bcachefs: Don't use a fixed size buffer for fs_usage_deltas

7 files changed, 69 insertions, 51 deletions

diff --git a/fs/bcachefs/btree_iter.c b/fs/bcachefs/btree_iter.c
index be3e9a8e43e7..e78c6cadeb96 100644
--- a/fs/bcachefs/btree_iter.c
+++ b/fs/bcachefs/btree_iter.c
@@ -1947,6 +1947,7 @@ void bch2_trans_init(struct btree_trans *trans, struct bch_fs *c,
 	trans->size		= ARRAY_SIZE(trans->iters_onstack);
 	trans->iters		= trans->iters_onstack;
 	trans->updates		= trans->updates_onstack;
+	trans->fs_usage_deltas	= NULL;
 
 	if (expected_nr_iters > trans->size)
 		bch2_trans_realloc_iters(trans, expected_nr_iters);
@@ -1959,6 +1960,7 @@ int bch2_trans_exit(struct btree_trans *trans)
 {
 	bch2_trans_unlock(trans);
 
+	kfree(trans->fs_usage_deltas);
 	kfree(trans->mem);
 	if (trans->used_mempool)
 		mempool_free(trans->iters, &trans->c->btree_iters_pool);
diff --git a/fs/bcachefs/btree_types.h b/fs/bcachefs/btree_types.h
index 945877e3c036..f2641d564e49 100644
--- a/fs/bcachefs/btree_types.h
+++ b/fs/bcachefs/btree_types.h
@@ -301,7 +301,7 @@ struct btree_trans {
 	struct btree_iter	iters_onstack[2];
 	struct btree_insert_entry updates_onstack[6];
 
-	struct replicas_delta_list fs_usage_deltas;
+	struct replicas_delta_list *fs_usage_deltas;
 };
 
 #define BTREE_FLAG(flag)						\
diff --git a/fs/bcachefs/btree_update_leaf.c b/fs/bcachefs/btree_update_leaf.c
index db506053200d..33e785763358 100644
--- a/fs/bcachefs/btree_update_leaf.c
+++ b/fs/bcachefs/btree_update_leaf.c
@@ -543,10 +543,11 @@ static inline int do_btree_insert_at(struct btree_trans *trans,
 	struct btree_insert_entry *i;
 	int ret;
 
-	if (likely(!(trans->flags & BTREE_INSERT_NO_CLEAR_REPLICAS))) {
-		memset(&trans->fs_usage_deltas.fs_usage, 0,
-		       sizeof(trans->fs_usage_deltas.fs_usage));
-		trans->fs_usage_deltas.top = trans->fs_usage_deltas.d;
+	if (likely(!(trans->flags & BTREE_INSERT_NO_CLEAR_REPLICAS)) &&
+	    trans->fs_usage_deltas) {
+		memset(&trans->fs_usage_deltas->fs_usage, 0,
+		       sizeof(trans->fs_usage_deltas->fs_usage));
+		trans->fs_usage_deltas->used = 0;
 	}
 
 	trans_for_each_update_iter(trans, i)
@@ -555,8 +556,7 @@ static inline int do_btree_insert_at(struct btree_trans *trans,
 	trans_for_each_update_iter(trans, i)
 		if (update_has_triggers(trans, i) &&
 		    update_triggers_transactional(trans, i)) {
-			ret = bch2_trans_mark_update(trans, i,
-						&trans->fs_usage_deltas);
+			ret = bch2_trans_mark_update(trans, i);
 			if (ret == -EINTR)
 				trace_trans_restart_mark(trans->ip);
 			if (ret)
@@ -626,9 +626,9 @@ static inline int do_btree_insert_at(struct btree_trans *trans,
 		    !update_triggers_transactional(trans, i))
 			bch2_mark_update(trans, i, fs_usage, 0);
 
