vendor: update c/common

Signed-off-by: Giuseppe Scrivano <gscrivan@redhat.com>
This commit is contained in:
Giuseppe Scrivano
2023-11-02 12:48:36 +01:00
parent d2a37222b9
commit 33753db47b
22 changed files with 647 additions and 1250 deletions

View File

@ -1,151 +0,0 @@
//go:build !linux
// +build !linux
package cgroups
import (
"bufio"
"errors"
"fmt"
"os"
"path/filepath"
"strconv"
"strings"
spec "github.com/opencontainers/runtime-spec/specs-go"
)
type blkioHandler struct{}
func getBlkioHandler() *blkioHandler {
return &blkioHandler{}
}
// Apply set the specified constraints
func (c *blkioHandler) Apply(ctr *CgroupControl, res *spec.LinuxResources) error {
if res.BlockIO == nil {
return nil
}
return fmt.Errorf("blkio apply function not implemented yet")
}
// Create the cgroup
func (c *blkioHandler) Create(ctr *CgroupControl) (bool, error) {
if ctr.cgroup2 {
return false, nil
}
return ctr.createCgroupDirectory(Blkio)
}
// Destroy the cgroup
func (c *blkioHandler) Destroy(ctr *CgroupControl) error {
return rmDirRecursively(ctr.getCgroupv1Path(Blkio))
}
// Stat fills a metrics structure with usage stats for the controller
func (c *blkioHandler) Stat(ctr *CgroupControl, m *Metrics) error {
var ioServiceBytesRecursive []BlkIOEntry
if ctr.cgroup2 {
// more details on the io.stat file format:X https://facebookmicrosites.github.io/cgroup2/docs/io-controller.html
values, err := readCgroup2MapFile(ctr, "io.stat")
if err != nil {
return err
}
for k, v := range values {
d := strings.Split(k, ":")
if len(d) != 2 {
continue
}
minor, err := strconv.ParseUint(d[0], 10, 0)
if err != nil {
return err
}
major, err := strconv.ParseUint(d[1], 10, 0)
if err != nil {
return err
}
for _, item := range v {
d := strings.Split(item, "=")
if len(d) != 2 {
continue
}
op := d[0]
// Accommodate the cgroup v1 naming
switch op {
case "rbytes":
op = "read"
case "wbytes":
op = "write"
}
value, err := strconv.ParseUint(d[1], 10, 0)
if err != nil {
return err
}
entry := BlkIOEntry{
Op: op,
Major: major,
Minor: minor,
Value: value,
}
ioServiceBytesRecursive = append(ioServiceBytesRecursive, entry)
}
}
} else {
BlkioRoot := ctr.getCgroupv1Path(Blkio)
p := filepath.Join(BlkioRoot, "blkio.throttle.io_service_bytes_recursive")
f, err := os.Open(p)
if err != nil {
if errors.Is(err, os.ErrNotExist) {
return nil
}
return fmt.Errorf("open %s: %w", p, err)
}
defer f.Close()
scanner := bufio.NewScanner(f)
for scanner.Scan() {
line := scanner.Text()
parts := strings.Fields(line)
if len(parts) < 3 {
continue
}
d := strings.Split(parts[0], ":")
if len(d) != 2 {
continue
}
minor, err := strconv.ParseUint(d[0], 10, 0)
if err != nil {
return err
}
major, err := strconv.ParseUint(d[1], 10, 0)
if err != nil {
return err
}
op := parts[1]
value, err := strconv.ParseUint(parts[2], 10, 0)
if err != nil {
return err
}
entry := BlkIOEntry{
Op: op,
Major: major,
Minor: minor,
Value: value,
}
ioServiceBytesRecursive = append(ioServiceBytesRecursive, entry)
}
if err := scanner.Err(); err != nil {
return fmt.Errorf("parse %s: %w", p, err)
}
}
m.Blkio = BlkioMetrics{IoServiceBytesRecursive: ioServiceBytesRecursive}
return nil
}

