fix: prevent segmentation fault in ipvs_group_sync_entry

[kazeburo]

Problem

When adding a new IP address to a virtual_server_group and reloading keepalived, the checker child segfaults if the virtual server does not have a sorry_server.

Sample configuration that triggers the crash:

global_defs {
        lvs_flush
}
virtual_server_group VG_80 {
        203.0.113.35 80
        203.0.113.32 80
        203.0.113.49 80
        203.0.113.56 80
        203.0.113.33 80
}

Stack trace:

#0  ipvs_group_sync_entry (keepalived + 0x20f6a)
#1  sync_service_vsg_entry (keepalived + 0x1d8d4)
#2  sync_service_vsg (keepalived + 0x1eb06)
#3  start_check (keepalived + 0xf591)
#4  reload_check_thread (keepalived + 0xf8c3)
...

Root Cause

In ipvs_group_sync_entry(), the weight of a real server is computed as:

drule.user.weight = ((rs->inhibit && !rs->alive) || vs->s_svr->alive) ? 0 : real_weight(rs->effective_weight);

If no sorry_server is configured, vs->s_svr is NULL, so accessing vs->s_svr->alive causes a null pointer dereference.

The later block in the same function already checks if (vs->s_svr && ...) before dereferencing it, so this was just a missing check.

Testing

• Reproduced the crash by adding a new IP to a VSG without sorry_server and reloading.
• Verified the crash no longer occurs after the fix.

[pqarmitage]

Many thanks for identifying the issue and providing the patch.