net/igmp: fix checksum validation that always dropped valid IGMP packets

[zhekunren]

Summary

igmp_input() rejected every well-formed IGMP packet as a checksum error because the validation compared net_chksum() against 0 instead of 0xffff. This breaks IGMP membership query/report processing entirely whenever CONFIG_NET_IGMP_CHECKSUMS is enabled.

Problem

In net/igmp/igmp_input.c, the IGMP checksum was validated as:

#ifdef CONFIG_NET_IGMP_CHECKSUMS
  if (net_chksum((FAR uint16_t *)igmp, IGMP_HDRLEN) != 0)
    {
      IGMP_STATINCR(g_netstats.igmp.chksum_errors);
      nwarn("WARNING: Checksum error\n");
      goto drop;
    }
#endif

With CONFIG_NET_IGMP_CHECKSUMS enabled, every valid IGMP message hits the goto drop path and is silently discarded, so the host never answers Membership Queries and never processes Membership Reports from neighbors.

Root Cause

net_chksum() (in net/utils/net_chksum.c) returns the raw one's complement sum of all 16-bit words over the buffer — it does not take the one's complement of that sum:

uint16_t net_chksum(FAR uint16_t *data, uint16_t len)
{
  return HTONS(chksum(0, (uint8_t *)data, len));
}

For a valid IGMP packet, the sender (net/igmp/igmp_send.c) stores igmp->chksum = ~igmp_chksum(...). Let S be the one's complement sum of the header with the checksum field zeroed. The received packet therefore contains ~S in the checksum field, and summing all 16-bit words yields:

S + ~S = 0xffff   (one's complement arithmetic)

So the correct validation is != 0xffff, not != 0. The old check != 0 is true for every legal IGMP packet (it can only be 0 in the degenerate case where the sum is exactly 0), so all valid packets were dropped.

This is the same Internet checksum principle used by IPv4 (RFC 1071) and TCP/UDP.

Fix

Compare against 0xffff:

  if (net_chksum((FAR uint16_t *)igmp, IGMP_HDRLEN) != 0xffff)

Consistency

This matches the convention already used by the other transport input handlers in the tree:

• net/devif/ipv4_input.c — (ipv4_chksum(IPv4BUF) != 0xffff)
• net/tcp/tcp_input.c — (tcp_chksum(dev) != 0xffff)

and is symmetric with the sender side in net/igmp/igmp_send.c, which writes igmp->chksum = ~igmp_chksum(...).

Impact

• Bug fix only, no behavioral change for malformed packets (they still go to goto drop).
• Restores IGMP receive path when CONFIG_NET_IGMP_CHECKSUMS=y: Membership Queries now correctly trigger the delaying-member state machine, and Membership Reports correctly cancel the report timer.
• No impact when CONFIG_NET_IGMP_CHECKSUMS is disabled (the block is compiled out).
• Related to the earlier fix f11df565 ("net/igmp: fix length check that always dropped valid IGMP packets") — together they restore a fully broken IGMP input path.

Testing

Verified by inspection against net_chksum() semantics and cross-checked with ipv4_input.c / tcp_input.c. Code path is gated behind CONFIG_NET_IGMP_CHECKSUMS so the default build is unchanged.

[github-actions]

MemBrowse Memory Report

No memory changes detected for:

• arduino-mega2560
• esp32-devkitc
• hifive1-revb
• mirtoo
• qemu-armv8a
• qemu-intel64
• rx65n-rsk2mb
• s698pm-dkit
• stm32-nucleo-f103rb

[acassis]

@zhekunren if you used AI to help you (based on long messages in the commit message), please include: Assisted-by: the name of AI

[zhekunren]

@zhekunren if you used AI to help you (based on long messages in the commit message), please include: Assisted-by: the name of AI
Ok. Commit tag Assisted‑by: GLM-5.2 noreply@z.ai is added as required.