View File

@ -1,614 +0,0 @@
//go:build !linux
// +build !linux
package cgroups
import (
"bufio"
"context"
"errors"
"fmt"
"math"
"os"
"path/filepath"
"strconv"
"strings"
"github.com/containers/storage/pkg/unshare"
systemdDbus "github.com/coreos/go-systemd/v22/dbus"
"github.com/godbus/dbus/v5"
spec "github.com/opencontainers/runtime-spec/specs-go"
"github.com/sirupsen/logrus"
)
var (
// ErrCgroupDeleted means the cgroup was deleted
ErrCgroupDeleted = errors.New("cgroup deleted")
// ErrCgroupV1Rootless means the cgroup v1 were attempted to be used in rootless environment
ErrCgroupV1Rootless = errors.New("no support for CGroups V1 in rootless environments")
ErrStatCgroup = errors.New("no cgroup available for gathering user statistics")
)
// CgroupControl controls a cgroup hierarchy
type CgroupControl struct {
cgroup2 bool
path string
systemd bool
// List of additional cgroup subsystems joined that
// do not have a custom handler.
additionalControllers []controller
}
// CPUUsage keeps stats for the CPU usage (unit: nanoseconds)
type CPUUsage struct {
Kernel uint64
Total uint64
PerCPU []uint64
}
// MemoryUsage keeps stats for the memory usage
type MemoryUsage struct {
Usage uint64
Limit uint64
}
// CPUMetrics keeps stats for the CPU usage
type CPUMetrics struct {
Usage CPUUsage
}
// BlkIOEntry describes an entry in the blkio stats
type BlkIOEntry struct {
Op string
Major uint64
Minor uint64
Value uint64
}
// BlkioMetrics keeps usage stats for the blkio cgroup controller
type BlkioMetrics struct {
IoServiceBytesRecursive []BlkIOEntry
}
// MemoryMetrics keeps usage stats for the memory cgroup controller
type MemoryMetrics struct {
Usage MemoryUsage
}
// PidsMetrics keeps usage stats for the pids cgroup controller
type PidsMetrics struct {
Current uint64
}
// Metrics keeps usage stats for the cgroup controllers
type Metrics struct {
CPU CPUMetrics
Blkio BlkioMetrics
Memory MemoryMetrics
Pids PidsMetrics
}
type controller struct {
name string
symlink bool
}
type controllerHandler interface {
Create(*CgroupControl) (bool, error)
Apply(*CgroupControl, *spec.LinuxResources) error
Destroy(*CgroupControl) error
Stat(*CgroupControl, *Metrics) error
}
const (
cgroupRoot = "/sys/fs/cgroup"
// CPU is the cpu controller
CPU = "cpu"
// CPUAcct is the cpuacct controller
CPUAcct = "cpuacct"
// CPUset is the cpuset controller
CPUset = "cpuset"
// Memory is the memory controller
Memory = "memory"
// Pids is the pids controller
Pids = "pids"
// Blkio is the blkio controller
Blkio = "blkio"
)
var handlers map[string]controllerHandler
func init() {
handlers = make(map[string]controllerHandler)
handlers[CPU] = getCPUHandler()
handlers[CPUset] = getCpusetHandler()
handlers[Memory] = getMemoryHandler()
handlers[Pids] = getPidsHandler()
handlers[Blkio] = getBlkioHandler()
}
// getAvailableControllers get the available controllers
func getAvailableControllers(exclude map[string]controllerHandler, cgroup2 bool) ([]controller, error) {
if cgroup2 {
controllers := []controller{}
controllersFile := cgroupRoot + "/cgroup.controllers"
// rootless cgroupv2: check available controllers for current user, systemd or servicescope will inherit
if unshare.IsRootless() {
userSlice, err := getCgroupPathForCurrentProcess()
if err != nil {
return controllers, err
}
// userSlice already contains '/' so not adding here
basePath := cgroupRoot + userSlice
controllersFile = fmt.Sprintf("%s/cgroup.controllers", basePath)
}
controllersFileBytes, err := os.ReadFile(controllersFile)
if err != nil {
return nil, fmt.Errorf("failed while reading controllers for cgroup v2: %w", err)
}
for _, controllerName := range strings.Fields(string(controllersFileBytes)) {
c := controller{
name: controllerName,
symlink: false,
}
controllers = append(controllers, c)
}
return controllers, nil
}
subsystems, _ := cgroupV1GetAllSubsystems()
controllers := []controller{}
// cgroupv1 and rootless: No subsystem is available: delegation is unsafe.
if unshare.IsRootless() {
return controllers, nil
}
for _, name := range subsystems {
if _, found := exclude[name]; found {
continue
}
fileInfo, err := os.Stat(cgroupRoot + "/" + name)
if err != nil {
continue
}
c := controller{
name: name,
symlink: !fileInfo.IsDir(),
}
controllers = append(controllers, c)
}
return controllers, nil
}
// GetAvailableControllers get string:bool map of all the available controllers
func GetAvailableControllers(exclude map[string]controllerHandler, cgroup2 bool) ([]string, error) {
availableControllers, err := getAvailableControllers(exclude, cgroup2)
if err != nil {
return nil, err
}
controllerList := []string{}
