting_for_journal.h

@@ -4,8 +4,8 @@
 
 #include "buckets_waiting_for_journal_types.h"
 
-bool bch2_bucket_needs_journal_commit(struct buckets_waiting_for_journal *,
-				      u64, unsigned, u64);
+u64 bch2_bucket_journal_seq_ready(struct buckets_waiting_for_journal *,
+				  unsigned, u64);
 int bch2_set_bucket_needs_journal_commit(struct buckets_waiting_for_journal *,
 					 u64, unsigned, u64, u64);
 

fs/bcachefs/journal.c

@@ -796,6 +796,7 @@ int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 	}
 
 	buf->must_flush = true;
+	j->flushing_seq = max(j->flushing_seq, seq);
 
 	if (parent && !closure_wait(&buf->wait, parent))
 		BUG();

fs/bcachefs/journal_types.h

@@ -237,6 +237,7 @@ struct journal {
 	/* seq, last_seq from the most recent journal entry successfully written */
 	u64			seq_ondisk;
 	u64			flushed_seq_ondisk;
+	u64			flushing_seq;
 	u64			last_seq_ondisk;
 	u64			err_seq;
 	u64			last_empty_seq;

fs/bcachefs/trace.h

@@ -727,7 +727,7 @@ DEFINE_EVENT(fs_str, bucket_alloc_fail,
 	TP_ARGS(c, str)
 );
 
-TRACE_EVENT(discard_buckets,
+DECLARE_EVENT_CLASS(discard_buckets_class,
 	TP_PROTO(struct bch_fs *c, u64 seen, u64 open,
 		 u64 need_journal_commit, u64 discarded, const char *err),
 	TP_ARGS(c, seen, open, need_journal_commit, discarded, err),
@@ -759,6 +759,18 @@ TRACE_EVENT(discard_buckets,
 		  __entry->err)
 );
 
+DEFINE_EVENT(discard_buckets_class, discard_buckets,
+	TP_PROTO(struct bch_fs *c, u64 seen, u64 open,
+		 u64 need_journal_commit, u64 discarded, const char *err),
+	TP_ARGS(c, seen, open, need_journal_commit, discarded, err)
+);
+
+DEFINE_EVENT(discard_buckets_class, discard_buckets_fast,
+	TP_PROTO(struct bch_fs *c, u64 seen, u64 open,
+		 u64 need_journal_commit, u64 discarded, const char *err),
+	TP_ARGS(c, seen, open, need_journal_commit, discarded, err)
+);
+
 TRACE_EVENT(bucket_invalidate,
 	TP_PROTO(struct bch_fs *c, unsigned dev, u64 bucket, u32 sectors),
 	TP_ARGS(c, dev, bucket, sectors),