My team has run into the following problem. We’ve discovered an operationally acceptable workaround which I will submit as an answer to this question, but we are hoping for a better solution, or at least a better explanation of what is going on.
We're trying to bring some Apple products into our test lab in order to define configurations for our customer. Examples and tests were performed on a MacBook Air, but confirmed the problem exists on some older model iPhone and iPads. We've put it on one of the lab’s Wireless Ethernet networks, where it receives DHCP settings from a Cisco 5921 ESR (software router) as expected, including IP, netmask, and default route. The MacBook is in the 192.168.1.0/24 network, and in the following test cases, the remote network is 192.168.2.0/24.
Full output of the MacBook’s ifconfig and netstat -nr are at the end of this question.
The problem is that while it can communicate to other devices on its local subnet, any non-local communication that should go to the gateway is failing. All types of network traffic to non-local destinations fail (ping, web, ssh). And wireshark running on the MacBook shows the expected ARP and IP traffic occurs when communicating with local nodes, even to the router itself. But absolutely no traffic is generated when trying to connect to non-local traffic, not even an ARP request. Yet the routing table seems to specify the correct interface and gateway address for the default entry.
Executing route get yields the following, first for local (which works) and non-local (which does not):
MacBook-Air:~ user1$ route get 192.168.1.1
route to: 192.168.1.1
destination: 192.168.1.1
interface: en0
flags: <UP,HOST,DONE,LLINFO,WASCLONED,IFSCOPE,IFREF,ROUTER>
recvpipe sendpipe ssthresh rtt,msec rttvar hopcount mtu expire
0 0 0 0 0 0 1500 1175
MacBook-Air:~ user1$ route get 192.168.2.80
route: writing to routing socket: not in table
And ping, first for local then non-local
MacBook-Air:~ user1$ ping 192.168.1.1
PING 192.168.1.1 (192.168.1.1): 56 data bytes
64 bytes from 192.168.1.1: icmp_seq=0 ttl=255 time=2.904 ms
64 bytes from 192.168.1.1: icmp_seq=1 ttl=255 time=3.963 ms
^C
--- 192.168.1.1 ping statistics ---
2 packets transmitted, 2 packets received, 0.0% packet loss
round-trip min/avg/max/stddev = 2.904/3.433/3.963/0.529 ms
MacBook-Air:~ user1$ ping 192.168.2.80
PING 192.168.2.80 (192.168.2.80): 56 data bytes
ping: sendto: No route to host
ping: sendto: No route to host
Request timeout for icmp_seq 0
ping: sendto: No route to host
Request timeout for icmp_seq 1
ping: sendto: No route to host
Request timeout for icmp_seq 2
^C
--- 192.168.2.80 ping statistics ---
4 packets transmitted, 0 packets received, 100.0% packet loss
After checking the basics (netmask, default routes, firewalls), we did some deeper digging that led us to this question
In it, the problem was that the default route is tagged with the letter I. On my system, this is also the case, as seen in this partial output to netstat -nr:
Routing tables
Internet:
Destination Gateway Flags Refs Use Netif Expire
default 192.168.1.1 UGScI 4 0 en0
According to the question above, the letter I means
I RTF_IFSCOPE Route is associated with an interface scope
I can confirm this by running ping with the -b flag, and running route get with the -ifscope flag (noticing this is what led us to searching RTF_IFSCOPE which led to the above stackexchange post). When these flags are used, both commands work as expected. Without their respective flags, both commands fail.
