Bug: Caddy.Admin.Request.do_recv/3 missing catch-all causes FunctionClauseError crash loop

[gsmlg]

Summary

Two bugs in Caddy.Admin.Request and Caddy.Admin.Api can cause the Caddy.Server.External GenServer to crash in a tight loop, which exhausts the Caddy.Supervisor max restart intensity (3 restarts in 5 seconds) and terminates the entire :caddy OTP application. Once terminated, all Caddy.* calls from the dashboard throw :exit and show "Unavailable / Not Ready / Not Configured".

---

Bug 1: do_recv/3 has no catch-all clause

File: lib/caddy/admin/request.ex

The do_recv/3 recursive function handles HTTP response parsing but only matches three patterns:

defp do_recv(socket, {:ok, {:http_response, {1, 1}, code, _}}, resp) -> ...
defp do_recv(socket, {:ok, {:http_header, _, h, _, v}}, resp) -> ...
defp do_recv(socket, {:ok, :http_eoh}, resp) -> ...

Missing patterns that cause FunctionClauseError:

• {:error, :timeout} — socket receive timeout (5 s limit is hit)
• {:error, :closed} — server closes connection before response is complete
• {:ok, {:http_response, {1, 0}, code, _}} — HTTP/1.0 response
• {:ok, {:http_error, msg}} — malformed HTTP response
• Any other unexpected value from :gen_tcp.recv/3

When any of these occur inside handle_info(:health_check, state) or handle_continue(:initial_setup, state), the exception propagates out of the GenServer callback and crashes Caddy.Server.External.

Fix:

defp do_recv(_socket, {:error, reason}, _resp) do
  {:error, reason}
end

defp do_recv(_socket, unexpected, _resp) do
  {:error, {:unexpected_response, unexpected}}
end

---

Bug 2: Api.load/1 crashes with FunctionClauseError when get_config() returns nil

File: lib/caddy/admin/api.ex

def load(conf) when is_map(conf) do
  get_config()       # Returns nil when Caddy Admin API is unreachable
  |> Map.merge(conf) # ** FunctionClauseError: no function clause matching Map.merge(nil, ...)
  |> Jason.encode!()
  |> load()
end

Api.get_config/0 returns nil on connection failure. Piping nil into Map.merge/2 throws FunctionClauseError, crashing the calling GenServer.

This is triggered from push_initial_config/0 in Caddy.Server.External — called during handle_continue(:initial_setup) and periodic handle_info(:health_check) — when Caddy briefly becomes reachable for the health check but is gone by the time the load request fires (race condition), or when the /adapt call succeeds but the subsequent get_config() inside load/1 fails.

Fix:

def load(conf) when is_map(conf) do
  case get_config() do
    nil ->
      conf |> Jason.encode!() |> load()
    existing when is_map(existing) ->
      existing |> Map.merge(conf) |> Jason.encode!() |> load()
  end
end

---

Crash cascade

Both bugs produce the same failure mode:

1. Caddy.Server.External crashes (FunctionClauseError in handle_info or handle_continue)
2. Caddy.Supervisor (:rest_for_one) restarts Caddy.Server.External
3. It crashes again immediately (same code path)
4. After 3 crashes in 5 seconds, Caddy.Supervisor terminates
5. Caddy.Application terminates — all Caddy.* processes are gone
6. Dashboard shows: Application State = Unavailable, Ready = Not Ready, Configuration = Not Configured, Sync Status = Unavailable

---

Environment

• caddy hex package version: 2.3.1
• Mode: :external, admin_url: "http://localhost:2019"
• Elixir 1.18, OTP 28