Files
ubicloud/prog/vnet/update_firewall_rules.rb
Furkan Sahin 8ce5c187ed Fix Load Balancer for IPv6
The Load Balancer was not working when there were only IPv6 only VMs in
the load balancing pool. The reason is, when there is only IPv6 only
VMs, the DNS resolves to an IPv6 address and the load balancer was
unable to route the traffic properly. Main reason of that bug is the
firewall rules implementation. When we started implementing firewalls,
we decided that the traffic in private network would be exempt from the
firewall rules. However, we made a concious decision to block packets
that blong to a NEW connection if they have the destination address of
the private ipv6. Because, it would be weird to have such a packet when
we are communicating in between different VMs. The whole private
networking logic depends on public IPs in pre forward hook + the IPSec
Tunnels.

Now, fast forward to the load balancer implementation, we DNAT the
packets destined to the public IPv6 of the VM and route them through the
Private IPv6 address. Therefore, when we evaluate the firewalls, just
before completing SNAT + IPSec Tunnelling, the firewalls see that there
is a NEW connection that is destined to the private IPv6 address of the
VM and drops the packet. With this commit, we change that logic to
accept such packets.
2024-08-21 10:47:15 +02:00

226 lines
8.4 KiB
Ruby

# frozen_string_literal: true
class Prog::Vnet::UpdateFirewallRules < Prog::Base
subject_is :vm
FirewallRuleObj = Struct.new(:cidr, :port_range)
label def update_firewall_rules
rules = vm.firewalls.map(&:firewall_rules).flatten
allowed_ingress_ip4 = NetAddr.summ_IPv4Net(rules.select { !_1.ip6? && !_1.port_range }.map { _1.cidr }).map(&:to_s)
allowed_ingress_ip6 = NetAddr.summ_IPv6Net(rules.select { _1.ip6? && !_1.port_range }.map { _1.cidr }).map(&:to_s)
allowed_ingress_ip4_port_set = consolidate_rules(rules.select { !_1.ip6? && _1.port_range })
allowed_ingress_ip6_port_set = consolidate_rules(rules.select { _1.ip6? && _1.port_range })
guest_ephemeral = subdivide_network(vm.ephemeral_net6).first.to_s
globally_blocked_ipv4s, globally_blocked_ipv6s = generate_globally_blocked_lists
vm.vm_host.sshable.cmd("sudo ip netns exec #{vm.inhost_name} nft --file -", stdin: <<TEMPLATE)
# An nftables idiom for idempotent re-create of a named entity: merge
# in an empty table (a no-op if the table already exists) and then
# delete, before creating with a new definition.
table inet fw_table;
delete table inet fw_table;
table inet fw_table {
set allowed_ipv4_cidrs {
type ipv4_addr;
flags interval;
#{allowed_ingress_ip4.any? ? "elements = {#{allowed_ingress_ip4.join(",")}}" : ""}
}
set allowed_ipv6_cidrs {
type ipv6_addr;
flags interval;
#{allowed_ingress_ip6.any? ? "elements = {#{allowed_ingress_ip6.join(",")}}" : ""}
}
set allowed_ipv4_port_tuple {
type ipv4_addr . inet_service;
flags interval;
#{allowed_ingress_ip4_port_set.empty? ? "" : "elements = {#{allowed_ingress_ip4_port_set}}"}
}
set allowed_ipv6_port_tuple {
type ipv6_addr . inet_service;
flags interval;
#{allowed_ingress_ip6_port_set.empty? ? "" : "elements = {#{allowed_ingress_ip6_port_set}}"}
}
set private_ipv4_cidrs {
type ipv4_addr;
flags interval;
elements = {
#{generate_private_ip4_list}
}
}
set private_ipv6_cidrs {
type ipv6_addr
flags interval
elements = { #{generate_private_ip6_list} }
}
set globally_blocked_ipv4s {
type ipv4_addr;
flags interval;
#{globally_blocked_ipv4s.empty? ? "" : "elements = {#{globally_blocked_ipv4s}}"}
}
set globally_blocked_ipv6s {
type ipv6_addr;
flags interval;
#{globally_blocked_ipv6s.empty? ? "" : "elements = {#{globally_blocked_ipv6s}}"}
}
flowtable ubi_flowtable {
hook ingress priority filter
devices = { #{vm.nics.map(&:ubid_to_tap_name).join(",")} }
}
chain forward_ingress {
type filter hook forward priority filter; policy drop;
meta l4proto { tcp, udp } flow offload @ubi_flowtable
meta l4proto { tcp, udp } th dport 111 drop
ip saddr @globally_blocked_ipv4s drop
ip6 saddr @globally_blocked_ipv6s drop
ip daddr @globally_blocked_ipv4s drop
ip6 daddr @globally_blocked_ipv6s drop
ip saddr @private_ipv4_cidrs ct state established,related,new counter accept
ip daddr @private_ipv4_cidrs ct state established,related counter accept
ip6 saddr @private_ipv6_cidrs ct state established,related,new counter accept
ip6 daddr @private_ipv6_cidrs ct state established,related,new counter accept
ip6 saddr #{guest_ephemeral} ct state established,related,new counter accept
ip6 daddr #{guest_ephemeral} ct state established,related,new counter accept
ip saddr @allowed_ipv4_cidrs ip daddr @private_ipv4_cidrs counter accept
