full diff: https://github.com/vishvananda/netns/compare/0a2b9b5464df8343199164a0321edf3313202f7e...db3c7e526aae966c4ccfa6c8189b693d6ac5d202
- Use golang.org/x/sys/unix instead of syscall
- Set O_CLOEXEC when opening a network namespace
- Fixes "the container‘s netns fds leak, causing the container netns to not
clean up successfully after the container stops"
- Allows to create and delete named network namespaces
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
(cherry picked from commit 818bad6ef21e473a80a27564bd2219449633dbb8)
Signed-off-by: Sebastiaan van Stijn <github@gone.nl>
| ... | ... |
@@ -50,7 +50,7 @@ github.com/hashicorp/go-sockaddr c7188e74f6acae5a989bdc959aa7 |
| 50 | 50 |
github.com/hashicorp/go-multierror 886a7fbe3eb1c874d46f623bfa70af45f425b3d1 # v1.0.0 |
| 51 | 51 |
github.com/hashicorp/serf 598c54895cc5a7b1a24a398d635e8c0ea0959870 |
| 52 | 52 |
github.com/docker/libkv 458977154600b9f23984d9f4b82e79570b5ae12b |
| 53 |
-github.com/vishvananda/netns 0a2b9b5464df8343199164a0321edf3313202f7e |
|
| 53 |
+github.com/vishvananda/netns db3c7e526aae966c4ccfa6c8189b693d6ac5d202 |
|
| 54 | 54 |
github.com/vishvananda/netlink a2ad57a690f3caf3015351d2d6e1c0b95c349752 |
| 55 | 55 |
|
| 56 | 56 |
# When updating, consider updating TOMLV_COMMIT in hack/dockerfile/install/tomlv.installer accordingly |
| ... | ... |
@@ -10,7 +10,8 @@ package netns |
| 10 | 10 |
|
| 11 | 11 |
import ( |
| 12 | 12 |
"fmt" |
| 13 |
- "syscall" |
|
| 13 |
+ |
|
| 14 |
+ "golang.org/x/sys/unix" |
|
| 14 | 15 |
) |
| 15 | 16 |
|
| 16 | 17 |
// NsHandle is a handle to a network namespace. It can be cast directly |
| ... | ... |
@@ -24,11 +25,11 @@ func (ns NsHandle) Equal(other NsHandle) bool {
|
| 24 | 24 |
if ns == other {
|
| 25 | 25 |
return true |
| 26 | 26 |
} |
| 27 |
- var s1, s2 syscall.Stat_t |
|
| 28 |
- if err := syscall.Fstat(int(ns), &s1); err != nil {
|
|
| 27 |
+ var s1, s2 unix.Stat_t |
|
| 28 |
+ if err := unix.Fstat(int(ns), &s1); err != nil {
|
|
| 29 | 29 |
return false |
| 30 | 30 |
} |
| 31 |
- if err := syscall.Fstat(int(other), &s2); err != nil {
|
|
| 31 |
+ if err := unix.Fstat(int(other), &s2); err != nil {
|
|
| 32 | 32 |
return false |
| 33 | 33 |
} |
| 34 | 34 |
return (s1.Dev == s2.Dev) && (s1.Ino == s2.Ino) |
| ... | ... |
@@ -36,11 +37,11 @@ func (ns NsHandle) Equal(other NsHandle) bool {
|
| 36 | 36 |
|
| 37 | 37 |
// String shows the file descriptor number and its dev and inode. |
| 38 | 38 |
func (ns NsHandle) String() string {
|
| 39 |
- var s syscall.Stat_t |
|
| 40 | 39 |
if ns == -1 {
|
| 41 | 40 |
return "NS(None)" |
| 42 | 41 |
} |
| 43 |
- if err := syscall.Fstat(int(ns), &s); err != nil {
|
|
| 42 |
+ var s unix.Stat_t |
|
| 43 |
+ if err := unix.Fstat(int(ns), &s); err != nil {
|
|
| 44 | 44 |
return fmt.Sprintf("NS(%d: unknown)", ns)
|
| 45 | 45 |
} |
| 46 | 46 |
return fmt.Sprintf("NS(%d: %d, %d)", ns, s.Dev, s.Ino)
|
| ... | ... |
@@ -49,11 +50,11 @@ func (ns NsHandle) String() string {
|
