bcachefs: Fix rebalance_work accounting

rebalance_work was keying off of the presence of rebelance_opts in the
extent - but that was incorrect, we keep those around after rebalance
for indirect extents since the inode's options are not directly
available

Fixes: 20ac515a9cc7 ("bcachefs: bch_acct_rebalance_work")
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

5 files changed, 98 insertions, 27 deletions

diff --git a/fs/bcachefs/bcachefs_format.h b/fs/bcachefs/bcachefs_format.h
index c75f2e0f32bb..14ce726bf5a3 100644
--- a/fs/bcachefs/bcachefs_format.h
+++ b/fs/bcachefs/bcachefs_format.h
@@ -677,7 +677,8 @@ struct bch_sb_field_ext {
 	x(bucket_stripe_sectors,	BCH_VERSION(1,  8))		\
 	x(disk_accounting_v2,		BCH_VERSION(1,  9))		\
 	x(disk_accounting_v3,		BCH_VERSION(1, 10))		\
-	x(disk_accounting_inum,		BCH_VERSION(1, 11))
+	x(disk_accounting_inum,		BCH_VERSION(1, 11))		\
+	x(rebalance_work_acct_fix,	BCH_VERSION(1, 12))
 
 enum bcachefs_metadata_version {
 	bcachefs_metadata_version_min = 9,
diff --git a/fs/bcachefs/buckets.c b/fs/bcachefs/buckets.c
index be2bbd248631..a2274429e7f4 100644
--- a/fs/bcachefs/buckets.c
+++ b/fs/bcachefs/buckets.c
@@ -699,7 +699,8 @@ err:
 static int __trigger_extent(struct btree_trans *trans,
 			    enum btree_id btree_id, unsigned level,
 			    struct bkey_s_c k,
-			    enum btree_iter_update_trigger_flags flags)
+			    enum btree_iter_update_trigger_flags flags,
+			    s64 *replicas_sectors)
 {
 	bool gc = flags & BTREE_TRIGGER_gc;
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
@@ -708,7 +709,6 @@ static int __trigger_extent(struct btree_trans *trans,
 	enum bch_data_type data_type = bkey_is_btree_ptr(k.k)
 		? BCH_DATA_btree
 		: BCH_DATA_user;
-	s64 replicas_sectors = 0;
 	int ret = 0;
 
 	struct disk_accounting_pos acc_replicas_key = {
@@ -739,7 +739,7 @@ static int __trigger_extent(struct btree_trans *trans,
 			if (ret)
 				return ret;
 		} else if (!p.has_ec) {
-			replicas_sectors       += disk_sectors;
+			*replicas_sectors       += disk_sectors;
 			acc_replicas_key.replicas.devs[acc_replicas_key.replicas.nr_devs++] = p.ptr.dev;
 		} else {
 			ret = bch2_trigger_stripe_ptr(trans, k, p, data_type, disk_sectors, flags);
@@ -777,7 +777,7 @@ static int __trigger_extent(struct btree_trans *trans,
 	}
 
 	if (acc_replicas_key.replicas.nr_devs) {
-		ret = bch2_disk_accounting_mod(trans, &acc_replicas_key, &replicas_sectors, 1, gc);
+		ret = bch2_disk_accounting_mod(trans, &acc_replicas_key, replicas_sectors, 1, gc);
 		if (ret)
 			return ret;
 	}
@@ -787,7 +787,7 @@ static int __trigger_extent(struct btree_trans *trans,
 			.type			= BCH_DISK_ACCOUNTING_snapshot,
 			.snapshot.id		= k.k->p.snapshot,
 		};
-		ret = bch2_disk_accounting_mod(trans, &acc_snapshot_key, &replicas_sectors, 1, gc);
+		ret = bch2_disk_accounting_mod(trans, &acc_snapshot_key, replicas_sectors, 1, gc);
 		if (ret)
 			return ret;
 	}
@@ -807,7 +807,7 @@ static int __trigger_extent(struct btree_trans *trans,
 			.type		= BCH_DISK_ACCOUNTING_btree,
 			.btree.id	= btree_id,
 		};
-		ret = bch2_disk_accounting_mod(trans, &acc_btree_key, &replicas_sectors, 1, gc);
+		ret = bch2_disk_accounting_mod(trans, &acc_btree_key, replicas_sectors, 1, gc);
 		if (ret)
 			return ret;
 	} else {
@@ -819,22 +819,13 @@ static int __trigger_extent(struct btree_trans *trans,
 		s64 v[3] = {
 			insert ? 1 : -1,
 			insert ? k.k->size : -((s64) k.k->size),
-			replicas_sectors,
+			*replicas_sectors,
 		};
 		ret = bch2_disk_accounting_mod(trans, &acc_inum_key, v, ARRAY_SIZE(v), gc);
 		if (ret)
 			return ret;
 	}
 