for _, controller := range availableControllers {
controllerList = append(controllerList, controller.name)
}
return controllerList, nil
}
func cgroupV1GetAllSubsystems() ([]string, error) {
f, err := os.Open("/proc/cgroups")
if err != nil {
return nil, err
}
defer f.Close()
subsystems := []string{}
s := bufio.NewScanner(f)
for s.Scan() {
text := s.Text()
if text[0] != '#' {
parts := strings.Fields(text)
if len(parts) >= 4 && parts[3] != "0" {
subsystems = append(subsystems, parts[0])
}
}
}
if err := s.Err(); err != nil {
return nil, err
}
return subsystems, nil
}
func getCgroupPathForCurrentProcess() (string, error) {
path := fmt.Sprintf("/proc/%d/cgroup", os.Getpid())
f, err := os.Open(path)
if err != nil {
return "", err
}
defer f.Close()
cgroupPath := ""
s := bufio.NewScanner(f)
for s.Scan() {
text := s.Text()
procEntries := strings.SplitN(text, "::", 2)
// set process cgroupPath only if entry is valid
if len(procEntries) > 1 {
cgroupPath = procEntries[1]
}
}
if err := s.Err(); err != nil {
return cgroupPath, err
}
return cgroupPath, nil
}
// getCgroupv1Path is a helper function to get the cgroup v1 path
func (c *CgroupControl) getCgroupv1Path(name string) string {
return filepath.Join(cgroupRoot, name, c.path)
}
// initialize initializes the specified hierarchy
func (c *CgroupControl) initialize() (err error) {
createdSoFar := map[string]controllerHandler{}
defer func() {
if err != nil {
for name, ctr := range createdSoFar {
if err := ctr.Destroy(c); err != nil {
logrus.Warningf("error cleaning up controller %s for %s", name, c.path)
}
}
}
}()
if c.cgroup2 {
if err := createCgroupv2Path(filepath.Join(cgroupRoot, c.path)); err != nil {
return fmt.Errorf("creating cgroup path %s: %w", c.path, err)
}
}
for name, handler := range handlers {
created, err := handler.Create(c)
if err != nil {
return err
}
if created {
createdSoFar[name] = handler
}
}
if !c.cgroup2 {
// We won't need to do this for cgroup v2
for _, ctr := range c.additionalControllers {
if ctr.symlink {
continue
}
path := c.getCgroupv1Path(ctr.name)
if err := os.MkdirAll(path, 0o755); err != nil {
return fmt.Errorf("creating cgroup path for %s: %w", ctr.name, err)
}
}
}
return nil
}
func readFileAsUint64(path string) (uint64, error) {
data, err := os.ReadFile(path)
if err != nil {
return 0, err
}
v := cleanString(string(data))
if v == "max" {
return math.MaxUint64, nil
}
ret, err := strconv.ParseUint(v, 10, 64)
if err != nil {
return ret, fmt.Errorf("parse %s from %s: %w", v, path, err)
}
return ret, nil
}
func readFileByKeyAsUint64(path, key string) (uint64, error) {
content, err := os.ReadFile(path)
if err != nil {
return 0, err
}
for _, line := range strings.Split(string(content), "\n") {
fields := strings.SplitN(line, " ", 2)
if fields[0] == key {
v := cleanString(string(fields[1]))
if v == "max" {
return math.MaxUint64, nil
}
ret, err := strconv.ParseUint(v, 10, 64)
if err != nil {
return ret, fmt.Errorf("parse %s from %s: %w", v, path, err)
}
return ret, nil
}
}
return 0, fmt.Errorf("no key named %s from %s", key, path)
}
// New creates a new cgroup control
func New(path string, resources *spec.LinuxResources) (*CgroupControl, error) {
cgroup2, err := IsCgroup2UnifiedMode()
if err != nil {
return nil, err
}
control := &CgroupControl{
cgroup2: cgroup2,
path: path,
}
if !cgroup2 {
controllers, err := getAvailableControllers(handlers, false)
if err != nil {
return nil, err
}
control.additionalControllers = controllers
}
if err := control.initialize(); err != nil {
return nil, err
}
return control, nil
}
// NewSystemd creates a new cgroup control
func NewSystemd(path string) (*CgroupControl, error) {
cgroup2, err := IsCgroup2UnifiedMode()
if err != nil {
return nil, err
}
control := &CgroupControl{
cgroup2: cgroup2,
path: path,
systemd: true,
}
return control, nil
}
// Load loads an existing cgroup control
func Load(path string) (*CgroupControl, error) {
cgroup2, err := IsCgroup2UnifiedMode()
if err != nil {
return nil, err
}
control := &CgroupControl{
cgroup2: cgroup2,
path: path,
systemd: false,
}
if !cgroup2 {
controllers, err := getAvailableControllers(handlers, false)
if err != nil {
return nil, err
}
control.additionalControllers = controllers
}
if !cgroup2 {
oneExists := false
// check that the cgroup exists at least under one controller
for name := range handlers {
p := control.getCgroupv1Path(name)
if _, err := os.Stat(p); err == nil {
oneExists = true
break
}
}
// if there is no controller at all, raise an error
if !oneExists {
if unshare.IsRootless() {
return nil, ErrCgroupV1Rootless
}
// compatible with the error code
