bcachefs: Account for stripe parity sectors separately

Instead of trying to charge EC parity to the data within the stripe
(which is subject to rounding errors), let's charge it to the stripe
itself. It should also make -ENOSPC issues easier to deal with if we
charge for parity blocks up front, and means we can also make more fine
grained accounting available to the user.
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>

fs/bcachefs/bcachefs_format.h

@@ -1036,7 +1036,8 @@ LE64_BITMASK(BCH_KDF_SCRYPT_P,	struct bch_sb_field_crypt, kdf_flags, 32, 48);
 	x(journal,	2)		\
 	x(btree,	3)		\
 	x(user,		4)		\
-	x(cached,	5)
+	x(cached,	5)		\
+	x(parity,	6)
 
 enum bch_data_type {
 #define x(t, n) BCH_DATA_##t,

fs/bcachefs/ec.c

@@ -343,12 +343,17 @@ static void ec_block_io(struct bch_fs *c, struct ec_stripe_buf *buf,
 	unsigned offset = 0, bytes = buf->size << 9;
 	struct bch_extent_ptr *ptr = &v->ptrs[idx];
 	struct bch_dev *ca = bch_dev_bkey_exists(c, ptr->dev);
+	enum bch_data_type data_type = idx < buf->key.v.nr_blocks - buf->key.v.nr_redundant
+		? BCH_DATA_user
+		: BCH_DATA_parity;
 
 	if (!bch2_dev_get_ioref(ca, rw)) {
 		clear_bit(idx, buf->valid);
 		return;
 	}
 
+	this_cpu_add(ca->io_done->sectors[rw][data_type], buf->size);
+
 	while (offset < bytes) {
 		unsigned nr_iovecs = min_t(size_t, BIO_MAX_VECS,
 					   DIV_ROUND_UP(bytes, PAGE_SIZE));
@@ -670,6 +675,7 @@ static void ec_stripe_delete_work(struct work_struct *work)
 /* stripe creation: */
 
 static int ec_stripe_bkey_insert(struct bch_fs *c,
+				 struct ec_stripe_new *s,
 				 struct bkey_i_stripe *stripe)
 {
 	struct btree_trans trans;
@@ -711,7 +717,7 @@ static int ec_stripe_bkey_insert(struct bch_fs *c,
 
 	bch2_trans_update(&trans, iter, &stripe->k_i, 0);
 
-	ret = bch2_trans_commit(&trans, NULL, NULL,
+	ret = bch2_trans_commit(&trans, &s->res, NULL,
 				BTREE_INSERT_NOFAIL);
 err:
 	bch2_trans_iter_put(&trans, iter);
@@ -858,8 +864,8 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 
 	ret = s->existing_stripe
 		? bch2_btree_insert(c, BTREE_ID_EC, &s->stripe.key.k_i,
-				    NULL, NULL, BTREE_INSERT_NOFAIL)
-		: ec_stripe_bkey_insert(c, &s->stripe.key);
+				    &s->res, NULL, BTREE_INSERT_NOFAIL)
+		: ec_stripe_bkey_insert(c, s, &s->stripe.key);
 	if (ret) {
 		bch_err(c, "error creating stripe: error creating stripe key");
 		goto err_put_writes;
@@ -886,6 +892,8 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 err_put_writes:
 	percpu_ref_put(&c->writes);
 err:
+	bch2_disk_reservation_put(c, &s->res);
+
 	open_bucket_for_each(c, &s->blocks, ob, i) {
 		ob->ec = NULL;
 		__bch2_open_bucket_put(c, ob);
@@ -1325,6 +1333,7 @@ struct ec_stripe_head *bch2_ec_stripe_head_get(struct bch_fs *c,
 	struct open_bucket *ob;
 	unsigned i, data_idx = 0;
 	s64 idx;
+	int ret;
 
 	closure_init_stack(&cl);
 
@@ -1356,6 +1365,22 @@ struct ec_stripe_head *bch2_ec_stripe_head_get(struct bch_fs *c,
 				}
 		}
 
+		if (!h->s->existing_stripe &&
+		    !h->s->res.sectors) {
+			ret = bch2_disk_reservation_get(c, &h->s->res,
+							h->blocksize,
+							h->s->nr_parity, 0);
+			if (ret) {
+				/* What should we do here? */
+				bch_err(c, "unable to create new stripe: %i", ret);
+				bch2_ec_stripe_head_put(c, h);
+				h = NULL;
+				goto out;
+
+			}
+
+		}
+
 		if (new_stripe_alloc_buckets(c, h)) {
 			bch2_ec_stripe_head_put(c, h);
 			h = NULL;

fs/bcachefs/ec.h

@@ -3,6 +3,7 @@
 #define _BCACHEFS_EC_H
 
 #include "ec_types.h"
+#include "buckets_types.h"
 #include "keylist_types.h"
 
 const char *bch2_stripe_invalid(const struct bch_fs *, struct bkey_s_c);
@@ -105,6 +106,7 @@ struct ec_stripe_new {
 	struct open_buckets	blocks;
 	u8			data_block_idx[EC_STRIPE_MAX];
 	struct open_buckets	parity;
+	struct disk_reservation	res;
 
 	struct keylist		keys;
 	u64			inline_keys[BKEY_U64s * 8];

fs/bcachefs/replicas.c

@@ -122,7 +122,7 @@ void bch2_bkey_to_replicas(struct bch_replicas_entry *e,
 		extent_to_replicas(k, e);
 		break;
 	case KEY_TYPE_stripe:
-		e->data_type = BCH_DATA_user;
+		e->data_type = BCH_DATA_parity;
 		stripe_to_replicas(k, e);
 		break;
 	}
@@ -449,7 +449,23 @@ static int __bch2_mark_bkey_replicas(struct bch_fs *c, struct bkey_s_c k,
 
 	bch2_bkey_to_replicas(&search.e, k);
 
-	return __bch2_mark_replicas(c, &search.e, check);
+	ret = __bch2_mark_replicas(c, &search.e, check);
+	if (ret)
+		return ret;
+
+	if (search.e.data_type == BCH_DATA_parity) {
+		search.e.data_type = BCH_DATA_cached;
+		ret = __bch2_mark_replicas(c, &search.e, check);
+		if (ret)
+			return ret;
+
+		search.e.data_type = BCH_DATA_user;
+		ret = __bch2_mark_replicas(c, &search.e, check);
+		if (ret)
+			return ret;
+	}
+
+	return 0;
 }
 
 bool bch2_bkey_replicas_marked(struct bch_fs *c,