-	if (fs_usage) {
+	if (fs_usage && trans->fs_usage_deltas) {
 		bch2_replicas_delta_list_apply(c, fs_usage,
-					       &trans->fs_usage_deltas);
+					       trans->fs_usage_deltas);
 		bch2_trans_fs_usage_apply(trans, fs_usage);
 	}
 
diff --git a/fs/bcachefs/buckets.c b/fs/bcachefs/buckets.c
index af358fc16afb..4235d7d5eb4d 100644
--- a/fs/bcachefs/buckets.c
+++ b/fs/bcachefs/buckets.c
@@ -1251,26 +1251,45 @@ void bch2_trans_fs_usage_apply(struct btree_trans *trans,
 
 /* trans_mark: */
 
-static inline void update_replicas_list(struct replicas_delta_list *d,
+static void replicas_deltas_realloc(struct btree_trans *trans)
+{
+	struct replicas_delta_list *d = trans->fs_usage_deltas;
+	unsigned new_size = d ? d->size * 2 : 128;
+
+	d = krealloc(d, sizeof(*d) + new_size, GFP_NOIO|__GFP_ZERO);
+	BUG_ON(!d);
+
+	d->size = new_size;
+	trans->fs_usage_deltas = d;
+}
+
+static inline void update_replicas_list(struct btree_trans *trans,
 					struct bch_replicas_entry *r,
 					s64 sectors)
 {
-	d->top->delta = sectors;
-	memcpy(&d->top->r, r, replicas_entry_bytes(r));
+	struct replicas_delta_list *d = trans->fs_usage_deltas;
+	struct replicas_delta *n;
+	unsigned b = replicas_entry_bytes(r) + 8;
 
-	d->top = (void *) d->top + replicas_entry_bytes(r) + 8;
+	if (!d || d->used + b > d->size) {
+		replicas_deltas_realloc(trans);
+		d = trans->fs_usage_deltas;
+	}
 
-	BUG_ON((void *) d->top > (void *) d->d + sizeof(d->pad));
+	n = (void *) d->d + d->used;
+	n->delta = sectors;
+	memcpy(&n->r, r, replicas_entry_bytes(r));
+	d->used += b;
 }
 
-static inline void update_cached_sectors_list(struct replicas_delta_list *d,
+static inline void update_cached_sectors_list(struct btree_trans *trans,
 					      unsigned dev, s64 sectors)
 {
 	struct bch_replicas_padded r;
 
 	bch2_replicas_entry_cached(&r.e, dev);
 
-	update_replicas_list(d, &r.e, sectors);
+	update_replicas_list(trans, &r.e, sectors);
 }
 
 void bch2_replicas_delta_list_apply(struct bch_fs *c,
@@ -1278,12 +1297,13 @@ void bch2_replicas_delta_list_apply(struct bch_fs *c,
 				    struct replicas_delta_list *r)
 {
 	struct replicas_delta *d = r->d;
+	struct replicas_delta *top = (void *) r->d + r->used;
 
 	acc_u64s((u64 *) fs_usage,
 		 (u64 *) &r->fs_usage, sizeof(*fs_usage) / sizeof(u64));
 
-	while (d != r->top) {
-		BUG_ON((void *) d > (void *) r->top);
+	while (d != top) {
+		BUG_ON((void *) d > (void *) top);
 
 		update_replicas(c, fs_usage, &d->r, d->delta);
 
@@ -1352,8 +1372,7 @@ static int trans_update_key(struct btree_trans *trans,
 
 static int bch2_trans_mark_pointer(struct btree_trans *trans,
 			struct extent_ptr_decoded p,
-			s64 sectors, enum bch_data_type data_type,
-			struct replicas_delta_list *d)
+			s64 sectors, enum bch_data_type data_type)
 {
 	struct bch_fs *c = trans->c;
 	struct bch_dev *ca = bch_dev_bkey_exists(c, p.ptr.dev);
@@ -1414,8 +1433,7 @@ out:
 
 static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
 			struct bch_extent_stripe_ptr p,
-			s64 sectors, enum bch_data_type data_type,
-			struct replicas_delta_list *d)
+			s64 sectors, enum bch_data_type data_type)
 {
 	struct bch_replicas_padded r;
 	struct btree_insert_entry *insert;
@@ -1460,7 +1478,7 @@ static int bch2_trans_mark_stripe_ptr(struct btree_trans *trans,
 
 	bch2_bkey_to_replicas(&r.e, s.s_c);
 