// used by containerd/cgroups
return nil, ErrCgroupDeleted
}
}
return control, nil
}
// CreateSystemdUnit creates the systemd cgroup
func (c *CgroupControl) CreateSystemdUnit(path string) error {
if !c.systemd {
return fmt.Errorf("the cgroup controller is not using systemd")
}
conn, err := systemdDbus.NewWithContext(context.TODO())
if err != nil {
return err
}
defer conn.Close()
return systemdCreate(path, conn)
}
// GetUserConnection returns a user connection to D-BUS
func GetUserConnection(uid int) (*systemdDbus.Conn, error) {
return systemdDbus.NewConnection(func() (*dbus.Conn, error) {
return dbusAuthConnection(uid, dbus.SessionBusPrivateNoAutoStartup)
})
}
// CreateSystemdUserUnit creates the systemd cgroup for the specified user
func (c *CgroupControl) CreateSystemdUserUnit(path string, uid int) error {
if !c.systemd {
return fmt.Errorf("the cgroup controller is not using systemd")
}
conn, err := GetUserConnection(uid)
if err != nil {
return err
}
defer conn.Close()
return systemdCreate(path, conn)
}
func dbusAuthConnection(uid int, createBus func(opts ...dbus.ConnOption) (*dbus.Conn, error)) (*dbus.Conn, error) {
conn, err := createBus()
if err != nil {
return nil, err
}
methods := []dbus.Auth{dbus.AuthExternal(strconv.Itoa(uid))}
err = conn.Auth(methods)
if err != nil {
conn.Close()
return nil, err
}
if err := conn.Hello(); err != nil {
return nil, err
}
return conn, nil
}
// Delete cleans a cgroup
func (c *CgroupControl) Delete() error {
return c.DeleteByPath(c.path)
}
// DeleteByPathConn deletes the specified cgroup path using the specified
// dbus connection if needed.
func (c *CgroupControl) DeleteByPathConn(path string, conn *systemdDbus.Conn) error {
if c.systemd {
return systemdDestroyConn(path, conn)
}
if c.cgroup2 {
return rmDirRecursively(filepath.Join(cgroupRoot, c.path))
}
var lastError error
for _, h := range handlers {
if err := h.Destroy(c); err != nil {
lastError = err
}
}
for _, ctr := range c.additionalControllers {
if ctr.symlink {
continue
}
p := c.getCgroupv1Path(ctr.name)
if err := rmDirRecursively(p); err != nil {
lastError = fmt.Errorf("remove %s: %w", p, err)
}
}
return lastError
}
// DeleteByPath deletes the specified cgroup path
func (c *CgroupControl) DeleteByPath(path string) error {
if c.systemd {
conn, err := systemdDbus.NewWithContext(context.TODO())
if err != nil {
return err
}
defer conn.Close()
return c.DeleteByPathConn(path, conn)
}
return c.DeleteByPathConn(path, nil)
}
// Update updates the cgroups
func (c *CgroupControl) Update(resources *spec.LinuxResources) error {
for _, h := range handlers {
if err := h.Apply(c, resources); err != nil {
return err
}
}
return nil
}
// AddPid moves the specified pid to the cgroup
func (c *CgroupControl) AddPid(pid int) error {
pidString := []byte(fmt.Sprintf("%d\n", pid))
if c.cgroup2 {
p := filepath.Join(cgroupRoot, c.path, "cgroup.procs")
if err := os.WriteFile(p, pidString, 0o644); err != nil {
return fmt.Errorf("write %s: %w", p, err)
}
return nil
}
names := make([]string, 0, len(handlers))
for n := range handlers {
names = append(names, n)
}
for _, c := range c.additionalControllers {
if !c.symlink {
names = append(names, c.name)
}
}
for _, n := range names {
// If we aren't using cgroup2, we won't write correctly to unified hierarchy
if !c.cgroup2 && n == "unified" {
continue
}
p := filepath.Join(c.getCgroupv1Path(n), "tasks")
if err := os.WriteFile(p, pidString, 0o644); err != nil {
return fmt.Errorf("write %s: %w", p, err)
}
}
return nil
}
// Stat returns usage statistics for the cgroup
func (c *CgroupControl) Stat() (*Metrics, error) {
m := Metrics{}
found := false
for _, h := range handlers {
if err := h.Stat(c, &m); err != nil {
if !errors.Is(err, os.ErrNotExist) {
return nil, err
}
logrus.Warningf("Failed to retrieve cgroup stats: %v", err)
continue
}
found = true
}
if !found {
return nil, ErrStatCgroup
}
return &m, nil
}
func readCgroupMapPath(path string) (map[string][]string, error) {
ret := map[string][]string{}
f, err := os.Open(path)
if err != nil {
if errors.Is(err, os.ErrNotExist) {
return ret, nil
}
return nil, fmt.Errorf("open file %s: %w", path, err)
}
defer f.Close()
scanner := bufio.NewScanner(f)
for scanner.Scan() {
line := scanner.Text()
parts := strings.Fields(line)
if len(parts) < 2 {
continue
}
ret[parts[0]] = parts[1:]
}
if err := scanner.Err(); err != nil {
return nil, fmt.Errorf("parsing file %s: %w", path, err)
}
return ret, nil
}
func readCgroup2MapFile(ctr *CgroupControl, name string) (map[string][]string, error) {
p := filepath.Join(cgroupRoot, ctr.path, name)
return readCgroupMapPath(p)
}

