Merge pull request #13583 from rhatdan/ipc

Add support for ipc namespace modes "none, private, sharable"

15 files changed, 177 insertions, 45 deletions

diff --git a/docs/source/markdown/podman-container-inspect.1.md b/docs/source/markdown/podman-container-inspect.1.md
index 9945fca7c..4e45bcc40 100644
--- a/docs/source/markdown/podman-container-inspect.1.md
+++ b/docs/source/markdown/podman-container-inspect.1.md
@@ -219,7 +219,7 @@ $ podman container inspect foobar
             "DnsSearch": [],
             "ExtraHosts": [],
             "GroupAdd": [],
-            "IpcMode": "private",
+            "IpcMode": "shareable",
             "Cgroup": "",
             "Cgroups": "default",
             "Links": null,
diff --git a/docs/source/markdown/podman-create.1.md b/docs/source/markdown/podman-create.1.md
index 9af7174fc..45d0d0b3e 100644
--- a/docs/source/markdown/podman-create.1.md
+++ b/docs/source/markdown/podman-create.1.md
@@ -504,10 +504,16 @@ To specify multiple static IPv6 addresses per container, set multiple networks u
 
 #### **--ipc**=*ipc*
 
-Default is to create a private IPC namespace (POSIX SysV IPC) for the container
-		`container:<name|id>`: reuses another container shared memory, semaphores and message queues
-		`host`: use the host shared memory,semaphores and message queues inside the container. Note: the host mode gives the container full access to local shared memory and is therefore considered insecure.
-		`ns:<path>` path to an IPC namespace to join.
+Set the IPC namespace mode for a container. The default is to create
+a private IPC namespace.
+
+- "": Use Podman's default, defined in containers.conf.
+- **container:**_id_: reuses another container's shared memory, semaphores, and message queues
+- **host**: use the host's shared memory, semaphores, and message queues inside the container. Note: the host mode gives the container full access to local shared memory and is therefore considered insecure.
+- **none**:  private IPC namespace, with /dev/shm not mounted.
+- **ns:**_path_: path to an IPC namespace to join.
+- **private**: private IPC namespace.
+= **shareable**: private IPC namespace with a possibility to share it with other containers.
 
 #### **--label**, **-l**=*label*
 
diff --git a/docs/source/markdown/podman-run.1.md b/docs/source/markdown/podman-run.1.md
index ecb24ba6d..5c276c04a 100644
--- a/docs/source/markdown/podman-run.1.md
+++ b/docs/source/markdown/podman-run.1.md
@@ -528,9 +528,13 @@ To specify multiple static IPv6 addresses per container, set multiple networks u
 Set the IPC namespace mode for a container. The default is to create
 a private IPC namespace.
 
+- "": Use Podman's default, defined in containers.conf.
 - **container:**_id_: reuses another container shared memory, semaphores and message queues
 - **host**: use the host shared memory,semaphores and message queues inside the container. Note: the host mode gives the container full access to local shared memory and is therefore considered insecure.
+- **none**:  private IPC namespace, with /dev/shm not mounted.
 - **ns:**_path_: path to an IPC namespace to join.
+- **private**: private IPC namespace.
+= **shareable**: private IPC namespace with a possibility to share it with  other containers.
 
 #### **--label**, **-l**=*key*=*value*
 
diff --git a/libpod/container.go b/libpod/container.go
index 578f16905..bc3cab439 100644
--- a/libpod/container.go
+++ b/libpod/container.go
@@ -291,6 +291,13 @@ func (c *Container) Config() *ContainerConfig {
 	return returnConfig
 }
 