-	if (bch2_bkey_rebalance_opts(k)) {
-		struct disk_accounting_pos acc = {
-			.type		= BCH_DISK_ACCOUNTING_rebalance_work,
-		};
-		ret = bch2_disk_accounting_mod(trans, &acc, &replicas_sectors, 1, gc);
-		if (ret)
-			return ret;
-	}
-
 	return 0;
 }
 
@@ -843,6 +834,7 @@ int bch2_trigger_extent(struct btree_trans *trans,
 			struct bkey_s_c old, struct bkey_s new,
 			enum btree_iter_update_trigger_flags flags)
 {
+	struct bch_fs *c = trans->c;
 	struct bkey_ptrs_c new_ptrs = bch2_bkey_ptrs_c(new.s_c);
 	struct bkey_ptrs_c old_ptrs = bch2_bkey_ptrs_c(old);
 	unsigned new_ptrs_bytes = (void *) new_ptrs.end - (void *) new_ptrs.start;
@@ -858,21 +850,53 @@ int bch2_trigger_extent(struct btree_trans *trans,
 		    new_ptrs_bytes))
 		return 0;
 
-	if (flags & BTREE_TRIGGER_transactional) {
-		struct bch_fs *c = trans->c;
-		int mod = (int) bch2_bkey_needs_rebalance(c, new.s_c) -
-			  (int) bch2_bkey_needs_rebalance(c, old);
+	if (flags & (BTREE_TRIGGER_transactional|BTREE_TRIGGER_gc)) {
+		s64 old_replicas_sectors = 0, new_replicas_sectors = 0;
+
+		if (old.k->type) {
+			int ret = __trigger_extent(trans, btree, level, old,
+						   flags & ~BTREE_TRIGGER_insert,
+						   &old_replicas_sectors);
+			if (ret)
+				return ret;
+		}
+
+		if (new.k->type) {
+			int ret = __trigger_extent(trans, btree, level, new.s_c,
+						   flags & ~BTREE_TRIGGER_overwrite,
+						   &new_replicas_sectors);
+			if (ret)
+				return ret;
+		}
+
+		int need_rebalance_delta = 0;
+		s64 need_rebalance_sectors_delta = 0;
+
+		s64 s = bch2_bkey_sectors_need_rebalance(c, old);
+		need_rebalance_delta -= s != 0;
+		need_rebalance_sectors_delta -= s;
 
-		if (mod) {
+		s = bch2_bkey_sectors_need_rebalance(c, old);
+		need_rebalance_delta += s != 0;
+		need_rebalance_sectors_delta += s;
+
+		if ((flags & BTREE_TRIGGER_transactional) && need_rebalance_delta) {
 			int ret = bch2_btree_bit_mod_buffered(trans, BTREE_ID_rebalance_work,
-							      new.k->p, mod > 0);
+							  new.k->p, need_rebalance_delta > 0);
 			if (ret)
 				return ret;
 		}
-	}
 
-	if (flags & (BTREE_TRIGGER_transactional|BTREE_TRIGGER_gc))
-		return trigger_run_overwrite_then_insert(__trigger_extent, trans, btree, level, old, new, flags);
+		if (need_rebalance_sectors_delta) {
+			struct disk_accounting_pos acc = {
+				.type		= BCH_DISK_ACCOUNTING_rebalance_work,
+			};
+			int ret = bch2_disk_accounting_mod(trans, &acc, &need_rebalance_sectors_delta, 1,
+							   flags & BTREE_TRIGGER_gc);
+			if (ret)
+				return ret;
+		}
+	}
 
 	return 0;
 }
diff --git a/fs/bcachefs/extents.c b/fs/bcachefs/extents.c
index 9406f82fc255..e317df3644a1 100644
--- a/fs/bcachefs/extents.c
+++ b/fs/bcachefs/extents.c
@@ -1379,6 +1379,45 @@ bool bch2_bkey_needs_rebalance(struct bch_fs *c, struct bkey_s_c k)
 	return r != NULL;
 }
 
