bcachefs: metadata version bucket_stripe_sectors

New on disk format version for bch_alloc->stripe_sectors and
BCH_DATA_unstriped - accounting for unstriped data in stripe buckets.

Upgrade/downgrade requires regenerating alloc info - but only if erasure
coding is in use.

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

fs/bcachefs/bcachefs.h

@@ -592,6 +592,7 @@ struct bch_dev {
 #define BCH_FS_FLAGS()			\
 	x(new_fs)			\
 	x(started)			\
+	x(btree_running)		\
 	x(may_go_rw)			\
 	x(rw)				\
 	x(was_rw)			\

fs/bcachefs/bcachefs_format.h

@@ -723,7 +723,8 @@ struct bch_sb_field_ext {
 	x(member_seq,			BCH_VERSION(1,  4))		\
 	x(subvolume_fs_parent,		BCH_VERSION(1,  5))		\
 	x(btree_subvolume_children,	BCH_VERSION(1,  6))		\
-	x(mi_btree_bitmap,		BCH_VERSION(1,  7))
+	x(mi_btree_bitmap,		BCH_VERSION(1,  7))		\
+	x(bucket_stripe_sectors,	BCH_VERSION(1,  8))
 
 enum bcachefs_metadata_version {
 	bcachefs_metadata_version_min = 9,

fs/bcachefs/btree_iter.h

@@ -866,6 +866,14 @@ __bch2_btree_iter_peek_and_restart(struct btree_trans *trans,
 	_p;								\
 })
 
+#define bch2_trans_run(_c, _do)						\
+({									\
+	struct btree_trans *trans = bch2_trans_get(_c);			\
+	int _ret = (_do);						\
+	bch2_trans_put(trans);						\
+	_ret;								\
+})
+
 void bch2_trans_updates_to_text(struct printbuf *, struct btree_trans *);
 void bch2_btree_path_to_text(struct printbuf *, struct btree_trans *, btree_path_idx_t);
 void bch2_trans_paths_to_text(struct printbuf *, struct btree_trans *);

fs/bcachefs/btree_update.h

@@ -178,14 +178,6 @@ static inline int bch2_trans_commit(struct btree_trans *trans,
 	nested_lockrestart_do(_trans, _do ?: bch2_trans_commit(_trans, (_disk_res),\
 					(_journal_seq), (_flags)))
 
-#define bch2_trans_run(_c, _do)						\
-({									\
-	struct btree_trans *trans = bch2_trans_get(_c);			\
-	int _ret = (_do);						\
-	bch2_trans_put(trans);						\
-	_ret;								\
-})
-
 #define bch2_trans_do(_c, _disk_res, _journal_seq, _flags, _do)		\
 	bch2_trans_run(_c, commit_do(trans, _disk_res, _journal_seq, _flags, _do))
 

fs/bcachefs/recovery.c

@@ -810,6 +810,10 @@ use_clean:
 	if (ret)
 		goto err;
 
+	set_bit(BCH_FS_btree_running, &c->flags);
+
+	ret = bch2_sb_set_upgrade_extra(c);
+
 	ret = bch2_run_recovery_passes(c);
 	if (ret)
 		goto err;
@@ -969,6 +973,7 @@ int bch2_fs_initialize(struct bch_fs *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);
 
 	for (unsigned i = 0; i < BTREE_ID_NR; i++)

fs/bcachefs/sb-downgrade.c

@@ -56,7 +56,9 @@
 	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
 	  BCH_FSCK_ERR_btree_bitmap_not_marked)
 
-#define DOWNGRADE_TABLE()
+#define DOWNGRADE_TABLE()					\
+	x(bucket_stripe_sectors,				\
+	  0)
 
 struct upgrade_downgrade_entry {
 	u64		recovery_passes;
@@ -80,6 +82,37 @@ UPGRADE_TABLE()
 #undef x
 };
 
+static int have_stripes(struct bch_fs *c)
+{
+	return !btree_node_fake(c->btree_roots_known[BTREE_ID_stripes].b);
+}
+
+int bch2_sb_set_upgrade_extra(struct bch_fs *c)
+{
+	unsigned old_version = c->sb.version_upgrade_complete ?: c->sb.version;
+	unsigned new_version = c->sb.version;
+	bool write_sb = false;
+	int ret = 0;
+
+	mutex_lock(&c->sb_lock);
+	struct bch_sb_field_ext *ext = bch2_sb_field_get(c->disk_sb.sb, ext);
+
+	if (old_version <  bcachefs_metadata_version_bucket_stripe_sectors &&
+	    new_version >= bcachefs_metadata_version_bucket_stripe_sectors &&
+	    (ret = have_stripes(c) > 0)) {
+		__set_bit_le64(BCH_RECOVERY_PASS_STABLE_check_allocations, ext->recovery_passes_required);
+		__set_bit_le64(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong, ext->errors_silent);
+		__set_bit_le64(BCH_FSCK_ERR_alloc_key_stripe_sectors_wrong, ext->errors_silent);
+		write_sb = true;
+	}
+
+	if (write_sb)
+		bch2_write_super(c);
+	mutex_unlock(&c->sb_lock);
+
+	return ret < 0 ? ret : 0;
+}
+
 void bch2_sb_set_upgrade(struct bch_fs *c,
 			 unsigned old_version,
 			 unsigned new_version)
@@ -101,16 +134,12 @@ void bch2_sb_set_upgrade(struct bch_fs *c,
 			ext->recovery_passes_required[0] |=
 				cpu_to_le64(bch2_recovery_passes_to_stable(passes));
 
