From e48fda6cdc2799ef4a63692f0f11240613067b38 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Mon, 18 Nov 2024 00:32:57 -0500
Subject: [PATCH 01/16] bcachefs: Fix check_backpointers_to_extents range
 limiting

bch2_get_btree_in_memory_pos() will return positions that refer directly
to the btree it's checking will fit in memory - i.e. backpointer
positions, not buckets.

This also means check_bp_exists() no longer has to refer to the device,
and we can delete some code.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/backpointers.c | 63 +++++++++++++++-----------------------
 1 file changed, 25 insertions(+), 38 deletions(-)

diff --git a/fs/bcachefs/backpointers.c b/fs/bcachefs/backpointers.c
index 24804f0bb2fd..5963217cd90c 100644
--- a/fs/bcachefs/backpointers.c
+++ b/fs/bcachefs/backpointers.c
@@ -352,8 +352,8 @@ int bch2_check_btree_backpointers(struct bch_fs *c)
 }
 
 struct extents_to_bp_state {
-	struct bpos	bucket_start;
-	struct bpos	bucket_end;
+	struct bpos	bp_start;
+	struct bpos	bp_end;
 	struct bkey_buf last_flushed;
 };
 
@@ -445,29 +445,16 @@ static int check_bp_exists(struct btree_trans *trans,
 			   struct bkey_s_c orig_k)
 {
 	struct bch_fs *c = trans->c;
-	struct btree_iter bp_iter = {};
 	struct btree_iter other_extent_iter = {};
 	struct printbuf buf = PRINTBUF;
-	struct bkey_s_c bp_k;
-	int ret = 0;
 
-	struct bch_dev *ca = bch2_dev_tryget_noerror(c, bp->k.p.inode);
-	if (!ca) {
-		prt_printf(&buf, "extent for nonexistent device %llu\n", bp->k.p.inode);
-		bch2_bkey_val_to_text(&buf, c, orig_k);
-		bch_err(c, "%s", buf.buf);
-		ret = -BCH_ERR_fsck_repair_unimplemented;
-		goto err;
-	}
-
-	struct bpos bucket = bp_pos_to_bucket(ca, bp->k.p);
-
-	if (bpos_lt(bucket, s->bucket_start) ||
-	    bpos_gt(bucket, s->bucket_end))
-		goto out;
+	if (bpos_lt(bp->k.p, s->bp_start) ||
+	    bpos_gt(bp->k.p, s->bp_end))
+		return 0;
 
-	bp_k = bch2_bkey_get_iter(trans, &bp_iter, BTREE_ID_backpointers, bp->k.p, 0);
-	ret = bkey_err(bp_k);
+	struct btree_iter bp_iter;
+	struct bkey_s_c bp_k = bch2_bkey_get_iter(trans, &bp_iter, BTREE_ID_backpointers, bp->k.p, 0);
+	int ret = bkey_err(bp_k);
 	if (ret)
 		goto err;
 
@@ -484,7 +471,6 @@ err:
 fsck_err:
 	bch2_trans_iter_exit(trans, &other_extent_iter);
 	bch2_trans_iter_exit(trans, &bp_iter);
-	bch2_dev_put(ca);
 	printbuf_exit(&buf);
 	return ret;
 check_existing_bp:
@@ -514,12 +500,13 @@ check_existing_bp:
 		bch_err(c, "%s", buf.buf);
 
 		if (other_extent.k->size <= orig_k.k->size) {
-			ret = drop_dev_and_update(trans, other_bp.v->btree_id, other_extent, bucket.inode);
+			ret = drop_dev_and_update(trans, other_bp.v->btree_id,
+						  other_extent, bp->k.p.inode);
 			if (ret)
 				goto err;
 			goto out;
 		} else {
-			ret = drop_dev_and_update(trans, bp->v.btree_id, orig_k, bucket.inode);
+			ret = drop_dev_and_update(trans, bp->v.btree_id, orig_k, bp->k.p.inode);
 			if (ret)
 				goto err;
 			goto missing;
@@ -529,7 +516,7 @@ check_existing_bp:
 	ret = check_extent_checksum(trans,
 				    other_bp.v->btree_id, other_extent,
 				    bp->v.btree_id, orig_k,
-				    bucket.inode);
+				    bp->k.p.inode);
 	if (ret < 0)
 		goto err;
 	if (ret) {
@@ -538,7 +525,7 @@ check_existing_bp:
 	}
 
 	ret = check_extent_checksum(trans, bp->v.btree_id, orig_k,
-				    other_bp.v->btree_id, other_extent, bucket.inode);
+				    other_bp.v->btree_id, other_extent, bp->k.p.inode);
 	if (ret < 0)
 		goto err;
 	if (ret) {
@@ -547,7 +534,7 @@ check_existing_bp:
 	}
 
 	printbuf_reset(&buf);
-	prt_printf(&buf, "duplicate extents pointing to same space on dev %llu\n  ", bucket.inode);
+	prt_printf(&buf, "duplicate extents pointing to same space on dev %llu\n  ", bp->k.p.inode);
 	bch2_bkey_val_to_text(&buf, c, orig_k);
 	prt_str(&buf, "\n  ");
 	bch2_bkey_val_to_text(&buf, c, other_extent);
@@ -811,7 +798,7 @@ static int bch2_check_extents_to_backpointers_pass(struct btree_trans *trans,
 int bch2_check_extents_to_backpointers(struct bch_fs *c)
 {
 	struct btree_trans *trans = bch2_trans_get(c);
-	struct extents_to_bp_state s = { .bucket_start = POS_MIN };
+	struct extents_to_bp_state s = { .bp_start = POS_MIN };
 	int ret;
 
 	bch2_bkey_buf_init(&s.last_flushed);
@@ -822,35 +809,35 @@ int bch2_check_extents_to_backpointers(struct bch_fs *c)
 		ret = bch2_get_btree_in_memory_pos(trans,
 				BIT_ULL(BTREE_ID_backpointers),
 				BIT_ULL(BTREE_ID_backpointers),
-				BBPOS(BTREE_ID_backpointers, s.bucket_start), &end);
+				BBPOS(BTREE_ID_backpointers, s.bp_start), &end);
 		if (ret)
 			break;
 
-		s.bucket_end = end.pos;
+		s.bp_end = end.pos;
 
-		if ( bpos_eq(s.bucket_start, POS_MIN) &&
-		    !bpos_eq(s.bucket_end, SPOS_MAX))
+		if ( bpos_eq(s.bp_start, POS_MIN) &&
+		    !bpos_eq(s.bp_end, SPOS_MAX))
 			bch_verbose(c, "%s(): alloc info does not fit in ram, running in multiple passes with %zu nodes per pass",
 				    __func__, btree_nodes_fit_in_ram(c));
 
-		if (!bpos_eq(s.bucket_start, POS_MIN) ||
-		    !bpos_eq(s.bucket_end, SPOS_MAX)) {
+		if (!bpos_eq(s.bp_start, POS_MIN) ||
+		    !bpos_eq(s.bp_end, SPOS_MAX)) {
 			struct printbuf buf = PRINTBUF;
 
 			prt_str(&buf, "check_extents_to_backpointers(): ");
-			bch2_bpos_to_text(&buf, s.bucket_start);
+			bch2_bpos_to_text(&buf, s.bp_start);
 			prt_str(&buf, "-");
-			bch2_bpos_to_text(&buf, s.bucket_end);
+			bch2_bpos_to_text(&buf, s.bp_end);
 
 			bch_verbose(c, "%s", buf.buf);
 			printbuf_exit(&buf);
 		}
 
 		ret = bch2_check_extents_to_backpointers_pass(trans, &s);
-		if (ret || bpos_eq(s.bucket_end, SPOS_MAX))
+		if (ret || bpos_eq(s.bp_end, SPOS_MAX))
 			break;
 
-		s.bucket_start = bpos_successor(s.bucket_end);
+		s.bp_start = bpos_successor(s.bp_end);
 	}
 	bch2_trans_put(trans);
 	bch2_bkey_buf_exit(&s.last_flushed, c);
-- 
2.51.0


From 1ab00b6cddc53d47e6f45002e69034941d40ae03 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sun, 17 Nov 2024 18:26:54 -0500
Subject: [PATCH 02/16] bcachefs: kill bch_backpointer.bucket_offset usage

bch_backpointer.bucket_offset is going away - it's no longer needed
since we no longer store backpointers in alloc keys, the same
information is in the key position itself.

And we'll be reclaiming the space in bch_backpointer for the bucket
generation number.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/backpointers.c | 15 +++++++--------
 fs/bcachefs/backpointers.h |  8 ++++++++
 2 files changed, 15 insertions(+), 8 deletions(-)

diff --git a/fs/bcachefs/backpointers.c b/fs/bcachefs/backpointers.c
index 5963217cd90c..620fa67db7a6 100644
--- a/fs/bcachefs/backpointers.c
+++ b/fs/bcachefs/backpointers.c
@@ -54,21 +54,20 @@ void bch2_backpointer_to_text(struct printbuf *out, struct bch_fs *c, struct bke
 	struct bkey_s_c_backpointer bp = bkey_s_c_to_backpointer(k);
 
 	rcu_read_lock();
-	struct bch_dev *ca = bch2_dev_rcu_noerror(c, k.k->p.inode);
+	struct bch_dev *ca = bch2_dev_rcu_noerror(c, bp.k->p.inode);
 	if (ca) {
-		struct bpos bucket = bp_pos_to_bucket(ca, k.k->p);
+		u32 bucket_offset;
+		struct bpos bucket = bp_pos_to_bucket_and_offset(ca, bp.k->p, &bucket_offset);
 		rcu_read_unlock();
-		prt_str(out, "bucket=");
-		bch2_bpos_to_text(out, bucket);
-		prt_str(out, " ");
+		prt_printf(out, "bucket=%llu:%llu:%u", bucket.inode, bucket.offset, bucket_offset);
 	} else {
 		rcu_read_unlock();
+		prt_printf(out, "sector=%llu:%llu", bp.k->p.inode, bp.k->p.offset >> MAX_EXTENT_COMPRESS_RATIO_SHIFT);
 	}
 
 	bch2_btree_id_level_to_text(out, bp.v->btree_id, bp.v->level);
-	prt_printf(out, " offset=%llu:%u len=%u pos=",
-		   (u64) (bp.v->bucket_offset >> MAX_EXTENT_COMPRESS_RATIO_SHIFT),
-		   (u32) bp.v->bucket_offset & ~(~0U << MAX_EXTENT_COMPRESS_RATIO_SHIFT),
+	prt_printf(out, " suboffset=%u len=%u pos=",
+		   (u32) bp.k->p.offset & ~(~0U << MAX_EXTENT_COMPRESS_RATIO_SHIFT),
 		   bp.v->bucket_len);
 	bch2_bpos_to_text(out, bp.v->pos);
 }
diff --git a/fs/bcachefs/backpointers.h b/fs/bcachefs/backpointers.h
index 5f34a25b599a..d8a15f5fa767 100644
--- a/fs/bcachefs/backpointers.h
+++ b/fs/bcachefs/backpointers.h
@@ -42,6 +42,14 @@ static inline struct bpos bp_pos_to_bucket(const struct bch_dev *ca, struct bpos
 	return POS(bp_pos.inode, sector_to_bucket(ca, bucket_sector));
 }
 