+static u64 __bch2_bkey_sectors_need_rebalance(struct bch_fs *c, struct bkey_s_c k,
+				       unsigned target, unsigned compression)
+{
+	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
+	const union bch_extent_entry *entry;
+	struct extent_ptr_decoded p;
+	u64 sectors = 0;
+
+	if (compression) {
+		unsigned compression_type = bch2_compression_opt_to_type(compression);
+
+		bkey_for_each_ptr_decode(k.k, ptrs, p, entry) {
+			if (p.crc.compression_type == BCH_COMPRESSION_TYPE_incompressible ||
+			    p.ptr.unwritten) {
+				sectors = 0;
+				goto incompressible;
+			}
+
+			if (!p.ptr.cached && p.crc.compression_type != compression_type)
+				sectors += p.crc.compressed_size;
+		}
+	}
+incompressible:
+	if (target && bch2_target_accepts_data(c, BCH_DATA_user, target)) {
+		bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
+			if (!p.ptr.cached && !bch2_dev_in_target(c, p.ptr.dev, target))
+				sectors += p.crc.compressed_size;
+	}
+
+	return sectors;
+}
+
+u64 bch2_bkey_sectors_need_rebalance(struct bch_fs *c, struct bkey_s_c k)
+{
+	const struct bch_extent_rebalance *r = bch2_bkey_rebalance_opts(k);
+
+	return r ? __bch2_bkey_sectors_need_rebalance(c, k, r->target, r->compression) : 0;
+}
+
 int bch2_bkey_set_needs_rebalance(struct bch_fs *c, struct bkey_i *_k,
 				  struct bch_io_opts *opts)
 {
diff --git a/fs/bcachefs/extents.h b/fs/bcachefs/extents.h
index 1a6ddee48041..709dd83183be 100644
--- a/fs/bcachefs/extents.h
+++ b/fs/bcachefs/extents.h
@@ -692,6 +692,7 @@ const struct bch_extent_rebalance *bch2_bkey_rebalance_opts(struct bkey_s_c);
 unsigned bch2_bkey_ptrs_need_rebalance(struct bch_fs *, struct bkey_s_c,
 				       unsigned, unsigned);
 bool bch2_bkey_needs_rebalance(struct bch_fs *, struct bkey_s_c);
+u64 bch2_bkey_sectors_need_rebalance(struct bch_fs *, struct bkey_s_c);
 
 int bch2_bkey_set_needs_rebalance(struct bch_fs *, struct bkey_i *,
 				  struct bch_io_opts *);
diff --git a/fs/bcachefs/sb-downgrade.c b/fs/bcachefs/sb-downgrade.c
index 650a1f77ca40..c7e4cdd3f6a5 100644
--- a/fs/bcachefs/sb-downgrade.c
+++ b/fs/bcachefs/sb-downgrade.c
@@ -75,6 +75,9 @@
 	  BCH_FSCK_ERR_accounting_key_junk_at_end)		\
 	x(disk_accounting_inum,					\
 	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
+	  BCH_FSCK_ERR_accounting_mismatch)			\
+	x(rebalance_work_acct_fix,				\
+	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
 	  BCH_FSCK_ERR_accounting_mismatch)
 
 #define DOWNGRADE_TABLE()					\
@@ -108,7 +111,10 @@
 	  BCH_FSCK_ERR_fs_usage_persistent_reserved_wrong,	\
 	  BCH_FSCK_ERR_fs_usage_replicas_wrong,			\
 	  BCH_FSCK_ERR_accounting_replicas_not_marked,		\
-	  BCH_FSCK_ERR_bkey_version_in_future)
+	  BCH_FSCK_ERR_bkey_version_in_future)			\
+	x(rebalance_work_acct_fix,				\
+	  BIT_ULL(BCH_RECOVERY_PASS_check_allocations),		\
+	  BCH_FSCK_ERR_accounting_mismatch)
 
 struct upgrade_downgrade_entry {
 	u64		recovery_passes;