commit 1deaf3979efdb3e27a57aeb2d7c93c0b7550639d
from: Omar Polo <op@omarpolo.com>
date: Sun Mar 14 18:10:27 2021 UTC

set the off to zero after flushing the buffer

this was the cause we had duplicated "chunks" in the pages: if we
enter copy_body with req->off != 0 but then tls_read returns
WANT_POLLIN/POLLOUT, the next time we flush the same chunk again.

commit - e430495420ca83b7c02175962e6e7fc38f028856
commit + 1deaf3979efdb3e27a57aeb2d7c93c0b7550639d
blob - c8a306c49553d05da19cacfd56928eda7bde96a2
blob + 479ee40078ca0f36bcd7a281d92b868e0d298a95
--- gemini.c
+++ gemini.c
@@ -384,6 +384,7 @@ copy_body(int fd, short ev, void *d)
 			imsg_compose(ibuf, IMSG_BUF, req->id, 0, -1,
 			    req->buf, req->off);
 			imsg_flush(ibuf);
+			req->off = 0;
 		}
 
 		switch (r = tls_read(req->ctx, req->buf, sizeof(req->buf))) {