+static inline struct bpos bp_pos_to_bucket_and_offset(const struct bch_dev *ca, struct bpos bp_pos,
+						      u32 *bucket_offset)
+{
+	u64 bucket_sector = bp_pos.offset >> MAX_EXTENT_COMPRESS_RATIO_SHIFT;
+
+	return POS(bp_pos.inode, sector_to_bucket_and_offset(ca, bucket_sector, bucket_offset));
+}
+
 static inline bool bp_pos_to_bucket_nodev_noerror(struct bch_fs *c, struct bpos bp_pos, struct bpos *bucket)
 {
 	rcu_read_lock();
-- 
2.51.0


From c80f33b752688392adf2aa71190636b8e1ff204b Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Thu, 14 Nov 2024 21:34:43 -0500
Subject: [PATCH 03/16] bcachefs: New backpointers helpers

- bch2_backpointer_del()
- bch2_backpointer_maybe_flush()

Kill a bit of open coding and make sure we're properly handling the
btree write buffer.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/backpointers.c | 58 +++++++++++++++++++++++++++++---------
 1 file changed, 45 insertions(+), 13 deletions(-)

diff --git a/fs/bcachefs/backpointers.c b/fs/bcachefs/backpointers.c
index 620fa67db7a6..cfd9b9ead473 100644
--- a/fs/bcachefs/backpointers.c
+++ b/fs/bcachefs/backpointers.c
@@ -199,6 +199,22 @@ err:
 	return ret;
 }
 
+static int bch2_backpointer_del(struct btree_trans *trans, struct bpos pos)
+{
+	return likely(!bch2_backpointers_no_use_write_buffer)
+	       ? bch2_btree_delete_at_buffered(trans, BTREE_ID_backpointers, pos)
+	       : bch2_btree_delete(trans, BTREE_ID_backpointers, pos, 0);
+}
+
+static int bch2_backpointers_maybe_flush(struct btree_trans *trans,
+					 struct bkey_s_c visiting_k,
+					 struct bkey_buf *last_flushed)
+{
+	return likely(!bch2_backpointers_no_use_write_buffer)
+		? bch2_btree_write_buffer_maybe_flush(trans, visiting_k, last_flushed)
+		: 0;
+}
+
 static void backpointer_target_not_found(struct btree_trans *trans,
 					 struct bkey_s_c_backpointer bp,
 					 struct bkey_s_c target_k)
@@ -300,9 +316,12 @@ err:
 	return b;
 }
 
-static int bch2_check_btree_backpointer(struct btree_trans *trans, struct btree_iter *bp_iter,
-					struct bkey_s_c k)
+static int bch2_check_backpointer_has_valid_bucket(struct btree_trans *trans, struct bkey_s_c k,
+						   struct bkey_buf *last_flushed)
 {
+	if (k.k->type != KEY_TYPE_backpointer)
+		return 0;
+
 	struct bch_fs *c = trans->c;
 	struct btree_iter alloc_iter = { NULL };
 	struct bkey_s_c alloc_k;
@@ -311,10 +330,14 @@ static int bch2_check_btree_backpointer(struct btree_trans *trans, struct btree_
 
 	struct bpos bucket;
 	if (!bp_pos_to_bucket_nodev_noerror(c, k.k->p, &bucket)) {
+		ret = bch2_backpointers_maybe_flush(trans, k, last_flushed);
+		if (ret)
+			goto out;
+
 		if (fsck_err(trans, backpointer_to_missing_device,
 			     "backpointer for missing device:\n%s",
 			     (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
-			ret = bch2_btree_delete_at(trans, bp_iter, 0);
+			ret = bch2_backpointer_del(trans, k.k->p);
 		goto out;
 	}
 
@@ -323,13 +346,16 @@ static int bch2_check_btree_backpointer(struct btree_trans *trans, struct btree_
 	if (ret)
 		goto out;
 
-	if (fsck_err_on(alloc_k.k->type != KEY_TYPE_alloc_v4,
-			trans, backpointer_to_missing_alloc,
-			"backpointer for nonexistent alloc key: %llu:%llu:0\n%s",
-			alloc_iter.pos.inode, alloc_iter.pos.offset,
-			(bch2_bkey_val_to_text(&buf, c, k), buf.buf))) {
-		ret = bch2_btree_delete_at(trans, bp_iter, 0);
-		goto out;
+	if (alloc_k.k->type != KEY_TYPE_alloc_v4) {
+		ret = bch2_backpointers_maybe_flush(trans, k, last_flushed);
+		if (ret)
+			goto out;
+
+		if (fsck_err(trans, backpointer_to_missing_alloc,
+			     "backpointer for nonexistent alloc key: %llu:%llu:0\n%s",
+			     alloc_iter.pos.inode, alloc_iter.pos.offset,
+			     (bch2_bkey_val_to_text(&buf, c, k), buf.buf)))
+			ret = bch2_backpointer_del(trans, k.k->p);
 	}
 out:
 fsck_err:
@@ -341,11 +367,17 @@ fsck_err:
 /* verify that every backpointer has a corresponding alloc key */
 int bch2_check_btree_backpointers(struct bch_fs *c)
 {
+	struct bkey_buf last_flushed;
+	bch2_bkey_buf_init(&last_flushed);
+	bkey_init(&last_flushed.k->k);
+
 	int ret = bch2_trans_run(c,
 		for_each_btree_key_commit(trans, iter,
 			BTREE_ID_backpointers, POS_MIN, 0, k,
 			NULL, NULL, BCH_TRANS_COMMIT_no_enospc,
-		  bch2_check_btree_backpointer(trans, &iter, k)));
+		  bch2_check_backpointer_has_valid_bucket(trans, k, &last_flushed)));
+
+	bch2_bkey_buf_exit(&last_flushed, c);
 	bch_err_fn(c, ret);
 	return ret;
 }
@@ -874,7 +906,7 @@ static int check_one_backpointer(struct btree_trans *trans,
 		return ret;
 
 	if (!k.k) {
-		ret = bch2_btree_write_buffer_maybe_flush(trans, bp.s_c, last_flushed);
+		ret = bch2_backpointers_maybe_flush(trans, bp.s_c, last_flushed);
 		if (ret)
 			goto out;
 
@@ -882,7 +914,7 @@ static int check_one_backpointer(struct btree_trans *trans,
 			     "backpointer for missing %s\n  %s",
 			     bp.v->level ? "btree node" : "extent",
 			     (bch2_bkey_val_to_text(&buf, c, bp.s_c), buf.buf))) {
-			ret = bch2_btree_delete_at_buffered(trans, BTREE_ID_backpointers, bp.k->p);
+			ret = bch2_backpointer_del(trans, bp.k->p);
 			goto out;
 		}
 	}
-- 
2.51.0


From c601e5d7daea102481bd7512b013018fdbf57034 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sat, 23 Nov 2024 16:27:47 -0500
Subject: [PATCH 04/16] bcachefs: Can now block journal activity without
 closing cur entry

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/journal.c       | 44 +++++++++++++++++++++++++++++++++++--
 fs/bcachefs/journal.h       |  3 ++-
 fs/bcachefs/journal_types.h |  2 ++
 3 files changed, 46 insertions(+), 3 deletions(-)

diff --git a/fs/bcachefs/journal.c b/fs/bcachefs/journal.c
index 2cf8f24d50cc..bfbb1ac60c3d 100644
--- a/fs/bcachefs/journal.c
+++ b/fs/bcachefs/journal.c
@@ -217,6 +217,12 @@ void bch2_journal_buf_put_final(struct journal *j, u64 seq)
 	if (__bch2_journal_pin_put(j, seq))
 		bch2_journal_reclaim_fast(j);
 	bch2_journal_do_writes(j);
+
+	/*
+	 * for __bch2_next_write_buffer_flush_journal_buf(), when quiescing an
+	 * open journal entry
+	 */
+	wake_up(&j->wait);
 }
 
 /*
@@ -251,6 +257,9 @@ static void __journal_entry_close(struct journal *j, unsigned closed_val, bool t
 	if (!__journal_entry_is_open(old))
 		return;
 
+	if (old.cur_entry_offset == JOURNAL_ENTRY_BLOCKED_VAL)
+		old.cur_entry_offset = j->cur_entry_offset_if_blocked;
+
 	/* Close out old buffer: */
 	buf->data->u64s		= cpu_to_le32(old.cur_entry_offset);
 
