diff --git a/libcontainer/init_linux.go b/libcontainer/init_linux.go index bd89b253d5a..7e1298c4d16 100644 --- a/libcontainer/init_linux.go +++ b/libcontainer/init_linux.go @@ -223,6 +223,12 @@ func containerInit(t initType, config *initConfig, pipe *syncSocket, consoleSock return 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: i := &linuxSetnsInit{ @@ -649,6 +655,15 @@ 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 { diff --git a/libcontainer/setns_init_linux.go b/libcontainer/setns_init_linux.go index 18745d658c8..de3a15f6e2f 100644 --- a/libcontainer/setns_init_linux.go +++ b/libcontainer/setns_init_linux.go @@ -49,6 +49,7 @@ func (l *linuxSetnsInit) Init() error { } } } + if l.config.CreateConsole { if err := setupConsole(l.consoleSocket, l.config, false); err != nil { return err diff --git a/libcontainer/system/linux.go b/libcontainer/system/linux.go index 368364baf04..5e24d44e112 100644 --- a/libcontainer/system/linux.go +++ b/libcontainer/system/linux.go @@ -8,6 +8,7 @@ import ( "io" "os" "strconv" + "sync/atomic" "syscall" "unsafe" @@ -15,6 +16,20 @@ import ( "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 {