X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libbcachefs%2Fec.c;h=3ebe699ff0a2e63abfab60ceadaec764124173a6;hb=46a6b9210c927ab46fd1227cb6f641be0b4a7505;hp=f18399906af5ee2d0258b1385f15b164c715d9dc;hpb=55e3496d06c9b112f93bb1dea942564f900c2f7d;p=bcachefs-tools-debian

diff --git a/libbcachefs/ec.c b/libbcachefs/ec.c
index f183999..3ebe699 100644
--- a/libbcachefs/ec.c
+++ b/libbcachefs/ec.c
@@ -4,10 +4,12 @@
 
 #include "bcachefs.h"
 #include "alloc_foreground.h"
+#include "backpointers.h"
 #include "bkey_buf.h"
 #include "bset.h"
 #include "btree_gc.h"
 #include "btree_update.h"
+#include "btree_write_buffer.h"
 #include "buckets.h"
 #include "disk_groups.h"
 #include "ec.h"
@@ -102,24 +104,34 @@ struct ec_bio {
 
 /* Stripes btree keys: */
 
-const char *bch2_stripe_invalid(const struct bch_fs *c, struct bkey_s_c k)
+int bch2_stripe_invalid(const struct bch_fs *c, struct bkey_s_c k,
+			unsigned flags, struct printbuf *err)
 {
 	const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
 
-	if (!bkey_cmp(k.k->p, POS_MIN))
-		return "stripe at pos 0";
+	if (bkey_eq(k.k->p, POS_MIN)) {
+		prt_printf(err, "stripe at POS_MIN");
+		return -BCH_ERR_invalid_bkey;
+	}
 
-	if (k.k->p.inode)
-		return "invalid stripe key";
+	if (k.k->p.inode) {
+		prt_printf(err, "nonzero inode field");
+		return -BCH_ERR_invalid_bkey;
+	}
 
-	if (bkey_val_bytes(k.k) < sizeof(*s))
-		return "incorrect value size";
+	if (bkey_val_bytes(k.k) < sizeof(*s)) {
+		prt_printf(err, "incorrect value size (%zu < %zu)",
+		       bkey_val_bytes(k.k), sizeof(*s));
+		return -BCH_ERR_invalid_bkey;
+	}
 
-	if (bkey_val_bytes(k.k) < sizeof(*s) ||
-	    bkey_val_u64s(k.k) < stripe_val_u64s(s))
-		return "incorrect value size";
+	if (bkey_val_u64s(k.k) < stripe_val_u64s(s)) {
+		prt_printf(err, "incorrect value size (%zu < %u)",
+		       bkey_val_u64s(k.k), stripe_val_u64s(s));
+		return -BCH_ERR_invalid_bkey;
+	}
 
-	return bch2_bkey_ptrs_invalid(c, k);
+	return bch2_bkey_ptrs_invalid(c, k, flags, err);
 }
 
 void bch2_stripe_to_text(struct printbuf *out, struct bch_fs *c,
@@ -128,7 +140,7 @@ void bch2_stripe_to_text(struct printbuf *out, struct bch_fs *c,
 	const struct bch_stripe *s = bkey_s_c_to_stripe(k).v;
 	unsigned i;
 
-	pr_buf(out, "algo %u sectors %u blocks %u:%u csum %u gran %u",
+	prt_printf(out, "algo %u sectors %u blocks %u:%u csum %u gran %u",
 	       s->algorithm,
 	       le16_to_cpu(s->sectors),
 	       s->nr_blocks - s->nr_redundant,
@@ -137,14 +149,14 @@ void bch2_stripe_to_text(struct printbuf *out, struct bch_fs *c,
 	       1U << s->csum_granularity_bits);
 
 	for (i = 0; i < s->nr_blocks; i++)
-		pr_buf(out, " %u:%llu:%u", s->ptrs[i].dev,
+		prt_printf(out, " %u:%llu:%u", s->ptrs[i].dev,
 		       (u64) s->ptrs[i].offset,
 		       stripe_blockcount_get(s, i));
 }
 