+// ConfigNoCopy returns the configuration used by the container.
+// Note that the returned value is not a copy and must hence
+// only be used in a reading fashion.
+func (c *Container) ConfigNoCopy() *ContainerConfig {
+	return c.config
+}
+
 // DeviceHostSrc returns the user supplied device to be passed down in the pod
 func (c *Container) DeviceHostSrc() []spec.LinuxDevice {
 	return c.config.DeviceHostSrc
diff --git a/libpod/container_config.go b/libpod/container_config.go
index 8500c6db9..371a1dec0 100644
--- a/libpod/container_config.go
+++ b/libpod/container_config.go
@@ -120,6 +120,10 @@ type ContainerRootFSConfig struct {
 	// with the size specified in ShmSize and populate this with the path of
 	// said tmpfs.
 	ShmDir string `json:"ShmDir,omitempty"`
+	// NoShmShare indicates whether /dev/shm can be shared with other containers
+	NoShmShare bool `json:"NOShmShare,omitempty"`
+	// NoShm indicates whether a tmpfs should be created and mounted on  /dev/shm
+	NoShm bool `json:"NoShm,omitempty"`
 	// ShmSize is the size of the container's SHM. Only used if ShmDir was
 	// not set manually at time of creation.
 	ShmSize int64 `json:"shmSize"`
diff --git a/libpod/container_inspect.go b/libpod/container_inspect.go
index c9d0b8a6c..14290ca0d 100644
--- a/libpod/container_inspect.go
+++ b/libpod/container_inspect.go
@@ -703,32 +703,31 @@ func (c *Container) generateInspectContainerHostConfig(ctrSpec *spec.Spec, named
 	}
 	hostConfig.CapAdd = capAdd
 	hostConfig.CapDrop = capDrop
-
-	// IPC Namespace mode
-	ipcMode := ""
-	if c.config.IPCNsCtr != "" {
-		ipcMode = fmt.Sprintf("container:%s", c.config.IPCNsCtr)
-	} else if ctrSpec.Linux != nil {
+	switch {
+	case c.config.IPCNsCtr != "":
+		hostConfig.IpcMode = fmt.Sprintf("container:%s", c.config.IPCNsCtr)
+	case ctrSpec.Linux != nil:
 		// Locate the spec's IPC namespace.
 		// If there is none, it's ipc=host.
 		// If there is one and it has a path, it's "ns:".
 		// If no path, it's default - the empty string.
-
 		for _, ns := range ctrSpec.Linux.Namespaces {
 			if ns.Type == spec.IPCNamespace {
 				if ns.Path != "" {
-					ipcMode = fmt.Sprintf("ns:%s", ns.Path)
+					hostConfig.IpcMode = fmt.Sprintf("ns:%s", ns.Path)
 				} else {
-					ipcMode = "private"
+					break
 				}
-				break
 			}
 		}
-		if ipcMode == "" {
-			ipcMode = "host"
-		}
+	case c.config.NoShm:
+		hostConfig.IpcMode = "none"
+	case c.config.NoShmShare:
+		hostConfig.IpcMode = "private"
+	}
+	if hostConfig.IpcMode == "" {
+		hostConfig.IpcMode = "shareable"
 	}
-	hostConfig.IpcMode = ipcMode
 
 	// Cgroup namespace mode
 	cgroupMode := ""
diff --git a/libpod/container_internal.go b/libpod/container_internal.go
index f1f467879..c7567a55e 100644
--- a/libpod/container_internal.go
+++ b/libpod/container_internal.go
@@ -1507,26 +1507,28 @@ func (c *Container) mountStorage() (_ string, deferredErr error) {
 		return c.state.Mountpoint, nil
 	}
 