View File

@ -5,6 +5,7 @@ package cgroups
import (
"bufio"
"bytes"
"context"
"errors"
"fmt"
@ -134,8 +135,8 @@ func getAvailableControllers(exclude map[string]controllerHandler, cgroup2 bool)
return controllers, nil
}
// GetAvailableControllers get string:bool map of all the available controllers
func GetAvailableControllers(exclude map[string]controllerHandler, cgroup2 bool) ([]string, error) {
// AvailableControllers get string:bool map of all the available controllers
func AvailableControllers(exclude map[string]controllerHandler, cgroup2 bool) ([]string, error) {
availableControllers, err := getAvailableControllers(exclude, cgroup2)
if err != nil {
return nil, err
@ -391,20 +392,13 @@ func (c *CgroupControl) CreateSystemdUnit(path string) error {
return systemdCreate(c.config.Resources, path, conn)
}
// GetUserConnection returns an user connection to D-BUS
func GetUserConnection(uid int) (*systemdDbus.Conn, error) {
return systemdDbus.NewConnection(func() (*dbus.Conn, error) {
return dbusAuthConnection(uid, dbus.SessionBusPrivateNoAutoStartup)
})
}
// CreateSystemdUserUnit creates the systemd cgroup for the specified user
func (c *CgroupControl) CreateSystemdUserUnit(path string, uid int) error {
if !c.systemd {
return fmt.Errorf("the cgroup controller is not using systemd")
}
conn, err := GetUserConnection(uid)
conn, err := UserConnection(uid)
if err != nil {
return err
}
@ -572,3 +566,170 @@ func readCgroup2MapFile(ctr *CgroupControl, name string) (map[string][]string, e
return readCgroupMapPath(p)
}
func (c *CgroupControl) createCgroupDirectory(controller string) (bool, error) {
cPath := c.getCgroupv1Path(controller)
_, err := os.Stat(cPath)
if err == nil {
return false, nil
}
if !errors.Is(err, os.ErrNotExist) {
return false, err
}
if err := os.MkdirAll(cPath, 0o755); err != nil {
return false, fmt.Errorf("creating cgroup for %s: %w", controller, err)
}
return true, nil
}
var TestMode bool
func createCgroupv2Path(path string) (deferredError error) {
if !strings.HasPrefix(path, cgroupRoot+"/") {
return fmt.Errorf("invalid cgroup path %s", path)
}
content, err := os.ReadFile(cgroupRoot + "/cgroup.controllers")
if err != nil {
return err
}
ctrs := bytes.Fields(content)
res := append([]byte("+"), bytes.Join(ctrs, []byte(" +"))...)
current := "/sys/fs"
elements := strings.Split(path, "/")
for i, e := range elements[3:] {
current = filepath.Join(current, e)
if i > 0 {
if err := os.Mkdir(current, 0o755); err != nil {
if !os.IsExist(err) {
return err
}
} else {
// If the directory was created, be sure it is not left around on errors.
defer func() {
if deferredError != nil {
os.Remove(current)
}
}()
}
}
// We enable the controllers for all the path components except the last one. It is not allowed to add
// PIDs if there are already enabled controllers.
if i < len(elements[3:])-1 {
if err := os.WriteFile(filepath.Join(current, "cgroup.subtree_control"), res, 0o755); err != nil {
return err
}
}
}
return nil
}
func cleanString(s string) string {
return strings.Trim(s, "\n")
}
func readAcct(ctr *CgroupControl, name string) (uint64, error) {
p := filepath.Join(ctr.getCgroupv1Path(CPUAcct), name)
return readFileAsUint64(p)
}
func readAcctList(ctr *CgroupControl, name string) ([]uint64, error) {
p := filepath.Join(ctr.getCgroupv1Path(CPUAcct), name)
data, err := os.ReadFile(p)
if err != nil {
return nil, err
}
r := []uint64{}
for _, s := range strings.Split(string(data), " ") {
s = cleanString(s)
if s == "" {
break
}
v, err := strconv.ParseUint(s, 10, 64)
if err != nil {
return nil, fmt.Errorf("parsing %s: %w", s, err)
}
r = append(r, v)
}
return r, nil
}
func cpusetCopyFromParent(path string, cgroupv2 bool) error {
for _, file := range []string{"cpuset.cpus", "cpuset.mems"} {
if _, err := cpusetCopyFileFromParent(path, file, cgroupv2); err != nil {
return err
}
}
return nil
}
func cpusetCopyFileFromParent(dir, file string, cgroupv2 bool) ([]byte, error) {
if dir == cgroupRoot {
return nil, fmt.Errorf("could not find parent to initialize cpuset %s", file)
}
path := filepath.Join(dir, file)
parentPath := path
if cgroupv2 {
parentPath = fmt.Sprintf("%s.effective", parentPath)
}
data, err := os.ReadFile(parentPath)
if err != nil {
// if the file doesn't exist, it is likely that the cpuset controller
// is not enabled in the kernel.
if os.IsNotExist(err) {
return nil, nil
}
return nil, err
}
if strings.Trim(string(data), "\n") != "" {
return data, nil
}
data, err = cpusetCopyFileFromParent(filepath.Dir(dir), file, cgroupv2)
if err != nil {
return nil, err
}
if err := os.WriteFile(path, data, 0o644); err != nil {
return nil, fmt.Errorf("write %s: %w", path, err)
}
return data, nil
}
// SystemCPUUsage returns the system usage for all the cgroups
func SystemCPUUsage() (uint64, error) {
cgroupv2, err := IsCgroup2UnifiedMode()
if err != nil {
return 0, err
}
if !cgroupv2 {
p := filepath.Join(cgroupRoot, CPUAcct, "cpuacct.usage")
return readFileAsUint64(p)
}
files, err := os.ReadDir(cgroupRoot)
if err != nil {
return 0, err
}
var total uint64
for _, file := range files {
if !file.IsDir() {
continue
}
p := filepath.Join(cgroupRoot, file.Name(), "cpu.stat")
values, err := readCgroupMapPath(p)
if err != nil {
return 0, err
}
if val, found := values["usage_usec"]; found {
v, err := strconv.ParseUint(cleanString(val[0]), 10, 64)
if err != nil {
return 0, err
}
total += v * 1000
}
}
return total, nil
}

View File

@ -15,6 +15,8 @@ import (
"syscall"
"time"
systemdDbus "github.com/coreos/go-systemd/v22/dbus"
"github.com/godbus/dbus/v5"
"golang.org/x/sys/unix"
)
@ -37,6 +39,13 @@ func IsCgroup2UnifiedMode() (bool, error) {
return isUnified, isUnifiedErr
}
// UserConnection returns an user connection to D-BUS
func UserConnection(uid int) (*systemdDbus.Conn, error) {
return systemdDbus.NewConnection(func() (*dbus.Conn, error) {
return dbusAuthConnection(uid, dbus.SessionBusPrivateNoAutoStartup)
})
}
// UserOwnsCurrentSystemdCgroup checks whether the current EUID owns the
// current cgroup.
func UserOwnsCurrentSystemdCgroup() (bool, error) {

View File

@ -4,7 +4,10 @@
package cgroups
import (
"fmt"
"os"
systemdDbus "github.com/coreos/go-systemd/v22/dbus"
)
// IsCgroup2UnifiedMode returns whether we are running in cgroup 2 cgroup2 mode.
@ -21,3 +24,8 @@ func UserOwnsCurrentSystemdCgroup() (bool, error) {
func rmDirRecursively(path string) error {
return os.RemoveAll(path)
}
// UserConnection returns an user connection to D-BUS
func UserConnection(uid int) (*systemdDbus.Conn, error) {
return nil, fmt.Errorf("systemd d-bus is not supported on this platform")
}

View File

@ -1,91 +0,0 @@
//go:build !linux
// +build !linux
package cgroups
import (
"errors"
"fmt"
"os"
"strconv"
spec "github.com/opencontainers/runtime-spec/specs-go"
)
type cpuHandler struct{}
func getCPUHandler() *cpuHandler {
return &cpuHandler{}
}
// Apply set the specified constraints
func (c *cpuHandler) Apply(ctr *CgroupControl, res *spec.LinuxResources) error {
if res.CPU == nil {
return nil
}
return fmt.Errorf("cpu apply not implemented yet")
}
// Create the cgroup
func (c *cpuHandler) Create(ctr *CgroupControl) (bool, error) {
if ctr.cgroup2 {
return false, nil
}
return ctr.createCgroupDirectory(CPU)
}
// Destroy the cgroup
func (c *cpuHandler) Destroy(ctr *CgroupControl) error {
return rmDirRecursively(ctr.getCgroupv1Path(CPU))
}
// Stat fills a metrics structure with usage stats for the controller
func (c *cpuHandler) Stat(ctr *CgroupControl, m *Metrics) error {
var err error
usage := CPUUsage{}
if ctr.cgroup2 {
values, err := readCgroup2MapFile(ctr, "cpu.stat")
if err != nil {
return err
}
if val, found := values["usage_usec"]; found {
usage.Total, err = strconv.ParseUint(cleanString(val[0]), 10, 64)
if err != nil {
return err
}
usage.Kernel *= 1000
}
if val, found := values["system_usec"]; found {
usage.Kernel, err = strconv.ParseUint(cleanString(val[0]), 10, 64)
if err != nil {
return err
}
usage.Total *= 1000
}
// FIXME: How to read usage.PerCPU?
} else {
usage.Total, err = readAcct(ctr, "cpuacct.usage")
if err != nil {
if !errors.Is(err, os.ErrNotExist) {
return err
}
usage.Total = 0
}
usage.Kernel, err = readAcct(ctr, "cpuacct.usage_sys")
if err != nil {
if !errors.Is(err, os.ErrNotExist) {
return err
}
usage.Kernel = 0
}
usage.PerCPU, err = readAcctList(ctr, "cpuacct.usage_percpu")
if err != nil {
if !errors.Is(err, os.ErrNotExist) {
return err
}
usage.PerCPU = nil
}
}
m.CPU = CPUMetrics{Usage: usage}
return nil
}

View File

@ -1,49 +0,0 @@
//go:build !linux
// +build !linux
package cgroups
import (
"fmt"
"path/filepath"
spec "github.com/opencontainers/runtime-spec/specs-go"
)
type cpusetHandler struct{}
func getCpusetHandler() *cpusetHandler {
return &cpusetHandler{}
}
// Apply set the specified constraints
func (c *cpusetHandler) Apply(ctr *CgroupControl, res *spec.LinuxResources) error {
if res.CPU == nil {
return nil
}
return fmt.Errorf("cpuset apply not implemented yet")
}
// Create the cgroup
func (c *cpusetHandler) Create(ctr *CgroupControl) (bool, error) {
if ctr.cgroup2 {
path := filepath.Join(cgroupRoot, ctr.path)
return true, cpusetCopyFromParent(path, true)
}
created, err := ctr.createCgroupDirectory(CPUset)
if !created || err != nil {
return created, err
}
return true, cpusetCopyFromParent(ctr.getCgroupv1Path(CPUset), false)
}
// Destroy the cgroup
func (c *cpusetHandler) Destroy(ctr *CgroupControl) error {
return rmDirRecursively(ctr.getCgroupv1Path(CPUset))
}
// Stat fills a metrics structure with usage stats for the controller
func (c *cpusetHandler) Stat(ctr *CgroupControl, m *Metrics) error {
return nil
}