 /* returns blocknr in stripe that we matched: */
-static int bkey_matches_stripe(struct bch_stripe *s,
-			       struct bkey_s_c k)
+static const struct bch_extent_ptr *bkey_matches_stripe(struct bch_stripe *s,
+						struct bkey_s_c k, unsigned *block)
 {
 	struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 	const struct bch_extent_ptr *ptr;
@@ -153,10 +165,12 @@ static int bkey_matches_stripe(struct bch_stripe *s,
 	bkey_for_each_ptr(ptrs, ptr)
 		for (i = 0; i < nr_data; i++)
 			if (__bch2_ptr_matches_stripe(&s->ptrs[i], ptr,
-						      le16_to_cpu(s->sectors)))
-				return i;
+						      le16_to_cpu(s->sectors))) {
+				*block = i;
+				return ptr;
+			}
 
-	return -1;
+	return NULL;
 }
 
 static bool extent_has_stripe_ptr(struct bkey_s_c k, u64 idx)
@@ -284,14 +298,15 @@ static void ec_validate_checksums(struct bch_fs *c, struct ec_stripe_buf *buf)
 			struct bch_csum got = ec_block_checksum(buf, i, offset);
 
 			if (bch2_crc_cmp(want, got)) {
-				char buf2[200];
+				struct printbuf buf2 = PRINTBUF;
 
-				bch2_bkey_val_to_text(&PBUF(buf2), c, bkey_i_to_s_c(&buf->key.k_i));
+				bch2_bkey_val_to_text(&buf2, c, bkey_i_to_s_c(&buf->key.k_i));
 
 				bch_err_ratelimited(c,
 					"stripe checksum error for %ps at %u:%u: csum type %u, expected %llx got %llx\n%s",
 					(void *) _RET_IP_, i, j, v->csum_type,
-					want.lo, got.lo, buf2);
+					want.lo, got.lo, buf2.buf);
+				printbuf_exit(&buf2);
 				clear_bit(i, buf->valid);
 				break;
 			}
@@ -399,7 +414,10 @@ static void ec_block_io(struct bch_fs *c, struct ec_stripe_buf *buf,
 				   nr_iovecs << PAGE_SHIFT);
 		struct ec_bio *ec_bio;
 
-		ec_bio = container_of(bio_alloc_bioset(GFP_KERNEL, nr_iovecs,
+		ec_bio = container_of(bio_alloc_bioset(ca->disk_sb.bdev,
+						       nr_iovecs,
+						       rw,
+						       GFP_KERNEL,
 						       &c->ec_bioset),
 				      struct ec_bio, bio);
 
@@ -407,9 +425,6 @@ static void ec_block_io(struct bch_fs *c, struct ec_stripe_buf *buf,
 		ec_bio->buf			= buf;
 		ec_bio->idx			= idx;
 
-		bio_set_dev(&ec_bio->bio, ca->disk_sb.bdev);
-		bio_set_op_attrs(&ec_bio->bio, rw, 0);
-
 		ec_bio->bio.bi_iter.bi_sector	= ptr->offset + buf->offset + (offset >> 9);
 		ec_bio->bio.bi_end_io		= ec_block_endio;
 		ec_bio->bio.bi_private		= cl;
@@ -559,18 +574,14 @@ static int ec_stripe_mem_alloc(struct btree_trans *trans,
 			       struct btree_iter *iter)
 {
 	size_t idx = iter->pos.offset;
-	int ret = 0;
 
 	if (!__ec_stripe_mem_alloc(trans->c, idx, GFP_NOWAIT|__GFP_NOWARN))
-		return ret;
+		return 0;
 
 	bch2_trans_unlock(trans);
-	ret = -EINTR;
-
-	if (!__ec_stripe_mem_alloc(trans->c, idx, GFP_KERNEL))
-		return ret;
 
-	return -ENOMEM;
+	return   __ec_stripe_mem_alloc(trans->c, idx, GFP_KERNEL) ?:
+		bch2_trans_relock(trans);
 }
 
 static ssize_t stripe_idx_to_delete(struct bch_fs *c)
@@ -663,9 +674,8 @@ void bch2_stripes_heap_update(struct bch_fs *c,
 
 	heap_verify_backpointer(c, idx);
 
-	if (stripe_idx_to_delete(c) >= 0 &&
-	    !percpu_ref_is_dying(&c->writes))
-		schedule_work(&c->ec_stripe_delete_work);
+	if (stripe_idx_to_delete(c) >= 0)
+		bch2_do_stripe_deletes(c);
 }
 
