1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276
|
package cgroups
import (
"errors"
"fmt"
"os"
"path/filepath"
"slices"
"strings"
"sync"
"syscall"
securejoin "github.com/cyphar/filepath-securejoin"
"github.com/moby/sys/mountinfo"
"golang.org/x/sys/unix"
)
// Code in this source file are specific to cgroup v1,
// and must not be used from any cgroup v2 code.
const (
CgroupNamePrefix = "name="
defaultPrefix = "/sys/fs/cgroup"
)
var (
errUnified = errors.New("not implemented for cgroup v2 unified hierarchy")
ErrV1NoUnified = errors.New("invalid configuration: cannot use unified on cgroup v1")
readMountinfoOnce sync.Once
readMountinfoErr error
cgroupMountinfo []*mountinfo.Info
)
type NotFoundError struct {
Subsystem string
}
func (e *NotFoundError) Error() string {
return fmt.Sprintf("mountpoint for %s not found", e.Subsystem)
}
func NewNotFoundError(sub string) error {
return &NotFoundError{
Subsystem: sub,
}
}
func IsNotFound(err error) bool {
var nfErr *NotFoundError
return errors.As(err, &nfErr)
}
func tryDefaultPath(cgroupPath, subsystem string) string {
if !strings.HasPrefix(defaultPrefix, cgroupPath) {
return ""
}
// remove possible prefix
subsystem = strings.TrimPrefix(subsystem, CgroupNamePrefix)
// Make sure we're still under defaultPrefix, and resolve
// a possible symlink (like cpu -> cpu,cpuacct).
path, err := securejoin.SecureJoin(defaultPrefix, subsystem)
if err != nil {
return ""
}
// (1) path should be a directory.
st, err := os.Lstat(path)
if err != nil || !st.IsDir() {
return ""
}
// (2) path should be a mount point.
pst, err := os.Lstat(filepath.Dir(path))
if err != nil {
return ""
}
if st.Sys().(*syscall.Stat_t).Dev == pst.Sys().(*syscall.Stat_t).Dev {
// parent dir has the same dev -- path is not a mount point
return ""
}
// (3) path should have 'cgroup' fs type.
fst := unix.Statfs_t{}
err = unix.Statfs(path, &fst)
if err != nil || fst.Type != unix.CGROUP_SUPER_MAGIC {
return ""
}
return path
}
// readCgroupMountinfo returns a list of cgroup v1 mounts (i.e. the ones
// with fstype of "cgroup") for the current running process.
//
// The results are cached (to avoid re-reading mountinfo which is relatively
// expensive), so it is assumed that cgroup mounts are not being changed.
func readCgroupMountinfo() ([]*mountinfo.Info, error) {
readMountinfoOnce.Do(func() {
// mountinfo.GetMounts uses /proc/thread-self, so we can use it without
// issues.
cgroupMountinfo, readMountinfoErr = mountinfo.GetMounts(
mountinfo.FSTypeFilter("cgroup"),
)
})
return cgroupMountinfo, readMountinfoErr
}
// https://www.kernel.org/doc/Documentation/cgroup-v1/cgroups.txt
func FindCgroupMountpoint(cgroupPath, subsystem string) (string, error) {
if IsCgroup2UnifiedMode() {
return "", errUnified
}
// If subsystem is empty, we look for the cgroupv2 hybrid path.
if len(subsystem) == 0 {
return hybridMountpoint, nil
}
// Avoid parsing mountinfo by trying the default path first, if possible.
if path := tryDefaultPath(cgroupPath, subsystem); path != "" {
return path, nil
}
mnt, _, err := FindCgroupMountpointAndRoot(cgroupPath, subsystem)
return mnt, err
}
func FindCgroupMountpointAndRoot(cgroupPath, subsystem string) (string, string, error) {
if IsCgroup2UnifiedMode() {
return "", "", errUnified
}
mi, err := readCgroupMountinfo()
if err != nil {
return "", "", err
}
return findCgroupMountpointAndRootFromMI(mi, cgroupPath, subsystem)
}
func findCgroupMountpointAndRootFromMI(mounts []*mountinfo.Info, cgroupPath, subsystem string) (string, string, error) {
for _, mi := range mounts {
if strings.HasPrefix(mi.Mountpoint, cgroupPath) {
if slices.Contains(strings.Split(mi.VFSOptions, ","), subsystem) {
return mi.Mountpoint, mi.Root, nil
}
}
}
return "", "", NewNotFoundError(subsystem)
}
func (m Mount) GetOwnCgroup(cgroups map[string]string) (string, error) {
if len(m.Subsystems) == 0 {
return "", errors.New("no subsystem for mount")
}
return getControllerPath(m.Subsystems[0], cgroups)
}
func getCgroupMountsHelper(ss map[string]bool, mounts []*mountinfo.Info, all bool) ([]Mount, error) {
res := make([]Mount, 0, len(ss))
numFound := 0
for _, mi := range mounts {
m := Mount{
Mountpoint: mi.Mountpoint,
Root: mi.Root,
}
for _, opt := range strings.Split(mi.VFSOptions, ",") {
seen, known := ss[opt]
if !known || (!all && seen) {
continue
}
ss[opt] = true
opt = strings.TrimPrefix(opt, CgroupNamePrefix)
m.Subsystems = append(m.Subsystems, opt)
numFound++
}
if len(m.Subsystems) > 0 || all {
res = append(res, m)
}
if !all && numFound >= len(ss) {
break
}
}
return res, nil
}
func getCgroupMountsV1(all bool) ([]Mount, error) {
mi, err := readCgroupMountinfo()
if err != nil {
return nil, err
}
// We don't need to use /proc/thread-self here because runc always runs
// with every thread in the same cgroup. This lets us avoid having to do
// runtime.LockOSThread.
allSubsystems, err := ParseCgroupFile("/proc/self/cgroup")
if err != nil {
return nil, err
}
allMap := make(map[string]bool)
for s := range allSubsystems {
allMap[s] = false
}
return getCgroupMountsHelper(allMap, mi, all)
}
// GetOwnCgroup returns the relative path to the cgroup docker is running in.
func GetOwnCgroup(subsystem string) (string, error) {
if IsCgroup2UnifiedMode() {
return "", errUnified
}
// We don't need to use /proc/thread-self here because runc always runs
// with every thread in the same cgroup. This lets us avoid having to do
// runtime.LockOSThread.
cgroups, err := ParseCgroupFile("/proc/self/cgroup")
if err != nil {
return "", err
}
return getControllerPath(subsystem, cgroups)
}
func GetOwnCgroupPath(subsystem string) (string, error) {
cgroup, err := GetOwnCgroup(subsystem)
if err != nil {
return "", err
}
// If subsystem is empty, we look for the cgroupv2 hybrid path.
if len(subsystem) == 0 {
return hybridMountpoint, nil
}
return getCgroupPathHelper(subsystem, cgroup)
}
func getCgroupPathHelper(subsystem, cgroup string) (string, error) {
mnt, root, err := FindCgroupMountpointAndRoot("", subsystem)
if err != nil {
return "", err
}
// This is needed for nested containers, because in /proc/self/cgroup we
// see paths from host, which don't exist in container.
relCgroup, err := filepath.Rel(root, cgroup)
if err != nil {
return "", err
}
return filepath.Join(mnt, relCgroup), nil
}
func getControllerPath(subsystem string, cgroups map[string]string) (string, error) {
if IsCgroup2UnifiedMode() {
return "", errUnified
}
if p, ok := cgroups[subsystem]; ok {
return p, nil
}
if p, ok := cgroups[CgroupNamePrefix+subsystem]; ok {
return p, nil
}
return "", NewNotFoundError(subsystem)
}
|