-	mounted, err := mount.Mounted(c.config.ShmDir)
-	if err != nil {
-		return "", errors.Wrapf(err, "unable to determine if %q is mounted", c.config.ShmDir)
-	}
-
-	if !mounted && !MountExists(c.config.Spec.Mounts, "/dev/shm") {
-		shmOptions := fmt.Sprintf("mode=1777,size=%d", c.config.ShmSize)
-		if err := c.mountSHM(shmOptions); err != nil {
-			return "", err
-		}
-		if err := os.Chown(c.config.ShmDir, c.RootUID(), c.RootGID()); err != nil {
-			return "", errors.Wrapf(err, "failed to chown %s", c.config.ShmDir)
+	if !c.config.NoShm {
+		mounted, err := mount.Mounted(c.config.ShmDir)
+		if err != nil {
+			return "", errors.Wrapf(err, "unable to determine if %q is mounted", c.config.ShmDir)
 		}
-		defer func() {
-			if deferredErr != nil {
-				if err := c.unmountSHM(c.config.ShmDir); err != nil {
-					logrus.Errorf("Unmounting SHM for container %s after mount error: %v", c.ID(), err)
-				}
+
+		if !mounted && !MountExists(c.config.Spec.Mounts, "/dev/shm") {
+			shmOptions := fmt.Sprintf("mode=1777,size=%d", c.config.ShmSize)
+			if err := c.mountSHM(shmOptions); err != nil {
+				return "", err
 			}
-		}()
+			if err := os.Chown(c.config.ShmDir, c.RootUID(), c.RootGID()); err != nil {
+				return "", errors.Wrapf(err, "failed to chown %s", c.config.ShmDir)
+			}
+			defer func() {
+				if deferredErr != nil {
+					if err := c.unmountSHM(c.config.ShmDir); err != nil {
+						logrus.Errorf("Unmounting SHM for container %s after mount error: %v", c.ID(), err)
+					}
+				}
+			}()
+		}
 	}
 
 	// We need to mount the container before volumes - to ensure the copyup
diff --git a/libpod/container_internal_linux.go b/libpod/container_internal_linux.go
index 9369b746c..9991003d6 100644
--- a/libpod/container_internal_linux.go
+++ b/libpod/container_internal_linux.go
@@ -1963,8 +1963,10 @@ func (c *Container) makeBindMounts() error {
 		}
 	}
 
-	// SHM is always added when we mount the container
-	c.state.BindMounts["/dev/shm"] = c.config.ShmDir
+	if c.config.ShmDir != "" {
+		// If ShmDir has a value SHM is always added when we mount the container
+		c.state.BindMounts["/dev/shm"] = c.config.ShmDir
+	}
 
 	if c.config.Passwd == nil || *c.config.Passwd {
 		newPasswd, newGroup, err := c.generatePasswdAndGroup()
diff --git a/libpod/options.go b/libpod/options.go
index 6c4b4cc42..ffd0e6037 100644
--- a/libpod/options.go
+++ b/libpod/options.go
@@ -559,6 +559,30 @@ func WithShmDir(dir string) CtrCreateOption {
 	}
 }
 
+// WithNOShmMount tells libpod whether to mount /dev/shm
+func WithNoShm(mount bool) CtrCreateOption {
+	return func(ctr *Container) error {
+		if ctr.valid {
+			return define.ErrCtrFinalized
+		}
+
+		ctr.config.NoShm = mount
+		return nil
+	}
+}
+
+// WithNoShmShare tells libpod whether to share containers /dev/shm with other containers
+func WithNoShmShare(share bool) CtrCreateOption {
+	return func(ctr *Container) error {
+		if ctr.valid {
+			return define.ErrCtrFinalized
+		}
+
+		ctr.config.NoShmShare = share
+		return nil
+	}
+}
+
 // WithSystemd turns on systemd mode in the container
 func WithSystemd() CtrCreateOption {
 	return func(ctr *Container) error {
diff --git a/libpod/runtime_ctr.go b/libpod/runtime_ctr.go
index f92898b1c..7edd49fd1 100644
--- a/libpod/runtime_ctr.go
+++ b/libpod/runtime_ctr.go
@@ -174,6 +174,8 @@ func (r *Runtime) initContainerVariables(rSpec *spec.Spec, config *ContainerConf
 			return nil, errors.Wrapf(err, "converting containers.conf ShmSize %s to an int", r.config.Containers.ShmSize)
 		}
 		ctr.config.ShmSize = size
+		ctr.config.NoShm = false
+		ctr.config.NoShmShare = false
 		ctr.config.StopSignal = 15
 
 		ctr.config.StopTimeout = r.config.Engine.StopTimeout
@@ -514,7 +516,7 @@ func (r *Runtime) setupContainer(ctx context.Context, ctr *Container) (_ *Contai
 		}
 	}
 