 /* stripe deletion */
@@ -674,8 +684,8 @@ static int ec_stripe_delete(struct bch_fs *c, size_t idx)
 {
 	return bch2_btree_delete_range(c, BTREE_ID_stripes,
 				       POS(0, idx),
-				       POS(0, idx + 1),
-				       NULL);
+				       POS(0, idx),
+				       0, NULL);
 }
 
 static void ec_stripe_delete_work(struct work_struct *work)
@@ -698,6 +708,15 @@ static void ec_stripe_delete_work(struct work_struct *work)
 		if (ec_stripe_delete(c, idx))
 			break;
 	}
+
+	bch2_write_ref_put(c, BCH_WRITE_REF_stripe_delete);
+}
+
+void bch2_do_stripe_deletes(struct bch_fs *c)
+{
+	if (bch2_write_ref_tryget(c, BCH_WRITE_REF_stripe_delete) &&
+	    !schedule_work(&c->ec_stripe_delete_work))
+		bch2_write_ref_put(c, BCH_WRITE_REF_stripe_delete);
 }
 
 /* stripe creation: */
@@ -713,26 +732,27 @@ static int ec_stripe_bkey_insert(struct btree_trans *trans,
 	struct bpos start_pos = bpos_max(min_pos, POS(0, c->ec_stripe_hint));
 	int ret;
 
-	for_each_btree_key(trans, iter, BTREE_ID_stripes, start_pos,
+	for_each_btree_key_norestart(trans, iter, BTREE_ID_stripes, start_pos,
 			   BTREE_ITER_SLOTS|BTREE_ITER_INTENT, k, ret) {
-		if (bkey_cmp(k.k->p, POS(0, U32_MAX)) > 0) {
+		if (bkey_gt(k.k->p, POS(0, U32_MAX))) {
 			if (start_pos.offset) {
 				start_pos = min_pos;
 				bch2_btree_iter_set_pos(&iter, start_pos);
 				continue;
 			}
 
-			ret = -ENOSPC;
+			ret = -BCH_ERR_ENOSPC_stripe_create;
 			break;
 		}
 
 		if (bkey_deleted(k.k))
-			goto found_slot;
+			break;
 	}
 
-	goto err;
-found_slot:
-	start_pos = iter.pos;
+	c->ec_stripe_hint = iter.pos.offset;
+
+	if (ret)
+		goto err;
 
 	ret = ec_stripe_mem_alloc(trans, &iter);
 	if (ret)
@@ -741,8 +761,6 @@ found_slot:
 	stripe->k.p = iter.pos;
 
 	ret = bch2_trans_update(trans, &iter, &stripe->k_i, 0);
-
-	c->ec_stripe_hint = start_pos.offset;
 err:
 	bch2_trans_iter_exit(trans, &iter);
 
@@ -809,73 +827,112 @@ static void extent_stripe_ptr_add(struct bkey_s_extent e,
 	};
 }
 
-static int ec_stripe_update_ptrs(struct bch_fs *c,
-				 struct ec_stripe_buf *s,
-				 struct bkey *pos)
+static int ec_stripe_update_extent(struct btree_trans *trans,
+				   struct btree_iter *iter,
+				   struct bkey_s_c k,
+				   struct ec_stripe_buf *s)
 {
-	struct btree_trans trans;
-	struct btree_iter iter;
-	struct bkey_s_c k;
-	struct bkey_s_extent e;
-	struct bkey_buf sk;
-	struct bpos next_pos;
-	int ret = 0, dev, block;
+	const struct bch_extent_ptr *ptr_c;
+	struct bch_extent_ptr *ptr, *ec_ptr = NULL;
+	struct bkey_i *n;
+	int ret, dev, block;
 
-	bch2_bkey_buf_init(&sk);
-	bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
+	if (extent_has_stripe_ptr(k, s->key.k.p.offset))
+		return 0;
 
-	/* XXX this doesn't support the reflink btree */
+	ptr_c = bkey_matches_stripe(&s->key.v, k, &block);
+	/*
+	 * It doesn't generally make sense to erasure code cached ptrs:
+	 * XXX: should we be incrementing a counter?
+	 */
+	if (!ptr_c || ptr_c->cached)
+		return 0;
+
+	dev = s->key.v.ptrs[block].dev;
+
+	n = bch2_bkey_make_mut(trans, k);
+	ret = PTR_ERR_OR_ZERO(n);
+	if (ret)
+		return ret;
 