View File

@ -1,69 +0,0 @@
//go:build !linux
// +build !linux
package cgroups
import (
"fmt"
"path/filepath"
spec "github.com/opencontainers/runtime-spec/specs-go"
)
type memHandler struct{}
func getMemoryHandler() *memHandler {
return &memHandler{}
}
// Apply set the specified constraints
func (c *memHandler) Apply(ctr *CgroupControl, res *spec.LinuxResources) error {
if res.Memory == nil {
return nil
}
return fmt.Errorf("memory apply not implemented yet")
}
// Create the cgroup
func (c *memHandler) Create(ctr *CgroupControl) (bool, error) {
if ctr.cgroup2 {
return false, nil
}
return ctr.createCgroupDirectory(Memory)
}
// Destroy the cgroup
func (c *memHandler) Destroy(ctr *CgroupControl) error {
return rmDirRecursively(ctr.getCgroupv1Path(Memory))
}
// Stat fills a metrics structure with usage stats for the controller
func (c *memHandler) Stat(ctr *CgroupControl, m *Metrics) error {
var err error
usage := MemoryUsage{}
var memoryRoot string
var limitFilename string
if ctr.cgroup2 {
memoryRoot = filepath.Join(cgroupRoot, ctr.path)
limitFilename = "memory.max"
if usage.Usage, err = readFileByKeyAsUint64(filepath.Join(memoryRoot, "memory.stat"), "anon"); err != nil {
return err
}
} else {
memoryRoot = ctr.getCgroupv1Path(Memory)
limitFilename = "memory.limit_in_bytes"
if usage.Usage, err = readFileAsUint64(filepath.Join(memoryRoot, "memory.usage_in_bytes")); err != nil {
return err
}
}
usage.Limit, err = readFileAsUint64(filepath.Join(memoryRoot, limitFilename))
if err != nil {
return err
}
m.Memory = MemoryMetrics{Usage: usage}
return nil
}

View File

@ -1,71 +0,0 @@
//go:build !linux
// +build !linux
package cgroups
import (
"fmt"
"os"
"path/filepath"
spec "github.com/opencontainers/runtime-spec/specs-go"
)
type pidHandler struct{}
func getPidsHandler() *pidHandler {
return &pidHandler{}
}
// Apply set the specified constraints
func (c *pidHandler) Apply(ctr *CgroupControl, res *spec.LinuxResources) error {
if res.Pids == nil {
return nil
}
var PIDRoot string
if ctr.cgroup2 {
PIDRoot = filepath.Join(cgroupRoot, ctr.path)
} else {
PIDRoot = ctr.getCgroupv1Path(Pids)
}
p := filepath.Join(PIDRoot, "pids.max")
return os.WriteFile(p, []byte(fmt.Sprintf("%d\n", res.Pids.Limit)), 0o644)
}
// Create the cgroup
func (c *pidHandler) Create(ctr *CgroupControl) (bool, error) {
if ctr.cgroup2 {
return false, nil
}
return ctr.createCgroupDirectory(Pids)
}
// Destroy the cgroup
func (c *pidHandler) Destroy(ctr *CgroupControl) error {
return rmDirRecursively(ctr.getCgroupv1Path(Pids))
}
// Stat fills a metrics structure with usage stats for the controller
func (c *pidHandler) Stat(ctr *CgroupControl, m *Metrics) error {
if ctr.path == "" {
// nothing we can do to retrieve the pids.current path
return nil
}
var PIDRoot string
if ctr.cgroup2 {
PIDRoot = filepath.Join(cgroupRoot, ctr.path)
} else {
PIDRoot = ctr.getCgroupv1Path(Pids)
}
current, err := readFileAsUint64(filepath.Join(PIDRoot, "pids.current"))
if err != nil {
return err
}
m.Pids = PidsMetrics{Current: current}
return nil
}

