Fix single active logblock allocation for logregion-only case

Should be irrelevant in practice, because we would not flush
a single block during startup.

src/fellow_log.c

@@ -2412,15 +2412,21 @@ logbuffer_flush(struct fellow_fd *ffd,
 		if (! cap(can, LBUF_CAN_REF))
 			return;
 
-		logbuffer_need_ioctx(ffd, lbuf);
-
 		// the active block is the head block
 		AZ(lbuf->n);
 
 		CHECK_BLK_SEQ(lbuf->active.block, lbuf->flush_seq);
 
-		if (lbuf->active.off == 0)
+		if (lbuf->active.off == 0 && cap(can, LBUF_CAN_FLUSH))
 			logbuffer_alloc_some(lbuf, &lbuf->active, 1);
+		if (lbuf->active.off == 0 &&
+		    logblocks_alloc_from_logregion(lbuf->logreg,
+		    &lbuf->active, 1) != 1) {
+			lbuf->state = LBUF_MEM;
+			return;
+		}
+
+		logbuffer_need_ioctx(ffd, lbuf);
 
 		while (! flush_active(ffd, lbuf, doclose ? FAIO_CLOSE : 0))
 			(void) fellow_io_log_wait_completions(