-	bch2_trans_iter_init(&trans, &iter, BTREE_ID_extents,
-			     bkey_start_pos(pos),
-			     BTREE_ITER_INTENT);
+	bch2_bkey_drop_ptrs(bkey_i_to_s(n), ptr, ptr->dev != dev);
+	ec_ptr = (void *) bch2_bkey_has_device(bkey_i_to_s_c(n), dev);
+	BUG_ON(!ec_ptr);
+
+	extent_stripe_ptr_add(bkey_i_to_s_extent(n), s, ec_ptr, block);
+
+	return bch2_trans_update(trans, iter, n, 0);
+}
+
+static int ec_stripe_update_bucket(struct btree_trans *trans, struct ec_stripe_buf *s,
+				   unsigned block)
+{
+	struct bch_fs *c = trans->c;
+	struct bch_extent_ptr bucket = s->key.v.ptrs[block];
+	struct bpos bucket_pos = PTR_BUCKET_POS(c, &bucket);
+	struct bch_backpointer bp;
+	struct btree_iter iter;
+	struct bkey_s_c k;
+	u64 bp_offset = 0;
+	int ret = 0;
 retry:
-	while (bch2_trans_begin(&trans),
-	       (k = bch2_btree_iter_peek(&iter)).k &&
-	       !(ret = bkey_err(k)) &&
-	       bkey_cmp(bkey_start_pos(k.k), pos->p) < 0) {
-		struct bch_extent_ptr *ptr, *ec_ptr = NULL;
-
-		if (extent_has_stripe_ptr(k, s->key.k.p.offset)) {
-			bch2_btree_iter_advance(&iter);
-			continue;
+	while (1) {
+		bch2_trans_begin(trans);
+
+		ret = bch2_get_next_backpointer(trans, bucket_pos, bucket.gen,
+						&bp_offset, &bp,
+						BTREE_ITER_CACHED);
+		if (ret)
+			break;
+		if (bp_offset == U64_MAX)
+			break;
+
+		if (bch2_fs_inconsistent_on(bp.level, c, "found btree node in erasure coded bucket!?")) {
+			ret = -EIO;
+			break;
 		}
 
-		block = bkey_matches_stripe(&s->key.v, k);
-		if (block < 0) {
-			bch2_btree_iter_advance(&iter);
+		k = bch2_backpointer_get_key(trans, &iter, bucket_pos, bp_offset, bp);
+		ret = bkey_err(k);
+		if (ret)
+			break;
+		if (!k.k)
 			continue;
-		}
 
-		dev = s->key.v.ptrs[block].dev;
+		ret = ec_stripe_update_extent(trans, &iter, k, s);
+		bch2_trans_iter_exit(trans, &iter);
+		if (ret)
+			break;
+
+		bp_offset++;
+	}
+
+	if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
+		goto retry;
 
-		bch2_bkey_buf_reassemble(&sk, c, k);
-		e = bkey_i_to_s_extent(sk.k);
+	return ret;
+}
 
-		bch2_bkey_drop_ptrs(e.s, ptr, ptr->dev != dev);
-		ec_ptr = (void *) bch2_bkey_has_device(e.s_c, dev);
-		BUG_ON(!ec_ptr);
+static int ec_stripe_update_extents(struct bch_fs *c, struct ec_stripe_buf *s)
+{
+	struct btree_trans trans;
+	struct bch_stripe *v = &s->key.v;
+	unsigned i, nr_data = v->nr_blocks - v->nr_redundant;
+	int ret = 0;
 
-		extent_stripe_ptr_add(e, s, ec_ptr, block);
+	bch2_trans_init(&trans, c, 0, 0);
 
-		bch2_btree_iter_set_pos(&iter, bkey_start_pos(&sk.k->k));
-		next_pos = sk.k->k.p;
+	ret = bch2_btree_write_buffer_flush(&trans);
+	if (ret)
+		goto err;
 
-		ret   = bch2_btree_iter_traverse(&iter) ?:
-			bch2_trans_update(&trans, &iter, sk.k, 0) ?:
-			bch2_trans_commit(&trans, NULL, NULL,
-					BTREE_INSERT_NOFAIL);
-		if (!ret)
-			bch2_btree_iter_set_pos(&iter, next_pos);
+	for (i = 0; i < nr_data; i++) {
+		ret = ec_stripe_update_bucket(&trans, s, i);
 		if (ret)
 			break;
 	}
-	if (ret == -EINTR)
-		goto retry;
-	bch2_trans_iter_exit(&trans, &iter);
-
+err:
 	bch2_trans_exit(&trans);
-	bch2_bkey_buf_exit(&sk, c);
 
 	return ret;
 }
@@ -887,7 +944,6 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 {
 	struct bch_fs *c = s->c;
 	struct open_bucket *ob;
-	struct bkey_i *k;
 	struct stripe *m;
 	struct bch_stripe *v = &s->new_stripe.key.v;
 	unsigned i, nr_data = v->nr_blocks - v->nr_redundant;
@@ -898,7 +954,7 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 	closure_sync(&s->iodone);
 
 	if (s->err) {
-		if (s->err != -EROFS)
+		if (!bch2_err_matches(s->err, EROFS))
 			bch_err(c, "error creating stripe: error writing data buckets");
 		goto err;
 	}
@@ -921,7 +977,7 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 
 	BUG_ON(!s->allocated);
 
