From 56ba714e1039ebc128e533c408b3b936dcd09b5d Mon Sep 17 00:00:00 2001
From: Josh Bleecher Snyder <josh@tailscale.com>
Date: Tue, 1 Jun 2021 22:11:40 -0700
Subject: [PATCH] smush together two return values from C

avoids a per-packet alloc

i will atone for my sins later

if only C let you return multiple values.
or Go let you pass in a pointer w/o it being on the heap.
---
 net/uring/io_uring.c        | 17 ++++++++++-------
 net/uring/io_uring_linux.go |  9 +++++----
 2 files changed, 15 insertions(+), 11 deletions(-)

diff --git a/net/uring/io_uring.c b/net/uring/io_uring.c
index e3eb73e06..5f8b4a099 100644
--- a/net/uring/io_uring.c
+++ b/net/uring/io_uring.c
@@ -20,7 +20,7 @@ typedef struct iovec go_iovec;
 typedef struct sockaddr_in go_sockaddr_in;
 
 // Wait for a completion to be available, fetch the data
-static int receive_into(struct io_uring *ring, size_t *idxptr) {
+static uint64_t receive_into(struct io_uring *ring) {
     struct io_uring_cqe *cqe;
 again:;
 
@@ -38,15 +38,18 @@ again:;
         fprintf(stderr, "recvmsg failed: %d.\n", cqe->res);
         return cqe->res;
     }
-    *idxptr = (size_t)(io_uring_cqe_get_data(cqe));
-    if (*idxptr < 0) {
+    size_t idxptr = (size_t)(io_uring_cqe_get_data(cqe));
+    uint32_t idxptr32 = (uint32_t)(idxptr);
+    uint64_t idxptr64 = (uint64_t)(idxptr32);
+    uint64_t n = cqe->res;
+    uint32_t n32 = (uint32_t)n;
+    uint64_t n64 = (uint64_t)n;
+    io_uring_cqe_seen(ring, cqe);
+    if (idxptr < 0) {
         fprintf(stderr, "received nop\n");
-        io_uring_cqe_seen(ring, cqe);
         return -1;
     }
-    int n = cqe->res;
-    io_uring_cqe_seen(ring, cqe);
-    return n;
+    return (n64 << 32) | idxptr64;
 }
 
 static uint32_t ip(struct sockaddr_in *sa) {
diff --git a/net/uring/io_uring_linux.go b/net/uring/io_uring_linux.go
index 7eeb2f5c6..44ff4be09 100644
--- a/net/uring/io_uring_linux.go
+++ b/net/uring/io_uring_linux.go
@@ -56,7 +56,7 @@ func NewUDPConn(conn *net.UDPConn) (*UDPConn, error) {
 	}
 	r := new(C.go_uring)
 
-	const queue_depth = 8 // TODO: What value to use here?
+	const queue_depth = 16 // TODO: What value to use here?
 	C.io_uring_queue_init(queue_depth, r, 0)
 	u := &UDPConn{
 		ptr:   r,
@@ -95,11 +95,12 @@ func (u *UDPConn) ReadFromNetaddr(buf []byte) (int, netaddr.IPPort, error) {
 	if u.fd == 0 {
 		return 0, netaddr.IPPort{}, errors.New("invalid uring.UDPConn")
 	}
-	var idx C.size_t
-	n := C.receive_into(u.ptr, &idx)
-	if n < 0 {
+	nidx := C.receive_into(u.ptr)
+	if int64(nidx) == -1 {
 		return 0, netaddr.IPPort{}, errors.New("something wrong")
 	}
+	idx := uint32(nidx)
+	n := uint32(nidx >> 32)
 	r := &u.reqs[int(idx)]
 	ip := C.ip(&r.sa)
 	var ip4 [4]byte