| 49 | 49 |
// UniqueId returns a string which uniquely identifies the namespace |
| 50 | 50 |
// associated with the network handle. |
| 51 | 51 |
func (ns NsHandle) UniqueId() string {
|
| 52 |
- var s syscall.Stat_t |
|
| 53 | 52 |
if ns == -1 {
|
| 54 | 53 |
return "NS(none)" |
| 55 | 54 |
} |
| 56 |
- if err := syscall.Fstat(int(ns), &s); err != nil {
|
|
| 55 |
+ var s unix.Stat_t |
|
| 56 |
+ if err := unix.Fstat(int(ns), &s); err != nil {
|
|
| 57 | 57 |
return "NS(unknown)" |
| 58 | 58 |
} |
| 59 | 59 |
return fmt.Sprintf("NS(%d:%d)", s.Dev, s.Ino)
|
| ... | ... |
@@ -67,7 +68,7 @@ func (ns NsHandle) IsOpen() bool {
|
| 67 | 67 |
// Close closes the NsHandle and resets its file descriptor to -1. |
| 68 | 68 |
// It is not safe to use an NsHandle after Close() is called. |
| 69 | 69 |
func (ns *NsHandle) Close() error {
|
| 70 |
- if err := syscall.Close(int(*ns)); err != nil {
|
|
| 70 |
+ if err := unix.Close(int(*ns)); err != nil {
|
|
| 71 | 71 |
return err |
| 72 | 72 |
} |
| 73 | 73 |
(*ns) = -1 |
| ... | ... |
@@ -6,46 +6,30 @@ import ( |
| 6 | 6 |
"fmt" |
| 7 | 7 |
"io/ioutil" |
| 8 | 8 |
"os" |
| 9 |
+ "path" |
|
| 9 | 10 |
"path/filepath" |
| 10 |
- "runtime" |
|
| 11 | 11 |
"strconv" |
| 12 | 12 |
"strings" |
| 13 | 13 |
"syscall" |
| 14 |
-) |
|
| 15 | 14 |
|
| 16 |
-// SYS_SETNS syscall allows changing the namespace of the current process. |
|
| 17 |
-var SYS_SETNS = map[string]uintptr{
|
|
| 18 |
- "386": 346, |
|
| 19 |
- "amd64": 308, |
|
| 20 |
- "arm64": 268, |
|
| 21 |
- "arm": 375, |
|
| 22 |
- "mips": 4344, |
|
| 23 |
- "mipsle": 4344, |
|
| 24 |
- "mips64le": 4344, |
|
| 25 |
- "ppc64": 350, |
|
| 26 |
- "ppc64le": 350, |
|
| 27 |
- "riscv64": 268, |
|
| 28 |
- "s390x": 339, |
|
| 29 |
-}[runtime.GOARCH] |
|
| 15 |
+ "golang.org/x/sys/unix" |
|
| 16 |
+) |
|
| 30 | 17 |
|
| 31 | 18 |
// Deprecated: use syscall pkg instead (go >= 1.5 needed). |
| 32 | 19 |
const ( |
| 33 |
- CLONE_NEWUTS = 0x04000000 /* New utsname group? */ |
|
| 34 |
- CLONE_NEWIPC = 0x08000000 /* New ipcs */ |
|
| 35 |
- CLONE_NEWUSER = 0x10000000 /* New user namespace */ |
|
| 36 |
- CLONE_NEWPID = 0x20000000 /* New pid namespace */ |
|
| 37 |
- CLONE_NEWNET = 0x40000000 /* New network namespace */ |
|
| 38 |
- CLONE_IO = 0x80000000 /* Get io context */ |
|
| 20 |
+ CLONE_NEWUTS = 0x04000000 /* New utsname group? */ |
|
| 21 |
+ CLONE_NEWIPC = 0x08000000 /* New ipcs */ |
|
| 22 |
+ CLONE_NEWUSER = 0x10000000 /* New user namespace */ |
|
| 23 |
+ CLONE_NEWPID = 0x20000000 /* New pid namespace */ |
|
| 24 |
+ CLONE_NEWNET = 0x40000000 /* New network namespace */ |
|
| 25 |
+ CLONE_IO = 0x80000000 /* Get io context */ |
|
| 26 |
+ bindMountPath = "/run/netns" /* Bind mount path for named netns */ |
|
| 39 | 27 |
) |
| 40 | 28 |
|
| 41 | 29 |
// Setns sets namespace using syscall. Note that this should be a method |
| 42 | 30 |
// in syscall but it has not been added. |
| 43 | 31 |
func Setns(ns NsHandle, nstype int) (err error) {
|
| 44 |
- _, _, e1 := syscall.Syscall(SYS_SETNS, uintptr(ns), uintptr(nstype), 0) |
|
| 45 |
- if e1 != 0 {
|
|
| 46 |
- err = e1 |
|
| 47 |
- } |
|
| 48 |
- return |
|
| 32 |
+ return unix.Setns(int(ns), nstype) |
|
| 49 | 33 |
} |
| 50 | 34 |
|
| 51 | 35 |
// Set sets the current network namespace to the namespace represented |
| ... | ... |
@@ -57,21 +41,64 @@ func Set(ns NsHandle) (err error) {
|
| 57 | 57 |
// New creates a new network namespace, sets it as current and returns |
| 58 | 58 |
// a handle to it. |
| 59 | 59 |
func New() (ns NsHandle, err error) {
|
| 60 |
- if err := syscall.Unshare(CLONE_NEWNET); err != nil {
|
|
| 60 |
+ if err := unix.Unshare(CLONE_NEWNET); err != nil {
|
|
| 61 | 61 |
return -1, err |
| 62 | 62 |
} |
| 63 | 63 |
return Get() |
| 64 | 64 |
} |
| 65 | 65 |
|
| 66 |
+// NewNamed creates a new named network namespace and returns a handle to it |
|
| 67 |
+func NewNamed(name string) (NsHandle, error) {
|
|
| 68 |
+ if _, err := os.Stat(bindMountPath); os.IsNotExist(err) {
|
|
| 69 |
+ err = os.MkdirAll(bindMountPath, 0755) |
|
| 70 |
+ if err != nil {
|
|
| 71 |
+ return None(), err |
|
| 72 |
+ } |
|
| 73 |
+ } |
|
| 74 |
+ |
|
| 75 |
+ newNs, err := New() |
|
| 76 |
+ if err != nil {
|
|
| 77 |
+ return None(), err |
|
| 78 |
+ } |
|
| 79 |
+ |
|
| 80 |
+ namedPath := path.Join(bindMountPath, name) |
|
| 81 |
+ |
|
| 82 |
+ f, err := os.OpenFile(namedPath, os.O_CREATE|os.O_EXCL, 0444) |
|
| 83 |
+ if err != nil {
|
|
| 84 |
+ return None(), err |
|
| 85 |
+ } |
|
| 86 |
+ f.Close() |
|
| 87 |
+ |
|
| 88 |
+ nsPath := fmt.Sprintf("/proc/%d/task/%d/ns/net", os.Getpid(), syscall.Gettid())
|
|
| 89 |
+ err = syscall.Mount(nsPath, namedPath, "bind", syscall.MS_BIND, "") |
|
| 90 |
+ if err != nil {
|
|
| 91 |
+ return None(), err |
|
| 92 |
+ } |
|
| 93 |
+ |
|
| 94 |
+ return newNs, nil |
|
| 95 |
+} |
|
| 96 |
+ |
|
| 97 |
+// DeleteNamed deletes a named network namespace |
|
| 98 |
+func DeleteNamed(name string) error {
|
|
| 99 |
+ namedPath := path.Join(bindMountPath, name) |
|
| 100 |
+ |
|
| 101 |
+ err := syscall.Unmount(namedPath, syscall.MNT_DETACH) |
|
| 102 |
+ if err != nil {
|
|
| 103 |
+ return err |
|
| 104 |
+ } |
|
| 105 |
+ |
|
| 106 |
+ return os.Remove(namedPath) |
|
| 107 |
+} |
|
| 108 |
+ |
|
| 66 | 109 |
// Get gets a handle to the current threads network namespace. |
| 67 | 110 |
func Get() (NsHandle, error) {
|
| 68 |
- return GetFromThread(os.Getpid(), syscall.Gettid()) |
|
| 111 |
+ return GetFromThread(os.Getpid(), unix.Gettid()) |
|
| 69 | 112 |
} |
| 70 | 113 |
|
| 71 | 114 |
// GetFromPath gets a handle to a network namespace |
| 72 | 115 |
// identified by the path |
| 73 | 116 |
func GetFromPath(path string) (NsHandle, error) {
|
| 74 |
- fd, err := syscall.Open(path, syscall.O_RDONLY, 0) |
|
| 117 |
+ fd, err := unix.Open(path, unix.O_RDONLY|unix.O_CLOEXEC, 0) |
|
| 75 | 118 |
if err != nil {
|
| 76 | 119 |
return -1, err |
| 77 | 120 |
} |