From b7b915d173e87f00ba91022d74097ce6ecfc008e Mon Sep 17 00:00:00 2001
From: Remi Gacogne <remi.gacogne@powerdns.com>
Date: Tue, 9 Jul 2024 10:33:48 +0200
Subject: [PATCH] dnsdist: Fix wrong assertions in AF_XDP/XSK code

(cherry picked from commit e380e9d1024d5d93e8d7bd1f737af64ebaff38cb)
---
 pdns/xsk.cc | 24 ++++++++++++++++--------
 pdns/xsk.hh |  6 ++++++
 2 files changed, 22 insertions(+), 8 deletions(-)

diff --git a/pdns/xsk.cc b/pdns/xsk.cc
index f2757ebe3c27..d21c075345bd 100644
--- a/pdns/xsk.cc
+++ b/pdns/xsk.cc
@@ -729,7 +729,7 @@ void XskPacket::changeDirectAndUpdateChecksum() noexcept
     // IPV6
     auto ipv6 = getIPv6Header();
     std::swap(ipv6.daddr, ipv6.saddr);
-    assert(ipv6.nexthdr == IPPROTO_UDP);
+    ipv6.nexthdr = IPPROTO_UDP;
 
     auto udp = getUDPHeader();
     std::swap(udp.dest, udp.source);
@@ -738,16 +738,18 @@ void XskPacket::changeDirectAndUpdateChecksum() noexcept
     /* needed to get the correct checksum */
     setIPv6Header(ipv6);
     setUDPHeader(udp);
-    udp.check = tcp_udp_v6_checksum(&ipv6);
+    // do not bother setting the UDP checksum: 0 is a valid value and most AF_XDP
+    // implementations do the same
+    // udp.check = tcp_udp_v6_checksum(&ipv6);
     rewriteIpv6Header(&ipv6, getFrameLen());
     setIPv6Header(ipv6);
     setUDPHeader(udp);
   }
-  else {
+  else if (ethHeader.h_proto == htons(ETH_P_IP)) {
     // IPV4
     auto ipv4 = getIPv4Header();
     std::swap(ipv4.daddr, ipv4.saddr);
-    assert(ipv4.protocol == IPPROTO_UDP);
+    ipv4.protocol = IPPROTO_UDP;
 
     auto udp = getUDPHeader();
     std::swap(udp.dest, udp.source);
@@ -756,7 +758,9 @@ void XskPacket::changeDirectAndUpdateChecksum() noexcept
     /* needed to get the correct checksum */
     setIPv4Header(ipv4);
     setUDPHeader(udp);
-    udp.check = tcp_udp_v4_checksum(&ipv4);
+    // do not bother setting the UDP checksum: 0 is a valid value and most AF_XDP
+    // implementations do the same
+    // udp.check = tcp_udp_v4_checksum(&ipv4);
     rewriteIpv4Header(&ipv4, getFrameLen());
     setIPv4Header(ipv4);
     setUDPHeader(udp);
@@ -908,9 +912,9 @@ void XskPacket::rewrite() noexcept
     auto ipHeader = getIPv4Header();
     ipHeader.daddr = to.sin4.sin_addr.s_addr;
     ipHeader.saddr = from.sin4.sin_addr.s_addr;
+    ipHeader.protocol = IPPROTO_UDP;
 
     auto udpHeader = getUDPHeader();
-    ipHeader.protocol = IPPROTO_UDP;
     udpHeader.source = from.sin4.sin_port;
     udpHeader.dest = to.sin4.sin_port;
     udpHeader.len = htons(getDataSize() + sizeof(udpHeader));
@@ -918,6 +922,8 @@ void XskPacket::rewrite() noexcept
     /* needed to get the correct checksum */
     setIPv4Header(ipHeader);
     setUDPHeader(udpHeader);
+    // do not bother setting the UDP checksum: 0 is a valid value and most AF_XDP
+    // implementations do the same
     // udpHeader.check = tcp_udp_v4_checksum(&ipHeader);
     rewriteIpv4Header(&ipHeader, getFrameLen());
     setIPv4Header(ipHeader);
@@ -929,9 +935,9 @@ void XskPacket::rewrite() noexcept
     auto ipHeader = getIPv6Header();
     memcpy(&ipHeader.daddr, &to.sin6.sin6_addr, sizeof(ipHeader.daddr));
     memcpy(&ipHeader.saddr, &from.sin6.sin6_addr, sizeof(ipHeader.saddr));
+    ipHeader.nexthdr = IPPROTO_UDP;
 
     auto udpHeader = getUDPHeader();
-    ipHeader.nexthdr = IPPROTO_UDP;
     udpHeader.source = from.sin6.sin6_port;
     udpHeader.dest = to.sin6.sin6_port;
     udpHeader.len = htons(getDataSize() + sizeof(udpHeader));
@@ -939,7 +945,9 @@ void XskPacket::rewrite() noexcept
     /* needed to get the correct checksum */
     setIPv6Header(ipHeader);
     setUDPHeader(udpHeader);
-    udpHeader.check = tcp_udp_v6_checksum(&ipHeader);
+    // do not bother setting the UDP checksum: 0 is a valid value and most AF_XDP
+    // implementations do the same
+    // udpHeader.check = tcp_udp_v6_checksum(&ipHeader);
     setIPv6Header(ipHeader);
     setUDPHeader(udpHeader);
   }
diff --git a/pdns/xsk.hh b/pdns/xsk.hh
index fdb9a24d1a0f..a10e4f716d1f 100644
--- a/pdns/xsk.hh
+++ b/pdns/xsk.hh
@@ -192,8 +192,10 @@ class XskPacket
 public:
   enum Flags : uint32_t
   {
+    /* whether the payload has been modified */
     UPDATE = 1 << 0,
     DELAY = 1 << 1,
+    /* whether the headers have already been updated */
     REWRITE = 1 << 2
   };
 
@@ -234,6 +236,7 @@ private:
   void setIPv6Header(const ipv6hdr& ipv6Header) noexcept;
   [[nodiscard]] udphdr getUDPHeader() const noexcept;
   void setUDPHeader(const udphdr& udpHeader) noexcept;
+  /* exchange the source and destination addresses (ethernet and IP) */
   void changeDirectAndUpdateChecksum() noexcept;
 
   constexpr static uint8_t DefaultTTL = 64;
@@ -250,10 +253,13 @@ public:
   [[nodiscard]] PacketBuffer cloneHeaderToPacketBuffer() const;
   void setAddr(const ComboAddress& from_, MACAddr fromMAC, const ComboAddress& to_, MACAddr toMAC) noexcept;
   bool setPayload(const PacketBuffer& buf);
+  /* rewrite the headers, usually after setAddr() and setPayload() have been called */
   void rewrite() noexcept;
   void setHeader(PacketBuffer& buf);
   XskPacket(uint8_t* frame, size_t dataSize, size_t frameSize);
   void addDelay(int relativeMilliseconds) noexcept;
+  /* if the payload have been updated, and the headers have not been rewritten, exchange the source
+     and destination addresses (ethernet and IP) and rewrite the headers */
   void updatePacket() noexcept;
   // parse IP and UDP payloads
   bool parse(bool fromSetHeader);