@@ -868,16 +877,44 @@ int bch2_journal_meta(struct journal *j)
 void bch2_journal_unblock(struct journal *j)
 {
 	spin_lock(&j->lock);
-	j->blocked--;
+	if (!--j->blocked &&
+	    j->cur_entry_offset_if_blocked < JOURNAL_ENTRY_CLOSED_VAL &&
+	    j->reservations.cur_entry_offset == JOURNAL_ENTRY_BLOCKED_VAL) {
+		union journal_res_state old, new;
+
+		old.v = atomic64_read(&j->reservations.counter);
+		do {
+			new.v = old.v;
+			new.cur_entry_offset = j->cur_entry_offset_if_blocked;
+		} while (!atomic64_try_cmpxchg(&j->reservations.counter, &old.v, new.v));
+	}
 	spin_unlock(&j->lock);
 
 	journal_wake(j);
 }
 
+static void __bch2_journal_block(struct journal *j)
+{
+	if (!j->blocked++) {
+		union journal_res_state old, new;
+
+		old.v = atomic64_read(&j->reservations.counter);
+		do {
+			j->cur_entry_offset_if_blocked = old.cur_entry_offset;
+
+			if (j->cur_entry_offset_if_blocked >= JOURNAL_ENTRY_CLOSED_VAL)
+				break;
+
+			new.v = old.v;
+			new.cur_entry_offset = JOURNAL_ENTRY_BLOCKED_VAL;
+		} while (!atomic64_try_cmpxchg(&j->reservations.counter, &old.v, new.v));
+	}
+}
+
 void bch2_journal_block(struct journal *j)
 {
 	spin_lock(&j->lock);
-	j->blocked++;
+	__bch2_journal_block(j);
 	spin_unlock(&j->lock);
 
 	journal_quiesce(j);
@@ -1481,6 +1518,9 @@ void __bch2_journal_debug_to_text(struct printbuf *out, struct journal *j)
 	case JOURNAL_ENTRY_CLOSED_VAL:
 		prt_printf(out, "closed\n");
 		break;
+	case JOURNAL_ENTRY_BLOCKED_VAL:
+		prt_printf(out, "blocked\n");
+		break;
 	default:
 		prt_printf(out, "%u/%u\n", s.cur_entry_offset, j->cur_entry_u64s);
 		break;
diff --git a/fs/bcachefs/journal.h b/fs/bcachefs/journal.h
index 2762be6f9814..6d3c839bbbef 100644
--- a/fs/bcachefs/journal.h
+++ b/fs/bcachefs/journal.h
@@ -285,7 +285,8 @@ static inline void bch2_journal_buf_put(struct journal *j, unsigned idx, u64 seq
 		spin_lock(&j->lock);
 		bch2_journal_buf_put_final(j, seq);
 		spin_unlock(&j->lock);
-	}
+	} else if (unlikely(s.cur_entry_offset == JOURNAL_ENTRY_BLOCKED_VAL))
+		wake_up(&j->wait);
 }
 
 /*
diff --git a/fs/bcachefs/journal_types.h b/fs/bcachefs/journal_types.h
index 19183fcf7ad7..425d1abb257e 100644
--- a/fs/bcachefs/journal_types.h
+++ b/fs/bcachefs/journal_types.h
@@ -112,6 +112,7 @@ union journal_res_state {
  */
 #define JOURNAL_ENTRY_OFFSET_MAX	((1U << 20) - 1)
 
+#define JOURNAL_ENTRY_BLOCKED_VAL	(JOURNAL_ENTRY_OFFSET_MAX - 2)
 #define JOURNAL_ENTRY_CLOSED_VAL	(JOURNAL_ENTRY_OFFSET_MAX - 1)
 #define JOURNAL_ENTRY_ERROR_VAL		(JOURNAL_ENTRY_OFFSET_MAX)
 
@@ -193,6 +194,7 @@ struct journal {
 	 * insufficient devices:
 	 */
 	enum journal_errors	cur_entry_error;
+	unsigned		cur_entry_offset_if_blocked;
 
 	unsigned		buf_size_want;
 	/*
-- 
2.51.0


From f4d67f6d5a4c042a28fa1d67f756e3d24ba60888 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sat, 23 Nov 2024 18:21:12 -0500
Subject: [PATCH 05/16] bcachefs: trivial btree write buffer refactoring

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/btree_write_buffer.c | 64 ++++++++++++++++----------------
 1 file changed, 31 insertions(+), 33 deletions(-)

diff --git a/fs/bcachefs/btree_write_buffer.c b/fs/bcachefs/btree_write_buffer.c
index 1639c60dffa0..1bd26221f156 100644
--- a/fs/bcachefs/btree_write_buffer.c
+++ b/fs/bcachefs/btree_write_buffer.c
@@ -19,8 +19,6 @@
 static int bch2_btree_write_buffer_journal_flush(struct journal *,
 				struct journal_entry_pin *, u64);
 
-static int bch2_journal_keys_to_write_buffer(struct bch_fs *, struct journal_buf *);
-
 static inline bool __wb_key_ref_cmp(const struct wb_key_ref *l, const struct wb_key_ref *r)
 {
 	return (cmp_int(l->hi, r->hi) ?:
@@ -481,13 +479,38 @@ err:
 	return ret;
 }
 
-static int fetch_wb_keys_from_journal(struct bch_fs *c, u64 seq)
+static int bch2_journal_keys_to_write_buffer(struct bch_fs *c, struct journal_buf *buf)
+{
+	struct journal_keys_to_wb dst;
+	int ret = 0;
+
+	bch2_journal_keys_to_write_buffer_start(c, &dst, le64_to_cpu(buf->data->seq));
+
+	for_each_jset_entry_type(entry, buf->data, BCH_JSET_ENTRY_write_buffer_keys) {
+		jset_entry_for_each_key(entry, k) {
+			ret = bch2_journal_key_to_wb(c, &dst, entry->btree_id, k);
+			if (ret)
+				goto out;
+		}
+
+		entry->type = BCH_JSET_ENTRY_btree_keys;
+	}
+
+	spin_lock(&c->journal.lock);
+	buf->need_flush_to_write_buffer = false;
+	spin_unlock(&c->journal.lock);
+out:
+	ret = bch2_journal_keys_to_write_buffer_end(c, &dst) ?: ret;
+	return ret;
+}
+
+static int fetch_wb_keys_from_journal(struct bch_fs *c, u64 max_seq)
 {
 	struct journal *j = &c->journal;
 	struct journal_buf *buf;
 	int ret = 0;
 
-	while (!ret && (buf = bch2_next_write_buffer_flush_journal_buf(j, seq))) {
+	while (!ret && (buf = bch2_next_write_buffer_flush_journal_buf(j, max_seq))) {
 		ret = bch2_journal_keys_to_write_buffer(c, buf);
 		mutex_unlock(&j->buf_lock);
 	}
@@ -495,7 +518,7 @@ static int fetch_wb_keys_from_journal(struct bch_fs *c, u64 seq)
 	return ret;
 }
 
-static int btree_write_buffer_flush_seq(struct btree_trans *trans, u64 seq,
+static int btree_write_buffer_flush_seq(struct btree_trans *trans, u64 max_seq,
 					bool *did_work)
 {
 	struct bch_fs *c = trans->c;
@@ -505,7 +528,7 @@ static int btree_write_buffer_flush_seq(struct btree_trans *trans, u64 seq,
 	do {
 		bch2_trans_unlock(trans);
 
-		fetch_from_journal_err = fetch_wb_keys_from_journal(c, seq);
+		fetch_from_journal_err = fetch_wb_keys_from_journal(c, max_seq);
 
 		*did_work |= wb->inc.keys.nr || wb->flushing.keys.nr;
 
@@ -518,8 +541,8 @@ static int btree_write_buffer_flush_seq(struct btree_trans *trans, u64 seq,
 		mutex_unlock(&wb->flushing.lock);
 	} while (!ret &&
 		 (fetch_from_journal_err ||
-		  (wb->inc.pin.seq && wb->inc.pin.seq <= seq) ||
-		  (wb->flushing.pin.seq && wb->flushing.pin.seq <= seq)));
+		  (wb->inc.pin.seq && wb->inc.pin.seq <= max_seq) ||
+		  (wb->flushing.pin.seq && wb->flushing.pin.seq <= max_seq)));
 
 	return ret;
 }
@@ -771,31 +794,6 @@ int bch2_journal_keys_to_write_buffer_end(struct bch_fs *c, struct journal_keys_
 	return ret;
 }
 
-static int bch2_journal_keys_to_write_buffer(struct bch_fs *c, struct journal_buf *buf)
-{
-	struct journal_keys_to_wb dst;
-	int ret = 0;
-
-	bch2_journal_keys_to_write_buffer_start(c, &dst, le64_to_cpu(buf->data->seq));
-
-	for_each_jset_entry_type(entry, buf->data, BCH_JSET_ENTRY_write_buffer_keys) {
-		jset_entry_for_each_key(entry, k) {
-			ret = bch2_journal_key_to_wb(c, &dst, entry->btree_id, k);
-			if (ret)
-				goto out;
-		}
-
-		entry->type = BCH_JSET_ENTRY_btree_keys;
-	}
-
-	spin_lock(&c->journal.lock);
-	buf->need_flush_to_write_buffer = false;
-	spin_unlock(&c->journal.lock);
-out:
-	ret = bch2_journal_keys_to_write_buffer_end(c, &dst) ?: ret;
-	return ret;
-}
-
 static int wb_keys_resize(struct btree_write_buffer_keys *wb, size_t new_size)
 {
 	if (wb->keys.size >= new_size)
-- 
2.51.0


From e1cb4f56dc4cfcf1cb1b90479cb35c5a304ff527 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sat, 23 Nov 2024 16:47:10 -0500
Subject: [PATCH 06/16] bcachefs: Bias reads more in favor of faster device

Per reports of performance issues on mixed multi device filesystems
where we're issuing too much IO to the spinning rust - tweak this
algorithm.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/extents.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c
index 98bb680b3860..83aeceb68847 100644
--- a/fs/bcachefs/extents.c
+++ b/fs/bcachefs/extents.c
@@ -89,6 +89,14 @@ static inline bool ptr_better(struct bch_fs *c,
 		u64 l1 = dev_latency(c, p1.ptr.dev);
 		u64 l2 = dev_latency(c, p2.ptr.dev);
 
