The Inside and Outside of NAT
Quite many people don't pay attention to the difference in handling packets on interfaces configured for NAT inside and outside. Here is an example to demonstrate how NAT "domains" interact with routing. Consider three routers connected in the following manner:
For this scenario we have no routing configured. Let's use static NAT to provide connectivity between R1 and R2. R2 would see R1 as a host on local connected segment with the IP address 155.1.23.1 and R1 would see R2 as a host on it's local segment with the IP address 155.1.13.2. This goal could be achieved with the following configuration:
R3:
!
interface Serial 1/0.301 point-to-point
ip address 155.1.13.3 255.255.255.0
ip nat inside
no ip route-cache
!
interface Serial 1/0.302 multipoint
ip address 155.1.23.3 255.255.255.0
frame-relay map ip 155.1.23.2 302
ip nat outside
no ip route-cache!
! Static NAT: translations are effectively bi-directional
!
ip nat inside source static 155.1.13.1 155.1.23.1
ip nat outside source static 155.1.23.2 155.1.13.2R2:
!
! Add a Frame-Relay mapping for the new IP (representing R1)
! so that R2 would know how to reach the address over multipoint FR interface
!
interface Serial 1/0.203 multipoint
ip address 155.1.23.2 255.255.255.0
frame-relay map ip 155.1.23.3 203
frame-relay map ip 155.1.23.2 203
Let's see how it's working. Note that we disabled route-cache on both interfaces to intercept packets via CPU.
Rack1R3#debug ip nat detailed
IP NAT detailed debugging is onRack1R3#debug ip packet detail
IP packet debugging is on (detailed)Rack1R2#ping 155.1.23.1
Type escape sequence to abort.
Sending 5, 100-byte ICMP Echos to 155.1.23.1, timeout is 2 seconds:
.....
Success rate is 0 percent (0/5)
Hmm...it fails. Look at the debugging output on R3:
Rack1R3#
!
! Packet on NAT outside (o - for outside) hits the interface
!
NAT*: o: icmp (155.1.23.2, 16) -> (155.1.23.1, 16) [84]!
! Source and destination for the packet rewritten according to NAT rules
!
NAT*: s=155.1.23.2->155.1.13.2, d=155.1.23.1 [84]
NAT*: s=155.1.13.2, d=155.1.23.1->155.1.13.1 [84]!
! The packet is routed after translation (with new source and destination IPs). Note that routing decision
! and the actual forwarding take place only after translation rules were triggered by NAT tables
!
P: tableid=0, s=155.1.13.2 (Serial1/0.302), d=155.1.13.1 (Serial1/0.301), routed via RIB
IP: s=155.1.13.2 (Serial1/0.302), d=155.1.13.1 (Serial1/0.301), g=155.1.13.1, len 100, forward
ICMP type=8, code=0
!
! The response packet from R1 comes in - to destination 155.1.13.2 - routed via RIB (to the same interface)
! But no NAT rules were triggered since the destination interface is the same as input interface!
!
IP: tableid=0, s=155.1.13.1 (Serial1/0.301), d=155.1.13.2 (Serial1/0.301), routed via RIB
IP: s=155.1.13.1 (Serial1/0.301), d=155.1.13.2 (Serial1/0.301), len 100, rcvd 3
ICMP type=0, code=0
OK hold here for a second.. Now we recall that for inside NAT routing is tried first, and only then the packet is translated according to the NAT rules. This is how the NAT order of operations works on the inside. So now it's clear: IOS first tries to route packet to 155.1.13.2 - which is the same interface as it came in.. therefore the inside->outside translation never occurs! To fix this, let's add a static route on R3:
R3:
ip route 155.1.13.2 255.255.255.255 155.1.23.2
Verification:
Rack1R2#ping 155.1.23.1
Type escape sequence to abort.
Sending 5, 100-byte ICMP Echos to 155.1.23.1, timeout is 2 seconds:
!!!!!
Success rate is 100 percent (5/5), round-trip min/avg/max = 12/33/52 msRack1R3#
!
! Outside: translate & route
!
NAT*: o: icmp (155.1.23.2, 17) -> (155.1.23.1, 17) [89]
NAT*: s=155.1.23.2->155.1.13.2, d=155.1.23.1 [89]
NAT*: s=155.1.13.2, d=155.1.23.1->155.1.13.1 [89]!
! Routing decision and forwarding
!
IP: tableid=0, s=155.1.13.2 (Serial1/0.302), d=155.1.13.1 (Serial1/0.301), routed via RIB
IP: s=155.1.13.2 (Serial1/0.302), d=155.1.13.1 (Serial1/0.301), g=155.1.13.1, len 100, forward
ICMP type=8, code=0
!