-	if (!percpu_ref_tryget(&c->writes))
+	if (!bch2_write_ref_tryget(c, BCH_WRITE_REF_stripe_create))
 		goto err;
 
 	ec_generate_ec(&s->new_stripe);
@@ -947,13 +1003,10 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 		goto err_put_writes;
 	}
 
-	for_each_keylist_key(&s->keys, k) {
-		ret = ec_stripe_update_ptrs(c, &s->new_stripe, &k->k);
-		if (ret) {
-			bch_err(c, "error creating stripe: error %i updating pointers", ret);
-			break;
-		}
-	}
+	ret = ec_stripe_update_extents(c, &s->new_stripe);
+	if (ret)
+		bch_err(c, "error creating stripe: error updating pointers: %s",
+			bch2_err_str(ret));
 
 	spin_lock(&c->ec_stripes_heap_lock);
 	m = genradix_ptr(&c->stripes, s->new_stripe.key.k.p.offset);
@@ -962,7 +1015,7 @@ static void ec_stripe_create(struct ec_stripe_new *s)
 	bch2_stripes_heap_insert(c, m, s->new_stripe.key.k.p.offset);
 	spin_unlock(&c->ec_stripes_heap_lock);
 err_put_writes:
-	percpu_ref_put(&c->writes);
+	bch2_write_ref_put(c, BCH_WRITE_REF_stripe_create);
 err:
 	bch2_disk_reservation_put(c, &s->res);
 
@@ -978,8 +1031,6 @@ err:
 			}
 		}
 
-	bch2_keylist_free(&s->keys, s->inline_keys);
-
 	ec_stripe_buf_exit(&s->existing_stripe);
 	ec_stripe_buf_exit(&s->new_stripe);
 	closure_debug_destroy(&s->iodone);
@@ -1056,36 +1107,12 @@ void *bch2_writepoint_ec_buf(struct bch_fs *c, struct write_point *wp)
 	if (!ob)
 		return NULL;
 
-	ca	= bch_dev_bkey_exists(c, ob->ptr.dev);
+	ca	= bch_dev_bkey_exists(c, ob->dev);
 	offset	= ca->mi.bucket_size - ob->sectors_free;
 
 	return ob->ec->new_stripe.data[ob->ec_idx] + (offset << 9);
 }
 
-void bch2_ob_add_backpointer(struct bch_fs *c, struct open_bucket *ob,
-			     struct bkey *k)
-{
-	struct ec_stripe_new *ec = ob->ec;
-
-	if (!ec)
-		return;
-
-	mutex_lock(&ec->lock);
-
-	if (bch2_keylist_realloc(&ec->keys, ec->inline_keys,
-				 ARRAY_SIZE(ec->inline_keys),
-				 BKEY_U64s)) {
-		BUG();
-	}
-
-	bkey_init(&ec->keys.top->k);
-	ec->keys.top->k.p	= k->p;
-	ec->keys.top->k.size	= k->size;
-	bch2_keylist_push(&ec->keys);
-
-	mutex_unlock(&ec->lock);
-}
-
 static int unsigned_cmp(const void *_l, const void *_r)
 {
 	unsigned l = *((const unsigned *) _l);
@@ -1145,7 +1172,7 @@ static void ec_stripe_key_init(struct bch_fs *c,
 	s->v.algorithm			= 0;
 	s->v.nr_blocks			= nr_data + nr_parity;
 	s->v.nr_redundant		= nr_parity;
-	s->v.csum_granularity_bits	= ilog2(c->sb.encoded_extent_max);
+	s->v.csum_granularity_bits	= ilog2(c->opts.encoded_extent_max >> 9);
 	s->v.csum_type			= BCH_CSUM_crc32c;
 	s->v.pad			= 0;
 