+		/*
+		 * Square the latencies, to bias more in favor of the faster
+		 * device - we never want to stop issuing reads to the slower
+		 * device altogether, so that we can update our latency numbers:
+		 */
+		l1 *= l1;
+		l2 *= l2;
+
 		/* Pick at random, biased in favor of the faster device: */
 
 		return bch2_rand_range(l1 + l2) > l1;
-- 
2.51.0


From bb61afebca3b5608842e2ac59bbe4eedf137050f Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Thu, 21 Nov 2024 20:09:45 -0500
Subject: [PATCH 07/16] bcachefs: discard fastpath now uses
 bch2_discard_one_bucket()

The discard bucket fastpath previously was using its own code for
discarding buckets and clearing them in the need_discard btree, which
didn't have any of the consistency checks of the main discard path.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/alloc_background.c | 75 +++++++++++++++++++---------------
 1 file changed, 41 insertions(+), 34 deletions(-)

diff --git a/fs/bcachefs/alloc_background.c b/fs/bcachefs/alloc_background.c
index ae9fdb5ad758..1e9f53db4bb8 100644
--- a/fs/bcachefs/alloc_background.c
+++ b/fs/bcachefs/alloc_background.c
@@ -1725,7 +1725,8 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 				   struct bch_dev *ca,
 				   struct btree_iter *need_discard_iter,
 				   struct bpos *discard_pos_done,
-				   struct discard_buckets_state *s)
+				   struct discard_buckets_state *s,
+				   bool fastpath)
 {
 	struct bch_fs *c = trans->c;
 	struct bpos pos = need_discard_iter->pos;
@@ -1782,10 +1783,12 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 		goto out;
 	}
 
-	if (discard_in_flight_add(ca, iter.pos.offset, true))
-		goto out;
+	if (!fastpath) {
+		if (discard_in_flight_add(ca, iter.pos.offset, true))
+			goto out;
 
-	discard_locked = true;
+		discard_locked = true;
+	}
 
 	if (!bkey_eq(*discard_pos_done, iter.pos) &&
 	    ca->mi.discard && !c->opts.nochanges) {
@@ -1799,6 +1802,7 @@ static int bch2_discard_one_bucket(struct btree_trans *trans,
 				     ca->mi.bucket_size,
 				     GFP_KERNEL);
 		*discard_pos_done = iter.pos;
+		s->discarded++;
 
 		ret = bch2_trans_relock_notrace(trans);
 		if (ret)
@@ -1819,12 +1823,12 @@ commit:
 		goto out;
 
 	count_event(c, bucket_discard);
-	s->discarded++;
 out:
 fsck_err:
 	if (discard_locked)
 		discard_in_flight_remove(ca, iter.pos.offset);
-	s->seen++;
+	if (!ret)
+		s->seen++;
 	bch2_trans_iter_exit(trans, &iter);
 	printbuf_exit(&buf);
 	return ret;
@@ -1848,7 +1852,7 @@ static void bch2_do_discards_work(struct work_struct *work)
 				   BTREE_ID_need_discard,
 				   POS(ca->dev_idx, 0),
 				   POS(ca->dev_idx, U64_MAX), 0, k,
-			bch2_discard_one_bucket(trans, ca, &iter, &discard_pos_done, &s)));
+			bch2_discard_one_bucket(trans, ca, &iter, &discard_pos_done, &s, false)));
 
 	trace_discard_buckets(c, s.seen, s.open, s.need_journal_commit, s.discarded,
 			      bch2_err_str(ret));
@@ -1881,27 +1885,31 @@ void bch2_do_discards(struct bch_fs *c)
 		bch2_dev_do_discards(ca);
 }
 
-static int bch2_clear_bucket_needs_discard(struct btree_trans *trans, struct bpos bucket)
+static int bch2_do_discards_fast_one(struct btree_trans *trans,
+				     struct bch_dev *ca,
+				     u64 bucket,
+				     struct bpos *discard_pos_done,
+				     struct discard_buckets_state *s)
 {
-	struct btree_iter iter;
-	bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc, bucket, BTREE_ITER_intent);
-	struct bkey_s_c k = bch2_btree_iter_peek_slot(&iter);
-	int ret = bkey_err(k);
-	if (ret)
-		goto err;
-
-	struct bkey_i_alloc_v4 *a = bch2_alloc_to_v4_mut(trans, k);
-	ret = PTR_ERR_OR_ZERO(a);
+	struct btree_iter need_discard_iter;
+	struct bkey_s_c discard_k = bch2_bkey_get_iter(trans, &need_discard_iter,
+					BTREE_ID_need_discard, POS(ca->dev_idx, bucket), 0);
+	int ret = bkey_err(discard_k);
 	if (ret)
-		goto err;
+		return ret;
 
-	BUG_ON(a->v.dirty_sectors);
-	SET_BCH_ALLOC_V4_NEED_DISCARD(&a->v, false);
-	alloc_data_type_set(&a->v, a->v.data_type);
+	if (log_fsck_err_on(discard_k.k->type != KEY_TYPE_set,
+			    trans, discarding_bucket_not_in_need_discard_btree,
+			    "attempting to discard bucket %u:%llu not in need_discard btree",
+			    ca->dev_idx, bucket)) {
+		/* log it in the superblock and continue: */
+		goto out;
+	}
 
-	ret = bch2_trans_update(trans, &iter, &a->k_i, 0);
-err:
-	bch2_trans_iter_exit(trans, &iter);
+	ret = bch2_discard_one_bucket(trans, ca, &need_discard_iter, discard_pos_done, s, true);
+out:
+fsck_err:
+	bch2_trans_iter_exit(trans, &need_discard_iter);
 	return ret;
 }
 
@@ -1909,6 +1917,10 @@ static void bch2_do_discards_fast_work(struct work_struct *work)
 {
 	struct bch_dev *ca = container_of(work, struct bch_dev, discard_fast_work);
 	struct bch_fs *c = ca->fs;
+	struct discard_buckets_state s = {};
+	struct bpos discard_pos_done = POS_MAX;
+	struct btree_trans *trans = bch2_trans_get(c);
+	int ret = 0;
 
 	while (1) {
 		bool got_bucket = false;
@@ -1929,16 +1941,8 @@ static void bch2_do_discards_fast_work(struct work_struct *work)
 		if (!got_bucket)
 			break;
 
-		if (ca->mi.discard && !c->opts.nochanges)
-			blkdev_issue_discard(ca->disk_sb.bdev,
-					     bucket_to_sector(ca, bucket),
-					     ca->mi.bucket_size,
-					     GFP_KERNEL);
-
-		int ret = bch2_trans_commit_do(c, NULL, NULL,
-			BCH_WATERMARK_btree|
-			BCH_TRANS_COMMIT_no_enospc,
-			bch2_clear_bucket_needs_discard(trans, POS(ca->dev_idx, bucket)));
+		ret = lockrestart_do(trans,
+			bch2_do_discards_fast_one(trans, ca, bucket, &discard_pos_done, &s));
 		bch_err_fn(c, ret);
 
 		discard_in_flight_remove(ca, bucket);
@@ -1947,6 +1951,9 @@ 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));
+
+	bch2_trans_put(trans);
 	percpu_ref_put(&ca->io_ref);
 	bch2_write_ref_put(c, BCH_WRITE_REF_discard_fast);
 }
-- 
2.51.0


From 0eaac0b44fa93a5bf8bd5d79c557ca5e04d2dc16 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Tue, 23 Apr 2024 02:18:18 -0400
Subject: [PATCH 08/16] bcachefs: btree_write_buffer_flush_seq() no longer
 closes journal

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/btree_write_buffer.c | 19 ++++++++++++++-----
 fs/bcachefs/journal.c            | 27 ++++++++++++++++++++-------
 fs/bcachefs/journal.h            |  2 +-
 3 files changed, 35 insertions(+), 13 deletions(-)

diff --git a/fs/bcachefs/btree_write_buffer.c b/fs/bcachefs/btree_write_buffer.c
index 1bd26221f156..49ce2d1e5c02 100644
--- a/fs/bcachefs/btree_write_buffer.c
+++ b/fs/bcachefs/btree_write_buffer.c
@@ -495,10 +495,6 @@ static int bch2_journal_keys_to_write_buffer(struct bch_fs *c, struct journal_bu
 
 		entry->type = BCH_JSET_ENTRY_btree_keys;
 	}
-
-	spin_lock(&c->journal.lock);
-	buf->need_flush_to_write_buffer = false;
-	spin_unlock(&c->journal.lock);
 out:
 	ret = bch2_journal_keys_to_write_buffer_end(c, &dst) ?: ret;
 	return ret;
@@ -508,11 +504,24 @@ static int fetch_wb_keys_from_journal(struct bch_fs *c, u64 max_seq)
 {
 	struct journal *j = &c->journal;
 	struct journal_buf *buf;
+	bool blocked;
 	int ret = 0;
 
-	while (!ret && (buf = bch2_next_write_buffer_flush_journal_buf(j, max_seq))) {
+	while (!ret && (buf = bch2_next_write_buffer_flush_journal_buf(j, max_seq, &blocked))) {
 		ret = bch2_journal_keys_to_write_buffer(c, buf);
+
+		if (!blocked && !ret) {
+			spin_lock(&j->lock);
+			buf->need_flush_to_write_buffer = false;
+			spin_unlock(&j->lock);
+		}
+
 		mutex_unlock(&j->buf_lock);
+
+		if (blocked) {
+			bch2_journal_unblock(j);
+			break;
+		}
 	}
 
 	return ret;
diff --git a/fs/bcachefs/journal.c b/fs/bcachefs/journal.c
index bfbb1ac60c3d..699db0d0749a 100644
--- a/fs/bcachefs/journal.c
+++ b/fs/bcachefs/journal.c
@@ -908,6 +908,8 @@ static void __bch2_journal_block(struct journal *j)
 			new.v = old.v;
 			new.cur_entry_offset = JOURNAL_ENTRY_BLOCKED_VAL;
 		} while (!atomic64_try_cmpxchg(&j->reservations.counter, &old.v, new.v));
+
+		journal_cur_buf(j)->data->u64s = cpu_to_le32(old.cur_entry_offset);
 	}
 }
 
