Bridged Network TCP Bug on macOS Host

[wallpunch]

Describe the bug
On a macOS host, TCP connections bound to an instance's bridged network interface hang after about 50kb data are received.

To Reproduce

1. multipass set local.bridged-network=en0
2. multipass launch -n test --bridged
3. multipass exec test -- curl https://canonical.com -o /dev/null --interface enp0s2
   (Sometimes need to repeat Step 3 a few times before it hangs. I don't think it's a curl bug as I've seen the same issue using netcat and even sockets directly.)

Expected behavior
All TCP data from the server should be received. (The curl request should complete.)

Logs
bug.log

Additional info

• macOS 14.2.1

• multipass version
  multipass 1.13.1+mac
  multipassd 1.13.1+mac

• multipass info
  Name: test
  State: Running
  Snapshots: 0
  IPv4: 192.168.64.65
  192.168.69.244
  Release: Ubuntu 22.04.4 LTS
  Image hash: 40ea1181447b (Ubuntu 22.04 LTS)
  CPU(s): 1
  Load: 0.27 0.10 0.03
  Disk usage: 1.6GiB out of 4.8GiB
  Memory usage: 147.7MiB out of 962.3MiB
  Mounts: --

• multipass get local.driver
  qemu

Additional context
The bug does not always occur. It seems more likely to occur when the incoming data arrives very rapidly. It only occurs when more than approximately 50kb are transmitted by the server.

The bug does not occur if the bridged interface is not explicitly bound to (i.e. curl 'https://canonical.com' -o /dev/null never hangs).

[luis4a0]

Hi @wallpunch, thanks for reporting. This seems to be another case of #3038. Unfortunately, we don't have a solution for this yet. But stay tuned to #3038 for news. Thanks!

[luis4a0]

Closing of duplicate of #3038.

[luis4a0]

Reopening because we are not 100% sure it's the same issue as #3038.