@@ -1178,8 +1205,6 @@ static int ec_new_stripe_alloc(struct bch_fs *c, struct ec_stripe_head *h)
 				BCH_BKEY_PTRS_MAX) - h->redundancy;
 	s->nr_parity	= h->redundancy;
 
-	bch2_keylist_init(&s->keys, s->inline_keys);
-
 	ec_stripe_key_init(c, &s->new_stripe.key, s->nr_data,
 			   s->nr_parity, h->blocksize);
 
@@ -1287,9 +1312,6 @@ static int new_stripe_alloc_buckets(struct bch_fs *c, struct ec_stripe_head *h,
 	BUG_ON(nr_have_data	> h->s->nr_data);
 	BUG_ON(nr_have_parity	> h->s->nr_parity);
 
-	percpu_down_read(&c->mark_lock);
-	rcu_read_lock();
-
 	buckets.nr = 0;
 	if (nr_have_parity < h->s->nr_parity) {
 		ret = bch2_bucket_alloc_set(c, &buckets,
@@ -1299,8 +1321,8 @@ static int new_stripe_alloc_buckets(struct bch_fs *c, struct ec_stripe_head *h,
 					    &nr_have_parity,
 					    &have_cache,
 					    h->copygc
-					    ? RESERVE_MOVINGGC
-					    : RESERVE_NONE,
+					    ? RESERVE_movinggc
+					    : RESERVE_none,
 					    0,
 					    cl);
 
@@ -1311,12 +1333,12 @@ static int new_stripe_alloc_buckets(struct bch_fs *c, struct ec_stripe_head *h,
 			BUG_ON(j >= h->s->nr_data + h->s->nr_parity);
 
 			h->s->blocks[j] = buckets.v[i];
-			h->s->new_stripe.key.v.ptrs[j] = ob->ptr;
+			h->s->new_stripe.key.v.ptrs[j] = bch2_ob_ptr(c, ob);
 			__set_bit(j, h->s->blocks_gotten);
 		}
 
 		if (ret)
-			goto err;
+			return ret;
 	}
 
 	buckets.nr = 0;
@@ -1328,8 +1350,8 @@ static int new_stripe_alloc_buckets(struct bch_fs *c, struct ec_stripe_head *h,
 					    &nr_have_data,
 					    &have_cache,
 					    h->copygc
-					    ? RESERVE_MOVINGGC
-					    : RESERVE_NONE,
+					    ? RESERVE_movinggc
+					    : RESERVE_none,
 					    0,
 					    cl);
 
@@ -1339,17 +1361,15 @@ static int new_stripe_alloc_buckets(struct bch_fs *c, struct ec_stripe_head *h,
 			BUG_ON(j >= h->s->nr_data);
 
 			h->s->blocks[j] = buckets.v[i];
-			h->s->new_stripe.key.v.ptrs[j] = ob->ptr;
+			h->s->new_stripe.key.v.ptrs[j] = bch2_ob_ptr(c, ob);
 			__set_bit(j, h->s->blocks_gotten);
 		}
 
 		if (ret)
-			goto err;
+			return ret;
 	}
-err:
-	rcu_read_unlock();
-	percpu_up_read(&c->mark_lock);
-	return ret;
+
+	return 0;
 }
 
 /* XXX: doesn't obey target: */
@@ -1395,10 +1415,8 @@ static int __bch2_ec_stripe_head_reuse(struct bch_fs *c,
 	int ret;
 
 	idx = get_existing_stripe(c, h);
-	if (idx < 0) {
-		bch_err(c, "failed to find an existing stripe");
-		return -ENOSPC;
-	}
+	if (idx < 0)
+		return -BCH_ERR_ENOSPC_stripe_reuse;
 
 	h->s->have_existing_stripe = true;
 	ret = get_stripe_key(c, idx, &h->s->existing_stripe);
@@ -1436,21 +1454,9 @@ static int __bch2_ec_stripe_head_reuse(struct bch_fs *c,
 static int __bch2_ec_stripe_head_reserve(struct bch_fs *c,
 							struct ec_stripe_head *h)
 {
-	int ret;
-
-	ret = bch2_disk_reservation_get(c, &h->s->res,
-			h->blocksize,
-			h->s->nr_parity, 0);
-
-	if (ret) {
-		/*
-		 * This means we need to wait for copygc to
-		 * empty out buckets from existing stripes:
-		 */
-		bch_err(c, "failed to reserve stripe");
-	}
-
-	return ret;
+	return bch2_disk_reservation_get(c, &h->s->res,
+					 h->blocksize,
+					 h->s->nr_parity, 0);
 }
 