! Inside: Routing decision - the packet is routed using our fixup static route
!
IP: tableid=0, s=155.1.13.1 (Serial1/0.301), d=155.1.13.2 (Serial1/0.302), routed via RIB!
! NAT rule (i - for inside) is triggered by the packet
!
NAT: i: icmp (155.1.13.1, 17) -> (155.1.13.2, 17) [89]!
! Source and destination addresses rewritten in the "opposite" direction
!
NAT: s=155.1.13.1->155.1.23.1, d=155.1.13.2 [89]
NAT: s=155.1.23.1, d=155.1.13.2->155.1.23.2 [89]!
! Packet is sent to R2 (with the new source and destination) - forwarding takes place
!
IP: s=155.1.23.1 (Serial1/0.301), d=155.1.23.2 (Serial1/0.302), g=155.1.23.2, len 100, forward
ICMP type=0, code=0
Nice. So now we know the difference for sure: packets on the NAT outside are first translated and then routed. On the inside interface routing decision kicks in first and only then translation rules get applied followed by forwarding. Before we finish with that, recall new 12.3T feature called NAT Virtual Interface. With this feature we can now configure any interface as "NAT enabled" an get rid of those "inside" and "outside" domains . All NAT traffic passed through new virtual interface called NVI, in symmetric manner. Let's reconfigure out task using this new concepts.
R3:
interface Serial 1/0.301 point-to-point
no ip nat inside
ip nat enable
!
interface Serial 1/0.302 multipoint
no ip nat outside
ip nat enable!
! Remove old rules
!
no ip nat inside source static 155.1.13.1 155.1.23.1
no ip nat outside source static 155.1.23.2 155.1.13.2!
! Add "domainless" rules
!
ip nat source static 155.1.13.1 155.1.23.1
ip nat source static 155.1.23.2 155.1.13.2no ip route 155.1.13.2 255.255.255.255 155.1.23.2
Verification:
Rack1R2#ping 155.1.23.1
Type escape sequence to abort.
Sending 5, 100-byte ICMP Echos to 155.1.23.1, timeout is 2 seconds:
!!!!!
Success rate is 100 percent (5/5), round-trip min/avg/max = 12/40/60 msRack1R3#
!
! Routing decision it taken: packet classified for NAT, since destination is in NAT table
! Note that no actual forwarding occurs, just routing decision to send packet
!
IP: tableid=0, s=155.1.23.2 (Serial1/0.302), d=155.1.23.1 (Serial1/0.302), routed via RIB!
! Packet translated according to NAT rules (note "i" for inside NAT)
!
NAT: i: icmp (155.1.23.2, 19) -> (155.1.23.1, 19) [95]
NAT: s=155.1.23.2->155.1.13.2, d=155.1.23.1 [95]
NAT: s=155.1.13.2, d=155.1.23.1->155.1.13.1 [95]!
! Another routing decision, for translated packet - now actual forwarding occurs
!
IP: tableid=0, s=155.1.13.2 (Serial1/0.302), d=155.1.13.1 (Serial1/0.301), routed via RIB
IP: s=155.1.13.2 (Serial1/0.302), d=155.1.13.1 (Serial1/0.301), g=155.1.13.1, len 100, forward
ICMP type=8, code=0!
! Response comes in, first routing decision - NAT table entry matched
!
IP: tableid=0, s=155.1.13.1 (Serial1/0.301), d=155.1.13.2 (Serial1/0.301), routed via RIB!
! Packet translated ("i" - inside NAT)
!
NAT: i: icmp (155.1.13.1, 19) -> (155.1.13.2, 19) [95]
NAT: s=155.1.13.1->155.1.23.1, d=155.1.13.2 [95]
NAT: s=155.1.23.1, d=155.1.13.2->155.1.23.2 [95]!
! Another routing decision, for post-translated packet, followed by forwarding
!
IP: tableid=0, s=155.1.23.1 (Serial1/0.301), d=155.1.23.2 (Serial1/0.302), routed via RIB
IP: s=155.1.23.1 (Serial1/0.301), d=155.1.23.2 (Serial1/0.302), g=155.1.23.2, len 100, forward
ICMP type=0, code=0
So what's the difference with NVI? First, we see that now NAT behaves symmetrically. Next, we see that NAT translation tables are used to take a "routing decision" to send packet to virtual interface. Packet is translated there and then another routing decision takes place, followed by packet forwarding. So the difference from the old model is that now routing decision is taken twice: before and after translation. This allows to get rid of any static routes needed by "legacy" NAT, since lookup is performed after translation.
To summarize: Domain-based NAT uses different orders of operations for inside and outside domain. NVI based NAT is symmetrical and performs routing lookup twice: first to send packet to NVI, second to route packet using the post-translated addresses.
Links:
