Merge pull request docker#18697 from jfrazelle/pids-cgroup

Add PIDs cgroup support to Docker

api/client/stats.go

@@ -164,7 +164,7 @@ func (cli *DockerCli) CmdStats(args ...string) error {
 			fmt.Fprint(cli.out, "\033[2J")
 			fmt.Fprint(cli.out, "\033[H")
 		}
-		io.WriteString(w, "CONTAINER\tCPU %\tMEM USAGE / LIMIT\tMEM %\tNET I/O\tBLOCK I/O\n")
+		io.WriteString(w, "CONTAINER\tCPU %\tMEM USAGE / LIMIT\tMEM %\tNET I/O\tBLOCK I/O\tPIDS\n")
 	}
 
 	for range time.Tick(500 * time.Millisecond) {

api/client/stats_helpers.go

@@ -24,6 +24,7 @@ type containerStats struct {
 	NetworkTx        float64
 	BlockRead        float64
 	BlockWrite       float64
+	PidsCurrent      uint64
 	mu               sync.RWMutex
 	err              error
 }
@@ -167,13 +168,14 @@ func (s *containerStats) Display(w io.Writer) error {
 	if s.err != nil {
 		return s.err
 	}
-	fmt.Fprintf(w, "%s\t%.2f%%\t%s / %s\t%.2f%%\t%s / %s\t%s / %s\n",
+	fmt.Fprintf(w, "%s\t%.2f%%\t%s / %s\t%.2f%%\t%s / %s\t%s / %s\t%d\n",
 		s.Name,
 		s.CPUPercentage,
 		units.HumanSize(s.Memory), units.HumanSize(s.MemoryLimit),
 		s.MemoryPercentage,
 		units.HumanSize(s.NetworkRx), units.HumanSize(s.NetworkTx),
-		units.HumanSize(s.BlockRead), units.HumanSize(s.BlockWrite))
+		units.HumanSize(s.BlockRead), units.HumanSize(s.BlockWrite),
+		s.PidsCurrent)
 	return nil
 }
 

api/client/stats_unit_test.go

@@ -19,14 +19,15 @@ func TestDisplay(t *testing.T) {
 		NetworkTx:        800 * 1024 * 1024,
 		BlockRead:        100 * 1024 * 1024,
 		BlockWrite:       800 * 1024 * 1024,
+		PidsCurrent:      1,
 		mu:               sync.RWMutex{},
 	}
 	var b bytes.Buffer
 	if err := c.Display(&b); err != nil {
 		t.Fatalf("c.Display() gave error: %s", err)
 	}
 	got := b.String()
-	want := "app\t30.00%\t104.9 MB / 2.147 GB\t4.88%\t104.9 MB / 838.9 MB\t104.9 MB / 838.9 MB\n"
+	want := "app\t30.00%\t104.9 MB / 2.147 GB\t4.88%\t104.9 MB / 838.9 MB\t104.9 MB / 838.9 MB\t1\n"
 	if got != want {
 		t.Fatalf("c.Display() = %q, want %q", got, want)
 	}

contrib/check-config.sh

@@ -202,6 +202,9 @@ echo 'Optional Features:'
 {
 	check_flags SECCOMP
 }
+{
+	check_flags CGROUP_PIDS
+}
 {
 	check_flags MEMCG_KMEM MEMCG_SWAP MEMCG_SWAP_ENABLED
 	if  is_set MEMCG_SWAP && ! is_set MEMCG_SWAP_ENABLED; then

contrib/completion/bash/docker

@@ -1637,6 +1637,7 @@ _docker_run() {
 		--net-alias
 		--oom-score-adj
 		--pid
+		--pids-limit
 		--publish -p
 		--restart
 		--security-opt

contrib/completion/zsh/_docker

@@ -534,6 +534,7 @@ __docker_subcommand() {
         "($help)*--net-alias=[Add network-scoped alias for the container]:alias: "
         "($help)--oom-kill-disable[Disable OOM Killer]"
         "($help)--oom-score-adj[Tune the host's OOM preferences for containers (accepts -1000 to 1000)]"
+        "($help)--pids-limit[Tune container pids limit (set -1 for unlimited)]"
         "($help -P --publish-all)"{-P,--publish-all}"[Publish all exposed ports]"
         "($help)*"{-p=,--publish=}"[Expose a container's port to the host]:port:_ports"
         "($help)--pid=[PID namespace to use]:PID: "