@@ -920,7 +922,8 @@ void bch2_journal_block(struct journal *j)
 	journal_quiesce(j);
 }
 
-static struct journal_buf *__bch2_next_write_buffer_flush_journal_buf(struct journal *j, u64 max_seq)
+static struct journal_buf *__bch2_next_write_buffer_flush_journal_buf(struct journal *j,
+						u64 max_seq, bool *blocked)
 {
 	struct journal_buf *ret = NULL;
 
@@ -937,13 +940,17 @@ static struct journal_buf *__bch2_next_write_buffer_flush_journal_buf(struct jou
 		struct journal_buf *buf = j->buf + idx;
 
 		if (buf->need_flush_to_write_buffer) {
-			if (seq == journal_cur_seq(j))
-				__journal_entry_close(j, JOURNAL_ENTRY_CLOSED_VAL, true);
-
 			union journal_res_state s;
 			s.v = atomic64_read_acquire(&j->reservations.counter);
 
-			ret = journal_state_count(s, idx)
+			unsigned open = seq == journal_cur_seq(j) && __journal_entry_is_open(s);
+
+			if (open && !*blocked) {
+				__bch2_journal_block(j);
+				*blocked = true;
+			}
+
+			ret = journal_state_count(s, idx) > open
 				? ERR_PTR(-EAGAIN)
 				: buf;
 			break;
@@ -956,11 +963,17 @@ static struct journal_buf *__bch2_next_write_buffer_flush_journal_buf(struct jou
 	return ret;
 }
 
-struct journal_buf *bch2_next_write_buffer_flush_journal_buf(struct journal *j, u64 max_seq)
+struct journal_buf *bch2_next_write_buffer_flush_journal_buf(struct journal *j,
+							     u64 max_seq, bool *blocked)
 {
 	struct journal_buf *ret;
+	*blocked = false;
+
+	wait_event(j->wait, (ret = __bch2_next_write_buffer_flush_journal_buf(j,
+						max_seq, blocked)) != ERR_PTR(-EAGAIN));
+	if (IS_ERR_OR_NULL(ret) && *blocked)
+		bch2_journal_unblock(j);
 
-	wait_event(j->wait, (ret = __bch2_next_write_buffer_flush_journal_buf(j, max_seq)) != ERR_PTR(-EAGAIN));
 	return ret;
 }
 
diff --git a/fs/bcachefs/journal.h b/fs/bcachefs/journal.h
index 6d3c839bbbef..71a50846967f 100644
--- a/fs/bcachefs/journal.h
+++ b/fs/bcachefs/journal.h
@@ -425,7 +425,7 @@ static inline void bch2_journal_set_replay_done(struct journal *j)
 
 void bch2_journal_unblock(struct journal *);
 void bch2_journal_block(struct journal *);
-struct journal_buf *bch2_next_write_buffer_flush_journal_buf(struct journal *j, u64 max_seq);
+struct journal_buf *bch2_next_write_buffer_flush_journal_buf(struct journal *, u64, bool *);
 
 void __bch2_journal_debug_to_text(struct printbuf *, struct journal *);
 void bch2_journal_debug_to_text(struct printbuf *, struct journal *);
-- 
2.51.0


From 375d21b76d9a20ff4bddf81aadadd6e57f38dee8 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sat, 23 Nov 2024 22:12:58 -0500
Subject: [PATCH 09/16] bcachefs: BCH_ERR_btree_node_read_error_cached

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/btree_cache.c | 6 +++---
 fs/bcachefs/errcode.h     | 1 +
 2 files changed, 4 insertions(+), 3 deletions(-)

diff --git a/fs/bcachefs/btree_cache.c b/fs/bcachefs/btree_cache.c
index a0a406b0c7bc..36dfa6a48aa6 100644
--- a/fs/bcachefs/btree_cache.c
+++ b/fs/bcachefs/btree_cache.c
@@ -1131,7 +1131,7 @@ retry:
 
 	if (unlikely(btree_node_read_error(b))) {
 		six_unlock_type(&b->c.lock, lock_type);
-		return ERR_PTR(-BCH_ERR_btree_node_read_error);
+		return ERR_PTR(-BCH_ERR_btree_node_read_err_cached);
 	}
 
 	EBUG_ON(b->c.btree_id != path->btree_id);
@@ -1221,7 +1221,7 @@ struct btree *bch2_btree_node_get(struct btree_trans *trans, struct btree_path *
 
 	if (unlikely(btree_node_read_error(b))) {
 		six_unlock_type(&b->c.lock, lock_type);
-		return ERR_PTR(-BCH_ERR_btree_node_read_error);
+		return ERR_PTR(-BCH_ERR_btree_node_read_err_cached);
 	}
 
 	EBUG_ON(b->c.btree_id != path->btree_id);
@@ -1303,7 +1303,7 @@ lock_node:
 
 	if (unlikely(btree_node_read_error(b))) {
 		six_unlock_read(&b->c.lock);
-		b = ERR_PTR(-BCH_ERR_btree_node_read_error);
+		b = ERR_PTR(-BCH_ERR_btree_node_read_err_cached);
 		goto out;
 	}
 
diff --git a/fs/bcachefs/errcode.h b/fs/bcachefs/errcode.h
index 2dda7f962e5b..131b9bef21a0 100644
--- a/fs/bcachefs/errcode.h
+++ b/fs/bcachefs/errcode.h
@@ -242,6 +242,7 @@
 	x(BCH_ERR_invalid,		invalid_bkey)				\
 	x(BCH_ERR_operation_blocked,    nocow_lock_blocked)			\
 	x(EIO,				btree_node_read_err)			\
+	x(BCH_ERR_btree_node_read_err,	btree_node_read_err_cached)		\
 	x(EIO,				sb_not_downgraded)			\
 	x(EIO,				btree_node_write_all_failed)		\
 	x(EIO,				btree_node_read_error)			\
-- 
2.51.0


From 525be09e63cc9ff4bee0b0f480551a4dc2e7b276 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sun, 24 Nov 2024 20:15:30 -0500
Subject: [PATCH 10/16] bcachefs: Use separate rhltable for
 bch2_inode_or_descendents_is_open()

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/bcachefs.h |  1 +
 fs/bcachefs/fs.c       | 39 ++++++++++++++++++++++++++++++---------
 fs/bcachefs/fs.h       |  1 +
 3 files changed, 32 insertions(+), 9 deletions(-)

diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h
index 11f9ed42a9da..f1d8c821d27a 100644
--- a/fs/bcachefs/bcachefs.h
+++ b/fs/bcachefs/bcachefs.h
@@ -1020,6 +1020,7 @@ struct bch_fs {
 	struct list_head	vfs_inodes_list;
 	struct mutex		vfs_inodes_lock;
 	struct rhashtable	vfs_inodes_table;
+	struct rhltable		vfs_inodes_by_inum_table;
 
 	/* VFS IO PATH - fs-io.c */
 	struct bio_set		writepage_bioset;
diff --git a/fs/bcachefs/fs.c b/fs/bcachefs/fs.c
index 50d323fca001..c6e7df7c67fa 100644
--- a/fs/bcachefs/fs.c
+++ b/fs/bcachefs/fs.c
@@ -39,6 +39,7 @@
 #include <linux/posix_acl.h>
 #include <linux/random.h>
 #include <linux/seq_file.h>
+#include <linux/siphash.h>
 #include <linux/statfs.h>
 #include <linux/string.h>
 #include <linux/xattr.h>
@@ -176,8 +177,9 @@ static bool subvol_inum_eq(subvol_inum a, subvol_inum b)
 static u32 bch2_vfs_inode_hash_fn(const void *data, u32 len, u32 seed)
 {
 	const subvol_inum *inum = data;
+	siphash_key_t k = { .key[0] = seed };
 
-	return jhash(&inum->inum, sizeof(inum->inum), seed);
+	return siphash_2u64(inum->subvol, inum->inum, &k);
 }
 
 static u32 bch2_vfs_inode_obj_hash_fn(const void *data, u32 len, u32 seed)
@@ -206,11 +208,18 @@ static const struct rhashtable_params bch2_vfs_inodes_params = {
 	.automatic_shrinking	= true,
 };
 
+static const struct rhashtable_params bch2_vfs_inodes_by_inum_params = {
+	.head_offset		= offsetof(struct bch_inode_info, by_inum_hash),
+	.key_offset		= offsetof(struct bch_inode_info, ei_inum.inum),
+	.key_len		= sizeof(u64),
+	.automatic_shrinking	= true,
+};
+
 int bch2_inode_or_descendents_is_open(struct btree_trans *trans, struct bpos p)
 {
 	struct bch_fs *c = trans->c;
-	struct rhashtable *ht = &c->vfs_inodes_table;
-	subvol_inum inum = (subvol_inum) { .inum = p.offset };
+	struct rhltable *ht = &c->vfs_inodes_by_inum_table;
+	u64 inum = p.offset;
 	DARRAY(u32) subvols;
 	int ret = 0;
 
@@ -235,15 +244,15 @@ restart_from_top:
 	struct rhash_lock_head __rcu *const *bkt;
 	struct rhash_head *he;
 	unsigned int hash;
-	struct bucket_table *tbl = rht_dereference_rcu(ht->tbl, ht);
+	struct bucket_table *tbl = rht_dereference_rcu(ht->ht.tbl, &ht->ht);
 restart:
-	hash = rht_key_hashfn(ht, tbl, &inum, bch2_vfs_inodes_params);
+	hash = rht_key_hashfn(&ht->ht, tbl, &inum, bch2_vfs_inodes_by_inum_params);
 	bkt = rht_bucket(tbl, hash);
 	do {
 		struct bch_inode_info *inode;
 
 		rht_for_each_entry_rcu_from(inode, he, rht_ptr_rcu(bkt), tbl, hash, hash) {
-			if (inode->ei_inum.inum == inum.inum) {
+			if (inode->ei_inum.inum == inum) {
 				ret = darray_push_gfp(&subvols, inode->ei_inum.subvol,
 						      GFP_NOWAIT|__GFP_NOWARN);
 				if (ret) {
@@ -264,7 +273,7 @@ restart:
 	/* Ensure we see any new tables. */
 	smp_rmb();
 
