+ return ob;
+}
+
+static struct open_bucket *bch2_bucket_alloc_freelist(struct btree_trans *trans,
+ struct bch_dev *ca,
+ enum bch_watermark watermark,
+ struct bucket_alloc_state *s,
+ struct closure *cl)
+{
+ struct btree_iter iter;
+ struct bkey_s_c k;
+ struct open_bucket *ob = NULL;
+ u64 alloc_start = max_t(u64, ca->mi.first_bucket, READ_ONCE(ca->alloc_cursor));
+ u64 alloc_cursor = alloc_start;
+ int ret;
+
+ BUG_ON(ca->new_fs_bucket_idx);
+again:
+ for_each_btree_key_norestart(trans, iter, BTREE_ID_freespace,
+ POS(ca->dev_idx, alloc_cursor), 0, k, ret) {
+ if (k.k->p.inode != ca->dev_idx)
+ break;
+
+ for (alloc_cursor = max(alloc_cursor, bkey_start_offset(k.k));
+ alloc_cursor < k.k->p.offset;
+ alloc_cursor++) {
+ ret = btree_trans_too_many_iters(trans);
+ if (ret) {
+ ob = ERR_PTR(ret);
+ break;
+ }
+
+ s->buckets_seen++;
+
+ ob = try_alloc_bucket(trans, ca, watermark,
+ alloc_cursor, s, k, cl);
+ if (ob) {
+ set_btree_iter_dontneed(&iter);
+ break;
+ }
+ }
+
+ if (ob || ret)
+ break;
+ }
+ bch2_trans_iter_exit(trans, &iter);
+
+ ca->alloc_cursor = alloc_cursor;
+
+ if (!ob && ret)
+ ob = ERR_PTR(ret);
+
+ if (!ob && alloc_start > ca->mi.first_bucket) {
+ alloc_cursor = alloc_start = ca->mi.first_bucket;
+ goto again;
+ }
+
+ return ob;
+}
+
+/**
+ * bch2_bucket_alloc_trans - allocate a single bucket from a specific device
+ * @trans: transaction object
+ * @ca: device to allocate from
+ * @watermark: how important is this allocation?
+ * @cl: if not NULL, closure to be used to wait if buckets not available
+ * @usage: for secondarily also returning the current device usage
+ *
+ * Returns: an open_bucket on success, or an ERR_PTR() on failure.
+ */
+static struct open_bucket *bch2_bucket_alloc_trans(struct btree_trans *trans,
+ struct bch_dev *ca,
+ enum bch_watermark watermark,
+ struct closure *cl,
+ struct bch_dev_usage *usage)
+{
+ struct bch_fs *c = trans->c;
+ struct open_bucket *ob = NULL;
+ bool freespace = READ_ONCE(ca->mi.freespace_initialized);
+ u64 avail;
+ struct bucket_alloc_state s = { 0 };
+ bool waiting = false;
+again:
+ bch2_dev_usage_read_fast(ca, usage);
+ avail = dev_buckets_free(ca, *usage, watermark);
+
+ if (usage->d[BCH_DATA_need_discard].buckets > avail)
+ bch2_do_discards(c);
+
+ if (usage->d[BCH_DATA_need_gc_gens].buckets > avail)
+ bch2_do_gc_gens(c);
+
+ if (should_invalidate_buckets(ca, *usage))
+ bch2_do_invalidates(c);
+
+ if (!avail) {
+ if (cl && !waiting) {
+ closure_wait(&c->freelist_wait, cl);
+ waiting = true;
+ goto again;
+ }
+
+ track_event_change(&c->times[BCH_TIME_blocked_allocate], true);
+
+ ob = ERR_PTR(-BCH_ERR_freelist_empty);
+ goto err;
+ }
+
+ if (waiting)
+ closure_wake_up(&c->freelist_wait);
+alloc:
+ ob = likely(freespace)
+ ? bch2_bucket_alloc_freelist(trans, ca, watermark, &s, cl)
+ : bch2_bucket_alloc_early(trans, ca, watermark, &s, cl);
+
+ if (s.skipped_need_journal_commit * 2 > avail)
+ bch2_journal_flush_async(&c->journal, NULL);
+
+ if (!ob && freespace && c->curr_recovery_pass <= BCH_RECOVERY_PASS_check_alloc_info) {
+ freespace = false;
+ goto alloc;
+ }
+err:
+ if (!ob)
+ ob = ERR_PTR(-BCH_ERR_no_buckets_found);
+
+ if (!IS_ERR(ob))
+ trace_and_count(c, bucket_alloc, ca,
+ bch2_watermarks[watermark],
+ ob->bucket,
+ usage->d[BCH_DATA_free].buckets,
+ avail,
+ bch2_copygc_wait_amount(c),
+ c->copygc_wait - atomic64_read(&c->io_clock[WRITE].now),
+ &s,
+ cl == NULL,
+ "");
+ else if (!bch2_err_matches(PTR_ERR(ob), BCH_ERR_transaction_restart))
+ trace_and_count(c, bucket_alloc_fail, ca,
+ bch2_watermarks[watermark],
+ 0,
+ usage->d[BCH_DATA_free].buckets,
+ avail,
+ bch2_copygc_wait_amount(c),
+ c->copygc_wait - atomic64_read(&c->io_clock[WRITE].now),
+ &s,
+ cl == NULL,
+ bch2_err_str(PTR_ERR(ob)));
+
+ return ob;
+}
+
+struct open_bucket *bch2_bucket_alloc(struct bch_fs *c, struct bch_dev *ca,
+ enum bch_watermark watermark,
+ struct closure *cl)
+{
+ struct bch_dev_usage usage;
+ struct open_bucket *ob;
+
+ bch2_trans_do(c, NULL, NULL, 0,
+ PTR_ERR_OR_ZERO(ob = bch2_bucket_alloc_trans(trans, ca, watermark,
+ cl, &usage)));