daemon/container_operations_unix.go

@@ -198,6 +198,7 @@ func (daemon *Daemon) populateCommand(c *container.Container, env []string) erro
 		BlkioThrottleWriteBpsDevice:  writeBpsDevice,
 		BlkioThrottleReadIOpsDevice:  readIOpsDevice,
 		BlkioThrottleWriteIOpsDevice: writeIOpsDevice,
+		PidsLimit:                    c.HostConfig.PidsLimit,
 		MemorySwappiness:             -1,
 	}
 

daemon/daemon_unix.go

@@ -291,6 +291,12 @@ func verifyContainerResources(resources *containertypes.Resources, sysInfo *sysi
 		resources.OomKillDisable = nil
 	}
 
+	if resources.PidsLimit != 0 && !sysInfo.PidsLimit {
+		warnings = append(warnings, "Your kernel does not support pids limit capabilities, pids limit discarded.")
+		logrus.Warnf("Your kernel does not support pids limit capabilities, pids limit discarded.")
+		resources.PidsLimit = 0
+	}
+
 	// cpu subsystem checks and adjustments
 	if resources.CPUShares > 0 && !sysInfo.CPUShares {
 		warnings = append(warnings, "Your kernel does not support CPU shares. Shares discarded.")

daemon/execdriver/driver_unix.go

@@ -50,6 +50,7 @@ type Resources struct {
 	CPUPeriod                    int64                      `json:"cpu_period"`
 	Rlimits                      []*units.Rlimit            `json:"rlimits"`
 	OomKillDisable               bool                       `json:"oom_kill_disable"`
+	PidsLimit                    int64                      `json:"pids_limit"`
 	MemorySwappiness             int64                      `json:"memory_swappiness"`
 }
 
@@ -202,6 +203,7 @@ func SetupCgroups(container *configs.Config, c *Command) error {
 		container.Cgroups.Resources.BlkioThrottleReadIOPSDevice = c.Resources.BlkioThrottleReadIOpsDevice
 		container.Cgroups.Resources.BlkioThrottleWriteIOPSDevice = c.Resources.BlkioThrottleWriteIOpsDevice
 		container.Cgroups.Resources.OomKillDisable = c.Resources.OomKillDisable
+		container.Cgroups.Resources.PidsLimit = c.Resources.PidsLimit
 		container.Cgroups.Resources.MemorySwappiness = c.Resources.MemorySwappiness
 	}
 

daemon/stats_linux.go

@@ -61,6 +61,10 @@ func convertStatsToAPITypes(ls *libcontainer.Stats) *types.StatsJSON {
 			Stats:    mem.Stats,
 			Failcnt:  mem.Usage.Failcnt,
 		}
+		pids := cs.PidsStats
+		s.PidsStats = types.PidsStats{
+			Current: pids.Current,
+		}
 	}
 
 	return s

docs/reference/api/docker_remote_api.md

@@ -123,6 +123,8 @@ This section lists each version from latest to oldest.  Each listing includes a
 * `POST /networks/create` now supports enabling ipv6 on the network by setting the `EnableIPv6` field (doing this with a label will no longer work).
 * `GET /info` now returns `CgroupDriver` field showing what cgroup driver the daemon is using; `cgroupfs` or `systemd`.
 * `GET /info` now returns `KernelMemory` field, showing if "kernel memory limit" is supported.
+* `POST /containers/create` now takes `PidsLimit` field, if the kernel is >= 4.3 and the pids cgroup is supported.
+* `GET /containers/(id or name)/stats` now returns `pids_stats`, if the kernel is >= 4.3 and the pids cgroup is supported.
 
 ### v1.22 API changes
 

docs/reference/api/docker_remote_api_v1.23.md

@@ -346,6 +346,7 @@ Json Parameters:
 -   **MemorySwappiness** - Tune a container's memory swappiness behavior. Accepts an integer between 0 and 100.
 -   **OomKillDisable** - Boolean value, whether to disable OOM Killer for the container or not.
 -   **OomScoreAdj** - An integer value containing the score given to the container in order to tune OOM killer preferences.
+-   **PidsLimit** - Tune a container's pids limit. Set -1 for unlimited.
 -   **AttachStdin** - Boolean value, attaches to `stdin`.
 -   **AttachStdout** - Boolean value, attaches to `stdout`.
 -   **AttachStderr** - Boolean value, attaches to `stderr`.
@@ -823,6 +824,9 @@ This endpoint returns a live stream of a container's resource usage statistics.
 