 struct ec_stripe_head *bch2_ec_stripe_head_get(struct bch_fs *c,
@@ -1492,8 +1498,10 @@ struct ec_stripe_head *bch2_ec_stripe_head_get(struct bch_fs *c,
 		ret = __bch2_ec_stripe_head_reserve(c, h);
 	if (ret && needs_stripe_new)
 		ret = __bch2_ec_stripe_head_reuse(c, h);
-	if (ret)
+	if (ret) {
+		bch_err_ratelimited(c, "failed to get stripe: %s", bch2_err_str(ret));
 		goto err;
+	}
 
 	if (!h->s->allocated) {
 		ret = new_stripe_alloc_buckets(c, h, cl);
@@ -1528,7 +1536,7 @@ void bch2_ec_stop_dev(struct bch_fs *c, struct bch_dev *ca)
 				continue;
 
 			ob = c->open_buckets + h->s->blocks[i];
-			if (ob->ptr.dev == ca->dev_idx)
+			if (ob->dev == ca->dev_idx)
 				goto found;
 		}
 		goto unlock;
@@ -1551,96 +1559,54 @@ void bch2_stripes_heap_start(struct bch_fs *c)
 			bch2_stripes_heap_insert(c, m, iter.pos);
 }
 
-static int bch2_stripes_read_fn(struct btree_trans *trans, struct bkey_s_c k)
+int bch2_stripes_read(struct bch_fs *c)
 {
+	struct btree_trans trans;
+	struct btree_iter iter;
+	struct bkey_s_c k;
 	const struct bch_stripe *s;
-	struct bch_fs *c = trans->c;
 	struct stripe *m;
 	unsigned i;
-	int ret = 0;
-
-	if (k.k->type != KEY_TYPE_stripe)
-		return 0;
+	int ret;
 
-	ret = __ec_stripe_mem_alloc(c, k.k->p.offset, GFP_KERNEL);
-	if (ret)
-		return ret;
+	bch2_trans_init(&trans, c, 0, 0);
 
-	s = bkey_s_c_to_stripe(k).v;
+	for_each_btree_key(&trans, iter, BTREE_ID_stripes, POS_MIN,
+			   BTREE_ITER_PREFETCH, k, ret) {
+		if (k.k->type != KEY_TYPE_stripe)
+			continue;
 
-	m = genradix_ptr(&c->stripes, k.k->p.offset);
-	m->alive	= true;
-	m->sectors	= le16_to_cpu(s->sectors);
-	m->algorithm	= s->algorithm;
-	m->nr_blocks	= s->nr_blocks;
-	m->nr_redundant	= s->nr_redundant;
-	m->blocks_nonempty = 0;
+		ret = __ec_stripe_mem_alloc(c, k.k->p.offset, GFP_KERNEL);
+		if (ret)
+			break;
 
-	for (i = 0; i < s->nr_blocks; i++)
-		m->blocks_nonempty += !!stripe_blockcount_get(s, i);
+		s = bkey_s_c_to_stripe(k).v;
 
-	spin_lock(&c->ec_stripes_heap_lock);
-	bch2_stripes_heap_update(c, m, k.k->p.offset);
-	spin_unlock(&c->ec_stripes_heap_lock);
+		m = genradix_ptr(&c->stripes, k.k->p.offset);
+		m->alive	= true;
+		m->sectors	= le16_to_cpu(s->sectors);
+		m->algorithm	= s->algorithm;
+		m->nr_blocks	= s->nr_blocks;
+		m->nr_redundant	= s->nr_redundant;
+		m->blocks_nonempty = 0;
 
-	return ret;
-}
+		for (i = 0; i < s->nr_blocks; i++)
+			m->blocks_nonempty += !!stripe_blockcount_get(s, i);
 
-int bch2_stripes_read(struct bch_fs *c)
-{
-	struct btree_trans trans;
-	int ret;
+		spin_lock(&c->ec_stripes_heap_lock);
+		bch2_stripes_heap_update(c, m, k.k->p.offset);
+		spin_unlock(&c->ec_stripes_heap_lock);
+	}
+	bch2_trans_iter_exit(&trans, &iter);
 