-	if !MountExists(ctr.config.Spec.Mounts, "/dev/shm") && ctr.config.ShmDir == "" {
+	if !MountExists(ctr.config.Spec.Mounts, "/dev/shm") && ctr.config.ShmDir == "" && !ctr.config.NoShm {
 		ctr.config.ShmDir = filepath.Join(ctr.bundlePath(), "shm")
 		if err := os.MkdirAll(ctr.config.ShmDir, 0700); err != nil {
 			if !os.IsExist(err) {
diff --git a/pkg/specgen/generate/container.go b/pkg/specgen/generate/container.go
index b38b0e695..f7ea2edfa 100644
--- a/pkg/specgen/generate/container.go
+++ b/pkg/specgen/generate/container.go
@@ -428,9 +428,12 @@ func ConfigToSpec(rt *libpod.Runtime, specg *specgen.SpecGenerator, contaierID s
 			case "cgroup":
 				specg.CgroupNS = specgen.Namespace{NSMode: specgen.Default} //default
 			case "ipc":
-				if conf.ShmDir == "/dev/shm" {
+				switch conf.ShmDir {
+				case "/dev/shm":
 					specg.IpcNS = specgen.Namespace{NSMode: specgen.Host}
-				} else {
+				case "":
+					specg.IpcNS = specgen.Namespace{NSMode: specgen.None}
+				default:
 					specg.IpcNS = specgen.Namespace{NSMode: specgen.Default} //default
 				}
 			case "uts":
diff --git a/pkg/specgen/generate/namespaces.go b/pkg/specgen/generate/namespaces.go
index 9ce45aaf0..05c2d1741 100644
--- a/pkg/specgen/generate/namespaces.go
+++ b/pkg/specgen/generate/namespaces.go
@@ -134,8 +134,17 @@ func namespaceOptions(ctx context.Context, s *specgen.SpecGenerator, rt *libpod.
 		if err != nil {
 			return nil, errors.Wrapf(err, "error looking up container to share ipc namespace with")
 		}
+		if ipcCtr.ConfigNoCopy().NoShmShare {
+			return nil, errors.Errorf("joining IPC of container %s is not allowed: non-shareable IPC (hint: use IpcMode:shareable for the donor container)", ipcCtr.ID())
+		}
 		toReturn = append(toReturn, libpod.WithIPCNSFrom(ipcCtr))
-		toReturn = append(toReturn, libpod.WithShmDir(ipcCtr.ShmDir()))
+		if !ipcCtr.ConfigNoCopy().NoShm {
+			toReturn = append(toReturn, libpod.WithShmDir(ipcCtr.ShmDir()))
+		}
+	case specgen.None:
+		toReturn = append(toReturn, libpod.WithNoShm(true))
+	case specgen.Private:
+		toReturn = append(toReturn, libpod.WithNoShmShare(true))
 	}
 
 	// UTS
diff --git a/pkg/specgen/generate/security.go b/pkg/specgen/generate/security.go
index 988c29832..ec52164ab 100644
--- a/pkg/specgen/generate/security.go
+++ b/pkg/specgen/generate/security.go
@@ -222,7 +222,7 @@ func securityConfigureGenerator(s *specgen.SpecGenerator, g *generate.Generator,
 	for sysctlKey, sysctlVal := range defaultSysctls {
 		// Ignore mqueue sysctls if --ipc=host
 		if noUseIPC && strings.HasPrefix(sysctlKey, "fs.mqueue.") {
-			logrus.Infof("Sysctl %s=%s ignored in containers.conf, since IPC Namespace set to host", sysctlKey, sysctlVal)
+			logrus.Infof("Sysctl %s=%s ignored in containers.conf, since IPC Namespace set to %q", sysctlKey, sysctlVal, s.IpcNS.NSMode)
 