-			for (const u16 *e = i->errors;
-			     e < i->errors + i->nr_errors;
-			     e++) {
-				__set_bit(*e, c->sb.errors_silent);
-				ext->errors_silent[*e / 64] |= cpu_to_le64(BIT_ULL(*e % 64));
-			}
+			for (const u16 *e = i->errors; e < i->errors + i->nr_errors; e++)
+				__set_bit_le64(*e, ext->errors_silent);
 		}
 }
 
-#define x(ver, passes, ...) static const u16 downgrade_ver_##errors[] = { __VA_ARGS__ };
+#define x(ver, passes, ...) static const u16 downgrade_##ver##_errors[] = { __VA_ARGS__ };
 DOWNGRADE_TABLE()
 #undef x
 
@@ -125,6 +154,37 @@ DOWNGRADE_TABLE()
 #undef x
 };
 
+static int downgrade_table_extra(struct bch_fs *c, darray_char *table)
+{
+	struct bch_sb_field_downgrade_entry *dst = (void *) &darray_top(*table);
+	unsigned bytes = sizeof(*dst) + sizeof(dst->errors[0]) * le16_to_cpu(dst->nr_errors);
+	int ret = 0;
+
+	unsigned nr_errors = le16_to_cpu(dst->nr_errors);
+
+	switch (le16_to_cpu(dst->version)) {
+	case bcachefs_metadata_version_bucket_stripe_sectors:
+		if (have_stripes(c)) {
+			bytes += sizeof(dst->errors[0]) * 2;
+
+			ret = darray_make_room(table, bytes);
+			if (ret)
+				return ret;
+
+			/* open coded __set_bit_le64, as dst is packed and
+			 * dst->recovery_passes is misaligned */
+			unsigned b = BCH_RECOVERY_PASS_STABLE_check_allocations;
+			dst->recovery_passes[b / 64] |= cpu_to_le64(BIT_ULL(b % 64));
+
+			dst->errors[nr_errors++] = cpu_to_le16(BCH_FSCK_ERR_alloc_key_dirty_sectors_wrong);
+		}
+		break;
+	}
+
+	dst->nr_errors = cpu_to_le16(nr_errors);
+	return ret;
+}
+
 static inline const struct bch_sb_field_downgrade_entry *
 downgrade_entry_next_c(const struct bch_sb_field_downgrade_entry *e)
 {
@@ -210,6 +270,9 @@ const struct bch_sb_field_ops bch_sb_field_ops_downgrade = {
 
 int bch2_sb_downgrade_update(struct bch_fs *c)
 {
+	if (!test_bit(BCH_FS_btree_running, &c->flags))
+		return 0;
+
 	darray_char table = {};
 	int ret = 0;
 
@@ -234,7 +297,14 @@ int bch2_sb_downgrade_update(struct bch_fs *c)
 		for (unsigned i = 0; i < src->nr_errors; i++)
 			dst->errors[i] = cpu_to_le16(src->errors[i]);
 
-		table.nr += bytes;
+		downgrade_table_extra(c, &table);
+
+		if (!dst->recovery_passes[0] &&
+		    !dst->recovery_passes[1] &&
+		    !dst->nr_errors)
+			continue;
+
+		table.nr += sizeof(*dst) + sizeof(dst->errors[0]) * le16_to_cpu(dst->nr_errors);
 	}
 
 	struct bch_sb_field_downgrade *d = bch2_sb_field_get(c->disk_sb.sb, downgrade);

fs/bcachefs/sb-downgrade.h

@@ -6,6 +6,7 @@ extern const struct bch_sb_field_ops bch_sb_field_ops_downgrade;
 
 int bch2_sb_downgrade_update(struct bch_fs *);
 void bch2_sb_set_upgrade(struct bch_fs *, unsigned, unsigned);
+int bch2_sb_set_upgrade_extra(struct bch_fs *);
 void bch2_sb_set_downgrade(struct bch_fs *, unsigned, unsigned);
 
 #endif /* _BCACHEFS_SB_DOWNGRADE_H */