View File

@ -1,179 +0,0 @@
package cgroups
import (
"bytes"
"errors"
"fmt"
"os"
"path/filepath"
"strconv"
"strings"
)
var TestMode bool
func cleanString(s string) string {
return strings.Trim(s, "\n")
}
func readAcct(ctr *CgroupControl, name string) (uint64, error) {
p := filepath.Join(ctr.getCgroupv1Path(CPUAcct), name)
return readFileAsUint64(p)
}
func readAcctList(ctr *CgroupControl, name string) ([]uint64, error) {
p := filepath.Join(ctr.getCgroupv1Path(CPUAcct), name)
data, err := os.ReadFile(p)
if err != nil {
return nil, err
}
r := []uint64{}
for _, s := range strings.Split(string(data), " ") {
s = cleanString(s)
if s == "" {
break
}
v, err := strconv.ParseUint(s, 10, 64)
if err != nil {
return nil, fmt.Errorf("parsing %s: %w", s, err)
}
r = append(r, v)
}
return r, nil
}
// GetSystemCPUUsage returns the system usage for all the cgroups
func GetSystemCPUUsage() (uint64, error) {
cgroupv2, err := IsCgroup2UnifiedMode()
if err != nil {
return 0, err
}
if !cgroupv2 {
p := filepath.Join(cgroupRoot, CPUAcct, "cpuacct.usage")
return readFileAsUint64(p)
}
files, err := os.ReadDir(cgroupRoot)
if err != nil {
return 0, err
}
var total uint64
for _, file := range files {
if !file.IsDir() {
continue
}
p := filepath.Join(cgroupRoot, file.Name(), "cpu.stat")
values, err := readCgroupMapPath(p)
if err != nil {
return 0, err
}
if val, found := values["usage_usec"]; found {
v, err := strconv.ParseUint(cleanString(val[0]), 10, 64)
if err != nil {
return 0, err
}
total += v * 1000
}
}
return total, nil
}
func cpusetCopyFileFromParent(dir, file string, cgroupv2 bool) ([]byte, error) {
if dir == cgroupRoot {
return nil, fmt.Errorf("could not find parent to initialize cpuset %s", file)
}
path := filepath.Join(dir, file)
parentPath := path
if cgroupv2 {
parentPath = fmt.Sprintf("%s.effective", parentPath)
}
data, err := os.ReadFile(parentPath)
if err != nil {
// if the file doesn't exist, it is likely that the cpuset controller
// is not enabled in the kernel.
if os.IsNotExist(err) {
return nil, nil
}
return nil, err
}
if strings.Trim(string(data), "\n") != "" {
return data, nil
}
data, err = cpusetCopyFileFromParent(filepath.Dir(dir), file, cgroupv2)
if err != nil {
return nil, err
}
if err := os.WriteFile(path, data, 0o644); err != nil {
return nil, fmt.Errorf("write %s: %w", path, err)
}
return data, nil
}
func cpusetCopyFromParent(path string, cgroupv2 bool) error {
for _, file := range []string{"cpuset.cpus", "cpuset.mems"} {
if _, err := cpusetCopyFileFromParent(path, file, cgroupv2); err != nil {
return err
}
}
return nil
}
// createCgroupv2Path creates the cgroupv2 path and enables all the available controllers
func createCgroupv2Path(path string) (deferredError error) {
if !strings.HasPrefix(path, cgroupRoot+"/") {
return fmt.Errorf("invalid cgroup path %s", path)
}
content, err := os.ReadFile(cgroupRoot + "/cgroup.controllers")
if err != nil {
return err
}
ctrs := bytes.Fields(content)
res := append([]byte("+"), bytes.Join(ctrs, []byte(" +"))...)
current := "/sys/fs"
elements := strings.Split(path, "/")
for i, e := range elements[3:] {
current = filepath.Join(current, e)
if i > 0 {
if err := os.Mkdir(current, 0o755); err != nil {
if !os.IsExist(err) {
return err
}
} else {
// If the directory was created, be sure it is not left around on errors.
defer func() {
if deferredError != nil {
os.Remove(current)
}
}()
}
}
// We enable the controllers for all the path components except the last one. It is not allowed to add
// PIDs if there are already enabled controllers.
if i < len(elements[3:])-1 {
if err := os.WriteFile(filepath.Join(current, "cgroup.subtree_control"), res, 0o755); err != nil {
return err
}
}
}
return nil
}
func (c *CgroupControl) createCgroupDirectory(controller string) (bool, error) {
cPath := c.getCgroupv1Path(controller)
_, err := os.Stat(cPath)
if err == nil {
return false, nil
}
if !errors.Is(err, os.ErrNotExist) {
return false, err
}
if err := os.MkdirAll(cPath, 0o755); err != nil {
return false, fmt.Errorf("creating cgroup for %s: %w", controller, err)
}
return true, nil
}

View File

@ -104,8 +104,8 @@ func ReadFile(dir, file string) (string, error) {
return buf.String(), err
}
// GetBlkioFiles gets the proper files for blkio weights
func GetBlkioFiles(cgroupPath string) (wtFile, wtDevFile string) {
// BlkioFiles gets the proper files for blkio weights
func BlkioFiles(cgroupPath string) (wtFile, wtDevFile string) {
var weightFile string
var weightDeviceFile string
// in this important since runc keeps these variables private, they won't be set

View File

@ -10,8 +10,8 @@ import (
"regexp"
"strings"
"github.com/container-orchestrated-devices/container-device-interface/pkg/parser"
units "github.com/docker/go-units"
"tags.cncf.io/container-device-interface/pkg/parser"
)
func (c *EngineConfig) validatePaths() error {

View File

@ -37,6 +37,30 @@ const (
)
var (
DefaultMaskedPaths = []string{
"/proc/acpi",
"/proc/kcore",
"/proc/keys",
"/proc/latency_stats",
"/proc/sched_debug",
"/proc/scsi",
"/proc/timer_list",
"/proc/timer_stats",
"/sys/dev/block",
"/sys/devices/virtual/powercap",
"/sys/firmware",
"/sys/fs/selinux",
}
DefaultReadOnlyPaths = []string{
"/proc/asound",
"/proc/bus",
"/proc/fs",
"/proc/irq",
"/proc/sys",
"/proc/sysrq-trigger",
}
// DefaultInfraImage is the default image to run as infrastructure containers in pods.
DefaultInfraImage = ""
// DefaultRootlessSHMLockPath is the default path for rootless SHM locks.