 			continue
 		}
diff --git a/pkg/specgenutil/specgen.go b/pkg/specgenutil/specgen.go
index 42b66d909..c86af7295 100644
--- a/pkg/specgenutil/specgen.go
+++ b/pkg/specgenutil/specgen.go
@@ -190,7 +190,7 @@ func setNamespaces(s *specgen.SpecGenerator, c *entities.ContainerCreateOptions)
 		}
 	}
 	if c.IPC != "" {
-		s.IpcNS, err = specgen.ParseNamespace(c.IPC)
+		s.IpcNS, err = specgen.ParseIPCNamespace(c.IPC)
 		if err != nil {
 			return err
 		}
diff --git a/test/system/190-run-ipcns.bats b/test/system/190-run-ipcns.bats
new file mode 100644
index 000000000..9327d8ec7
--- /dev/null
+++ b/test/system/190-run-ipcns.bats
@@ -0,0 +1,70 @@
+#!/usr/bin/env bats   -*- bats -*-
+# shellcheck disable=SC2096
+#
+# Tests for podman build
+#
+
+load helpers
+
+@test "podman --ipc=host" {
+    run readlink /proc/self/ns/ipc
+    hostipc=$output
+    run_podman run --rm --ipc=host $IMAGE readlink /proc/self/ns/ipc
+    is "$output" "$hostipc" "HostIPC and container IPC should be same"
+}
+
+@test "podman --ipc=none" {
+    run readlink /proc/self/ns/ipc
+    hostipc=$output
+    run_podman run --rm --ipc=none $IMAGE readlink /proc/self/ns/ipc
+    if [[ $output == "$hostipc" ]]; then
+       die "hostipc and containeripc should be different"
+    fi
+    run_podman 1 run --rm --ipc=none $IMAGE ls /dev/shm
+    is "$output" "ls: /dev/shm: No such file or directory" "Should fail with missing /dev/shm"
+}
+
+@test "podman --ipc=private" {
+    run readlink /proc/self/ns/ipc
+    hostipc=$output
+    run_podman run -d --ipc=private --name test $IMAGE sleep 100
+    if [[ $output == "$hostipc" ]]; then
+       die "hostipc and containeripc should be different"
+    fi
+    run_podman 125 run --ipc=container:test --rm $IMAGE readlink /proc/self/ns/ipc
+    is "$output" ".*is not allowed: non-shareable IPC (hint: use IpcMode:shareable for the donor container)" "Containers should not share private ipc namespace"
+    run_podman stop -t 0 test
+    run_podman rm test
+}
+
+@test "podman --ipc=shareable" {
+    run readlink /proc/self/ns/ipc
+    hostipc=$output
+    run_podman run -d --ipc=shareable --name test $IMAGE sleep 100
+    if [[ $output == "$hostipc" ]]; then
+       die "hostipc and containeripc should be different"
+    fi
+    run_podman run --ipc=container:test --rm $IMAGE readlink /proc/self/ns/ipc
+    if [[ $output == "$hostipc" ]]; then
+       die "hostipc and containeripc should be different"
+    fi
+    run_podman stop -t 0 test
+    run_podman rm test
+}
+
+@test "podman --ipc=container@test" {
+    run readlink /proc/self/ns/ipc
+    hostipc=$output
+    run_podman run -d --name test $IMAGE sleep 100
+    run_podman exec test readlink /proc/self/ns/ipc
+    if [[ $output == "$hostipc" ]]; then
+       die "hostipc and containeripc should be different"
+    fi
+    testipc=$output
+    run_podman run --ipc=container:test --rm $IMAGE readlink /proc/self/ns/ipc
+    is "$output" "$testipc" "Containers should share ipc namespace"
+    run_podman stop -t 0 test
+    run_podman rm test
+}
+
+# vim: filetype=sh