-	tbl = rht_dereference_rcu(tbl->future_tbl, ht);
+	tbl = rht_dereference_rcu(tbl->future_tbl, &ht->ht);
 	if (unlikely(tbl))
 		goto restart;
 	rcu_read_unlock();
@@ -343,7 +352,11 @@ static void bch2_inode_hash_remove(struct bch_fs *c, struct bch_inode_info *inod
 	spin_unlock(&inode->v.i_lock);
 
 	if (remove) {
-		int ret = rhashtable_remove_fast(&c->vfs_inodes_table,
+		int ret = rhltable_remove(&c->vfs_inodes_by_inum_table,
+					&inode->by_inum_hash, bch2_vfs_inodes_by_inum_params);
+		BUG_ON(ret);
+
+		ret = rhashtable_remove_fast(&c->vfs_inodes_table,
 					&inode->hash, bch2_vfs_inodes_params);
 		BUG_ON(ret);
 		inode->v.i_hash.pprev = NULL;
@@ -388,6 +401,11 @@ retry:
 		discard_new_inode(&inode->v);
 		return old;
 	} else {
+		int ret = rhltable_insert(&c->vfs_inodes_by_inum_table,
+					  &inode->by_inum_hash,
+					  bch2_vfs_inodes_by_inum_params);
+		BUG_ON(ret);
+
 		inode_fake_hash(&inode->v);
 
 		inode_sb_list_add(&inode->v);
@@ -2359,13 +2377,16 @@ static int bch2_init_fs_context(struct fs_context *fc)
 
 void bch2_fs_vfs_exit(struct bch_fs *c)
 {
+	if (c->vfs_inodes_by_inum_table.ht.tbl)
+		rhltable_destroy(&c->vfs_inodes_by_inum_table);
 	if (c->vfs_inodes_table.tbl)
 		rhashtable_destroy(&c->vfs_inodes_table);
 }
 
 int bch2_fs_vfs_init(struct bch_fs *c)
 {
-	return rhashtable_init(&c->vfs_inodes_table, &bch2_vfs_inodes_params);
+	return rhashtable_init(&c->vfs_inodes_table, &bch2_vfs_inodes_params) ?:
+		rhltable_init(&c->vfs_inodes_by_inum_table, &bch2_vfs_inodes_by_inum_params);
 }
 
 static struct file_system_type bcache_fs_type = {
diff --git a/fs/bcachefs/fs.h b/fs/bcachefs/fs.h
index 59f9f7ae728d..dd2198541455 100644
--- a/fs/bcachefs/fs.h
+++ b/fs/bcachefs/fs.h
@@ -14,6 +14,7 @@
 struct bch_inode_info {
 	struct inode		v;
 	struct rhash_head	hash;
+	struct rhlist_head	by_inum_hash;
 	subvol_inum		ei_inum;
 
 	struct list_head	ei_vfs_inode_list;
-- 
2.51.0


From 6534a404d4924820bd1c06fb4412dd4444234f79 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sun, 24 Nov 2024 21:49:08 -0500
Subject: [PATCH 11/16] bcachefs: errcode cleanup: journal errors

Instead of throwing standard error codes, we should be throwing
dedicated private error codes, this greatly improves debugability.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/errcode.h | 2 ++
 fs/bcachefs/journal.c | 4 ++--
 fs/bcachefs/journal.h | 2 +-
 3 files changed, 5 insertions(+), 3 deletions(-)

diff --git a/fs/bcachefs/errcode.h b/fs/bcachefs/errcode.h
index 131b9bef21a0..c989ce4f715f 100644
--- a/fs/bcachefs/errcode.h
+++ b/fs/bcachefs/errcode.h
@@ -241,6 +241,8 @@
 	x(BCH_ERR_invalid_sb,		invalid_sb_downgrade)			\
 	x(BCH_ERR_invalid,		invalid_bkey)				\
 	x(BCH_ERR_operation_blocked,    nocow_lock_blocked)			\
+	x(EIO,				journal_shutdown)			\
+	x(EIO,				journal_flush_err)			\
 	x(EIO,				btree_node_read_err)			\
 	x(BCH_ERR_btree_node_read_err,	btree_node_read_err_cached)		\
 	x(EIO,				sb_not_downgraded)			\
diff --git a/fs/bcachefs/journal.c b/fs/bcachefs/journal.c
index 699db0d0749a..bbdd0b17ae69 100644
--- a/fs/bcachefs/journal.c
+++ b/fs/bcachefs/journal.c
@@ -673,7 +673,7 @@ out:
  * @seq:	seq to flush
  * @parent:	closure object to wait with
  * Returns:	1 if @seq has already been flushed, 0 if @seq is being flushed,
- *		-EIO if @seq will never be flushed
+ *		-BCH_ERR_journal_flush_err if @seq will never be flushed
  *
  * Like bch2_journal_wait_on_seq, except that it triggers a write immediately if
  * necessary
@@ -696,7 +696,7 @@ int bch2_journal_flush_seq_async(struct journal *j, u64 seq,
 
 	/* Recheck under lock: */
 	if (j->err_seq && seq >= j->err_seq) {
-		ret = -EIO;
+		ret = -BCH_ERR_journal_flush_err;
 		goto out;
 	}
 
diff --git a/fs/bcachefs/journal.h b/fs/bcachefs/journal.h
index 71a50846967f..a6a2e888c59b 100644
--- a/fs/bcachefs/journal.h
+++ b/fs/bcachefs/journal.h
@@ -412,7 +412,7 @@ void bch2_journal_halt(struct journal *);
 static inline int bch2_journal_error(struct journal *j)
 {
 	return j->reservations.cur_entry_offset == JOURNAL_ENTRY_ERROR_VAL
-		? -EIO : 0;
+		? -BCH_ERR_journal_shutdown : 0;
 }
 
 struct bch_dev;
-- 
2.51.0


From a7ecd5f2ccabb97174f8264b965c9597ea854e8f Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sun, 24 Nov 2024 22:23:41 -0500
Subject: [PATCH 12/16] bcachefs: disk_accounting: bch2_dev_rcu ->
 bch2_dev_rcu_noerror

Accounting keys that reference invalid devices are corrected by fsck,
they shouldn't cause an emergency shutdown.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/disk_accounting.c | 4 ++--
 fs/bcachefs/disk_accounting.h | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/fs/bcachefs/disk_accounting.c b/fs/bcachefs/disk_accounting.c
index 55a00018dc8b..fa821d278c45 100644
--- a/fs/bcachefs/disk_accounting.c
+++ b/fs/bcachefs/disk_accounting.c
@@ -805,7 +805,7 @@ int bch2_accounting_read(struct bch_fs *c)
 			break;
 		case BCH_DISK_ACCOUNTING_dev_data_type:
 			rcu_read_lock();
-			struct bch_dev *ca = bch2_dev_rcu(c, k.dev_data_type.dev);
+			struct bch_dev *ca = bch2_dev_rcu_noerror(c, k.dev_data_type.dev);
 			if (ca) {
 				struct bch_dev_usage_type __percpu *d = &ca->usage->d[k.dev_data_type.data_type];
 				percpu_u64_set(&d->buckets,	v[0]);
@@ -911,7 +911,7 @@ void bch2_verify_accounting_clean(struct bch_fs *c)
 				break;
 			case BCH_DISK_ACCOUNTING_dev_data_type: {
 				rcu_read_lock();
-				struct bch_dev *ca = bch2_dev_rcu(c, acc_k.dev_data_type.dev);
+				struct bch_dev *ca = bch2_dev_rcu_noerror(c, acc_k.dev_data_type.dev);
 				if (!ca) {
 					rcu_read_unlock();
 					continue;
diff --git a/fs/bcachefs/disk_accounting.h b/fs/bcachefs/disk_accounting.h
index 6639535dc91c..8b2b2f83e6a4 100644
--- a/fs/bcachefs/disk_accounting.h
+++ b/fs/bcachefs/disk_accounting.h
@@ -142,7 +142,7 @@ static inline int bch2_accounting_mem_mod_locked(struct btree_trans *trans,
 			break;
 		case BCH_DISK_ACCOUNTING_dev_data_type:
 			rcu_read_lock();
-			struct bch_dev *ca = bch2_dev_rcu(c, acc_k.dev_data_type.dev);
+			struct bch_dev *ca = bch2_dev_rcu_noerror(c, acc_k.dev_data_type.dev);
 			if (ca) {
 				this_cpu_add(ca->usage->d[acc_k.dev_data_type.data_type].buckets, a.v->d[0]);
 				this_cpu_add(ca->usage->d[acc_k.dev_data_type.data_type].sectors, a.v->d[1]);
-- 
2.51.0


From b71d89bd7b1fbe0a4569d072a0069110f60f9ec9 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sun, 24 Nov 2024 22:28:41 -0500
Subject: [PATCH 13/16] bcachefs: Fix accounting_read when we rewind

If we rewind recovery to run topology repair, that causes
accounting_read to run twice.

This fixes accounting being double counted.

