Understanding Packet Flow

More advanced firewall setups, or complicated tasks, such as traffic prioritization, routing policies, where it is necessary to utilize more than one RouterOS facility, requires knowledge: How these facilities work together? What happens when and why?

RouterOS packet flow diagram and flow examples will try to answers these questions.

It would be very complicated to represent what is going on with the packet in one diagram, therefore a packet flow diagram is divided into three parts:

• overall diagram;
• detailed bridging, routing, and MPLS flow diagram;
• diagram that shows what facilities and in what order is included in prerouting, input, forward, output and postrouting.

Overall Packetflow Diagram

Let's look at the overall diagram. It looks complicated at first, but after we go through the diagram with examples it will become much clearer.

There are 4 boxes in the center of the diagram: Bridging, Routing, Mpls decisions, and local router processes. So for example, if the packet needs to be routed over the router, a packet will flow as illustrated in the image below. Without looking deeper into each facility, packet enters in-interface, the router determines that it is IP traffic and needs to be routed, packet goes through all routing processes and exits out-interface.

Let's take a look at another example that will illustrate what happens if the packet's destination is a router. For example, the in-interface receives ICMP (ping) packet, its destination is the router itself, so packet will go for local-in processing. After packet is processed ICMP (ping) reply is generated inside the router (local-out processing) and will be sent out over the out-interface.

A simple explanation of each box before we go further with examples:

• physical in-interface - the starting point of the packet received by the router;
• logical in-interface - the starting point of the decapsulated packet (from tunnels, IPsec, etc);
• local in - the last point of packet destined to router itself;
• interface HTB (Hierarchical Token Bucket) - interface queue;
• physical out-interface - last point of the packet before it is actually sent out;
• logical out-interface - last point of the packet before encapsulation (to tunnels, IPsec, etc);
• local out - the starting point of a packet generated by the router;

Now it is time to take a deeper look at what is happening inside bridging, MPLS, and routing flows.

A simple explanation of each box before we go further with examples:

• routing decision - go through routes in the routing table to find a match for the destination IP address of the packet. When match is found - packet will be processed further, in case of no match - packet will be discarded.;
• mpls decision - what to do with the packet based on MPLS forwarding tables;
• bridging decision - bridge goes through the MAC address table to find a match for destination MAC address of the packet. When a match is found - packet will be processed further, in case of no match - multiple copies of packet will be created and packets will be flooded (sent out via all bridge ports). A single packet copy will also reach a bridge input chain as the bridge interface itself is one of the many destinations.
• use-ip-firewall - whether a use-ip-firewall option is enabled in bridge settings;
• ipsec-policy - whether packet matches any of configured IPsec policies;

Chains

RouterOS consist of a few default chains. These chains allow you to filter packets at various points:

• The PREROUTING chain: Rules in this chain apply to packets as they just arrive on the network interface. This chain is present in the nat, mangle and raw tables.
• The INPUT chain: Rules in this chain apply to packets just before they’re given to a local process. This chain is present in the mangle and filter tables.
• The OUTPUT chain: The rules here apply to packets just after they’ve been produced by a process. This chain is present in the raw, mangle, nat and filter tables.
• The FORWARD chain: The rules here apply to any packets that are routed through the current host. This chain is only present in the mangle and filter tables.
• The POSTROUTING chain: The rules in this chain apply to packets as they just leave the network interface. This chain is present in the nat and mangle tables.

Each of prerouting, input, forward, output and postrouting blocks contain even more facilities, which are illustrated in the third part of the packet flow diagram:

A simple explanation of each box before we go further with examples:

• Hotspot-in - allow to capture traffic which otherwise would be discarded by connection tracking - this way our Hotspot feature is able to provide connectivity even if networks settings are an incomplete mess ;
• RAW Prerouting - RAW table prerouting chain;
• Connection tracking - packet is processed by connection tracking;
• Mangle Prerouting - Mangle prerouting chain;
• Mangle Input - Mangle input chain;
• Filter Input - Firewall filter input chain;
• HTB Global - Queue tree;
• Simple Queues - ;
• TTL - indicates exact place where Time To Live (TTL) of the routed packet is reduced by 1 if TTL becomes 0 packet will be discarded;
• Mangle Forward - Mangle forward chain;
• Filter Forward - Mangle forward chain;
• Accounting - Authentication, Authorization and Accounting feature processing;
• RAW Output - RAW table output chain;
• Mangle Output - Mangle output chain;
• Filter Output - Firewall filter output chain;
• Routing Adjustment - this is a workaround that allows to set-up policy routing in mangle chain output (routing-mark) ;
• Mangle Postrouting - Mangle postrouting chain;
• Src Nat - Network Address Translation srcnat chain;
• Dst Nat - Network Address Translation dstnat chain;
• Hotspot-out - undo all that was done by hotspot-in for the packets that are going back to the client;

Flow of Routed Packet

Forward

Now, let's take our first example where packet gets routed over the router and look deeper through what facilities packet goes:

Section

Column width 400px We already learned that packet goes into in-interface, the router determines that it is an IP packet and need to be routed, and here starts the complicated process:  Packet enters prerouting processing: check if there is a hotspot and modify the packet for hotspot use process packet through RAW prerouting chain; send packet through connection tracking; process packet through Mangle prerouting chain;  process packet through NATs dst-nat chain; Run packet through routing table to make routing decision; Packet enters forward process; check TTL value; process packet through Mangle forward chain; process packet through Filter forward chain; send packet to accounting processes; A packet enters postrouting process; process packet through Mangle postrouting chain; process packet through NATs src-nat chain; if there is a hotspot undo any modifications made in hotspot-in; process packet through queue tree (HTB Global); process packet through simple queues; Check if there is IPsec and process through IPsec policies; Column

Input

We already learned that packet goes into in-interface, the router determines that it is an IP packet and need to be routed, and here starts the complicated process:

Output

Or when a packet is originated from the router (routing output):

Flow of Bridged Packet

Below is discussed a general bridging process in RouterOS. Most of the packets will always follow the same processing path, but in certain configurations (e.g. with enabled VLAN filtering, horizon, STP, DHCP or IGMP snooping) some packets can be treated differently. Please visit the bridging manual for more specific information.

Bridge Forward

Bridge forward is a process that takes place when a packet is forwarded from one bridge port to another, essentially connecting multiple devices on the same network. After receiving a packet on in-interface, the device determines that in-interface is a bridge port, so it gets passed through the bridging process:

Bridge Input

Bridge input is a process that takes place when a packet is destined to the bridge interface. Most commonly this happens when you need to reach some services that are running on the bridge interface (e.g. a DHCP server) or you need to route traffic to other networks. The very first steps are similar to the bridge forward process - after receiving a packet on in-interface, the device determines that in-interface is a bridge port, so it gets passed through the bridging process:

Bridge Output

Bridge output is a process that takes place when a packet should exit the device through one or multiple bridge ports. Most commonly this happens when a bridge interface itself tries to reach a device connected to a certain bridge port (e.g. when a DHCP server running on a bridge interface is responding to a DHCP client). After a packet is processed on other higher-level RouterOS processes and device finally determines that the output interface is a bridge, the packet gets passed through the bridging process:

Forward With Firewall Enabled

In certain network configurations, you might need to enable additional processing on routing chains for bridged traffic, for example, to use simple queues or IP firewall. This can be done when use-ip-firewall is enabled under the bridge settings. Note that additional processing will consume more CPU resources to handle these packets. All the steps were already discussed in previous points, below is a recap:

Flow of Hardware Offloaded Packet

On the previous topic, we solely discussed a software bridging that requires the main CPU processing to forward packets through the correct bridge port. Most of the MikroTik devices are equipped with dedicated switching hardware, so-called switch chip or switch ASIC. This allows us to offload some of the bridging functions, like packet forwarding between bridge ports or packet filtering, to this specialized hardware chip without consuming any CPU resources. In RouterOS, we have named this function as Bridge Hardware (HW) Offloading. Different MikroTik devices might have different switch chips and each chip has a different set of features available, so make sure to visit this article to get more details - Bridge Hardware Offloading.