-	update_replicas_list(d, &r.e, sectors);
+	update_replicas_list(trans, &r.e, sectors);
 out:
 	bch2_trans_iter_put(trans, iter);
 	return ret;
@@ -1468,8 +1486,7 @@ out:
 
 static int bch2_trans_mark_extent(struct btree_trans *trans,
 			struct bkey_s_c k,
-			s64 sectors, enum bch_data_type data_type,
-			struct replicas_delta_list *d)
+			s64 sectors, enum bch_data_type data_type)
 {
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	const union bch_extent_entry *entry;
@@ -1492,7 +1509,7 @@ static int bch2_trans_mark_extent(struct btree_trans *trans,
 			: ptr_disk_sectors_delta(p, sectors);
 
 		ret = bch2_trans_mark_pointer(trans, p, disk_sectors,
-					      data_type, d);
+					      data_type);
 		if (ret < 0)
 			return ret;
 
@@ -1500,7 +1517,7 @@ static int bch2_trans_mark_extent(struct btree_trans *trans,
 
 		if (p.ptr.cached) {
 			if (disk_sectors && !stale)
-				update_cached_sectors_list(d, p.ptr.dev,
+				update_cached_sectors_list(trans, p.ptr.dev,
 							   disk_sectors);
 		} else if (!p.ec_nr) {
 			dirty_sectors	       += disk_sectors;
@@ -1508,7 +1525,7 @@ static int bch2_trans_mark_extent(struct btree_trans *trans,
 		} else {
 			for (i = 0; i < p.ec_nr; i++) {
 				ret = bch2_trans_mark_stripe_ptr(trans, p.ec[i],
-						disk_sectors, data_type, d);
+						disk_sectors, data_type);
 				if (ret)
 					return ret;
 			}
@@ -1518,16 +1535,16 @@ static int bch2_trans_mark_extent(struct btree_trans *trans,
 	}
 
 	if (dirty_sectors)
-		update_replicas_list(d, &r.e, dirty_sectors);
+		update_replicas_list(trans, &r.e, dirty_sectors);
 
 	return 0;
 }
 
 int bch2_trans_mark_key(struct btree_trans *trans,
 			struct bkey_s_c k,
-			bool inserting, s64 sectors,
-			struct replicas_delta_list *d)
+			bool inserting, s64 sectors)
 {
+	struct replicas_delta_list *d;
 	struct bch_fs *c = trans->c;
 
 	switch (k.k->type) {
@@ -1535,11 +1552,15 @@ int bch2_trans_mark_key(struct btree_trans *trans,
 		return bch2_trans_mark_extent(trans, k, inserting
 				?  c->opts.btree_node_size
 				: -c->opts.btree_node_size,
-				BCH_DATA_BTREE, d);
+				BCH_DATA_BTREE);
 	case KEY_TYPE_extent:
 		return bch2_trans_mark_extent(trans, k,
-				sectors, BCH_DATA_USER, d);
+				sectors, BCH_DATA_USER);
 	case KEY_TYPE_inode:
+		if (!trans->fs_usage_deltas)
+			replicas_deltas_realloc(trans);
+		d = trans->fs_usage_deltas;
+
 		if (inserting)
 			d->fs_usage.nr_inodes++;
 		else
@@ -1548,6 +1569,10 @@ int bch2_trans_mark_key(struct btree_trans *trans,
 	case KEY_TYPE_reservation: {
 		unsigned replicas = bkey_s_c_to_reservation(k).v->nr_replicas;
 