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/disk_accounting.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/fs/bcachefs/disk_accounting.c b/fs/bcachefs/disk_accounting.c
index fa821d278c45..bb5dbbf71d04 100644
--- a/fs/bcachefs/disk_accounting.c
+++ b/fs/bcachefs/disk_accounting.c
@@ -700,6 +700,21 @@ int bch2_accounting_read(struct bch_fs *c)
 	struct btree_trans *trans = bch2_trans_get(c);
 	struct printbuf buf = PRINTBUF;
 
+	/*
+	 * We might run more than once if we rewind to start topology repair or
+	 * btree node scan - and those might cause us to get different results,
+	 * so we can't just skip if we've already run.
+	 *
+	 * Instead, zero out any accounting we have:
+	 */
+	percpu_down_write(&c->mark_lock);
+	darray_for_each(acc->k, e)
+		percpu_memset(e->v[0], 0, sizeof(u64) * e->nr_counters);
+	for_each_member_device(c, ca)
+		percpu_memset(ca->usage, 0, sizeof(*ca->usage));
+	percpu_memset(c->usage, 0, sizeof(*c->usage));
+	percpu_up_write(&c->mark_lock);
+
 	int ret = for_each_btree_key(trans, iter,
 				BTREE_ID_accounting, POS_MIN,
 				BTREE_ITER_prefetch|BTREE_ITER_all_snapshots, k, ({
-- 
2.51.0


From 427db7ffe9a91bcad00a7bac4edcd74d75007a11 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sun, 24 Nov 2024 22:45:25 -0500
Subject: [PATCH 14/16] bcachefs: backpointer_to_missing_ptr is now autofix

Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/sb-errors_format.h | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/bcachefs/sb-errors_format.h b/fs/bcachefs/sb-errors_format.h
index 9e3425f533bc..d45d0789f1b1 100644
--- a/fs/bcachefs/sb-errors_format.h
+++ b/fs/bcachefs/sb-errors_format.h
@@ -141,7 +141,7 @@ enum bch_fsck_flags {
 	x(backpointer_dev_bad,					297,	0)		\
 	x(backpointer_to_missing_device,			126,	0)		\
 	x(backpointer_to_missing_alloc,				127,	0)		\
-	x(backpointer_to_missing_ptr,				128,	0)		\
+	x(backpointer_to_missing_ptr,				128,	FSCK_AUTOFIX)	\
 	x(lru_entry_at_time_0,					129,	FSCK_AUTOFIX)	\
 	x(lru_entry_to_invalid_bucket,				130,	FSCK_AUTOFIX)	\
 	x(lru_entry_bad,					131,	FSCK_AUTOFIX)	\
-- 
2.51.0


From abf23afa36eb425fb75d47c26fc665dbab2a9ae1 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sun, 24 Nov 2024 22:57:01 -0500
Subject: [PATCH 15/16] bcachefs: Fix btree node scan when unknown btree IDs
 are present

btree_root entries for unknown btree IDs are created during recovery,
before reading those btree roots.

But btree_node_scan may find btree nodes with unknown btree IDs when we
haven't seen roots for those btrees.

Reported-by: syzbot+1f202d4da221ec6ebf8e@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/btree_cache.c | 11 ++++++++---
 fs/bcachefs/btree_cache.h |  9 +++++++--
 2 files changed, 15 insertions(+), 5 deletions(-)

diff --git a/fs/bcachefs/btree_cache.c b/fs/bcachefs/btree_cache.c
index 36dfa6a48aa6..1f06e24e53fc 100644
--- a/fs/bcachefs/btree_cache.c
+++ b/fs/bcachefs/btree_cache.c
@@ -1406,9 +1406,14 @@ void bch2_btree_id_level_to_text(struct printbuf *out, enum btree_id btree, unsi
 void bch2_btree_pos_to_text(struct printbuf *out, struct bch_fs *c, const struct btree *b)
 {
 	bch2_btree_id_to_text(out, b->c.btree_id);
-	prt_printf(out, " level %u/%u\n  ",
-		   b->c.level,
-		   bch2_btree_id_root(c, b->c.btree_id)->level);
+	prt_printf(out, " level %u/", b->c.level);
+	struct btree_root *r = bch2_btree_id_root(c, b->c.btree_id);
+	if (r)
+		prt_printf(out, "%u", r->level);
+	else
+		prt_printf(out, "(unknown)");
+	prt_printf(out, "\n  ");
+
 	bch2_bkey_val_to_text(out, c, bkey_i_to_s_c(&b->key));
 }
 
diff --git a/fs/bcachefs/btree_cache.h b/fs/bcachefs/btree_cache.h
index 6cfacacb6769..dcc34fe4996d 100644
--- a/fs/bcachefs/btree_cache.h
+++ b/fs/bcachefs/btree_cache.h
@@ -128,14 +128,19 @@ static inline struct btree_root *bch2_btree_id_root(struct bch_fs *c, unsigned i
 	} else {
 		unsigned idx = id - BTREE_ID_NR;
 
-		EBUG_ON(idx >= c->btree_roots_extra.nr);
+		/* This can happen when we're called from btree_node_scan */
+		if (idx >= c->btree_roots_extra.nr)
+			return NULL;
+
 		return &c->btree_roots_extra.data[idx];
 	}
 }
 
 static inline struct btree *btree_node_root(struct bch_fs *c, struct btree *b)
 {
-	return bch2_btree_id_root(c, b->c.btree_id)->b;
+	struct btree_root *r = bch2_btree_id_root(c, b->c.btree_id);
+
+	return r ? r->b : NULL;
 }
 
 const char *bch2_btree_id_str(enum btree_id);	/* avoid */
-- 
2.51.0


From 828552ca74a45877dbf139b34c47d0f600a1e852 Mon Sep 17 00:00:00 2001
From: Kent Overstreet <kent.overstreet@linux.dev>
Date: Sun, 24 Nov 2024 23:28:21 -0500
Subject: [PATCH 16/16] bcachefs: Kill bch2_bucket_alloc_new_fs()

The early-early allocation path, bch2_bucket_alloc_new_fs(), is no
longer needed - and inconsistencies around new_fs_bucket_idx have been a
frequent source of bugs.

Reported-by: syzbot+592425844580a6598410@syzkaller.appspotmail.com
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
---
 fs/bcachefs/alloc_foreground.c | 40 ++++++++++++++--------------------
 fs/bcachefs/alloc_foreground.h |  2 --
 fs/bcachefs/bcachefs.h         |  1 -
 fs/bcachefs/buckets.c          | 25 +++++++++++++++++++++
 fs/bcachefs/buckets.h          | 21 +-----------------
 fs/bcachefs/journal.c          | 34 +++++++++++++----------------
 fs/bcachefs/journal_reclaim.c  |  3 +++
 fs/bcachefs/recovery.c         |  5 +----
 fs/bcachefs/super.c            | 12 +++++-----
 9 files changed, 66 insertions(+), 77 deletions(-)

diff --git a/fs/bcachefs/alloc_foreground.c b/fs/bcachefs/alloc_foreground.c
index 6d665b720f72..4d1ff7f1f302 100644
--- a/fs/bcachefs/alloc_foreground.c
+++ b/fs/bcachefs/alloc_foreground.c
@@ -156,6 +156,14 @@ static struct open_bucket *bch2_open_bucket_alloc(struct bch_fs *c)
 	return ob;
 }
 
+static inline bool is_superblock_bucket(struct bch_fs *c, struct bch_dev *ca, u64 b)
+{
+	if (c->curr_recovery_pass > BCH_RECOVERY_PASS_trans_mark_dev_sbs)
+		return false;
+
+	return bch2_is_superblock_bucket(ca, b);
+}
+
 static void open_bucket_free_unused(struct bch_fs *c, struct open_bucket *ob)
 {
 	BUG_ON(c->open_buckets_partial_nr >=
@@ -175,20 +183,6 @@ static void open_bucket_free_unused(struct bch_fs *c, struct open_bucket *ob)
 	closure_wake_up(&c->freelist_wait);
 }
 
-/* _only_ for allocating the journal on a new device: */
-long bch2_bucket_alloc_new_fs(struct bch_dev *ca)
-{
-	while (ca->new_fs_bucket_idx < ca->mi.nbuckets) {
-		u64 b = ca->new_fs_bucket_idx++;
-
-		if (!is_superblock_bucket(ca, b) &&
-		    (!ca->buckets_nouse || !test_bit(b, ca->buckets_nouse)))
-			return b;
-	}
-
-	return -1;
-}
-
 static inline unsigned open_buckets_reserved(enum bch_watermark watermark)
 {
 	switch (watermark) {
@@ -214,6 +208,9 @@ static struct open_bucket *__try_alloc_bucket(struct bch_fs *c, struct bch_dev *
 {
 	struct open_bucket *ob;
 
+	if (unlikely(is_superblock_bucket(c, ca, bucket)))
+		return NULL;
+
 	if (unlikely(ca->buckets_nouse && test_bit(bucket, ca->buckets_nouse))) {
 		s->skipped_nouse++;
 		return NULL;
@@ -295,9 +292,6 @@ static struct open_bucket *try_alloc_bucket(struct btree_trans *trans, struct bc
 
 /*
  * This path is for before the freespace btree is initialized:
- *
- * If ca->new_fs_bucket_idx is nonzero, we haven't yet marked superblock &
- * journal buckets - journal buckets will be < ca->new_fs_bucket_idx
  */
 static noinline struct open_bucket *
 bch2_bucket_alloc_early(struct btree_trans *trans,
@@ -309,7 +303,7 @@ bch2_bucket_alloc_early(struct btree_trans *trans,
 	struct btree_iter iter, citer;
 	struct bkey_s_c k, ck;
 	struct open_bucket *ob = NULL;
-	u64 first_bucket = max_t(u64, ca->mi.first_bucket, ca->new_fs_bucket_idx);
+	u64 first_bucket = ca->mi.first_bucket;
 	u64 *dev_alloc_cursor = &ca->alloc_cursor[s->btree_bitmap];
 	u64 alloc_start = max(first_bucket, *dev_alloc_cursor);
 	u64 alloc_cursor = alloc_start;
@@ -332,10 +326,6 @@ again:
 		if (bkey_ge(k.k->p, POS(ca->dev_idx, ca->mi.nbuckets)))
 			break;
 