MacBook-Air:~ user1$ route get -ifscope en0 192.168.2.80
route to: 192.168.2.80
destination: 192.168.2.80
gateway: 192.168.1.1
interface: en0
flags: <UP,GATEWAY,HOST,DONE,WASCLONED,IFSCOPE,IFREF>
recvpipe sendpipe ssthresh rtt,msec rttvar hopcount mtu expire
0 0 0 0 0 0 1500 0
MacBook-Air:~ user1$ ping -b en0 192.168.2.80
PING 192.168.2.80 (192.168.2.80): 56 data bytes
64 bytes from 192.168.2.80: icmp_seq=0 ttl=61 time=3.206 ms
64 bytes from 192.168.2.80: icmp_seq=1 ttl=61 time=2.713 ms
64 bytes from 192.168.2.80: icmp_seq=2 ttl=61 time=3.246 ms
64 bytes from 192.168.2.80: icmp_seq=3 ttl=61 time=3.001 ms
64 bytes from 192.168.2.80: icmp_seq=4 ttl=61 time=2.444 ms
64 bytes from 192.168.2.80: icmp_seq=5 ttl=61 time=2.426 ms
64 bytes from 192.168.2.80: icmp_seq=6 ttl=61 time=2.205 ms
64 bytes from 192.168.2.80: icmp_seq=7 ttl=61 time=4.701 ms
64 bytes from 192.168.2.80: icmp_seq=8 ttl=61 time=4.964 ms
64 bytes from 192.168.2.80: icmp_seq=9 ttl=61 time=2.674 ms
^C
--- 192.168.2.80 ping statistics ---
10 packets transmitted, 10 packets received, 0.0% packet loss
round-trip min/avg/max/stddev = 2.205/3.158/4.964/0.898 ms
So, RTF_IFSCOPE looks like the BSD version of Linux Network Namespaces or Cisco IOS VRFs (I realize it may predate those). But in my case, the default route is set up by DHCP, so it’s not clear how or why this came to be, nor how to change it to the default scope (or, alternatively, make all applications utilize the special scope).
As a test, I manually added a second default route:
sudo route -r add default 192.168.1.1
The resulting route does not get an I flag associated with it:
Routing tables
Internet:
Destination Gateway Flags Refs Use Netif Expire
default 192.168.1.1 UGSc 4 0 en0
default 192.168.1.1 UGScI 4 0 en0
This does fix the problem, now allowing all non-local network communication to work. We don’t consider this an operationally acceptable workaround, especially for iPhone and iPads, since as far as we can tell it would require adding apps to the devices in order to add manual routes.
We suspect that something about our network environment is causing this behavior. But we aren’t sure what that is.
The following output of ifconfig and netstat -nr are from the MacBook Air:
MacBook-Air:~ user1$ ifconfig
lo0: flags=8049<UP,LOOPBACK,RUNNING,MULTICAST> mtu 16384
options=1203<RXCSUM,TXCSUM,TXSTATUS,SW_TIMESTAMP>
inet 127.0.0.1 netmask 0xff000000
inet6 ::1 prefixlen 128
inet6 fe80::1%lo0 prefixlen 64 scopeid 0x1
nd6 options=201<PERFORMNUD,DAD>
gif0: flags=8010<POINTOPOINT,MULTICAST> mtu 1280
stf0: flags=0<> mtu 1280
OHC4: flags=0<> mtu 0
EHC36: flags=0<> mtu 0
OHC6: flags=0<> mtu 0
EHC38: flags=0<> mtu 0
en0: flags=8963<UP,BROADCAST,SMART,RUNNING,PROMISC,SIMPLEX,MULTICAST> mtu 1500
ether 04:0c:ce:cf:be:ac
inet 192.168.1.52 netmask 0xffffff00 broadcast 192.168.1.255
nd6 options=201<PERFORMNUD,DAD>
media: autoselect
status: active
p2p0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> mtu 2304
ether 06:0c:ce:cf:be:ac
media: autoselect
status: inactive
utun0: flags=8051<UP,POINTOPOINT,RUNNING,MULTICAST> mtu 2000
inet6 fe80::d380:8cb1:83eb:3753%utun0 prefixlen 64 scopeid 0xa
nd6 options=201<PERFORMNUD,DAD>
MacBook-Air:~ user1$ netstat -nr
Routing tables
Internet:
Destination Gateway Flags Refs Use Netif Expire
default 192.168.1.1 UGScI 4 0 en0
127 127.0.0.1 UCS 0 0 lo0
127.0.0.1 127.0.0.1 UH 1 130 lo0
169.254 link#8 UCS 0 0 en0
192.168.1 link#8 UCS 3 0 en0
192.168.1.1/32 link#8 UCS 1 0 en0
192.168.1.1 0:1b:ac:1:79:9a UHLWIir 3 60 en0