+		if (!trans->fs_usage_deltas)
+			replicas_deltas_realloc(trans);
+		d = trans->fs_usage_deltas;
+
 		sectors *= replicas;
 		replicas = clamp_t(unsigned, replicas, 1,
 				   ARRAY_SIZE(d->fs_usage.persistent_reserved));
@@ -1562,8 +1587,7 @@ int bch2_trans_mark_key(struct btree_trans *trans,
 }
 
 int bch2_trans_mark_update(struct btree_trans *trans,
-			   struct btree_insert_entry *insert,
-			   struct replicas_delta_list *d)
+			   struct btree_insert_entry *insert)
 {
 	struct btree_iter	*iter = insert->iter;
 	struct btree		*b = iter->l[0].b;
@@ -1577,7 +1601,7 @@ int bch2_trans_mark_update(struct btree_trans *trans,
 	ret = bch2_trans_mark_key(trans,
 			bkey_i_to_s_c(insert->k), true,
 			bpos_min(insert->k->k.p, b->key.k.p).offset -
-			bkey_start_offset(&insert->k->k), d);
+			bkey_start_offset(&insert->k->k));
 	if (ret)
 		return ret;
 
@@ -1612,7 +1636,7 @@ int bch2_trans_mark_update(struct btree_trans *trans,
 				BUG_ON(sectors <= 0);
 
 				ret = bch2_trans_mark_key(trans, k, true,
-							  sectors, d);
+							  sectors);
 				if (ret)
 					return ret;
 
@@ -1624,7 +1648,7 @@ int bch2_trans_mark_update(struct btree_trans *trans,
 			BUG_ON(sectors >= 0);
 		}
 
-		ret = bch2_trans_mark_key(trans, k, false, sectors, d);
+		ret = bch2_trans_mark_key(trans, k, false, sectors);
 		if (ret)
 			return ret;
 
diff --git a/fs/bcachefs/buckets.h b/fs/bcachefs/buckets.h
index 43b2b5bf3c31..e31fe3149a2d 100644
--- a/fs/bcachefs/buckets.h
+++ b/fs/bcachefs/buckets.h
@@ -270,11 +270,9 @@ int bch2_mark_update(struct btree_trans *, struct btree_insert_entry *,
 void bch2_replicas_delta_list_apply(struct bch_fs *,
 				    struct bch_fs_usage *,
 				    struct replicas_delta_list *);
-int bch2_trans_mark_key(struct btree_trans *, struct bkey_s_c,
-			bool, s64, struct replicas_delta_list *);
+int bch2_trans_mark_key(struct btree_trans *, struct bkey_s_c, bool, s64);
 int bch2_trans_mark_update(struct btree_trans *,
-			   struct btree_insert_entry *,
-			   struct replicas_delta_list *);
+			   struct btree_insert_entry *);
 void bch2_trans_fs_usage_apply(struct btree_trans *, struct bch_fs_usage *);
 
 /* disk reservations: */
diff --git a/fs/bcachefs/buckets_types.h b/fs/bcachefs/buckets_types.h
index a333b9ec9302..309a5fb61e98 100644
--- a/fs/bcachefs/buckets_types.h
+++ b/fs/bcachefs/buckets_types.h
@@ -100,11 +100,10 @@ struct replicas_delta {
 } __packed;
 
 struct replicas_delta_list {
+	unsigned		size;
+	unsigned		used;
 	struct bch_fs_usage	fs_usage;
-
-	struct replicas_delta	*top;
 	struct replicas_delta	d[0];
-	u8			pad[256];
 };
 
 /*
diff --git a/fs/bcachefs/recovery.c b/fs/bcachefs/recovery.c
index afa7e9ee0fd4..5baf002cb156 100644
--- a/fs/bcachefs/recovery.c
+++ b/fs/bcachefs/recovery.c
@@ -257,13 +257,8 @@ retry:
 	} while (bkey_cmp(iter->pos, k->k.p) < 0);
 
 	if (split_compressed) {
-		memset(&trans.fs_usage_deltas.fs_usage, 0,
-		       sizeof(trans.fs_usage_deltas.fs_usage));
-		trans.fs_usage_deltas.top = trans.fs_usage_deltas.d;
-
 		ret = bch2_trans_mark_key(&trans, bkey_i_to_s_c(k), false,
-					  -((s64) k->k.size),
-					  &trans.fs_usage_deltas) ?:
+					  -((s64) k->k.size)) ?:
 		      bch2_trans_commit(&trans, &disk_res, NULL,
 					BTREE_INSERT_ATOMIC|
 					BTREE_INSERT_NOFAIL|