• switching decision - widely depends on the switch model. This block controls all the switching related tasks, like host learning, packet forwarding, filtering, rate-limiting, VLAN tagging/untagging, mirroring, etc. Certain switch configuration can alter the packet flow;
• switch-cpu port - a special purpose switch port for communication between the main CPU and other switch ports. Note that the switch-cpu port does not show up anywhere on RouterOS except for the switch menu, none of the software related configuration (e.g. interface-list) can be applied to this port. Packets that reach the CPU are automatically associated with the physical in-interface.

The hardware offloading, however, does not restrict a device to only hardware limited features, rather it is possible to take advantage of the hardware and software processing at the same time. This does require a profound understanding of how packets travel through switch chip and when exactly they are passed to the main CPU.

Switch

Forward

We will further discuss a packet flow when a bridge hardware offloading is enabled and a packet is forwarded between two switched ports on a single switch chip. This is the most common and also the simplest example:

Switch to CPU

Input

This process takes place when a packet is received on a physical interface and it is destined to switch-cpu port for further software processing. There are two paths to the switch-cpu. One where hardware offloading and switching is not even used (e.g. a standalone interface for routing or a bridged interface but with deliberately disabled HW offloading), so the packet is simply passed further for software processing. Another path is taken when hardware offloading is active on the in-interface. This will cause the packet to pass through the switching decision and there are various reasons why switch might forward the packet to the switch-cpu port:

• a packet's destination MAC address match with a local MAC address, e.g. when a packet is destined to a local bridge interface;
• a packet might get flooded to all switch ports including the switch-cpu port, e.g. when broadcast, multicast or unknown unicast traffic is received;
• a switch might have learned that some hosts can only be reached through the CPU (switch-cpu port learning is discussed in the next section), e.g. when a bridge contains HW and non-HW offloaded interfaces, such as wireless, EoIP and even Ethernet interfaces;
• a packet is intentionally copied to the switch-cpu, e.g. for a packet inspection;
• a packet is triggered by the switch configuration and should be processed in software, e.g. a DHCP or IGMP snooping.

See the packet walkthrough when an in-interface is hardware offloaded:

CPU

Output to

Switch

This process takes place when a packet exits the RouterOS software processing and it is received on the switch-cpu port. Again, there are two paths the packet can take. One where hardware offloading and switching is not even used (e.g. a standalone interface for routing or a bridged interface but with deliberately disabled HW offloading), so the packet is simply sent out through the physical out-interface. Another path is taken when hardware offloading is active on the out-interface. This will cause the packet to pass through the switching decision. Just like any other switch port, the switch will learn the source MAC addresses from packets that are received on the switch-cpu port. This does come handy when a bridge contains HW and non-HW offloaded interfaces, so the switch can learn which frames should be forwarded to the CPU. See the packet walkthrough when an out-interface is hardware offloaded:

Flow of MPLS Packet

Pop Label

Switch Label

Push Label

Logical Interfaces

So far we looked at examples when in or out interfaces are actual physical interfaces (Ethernet, wireless), but how packets will flow if the router receives tunnel encapsulated packet?

Let's assume that there is an IPIP packet coming in the router. Since it is a regular ipv4 packet it will be processed through all routing-related facilities ( until "J" in the diagram). Then the router will look if packet needs to be decapsulated., in this case, it is an IPIP packet so "yes" send packet to decapsulation. After that packet will go another loop through all the facilities but this time as a decapsulated IPv4 packet.

It is very important because the packet actually travels through the firewall twice, so if there is a strict firewall, then there should accept rules for IPIP encapsulated packet as well as a decapsulated IP packet.

IPSec Policies

Let's take a look at another tunnel type - IPSec. This type of VPN does not have logical interfaces but is processed in a similar manner.

Fast Path

From what we learned so far, it is quite obvious that such packet processing takes a lot of CPU resources. To fast things up FastPath was introduced since the first RouterOS v6. What it does is it skips processing in the Linux kernel, basically trade some RouterOS functionality for performance. For FastPath to work, interface driver support and specific configuration conditions are required.