ip6 saddr @allowed_ipv6_cidrs ip6 daddr #{guest_ephemeral} counter accept
ip saddr . tcp dport @allowed_ipv4_port_tuple ip daddr @private_ipv4_cidrs counter accept
ip6 saddr . tcp dport @allowed_ipv6_port_tuple ip6 daddr #{guest_ephemeral} counter accept
ip saddr . udp dport @allowed_ipv4_port_tuple ip daddr @private_ipv4_cidrs counter accept
ip6 saddr . udp dport @allowed_ipv6_port_tuple ip6 daddr #{guest_ephemeral} counter accept
ip saddr 0.0.0.0/0 icmp type echo-request counter accept
ip6 saddr ::/0 icmpv6 type echo-request counter accept
}
}
TEMPLATE
pop "firewall rule is added"
end
def generate_globally_blocked_lists
globally_blocked_ipv4s = []
globally_blocked_ipv6s = []
GloballyBlockedDnsname.each do |globally_blocked_dnsname|
ips = globally_blocked_dnsname.ip_list || []
ips.each do |ip|
globally_blocked_ipv4s << "#{ip}/32" if ip.ipv4?
globally_blocked_ipv6s << "#{ip}/128" if ip.ipv6?
end
end
summ_ipv4 = NetAddr.summ_IPv4Net(globally_blocked_ipv4s.map { NetAddr::IPv4Net.parse(_1.to_s) })
summ_ipv6 = NetAddr.summ_IPv6Net(globally_blocked_ipv6s.map { NetAddr::IPv6Net.parse(_1.to_s) })
[summ_ipv4.join(", "), summ_ipv6.join(", ")]
end
# This method is needed to properly consolidate port_ranges + cidrs.
# For example, if we have the following rules:
# 1. 10.10.10.8/29 . 80-8080
# 2. 10.10.10.0/27 . 5432-10000
#
# We can't just merge the cidrs because the port ranges overlap. We need to
# first identify where the overlap is in the port ranges and then merge the
# cidrs for the overlapping port ranges. The result should be:
# 1. 10.10.10.8/29 . 80-5431
# 2. 10.10.10.0/27 . 5432-10000
#
# In the processing of these 2 rules, we first identify the port segments as;
# 1. 80-5431
# 2. 5432-8080
# 3. 8081-10000
#
# Then we identify the cidrs for each segment:
# 1. 10.10.10.8/29 (This is simply used from the first rule because the first
# rule is the only rule that has a cidr that overlaps with this segment)
# 2. 10.10.10.8/29 + 10.10.10.0/27: The combination of these will result in
# 10.10.10.0/27
# 3. 10.10.10.0/27 (This is simply used from the second rule because the
# second rule is the only rule that has a cidr that overlaps with this
# segment)
#
# For the combination of the cidrs, we use the summ_IPv4/6Net method from the
# netaddr gem. This method will combine the cidrs and remove any duplicates.
# If we don't perform this combination, we will end up with an error from
# nftables saying file exists.
#
# Additionally, the customers may have thousands of rules and possibly, they
# overlap. We want to minimize the number of rules that we create on the
# nftables side to avoid performance issues.
def consolidate_rules(rules)
port_segments = create_port_segments(rules)
consolidated_rules = []
port_segments.each do |segment|
# Find rules that overlap with the current segment
overlapping_rules = rules.select do |r|
r.port_range.begin <= segment[:end] && r.port_range.end - 1 >= segment[:begin]
end
# Merge cidrs for overlapping rules
merged_cidrs = if rules.first.cidr.version == 4
NetAddr.summ_IPv4Net(overlapping_rules.map(&:cidr))
else
NetAddr.summ_IPv6Net(overlapping_rules.map(&:cidr))
end
merged_cidrs.each do |cidr|
consolidated_rules << FirewallRuleObj.new(cidr, {begin: segment[:begin], end: segment[:end] + 1})
end
end
combined_rules = combine_continuous_ranges_for_same_subnet(consolidated_rules)
combined_rules.map do |r|
if r.port_range[:begin] != r.port_range[:end] - 1
"#{r.cidr} . #{r.port_range[:begin]}-#{r.port_range[:end] - 1}"
else
"#{r.cidr} . #{r.port_range[:begin]}"
end
end.join(",")
end
def combine_continuous_ranges_for_same_subnet(rules)
rules.sort_by { |r| [r.cidr.to_s, r.port_range[:begin]] }.chunk_while { |a, b| a.cidr.to_s == b.cidr.to_s && a.port_range[:end] == b.port_range[:begin] }.map do |chunk|
if chunk.size > 1
FirewallRuleObj.new(chunk.first.cidr, {begin: chunk.first.port_range[:begin], end: chunk.last.port_range[:end]})
else
chunk.first
end
end
end
def create_port_segments(rules)
# Extract unique start and end points from port ranges
points = rules.flat_map { |r| [r.port_range.begin.to_i, r.port_range.end.to_i] }.uniq.sort
segments = []
# Create segments based on unique points
points.each_cons(2) do |start_point, end_point|
segments << {begin: start_point, end: end_point - 1}
end
segments
end
def generate_private_ip4_list
vm.nics.map { NetAddr::IPv4Net.parse(_1.private_ipv4.network.to_s + "/26").to_s }.join(",")
end
def generate_private_ip6_list
vm.nics.map { NetAddr::IPv6Net.parse(_1.private_ipv6.network.to_s + "/64").to_s }.join(",")
end
def subdivide_network(net)
prefix = net.netmask.prefix_len + 1
halved = net.resize(prefix)
[halved, halved.next_sib]
end
end