Merge PR #930 (Supervisor, no snabb gc) into kbara-next

src/core/main.lua

@@ -10,10 +10,11 @@ local STP = require("lib.lua.StackTracePlus")
 local ffi = require("ffi")
 local zone = require("jit.zone")
 local lib = require("core.lib")
+local shm = require("core.shm")
 local C   = ffi.C
 -- Load ljsyscall early to help detect conflicts
 -- (e.g. FFI type name conflict between Snabb and ljsyscall)
-require("syscall")
+local S = require("syscall")
 
 require("lib.lua.strict")
 require("lib.lua.class")
@@ -28,8 +29,8 @@ ffi.cdef[[
       extern char** argv;
 ]]
 
-_G.developer_debug = false
-debug_on_error = false
+_G.developer_debug = lib.getenv("SNABB_DEBUG") and true
+debug_on_error = _G.developer_debug
 
 function main ()
    zone("startup")
@@ -130,6 +131,13 @@ function handler (reason)
    os.exit(1)
 end
 
+-- Cleanup after Snabb process.
+function shutdown (pid)
+   if not _G.developer_debug then
+      shm.unlink("//"..pid)
+   end
+end
+
 function selftest ()
    print("selftest")
    assert(programname("/bin/snabb-1.0") == "snabb",
@@ -154,4 +162,46 @@ function selftest ()
       "Incorrect program name selected")
 end
 
-xpcall(main, handler)
+-- Fork into worker process and supervisor
+local worker_pid = S.fork()
+if worker_pid == 0 then
+   -- Worker: Use prctl to ensure we are killed (SIGHUP) when our parent quits
+   -- and run main.
+   S.prctl("set_pdeathsig", "hup")
+   xpcall(main, handler)
+else
+   -- Supervisor: Queue exit_signals using signalfd, prevent them from killing
+   -- us instantly using sigprocmask.
+   local exit_signals = "hup, int, quit, term, chld"
+   local signalfd = S.signalfd(exit_signals)
+   S.sigprocmask("block", exit_signals)
+   while true do
+      -- Read signals from signalfd. Only process the first signal because any
+      -- signal causes shutdown.
+      local signals, err = S.util.signalfd_read(signalfd)
+      assert(signals, tostring(err))
+      for i = 1, #signals do
+         local exit_status
+         if signals[i].chld then
+            -- SIGCHILD means worker state changed: retrieve its status using
+            -- waitpid and set exit status accordingly.
+            local status, err, worker =
+               S.waitpid(worker_pid, "stopped,continued")
+            assert(status, tostring(err))
+            if     worker.WIFEXITED   then exit_status = worker.EXITSTATUS
+            elseif worker.WIFSIGNALED then exit_status = 128 + worker.WTERMSIG
+            -- WIFSTOPPED and WIFCONTINUED are ignored.
+            else goto ignore_signal end
+         else
+            -- Supervisor received exit signal: kill worker by sending SIGHUP
+            -- and and set exit status accordingly.
+            S.kill(worker_pid, "hup")
+            exit_status = 128 + signals[i].signo
+         end
+         -- Run shutdown routine and exit.
+         shutdown(worker_pid)
+         os.exit(exit_status)
+         ::ignore_signal::
+      end
+   end
+end