How Fast Path Works

FastPath is an interface driver extension, that allows a driver to talk directly to specific RouterOS facilities and skipping all others.

Packet can be forwarded by a fast path handler only if at least source interface supports a fast path. For complete fast-forwarding, destination interface support is also required.

Currently, RouterOS has following FastPath handlers:

• IPv4
• IPv4 FastTrack
• Traffic Generator
• MPLS
• Bridge

IPv4 FastPath handler is used if the following conditions are met:

• firewall rules are not configured;
• firewall address lists are not configured;
• simple and queue trees with parent=global are not configured;
• no mesh, metarouter interface configuration;
• sniffer, torch and traffic generator is not running;
• connection tracking is not active;
• ip accounting is disabled;
• VRFs are not configured (/ip route vrf is empty);
• Hotspot is not used (/ip hotspot has no interfaces);
• IPSec policies are not configured;
• /tool mac-scan is not actively used;
• /tool ip-scan is not actively used.

Traffic Generator and MPLS automatically use FastPath if interface support this feature. Currently, MPLS fast-path applies only to MPLS switched traffic (frames that enter router as MPLS and must leave router as MPLS) - MPLS ingress and egress (including VPLS tunnel endpoints that do VPLS encap/decap) will operate as before.

Bridge handler is used if the following conditions are met:

• there are no bridge firewall rules;
• use-ip-firewall is disabled;
• no mesh, MetaRouter interface configuration;
• sniffer, torch and traffic generator is not running.

Interfaces that support FastPath:

EoIP, Gre, IPIP and L2TP interfaces have per-interface setting allow-fast-path. Allowing fast path on these interfaces have side effect of bypassing firewall, connection tracking, simple queues, queue tree with parent=global, IP accounting, IPsec, hotspot universal client, vrf assignment for encapsulated packets that go trough fast path. Also, packet fragments cannot be received in FastPath.

Only interface queue that guarantees FastPath is only-hardware-queue. If you need interface queue other than hardware then the packet will not go fully FastPath, but there is not a big impact on performance, as "interface queue" is the last step in the packet flow.

Packet may go Half-FastPath by switching from FastPath to SlowPath, but not the other way around. So, for example, if the receiving interface has FastPath support, but out interface does not, then the router will process the packet by FastPath handlers as far as it can and then proceed with SlowPath. If the receiving interface does not support FastPath but out interface does, packet will be processed by SlowPath all the way through the router.

Fasttrack

Fasttrack can be decoded as Fast Path + Connection Tracking. It allows to mark connections as "fasttracked", after marking packets that belong to fastracked connection will be sent fastpath way. Connection table entry for such connection now will have fasttracked flag.

To mark a connection as fasttracked new action was implemented fasttrack-connection for firewall filter and mangle. Currently, only IPv4 TCP and UDP connections can be fasttracked and to maintain connection tracking entries some random packets will still be sent to slow path. This must be taken into consideration when designing firewalls with enabled fasttrack.

FastTrack handler also supports source and destination NAT, so special exceptions for NATed connections are not required.

Easiest way to start using this feature on home routers is to enable fasttrack for all established, related connections:

/ip firewall filter 
add chain=forward action=fasttrack-connection connection-state=established,related \
  comment="fasttrack established/related"
add chain=forward action=accept connection-state=established,related \
  comment="accept established/related"

Notice that first rule marks established/related connections as fasttracked, second rule is still required to accept packets belonging to those connections. The reason for this is that, as was mentioned earlier, some random packets from fasttracked connections are still sent the slow path way and only UDP and TCP are fasttracked, but we still want to accept packets for other protocols.

After adding fasttrack rule special dummy rule appeared at the top of the list. This is not an actual rule, it is for visual information showing that some of the traffic is traveling FastPath and will not reach other firewall rules.

These rules appear as soon as there is at least one "fasttracked" connection tracking entry and will disappear after the last "fasttracked" connection times out in the connection table.

Packet flow for visually impaired

The following document in DOCX format describes the diagram in a way optimized for visually impaired people. The descriptions are by Apex CoVantage care of Benetech. They are not being updated.

• Packet flow, optimized document.