       {
          "read" : "2015-01-08T22:57:31.547920715Z",
+         "pids_stats": {
+            "current": 3
+         },
          "networks": {
                  "eth0": {
                      "rx_bytes": 5338,

docs/reference/commandline/create.md

@@ -74,6 +74,7 @@ Creates a new container.
       -P, --publish-all             Publish all exposed ports to random ports
       -p, --publish=[]              Publish a container's port(s) to the host
       --pid=""                      PID namespace to use
+      --pids-limit=-1                Tune container pids limit (set -1 for unlimited), kernel >= 4.3
       --privileged                  Give extended privileges to this container
       --read-only                   Mount the container's root filesystem as read only
       --restart="no"                Restart policy (no, on-failure[:max-retry], always, unless-stopped)

docs/reference/commandline/run.md

@@ -74,6 +74,7 @@ parent = "smn_cli"
       -P, --publish-all             Publish all exposed ports to random ports
       -p, --publish=[]              Publish a container's port(s) to the host
       --pid=""                      PID namespace to use
+      --pids-limit=-1                Tune container pids limit (set -1 for unlimited), kernel >= 4.3
       --privileged                  Give extended privileges to this container
       --read-only                   Mount the container's root filesystem as read only
       --restart="no"                Restart policy (no, on-failure[:max-retry], always, unless-stopped)

integration-cli/docker_cli_run_unix_test.go

@@ -968,3 +968,15 @@ func (s *DockerSuite) TestRunDeviceSymlink(c *check.C) {
 	c.Assert(err, check.NotNil)
 	c.Assert(strings.Trim(out, "\r\n"), checker.Contains, "not a device node", check.Commentf("expected output 'not a device node'"))
 }
+
+// TestRunPidsLimit makes sure the pids cgroup is set with --pids-limit
+func (s *DockerSuite) TestRunPidsLimit(c *check.C) {
+	testRequires(c, pidsLimit)
+
+	file := "/sys/fs/cgroup/pids/pids.max"
+	out, _ := dockerCmd(c, "run", "--name", "skittles", "--pids-limit", "2", "busybox", "cat", file)
+	c.Assert(strings.TrimSpace(out), checker.Equals, "2")
+
+	out = inspectField(c, "skittles", "HostConfig.PidsLimit")
+	c.Assert(out, checker.Equals, "2", check.Commentf("setting the pids limit failed"))
+}

integration-cli/requirements_unix.go

@@ -33,6 +33,12 @@ var (
 		},
 		"Test requires Oom control enabled.",
 	}
+	pidsLimit = testRequirement{
+		func() bool {
+			return SysInfo.PidsLimit
+		},
+		"Test requires pids limit enabled.",
+	}
 	kernelMemorySupport = testRequirement{
 		func() bool {
 			return SysInfo.KernelMemory

man/docker-create.1.md

@@ -58,6 +58,7 @@ docker-create - Create a new container
 [**-P**|**--publish-all**]
 [**-p**|**--publish**[=*[]*]]
 [**--pid**[=*[]*]]
+[**--pids-limit**[=*PIDS_LIMIT*]]
 [**--privileged**]
 [**--read-only**]
 [**--restart**[=*RESTART*]]
@@ -290,6 +291,9 @@ unit, `b` is used. Set LIMIT to `-1` to enable unlimited swap.
      **host**: use the host's PID namespace inside the container.
      Note: the host mode gives the container full access to local PID and is therefore considered insecure.
 
+**--pids-limit**=""
+   Tune the container's pids limit. Set `-1` to have unlimited pids for the container.
+
 **--privileged**=*true*|*false*
    Give extended privileges to this container. The default is *false*.
 

man/docker-run.1.md

@@ -60,6 +60,7 @@ docker-run - Run a command in a new container
 [**-P**|**--publish-all**]
 [**-p**|**--publish**[=*[]*]]
 [**--pid**[=*[]*]]
+[**--pids-limit**[=*PIDS_LIMIT*]]
 [**--privileged**]
 [**--read-only**]
 [**--restart**[=*RESTART*]]
@@ -420,6 +421,9 @@ Use `docker port` to see the actual mapping: `docker port CONTAINER $CONTAINERPO
      **host**: use the host's PID namespace inside the container.
      Note: the host mode gives the container full access to local PID and is therefore considered insecure.
 