-	bch2_trans_init(&trans, c, 0, 0);
-	ret = bch2_btree_and_journal_walk(&trans, BTREE_ID_stripes,
-					  bch2_stripes_read_fn);
 	bch2_trans_exit(&trans);
+
 	if (ret)
 		bch_err(c, "error reading stripes: %i", ret);
 
 	return ret;
 }
 
-int bch2_ec_mem_alloc(struct bch_fs *c, bool gc)
-{
-	struct btree_trans trans;
-	struct btree_iter iter;
-	struct bkey_s_c k;
-	size_t i, idx = 0;
-	int ret = 0;
-
-	bch2_trans_init(&trans, c, 0, 0);
-	bch2_trans_iter_init(&trans, &iter, BTREE_ID_stripes, POS(0, U64_MAX), 0);
-
-	k = bch2_btree_iter_prev(&iter);
-	ret = bkey_err(k);
-	if (!ret && k.k)
-		idx = k.k->p.offset + 1;
-
-	bch2_trans_iter_exit(&trans, &iter);
-	bch2_trans_exit(&trans);
-	if (ret)
-		return ret;
-
-	if (!idx)
-		return 0;
-
-	if (!gc &&
-	    !init_heap(&c->ec_stripes_heap, roundup_pow_of_two(idx),
-		       GFP_KERNEL))
-		return -ENOMEM;
-#if 0
-	ret = genradix_prealloc(&c->stripes[gc], idx, GFP_KERNEL);
-#else
-	for (i = 0; i < idx; i++)
-		if (!gc
-		    ? !genradix_ptr_alloc(&c->stripes, i, GFP_KERNEL)
-		    : !genradix_ptr_alloc(&c->gc_stripes, i, GFP_KERNEL))
-			return -ENOMEM;
-#endif
-	return 0;
-}
-
 void bch2_stripes_heap_to_text(struct printbuf *out, struct bch_fs *c)
 {
 	ec_stripes_heap *h = &c->ec_stripes_heap;
@@ -1651,7 +1617,7 @@ void bch2_stripes_heap_to_text(struct printbuf *out, struct bch_fs *c)
 	for (i = 0; i < min_t(size_t, h->used, 20); i++) {
 		m = genradix_ptr(&c->stripes, h->data[i].idx);
 
-		pr_buf(out, "%zu %u/%u+%u\n", h->data[i].idx,
+		prt_printf(out, "%zu %u/%u+%u\n", h->data[i].idx,
 		       h->data[i].blocks_nonempty,
 		       m->nr_blocks - m->nr_redundant,
 		       m->nr_redundant);
@@ -1666,11 +1632,11 @@ void bch2_new_stripes_to_text(struct printbuf *out, struct bch_fs *c)
 
 	mutex_lock(&c->ec_stripe_head_lock);
 	list_for_each_entry(h, &c->ec_stripe_head_list, list) {
-		pr_buf(out, "target %u algo %u redundancy %u:\n",
+		prt_printf(out, "target %u algo %u redundancy %u:\n",
 		       h->target, h->algo, h->redundancy);
 
 		if (h->s)
-			pr_buf(out, "\tpending: blocks %u+%u allocated %u\n",
+			prt_printf(out, "\tpending: blocks %u+%u allocated %u\n",
 			       h->s->nr_data, h->s->nr_parity,
 			       bitmap_weight(h->s->blocks_allocated,
 					     h->s->nr_data));
@@ -1679,7 +1645,7 @@ void bch2_new_stripes_to_text(struct printbuf *out, struct bch_fs *c)
 
 	mutex_lock(&c->ec_stripe_new_lock);
 	list_for_each_entry(s, &c->ec_stripe_new_list, list) {
-		pr_buf(out, "\tin flight: blocks %u+%u pin %u\n",
+		prt_printf(out, "\tin flight: blocks %u+%u pin %u\n",
 		       s->nr_data, s->nr_parity,
 		       atomic_read(&s->pin));
 	}
@@ -1711,11 +1677,14 @@ void bch2_fs_ec_exit(struct bch_fs *c)
 	bioset_exit(&c->ec_bioset);
 }
 
-int bch2_fs_ec_init(struct bch_fs *c)
+void bch2_fs_ec_init_early(struct bch_fs *c)
 {
 	INIT_WORK(&c->ec_stripe_create_work, ec_stripe_create_work);
 	INIT_WORK(&c->ec_stripe_delete_work, ec_stripe_delete_work);
+}
 
+int bch2_fs_ec_init(struct bch_fs *c)
+{
 	return bioset_init(&c->ec_bioset, 1, offsetof(struct ec_bio, bio),
 			   BIOSET_NEED_BVECS);
 }