libcontainer:clean cached rlimit nofile in go runtime

As reported in issue opencontainers#4195, the new version of go runtime will
cache rlimit-nofile. before executing exec, the rlimit-nofile
of the process will be updated with the cache. in runc, this will
cause the rlimit-nofile set by the parent process for the container
to become invalid. this can be solved by clearing the cache.

Signed-off-by: ls-ggg <[email protected]>

Author: ls-ggg

libcontainer/setns_init_linux.go

@@ -49,6 +49,20 @@ func (l *linuxSetnsInit) Init() error {
 			}
 		}
 	}
+
+	// Set RLIMIT_NOFILE again to refresh the cache in go runtime
+	// The problem originates from https://github.com/golang/go/commit/f5eef58e4381259cbd84b3f2074c79607fb5c821
+	for _, rlimit := range l.config.Rlimits {
+		if rlimit.Type == unix.RLIMIT_NOFILE {
+			if err := unix.Setrlimit(rlimit.Type, &unix.Rlimit{
+				Cur: rlimit.Soft,
+				Max: rlimit.Hard,
+			}); err != nil {
+				return fmt.Errorf("failed to re-apply nofile rlimit: %w", err)
+			}
+		}
+	}
+
 	if l.config.CreateConsole {
 		if err := setupConsole(l.consoleSocket, l.config, false); err != nil {
 			return err

libcontainer/standard_init_linux.go

@@ -76,6 +76,18 @@ func (l *linuxStandardInit) Init() error {
 			}
 		}
 	}
+	// Set RLIMIT_NOFILE again to refresh the cache in go runtime
+	// The problem originates from https://github.com/golang/go/commit/f5eef58e4381259cbd84b3f2074c79607fb5c821
+	for _, rlimit := range l.config.Rlimits {
+		if rlimit.Type == unix.RLIMIT_NOFILE {
+			if err := unix.Setrlimit(rlimit.Type, &unix.Rlimit{
+				Cur: rlimit.Soft,
+				Max: rlimit.Hard,
+			}); err != nil {
+				return fmt.Errorf("failed to re-apply nofile rlimit: %w", err)
+			}
+		}
+	}
 
 	if err := setupNetwork(l.config); err != nil {
 		return err

tests/integration/resources.bats

@@ -0,0 +1,22 @@
+#!/usr/bin/env bats
+
+load helpers
+
+function setup() {
+	setup_busybox
+}
+
+function teardown() {
+	teardown_bundle
+}
+
+@test "runc run with RLIMIT_NOFILE" {
+	update_config '.process.args = ["/bin/sh", "-c", "ulimit -n"]'
+	update_config '.process.capabilities.bounding = ["CAP_SYS_RESOURCE"]'
+	update_config '.process.rlimits = [{"type": "RLIMIT_NOFILE", "hard": 10000, "soft": 10000}]'
+
+	runc run test_hello
+	[ "$status" -eq 0 ]
+
+	[[ "${output}" == "10000" ]]
+}