192.168.1.52/32 link#8 UCS 0 0 en0
192.168.1.53 2c:be:8:aa:47:98 UHLWI 0 2 en0 472
192.168.1.54 88:1f:a1:7a:6d:5a UHLWI 0 0 en0 1200
192.168.1.56 2c:be:8:9d:ff:78 UHLWI 0 10 en0 817
224.0.0/4 link#8 UmCS 1 0 en0
224.0.0.251 1:0:5e:0:0:fb UHmLWI 0 0 en0
255.255.255.255/32 link#8 UCS 0 0 en0
Internet6:
Destination Gateway Flags Netif Expire
default fe80::%utun0 UGcI utun0
::1 ::1 UHL lo0
fe80::%lo0/64 fe80::1%lo0 UcI lo0
fe80::1%lo0 link#1 UHLI lo0
fe80::%utun0/64 fe80::d380:8cb1:83eb:3753%utun0 UcI utun0
fe80::d380:8cb1:83eb:3753%utun0 link#10 UHLI lo0
ff01::%lo0/32 ::1 UmCI lo0
ff01::%utun0/32 fe80::d380:8cb1:83eb:3753%utun0 UmCI utun0
ff02::%lo0/32 ::1 UmCI lo0
ff02::%utun0/32 fe80::d380:8cb1:83eb:3753%utun0 UmCI utun0
Best Answer
I think the missing piece in your puzzle is that Apple uses a slightly different way of handling interface priority than other popular operating systems.
For example, if we take a normal home user scenario, Apple products will try to prioritise traffic over the best available Internet connection. If you have an Internet connection over both Ethernet and WiFi, it will use the cabled connection rather than the wireless connection.
The way this prioritisation is done technically is via the RTF_IFSCOPE flag, you've experienced. Essentially the system will mark the lower prioritized route with the RTF_IFSCOPE flag to ensure that it is only used if is specifically requested (i.e. for example if you have a VPN program that uses a specific connection, it will be able to do so).
You can observe the same thing if you connect to a WiFi without an Internet connection or a WiFi with a captive portal where you haven't signed in yet. On an iOS device for example, you'll see that the WiFi symbol at the top changes to a the waves overlaid with an exclamation mark (!). You can still use the interface and communicate with devices on that specific subnet, but it will not be used as a default gateway - unless specifically requested.
So in your case it seems that you have already got a default gateway on the Mac before you connect it to the WiFi. As the DHCP server on the WiFi does not give out an option 6 (i.e. list of DNS servers), the system will give this connection a lower priority.
It seems that you should be able to solve the problem by not having an existing default route before you connect to the WiFi.
A more common solution would be to simply specify the priorisation you want instead of relying on automatic defaults. This is done on the Mac in System Preferences > Network. Click the Gear icon below the list on the left, and choose "Set Service Order". Now drag the Wi-Fi interface above the other default route interface you have (for example Ethernet). This removes the REF_IFSCOPE flag from your WiFi. You can also do this via command line, or from a central management server, ofcourse - but the GUI route is the easiest way to quickly test this.
Apart from that, I would say that your setup sounds relatively uncommon. I don't agree with your choice in specifically foregoing DNS servers simply to avoid reconfiguring web servers from doing reverse lookups on IP addresses. The reverse lookups will still be done, but they'll just be handled locally instead.
It seems it would be a better choice to have DNS servers and configure web servers reasonably instead. You can configure a DNS server to statically reply to all reverse lookups instead of looking anything up on the Internet or otherwise spending time providing names.