+**--pids-limit**=""
+   Tune the container's pids limit. Set `-1` to have unlimited pids for the container.
+
 **--uts**=*host*
    Set the UTS mode for the container
      **host**: use the host's UTS namespace inside the container.

pkg/sysinfo/sysinfo.go

@@ -14,6 +14,7 @@ type SysInfo struct {
 	cgroupCPUInfo
 	cgroupBlkioInfo
 	cgroupCpusetInfo
+	cgroupPids
 
 	// Whether IPv4 forwarding is supported or not, if this was disabled, networking will not work
 	IPv4ForwardingDisabled bool
@@ -90,6 +91,11 @@ type cgroupCpusetInfo struct {
 	Mems string
 }
 
+type cgroupPids struct {
+	// Whether Pids Limit is supported or not
+	PidsLimit bool
+}
+
 // IsCpusetCpusAvailable returns `true` if the provided string set is contained
 // in cgroup's cpuset.cpus set, `false` otherwise.
 // If error is not nil a parsing error occurred.

pkg/sysinfo/sysinfo_linux.go

@@ -44,6 +44,7 @@ func New(quiet bool) *SysInfo {
 		sysInfo.cgroupCPUInfo = checkCgroupCPU(cgMounts, quiet)
 		sysInfo.cgroupBlkioInfo = checkCgroupBlkioInfo(cgMounts, quiet)
 		sysInfo.cgroupCpusetInfo = checkCgroupCpusetInfo(cgMounts, quiet)
+		sysInfo.cgroupPids = checkCgroupPids(quiet)
 	}
 
 	_, ok := cgMounts["devices"]
@@ -216,6 +217,21 @@ func checkCgroupCpusetInfo(cgMounts map[string]string, quiet bool) cgroupCpusetI
 	}
 }
 
+// checkCgroupPids reads the pids information from the pids cgroup mount point.
+func checkCgroupPids(quiet bool) cgroupPids {
+	_, err := cgroups.FindCgroupMountpoint("pids")
+	if err != nil {
+		if !quiet {
+			logrus.Warn(err)
+		}
+		return cgroupPids{}
+	}
+
+	return cgroupPids{
+		PidsLimit: true,
+	}
+}
+
 func cgroupEnabled(mountPoint, name string) bool {
 	_, err := os.Stat(path.Join(mountPoint, name))
 	return err == nil

runconfig/opts/parse.go

@@ -85,6 +85,7 @@ func Parse(cmd *flag.FlagSet, args []string) (*container.Config, *container.Host
 		flIPv4Address       = cmd.String([]string{"-ip"}, "", "Container IPv4 address (e.g. 172.30.100.104)")
 		flIPv6Address       = cmd.String([]string{"-ip6"}, "", "Container IPv6 address (e.g. 2001:db8::33)")
 		flIpcMode           = cmd.String([]string{"-ipc"}, "", "IPC namespace to use")
+		flPidsLimit         = cmd.Int64([]string{"-pids-limit"}, 0, "Tune container pids limit (set -1 for unlimited)")
 		flRestartPolicy     = cmd.String([]string{"-restart"}, "no", "Restart policy to apply when a container exits")
 		flReadonlyRootfs    = cmd.Bool([]string{"-read-only"}, false, "Mount the container's root filesystem as read only")
 		flLoggingDriver     = cmd.String([]string{"-log-driver"}, "", "Logging driver for container")
@@ -343,6 +344,7 @@ func Parse(cmd *flag.FlagSet, args []string) (*container.Config, *container.Host
 		CpusetCpus:           *flCpusetCpus,
 		CpusetMems:           *flCpusetMems,
 		CPUQuota:             *flCPUQuota,
+		PidsLimit:            *flPidsLimit,
 		BlkioWeight:          *flBlkioWeight,
 		BlkioWeightDevice:    flBlkioWeightDevice.GetList(),
 		BlkioDeviceReadBps:   flDeviceReadBps.GetList(),