libct: clean cached rlimit nofile in go runtime

As reported in issue opencontainers#4195, the new version(since 1.19) of go runtime
will cache rlimit-nofile. Before executing execve, the rlimit-nofile
of the process will be restored with the cache. In runc, this will
cause the rlimit-nofile set by the parent process for the container
to become invalid. It can be solved by clearing the cache.

Signed-off-by: ls-ggg <[email protected]>
(cherry picked from commit f9f8abf)
Signed-off-by: lifubang <[email protected]>
(cherry picked from commit da68c8e)
Signed-off-by: lifubang <[email protected]>

Author: ls-ggg

libcontainer/init_linux.go

@@ -84,6 +84,13 @@ func newContainerInit(t initType, pipe *os.File, consoleSocket *os.File, fifoFd,
 	if err := populateProcessEnvironment(config.Env); err != nil {
 		return nil, err
 	}
+
+	// Clean the RLIMIT_NOFILE cache in go runtime.
+	// Issue: https://github.com/opencontainers/runc/issues/4195
+	if containsRlimit(config.Rlimits, unix.RLIMIT_NOFILE) {
+		system.ClearRlimitNofileCache()
+	}
+
 	switch t {
 	case initSetns:
 		// mountFds must be nil in this case. We don't mount while doing runc exec.
@@ -518,6 +525,15 @@ func setupRoute(config *configs.Config) error {
 	return nil
 }
 
+func containsRlimit(limits []configs.Rlimit, resource int) bool {
+	for _, rlimit := range limits {
+		if rlimit.Type == resource {
+			return true
+		}
+	}
+	return false
+}
+
 func setupRlimits(limits []configs.Rlimit, pid int) error {
 	for _, rlimit := range limits {
 		if err := unix.Prlimit(pid, rlimit.Type, &unix.Rlimit{Max: rlimit.Hard, Cur: rlimit.Soft}, nil); err != nil {

libcontainer/setns_init_linux.go

@@ -48,6 +48,7 @@ func (l *linuxSetnsInit) Init() error {
 			}
 		}
 	}
+
 	if l.config.CreateConsole {
 		if err := setupConsole(l.consoleSocket, l.config, false); err != nil {
 			return err

libcontainer/system/linux.go

@@ -5,11 +5,27 @@ package system
 
 import (
 	"os"
+	"sync/atomic"
+	"syscall"
 	"unsafe"
 
 	"golang.org/x/sys/unix"
 )
 
+//go:linkname syscallOrigRlimitNofile syscall.origRlimitNofile
+var syscallOrigRlimitNofile atomic.Pointer[syscall.Rlimit]
+
+// As reported in issue #4195, the new version of go runtime(since 1.19)
+// will cache rlimit-nofile. Before executing execve, the rlimit-nofile
+// of the process will be restored with the cache. In runc, this will
+// cause the rlimit-nofile setting by the parent process for the container
+// to become invalid. It can be solved by clearing this cache. But
+// unfortunately, go stdlib doesn't provide such function, so we need to
+// link to the private var `origRlimitNofile` in package syscall to hack.
+func ClearRlimitNofileCache() {
+	syscallOrigRlimitNofile.Store(nil)
+}
+
 type ParentDeathSignal int
 
 func (p ParentDeathSignal) Restore() error {