-		if (ca->new_fs_bucket_idx &&
-		    is_superblock_bucket(ca, k.k->p.offset))
-			continue;
-
 		if (s->btree_bitmap != BTREE_BITMAP_ANY &&
 		    s->btree_bitmap != bch2_dev_btree_bitmap_marked_sectors(ca,
 				bucket_to_sector(ca, bucket), ca->mi.bucket_size)) {
@@ -406,8 +396,6 @@ static struct open_bucket *bch2_bucket_alloc_freelist(struct btree_trans *trans,
 	u64 alloc_start = max_t(u64, ca->mi.first_bucket, READ_ONCE(*dev_alloc_cursor));
 	u64 alloc_cursor = alloc_start;
 	int ret;
-
-	BUG_ON(ca->new_fs_bucket_idx);
 again:
 	for_each_btree_key_max_norestart(trans, iter, BTREE_ID_freespace,
 					 POS(ca->dev_idx, alloc_cursor),
@@ -551,6 +539,10 @@ again:
 		bch2_dev_do_invalidates(ca);
 
 	if (!avail) {
+		if (watermark > BCH_WATERMARK_normal &&
+		    c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_allocations)
+			goto alloc;
+
 		if (cl && !waiting) {
 			closure_wait(&c->freelist_wait, cl);
 			waiting = true;
diff --git a/fs/bcachefs/alloc_foreground.h b/fs/bcachefs/alloc_foreground.h
index 1a16fd5bd4f8..4f87745df97e 100644
--- a/fs/bcachefs/alloc_foreground.h
+++ b/fs/bcachefs/alloc_foreground.h
@@ -28,8 +28,6 @@ struct dev_alloc_list bch2_dev_alloc_list(struct bch_fs *,
 					  struct bch_devs_mask *);
 void bch2_dev_stripe_increment(struct bch_dev *, struct dev_stripe_state *);
 
-long bch2_bucket_alloc_new_fs(struct bch_dev *);
-
 static inline struct bch_dev *ob_dev(struct bch_fs *c, struct open_bucket *ob)
 {
 	return bch2_dev_have_ref(c, ob->dev);
diff --git a/fs/bcachefs/bcachefs.h b/fs/bcachefs/bcachefs.h
index f1d8c821d27a..a85b3bcc6383 100644
--- a/fs/bcachefs/bcachefs.h
+++ b/fs/bcachefs/bcachefs.h
@@ -560,7 +560,6 @@ struct bch_dev {
 	struct bch_dev_usage __percpu	*usage;
 
 	/* Allocator: */
-	u64			new_fs_bucket_idx;
 	u64			alloc_cursor[3];
 
 	unsigned		nr_open_buckets;
diff --git a/fs/bcachefs/buckets.c b/fs/bcachefs/buckets.c
index 1547141ba2a0..afd35c93fcfb 100644
--- a/fs/bcachefs/buckets.c
+++ b/fs/bcachefs/buckets.c
@@ -1161,6 +1161,31 @@ int bch2_trans_mark_dev_sbs(struct bch_fs *c)
 	return bch2_trans_mark_dev_sbs_flags(c, BTREE_TRIGGER_transactional);
 }
 
+bool bch2_is_superblock_bucket(struct bch_dev *ca, u64 b)
+{
+	struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
+	u64 b_offset	= bucket_to_sector(ca, b);
+	u64 b_end	= bucket_to_sector(ca, b + 1);
+	unsigned i;
+
+	if (!b)
+		return true;
+
+	for (i = 0; i < layout->nr_superblocks; i++) {
+		u64 offset = le64_to_cpu(layout->sb_offset[i]);
+		u64 end = offset + (1 << layout->sb_max_size_bits);
+
+		if (!(offset >= b_end || end <= b_offset))
+			return true;
+	}
+
+	for (i = 0; i < ca->journal.nr; i++)
+		if (b == ca->journal.buckets[i])
+			return true;
+
+	return false;
+}
+
 /* Disk reservations: */
 
 #define SECTORS_CACHE	1024
diff --git a/fs/bcachefs/buckets.h b/fs/bcachefs/buckets.h
index ccc78bfe2fd4..3bebc4c3044f 100644
--- a/fs/bcachefs/buckets.h
+++ b/fs/bcachefs/buckets.h
@@ -308,26 +308,7 @@ int bch2_trans_mark_dev_sbs_flags(struct bch_fs *,
 				    enum btree_iter_update_trigger_flags);
 int bch2_trans_mark_dev_sbs(struct bch_fs *);
 
-static inline bool is_superblock_bucket(struct bch_dev *ca, u64 b)
-{
-	struct bch_sb_layout *layout = &ca->disk_sb.sb->layout;
-	u64 b_offset	= bucket_to_sector(ca, b);
-	u64 b_end	= bucket_to_sector(ca, b + 1);
-	unsigned i;
-
-	if (!b)
-		return true;
-
-	for (i = 0; i < layout->nr_superblocks; i++) {
-		u64 offset = le64_to_cpu(layout->sb_offset[i]);
-		u64 end = offset + (1 << layout->sb_max_size_bits);
-
-		if (!(offset >= b_end || end <= b_offset))
-			return true;
-	}
-
-	return false;
-}
+bool bch2_is_superblock_bucket(struct bch_dev *, u64);
 
 static inline const char *bch2_data_type_str(enum bch_data_type type)
 {
diff --git a/fs/bcachefs/journal.c b/fs/bcachefs/journal.c
index bbdd0b17ae69..95cccda3b22c 100644
--- a/fs/bcachefs/journal.c
+++ b/fs/bcachefs/journal.c
@@ -1002,19 +1002,17 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 	}
 
 	for (nr_got = 0; nr_got < nr_want; nr_got++) {
-		if (new_fs) {
-			bu[nr_got] = bch2_bucket_alloc_new_fs(ca);
-			if (bu[nr_got] < 0) {
-				ret = -BCH_ERR_ENOSPC_bucket_alloc;
-				break;
-			}
-		} else {
-			ob[nr_got] = bch2_bucket_alloc(c, ca, BCH_WATERMARK_normal,
-						       BCH_DATA_journal, cl);
-			ret = PTR_ERR_OR_ZERO(ob[nr_got]);
-			if (ret)
-				break;
+		enum bch_watermark watermark = new_fs
+			? BCH_WATERMARK_btree
+			: BCH_WATERMARK_normal;
 
+		ob[nr_got] = bch2_bucket_alloc(c, ca, watermark,
+					       BCH_DATA_journal, cl);
+		ret = PTR_ERR_OR_ZERO(ob[nr_got]);
+		if (ret)
+			break;
+
+		if (!new_fs) {
 			ret = bch2_trans_run(c,
 				bch2_trans_mark_metadata_bucket(trans, ca,
 						ob[nr_got]->bucket, BCH_DATA_journal,
@@ -1024,9 +1022,9 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 				bch_err_msg(c, ret, "marking new journal buckets");
 				break;
 			}
-
-			bu[nr_got] = ob[nr_got]->bucket;
 		}
+
+		bu[nr_got] = ob[nr_got]->bucket;
 	}
 
 	if (!nr_got)
@@ -1066,8 +1064,7 @@ static int __bch2_set_nr_journal_buckets(struct bch_dev *ca, unsigned nr,
 	if (ret)
 		goto err_unblock;
 
-	if (!new_fs)
-		bch2_write_super(c);
+	bch2_write_super(c);
 
 	/* Commit: */
 	if (c)
@@ -1101,9 +1098,8 @@ err_unblock:
 						bu[i], BCH_DATA_free, 0,
 						BTREE_TRIGGER_transactional));
 err_free:
-	if (!new_fs)
-		for (i = 0; i < nr_got; i++)
-			bch2_open_bucket_put(c, ob[i]);
+	for (i = 0; i < nr_got; i++)
+		bch2_open_bucket_put(c, ob[i]);
 
 	kfree(new_bucket_seq);
 	kfree(new_buckets);
diff --git a/fs/bcachefs/journal_reclaim.c b/fs/bcachefs/journal_reclaim.c
index 3d8fc2642425..1aabbbe328d9 100644
--- a/fs/bcachefs/journal_reclaim.c
+++ b/fs/bcachefs/journal_reclaim.c
@@ -38,6 +38,9 @@ unsigned bch2_journal_dev_buckets_available(struct journal *j,
 					    struct journal_device *ja,
 					    enum journal_space_from from)
 {
+	if (!ja->nr)
+		return 0;
+
 	unsigned available = (journal_space_from(ja, from) -
 			      ja->cur_idx - 1 + ja->nr) % ja->nr;
 
diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c
index 7086a7226989..547c78a323f7 100644
--- a/fs/bcachefs/recovery.c
+++ b/fs/bcachefs/recovery.c
@@ -1070,7 +1070,6 @@ int bch2_fs_initialize(struct bch_fs *c)
 	bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
 
-	c->curr_recovery_pass = BCH_RECOVERY_PASS_NR;
 	set_bit(BCH_FS_btree_running, &c->flags);
 	set_bit(BCH_FS_may_go_rw, &c->flags);
 
@@ -1111,9 +1110,6 @@ int bch2_fs_initialize(struct bch_fs *c)
 	if (ret)
 		goto err;
 
-	for_each_online_member(c, ca)
-		ca->new_fs_bucket_idx = 0;
-
 	ret = bch2_fs_freespace_init(c);
 	if (ret)
 		goto err;
@@ -1172,6 +1168,7 @@ int bch2_fs_initialize(struct bch_fs *c)
 	bch2_write_super(c);
 	mutex_unlock(&c->sb_lock);
 
+	c->curr_recovery_pass = BCH_RECOVERY_PASS_NR;
 	return 0;
 err:
 	bch_err_fn(c, ret);
diff --git a/fs/bcachefs/super.c b/fs/bcachefs/super.c
index 08170a3d524f..14157820705d 100644
--- a/fs/bcachefs/super.c
+++ b/fs/bcachefs/super.c
@@ -1750,11 +1750,6 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 	if (ret)
 		goto err;
 
-	ret = bch2_dev_journal_alloc(ca, true);
-	bch_err_msg(c, ret, "allocating journal");
-	if (ret)
-		goto err;
-
 	down_write(&c->state_lock);
 	mutex_lock(&c->sb_lock);
 
@@ -1805,11 +1800,14 @@ int bch2_dev_add(struct bch_fs *c, const char *path)
 	if (ret)
 		goto err_late;
 
-	ca->new_fs_bucket_idx = 0;
-
 	if (ca->mi.state == BCH_MEMBER_STATE_rw)
 		__bch2_dev_read_write(c, ca);
 
+	ret = bch2_dev_journal_alloc(ca, false);
+	bch_err_msg(c, ret, "allocating journal");
+	if (ret)
+		goto err_late;
+
 	up_write(&c->state_lock);
 	return 0;
 
-- 
2.51.0