Fix failure with rw bind mount of a ro fuse

As reported in [1], in a case where read-only fuse (sshfs) mount
is used as a volume without specifying ro flag, the kernel fails
to remount it (when adding various flags such as nosuid and nodev),
returning EPERM.

Here's the relevant strace line:

> [pid 333966] mount("/tmp/bats-run-PRVfWc/runc.RbNv8g/bundle/mnt", "/proc/self/fd/7", 0xc0001e9164, MS_NOSUID|MS_NODEV|MS_REMOUNT|MS_BIND|MS_REC, NULL) = -1 EPERM (Operation not permitted)

I was not able to reproduce it with other read-only mounts as the source
(tried tmpfs, read-only bind mount, and an ext2 mount), so somehow this
might be specific to fuse.

The fix is to check whether the source has RDONLY flag, and retry the
remount with this flag added.

A test case (which was kind of hard to write) is added, and it fails
without the fix. Note that rootless user need to be able to ssh to
rootless@localhost in order to sshfs to work -- amend setup scripts
to make it work, and skip the test if the setup is not working.

[1] https://github.com/containers/podman/issues/12205

Signed-off-by: Kir Kolyshkin <kolyshkin@gmail.com>

.cirrus.yml

@@ -99,7 +99,7 @@ task:
     # Work around dnf mirror failures by retrying a few times.
     for i in $(seq 0 2); do
       sleep $i
-      yum install -y -q gcc git iptables jq glibc-static libseccomp-devel make criu && break
+      yum install -y -q gcc git iptables jq glibc-static libseccomp-devel make criu fuse-sshfs && break
     done
     [ $? -eq 0 ] # fail if yum failed
     # install Go
@@ -113,6 +113,12 @@ task:
     cd -
     # Add a user for rootless tests
     useradd -u2000 -m -d/home/rootless -s/bin/bash rootless
+    # Allow root and rootless itself to execute `ssh rootless@localhost` in tests/rootless.sh
+    ssh-keygen -t ecdsa -N "" -f /root/rootless.key
+    mkdir -m 0700 -p /home/rootless/.ssh
+    cp /root/rootless.key /home/rootless/.ssh/id_ecdsa
+    cat /root/rootless.key.pub >> /home/rootless/.ssh/authorized_keys
+    chown -R rootless.rootless /home/rootless
     # set PATH
     echo 'export PATH=/usr/local/go/bin:/usr/local/bin:$PATH' >> /root/.bashrc
     # Setup ssh localhost for terminal emulation (script -e did not work)

.github/workflows/test.yml

@@ -46,13 +46,13 @@ jobs:
         curl -fSsl $REPO/Release.key | sudo apt-key add -
         echo "deb $REPO/ /" | sudo tee /etc/apt/sources.list.d/criu.list
         sudo apt update
-        sudo apt install libseccomp-dev criu
+        sudo apt install libseccomp-dev criu sshfs
 
     - name: install deps (criu ${{ matrix.criu }})
       if: matrix.criu != ''
       run: |
         sudo apt -q update
-        sudo apt -q install libseccomp-dev \
+        sudo apt -q install libseccomp-dev sshfs \
           libcap-dev libnet1-dev libnl-3-dev \
           libprotobuf-c-dev libprotobuf-dev protobuf-c-compiler protobuf-compiler
         git clone https://github.com/checkpoint-restore/criu.git ~/criu
@@ -81,9 +81,10 @@ jobs:
       if: matrix.rootless == 'rootless'
       run: |
         sudo useradd -u2000 -m -d/home/rootless -s/bin/bash rootless
-        # Allow root to execute `ssh rootless@localhost` in tests/rootless.sh
+        # Allow root and rootless itself to execute `ssh rootless@localhost` in tests/rootless.sh
         ssh-keygen -t ecdsa -N "" -f $HOME/rootless.key
         sudo mkdir -m 0700 -p /home/rootless/.ssh
+        sudo cp $HOME/rootless.key /home/rootless/.ssh/id_ecdsa
         sudo cp $HOME/rootless.key.pub /home/rootless/.ssh/authorized_keys
         sudo chown -R rootless.rootless /home/rootless
 

Dockerfile

@@ -31,6 +31,7 @@ RUN KEYFILE=/usr/share/keyrings/criu-repo-keyring.gpg; \
         kmod \
         pkg-config \
         python3-minimal \
+        sshfs \
         sudo \
         uidmap \
     && apt-get clean \

Vagrantfile.fedora

@@ -27,7 +27,7 @@ Vagrant.configure("2") do |config|
       cat << EOF | dnf -y --exclude=kernel,kernel-core,systemd,systemd-* shell && break
 config install_weak_deps false
 update
-install iptables gcc make golang-go glibc-static libseccomp-devel bats jq git-core criu
+install iptables gcc make golang-go glibc-static libseccomp-devel bats jq git-core criu fuse-sshfs
 ts run
 EOF
     done
@@ -36,9 +36,10 @@ EOF
     # Add a user for rootless tests
     useradd -u2000 -m -d/home/rootless -s/bin/bash rootless
 
-    # Allow root to execute `ssh rootless@localhost` in tests/rootless.sh
+    # Allow root and rootless itself to execute `ssh rootless@localhost` in tests/rootless.sh
     ssh-keygen -t ecdsa -N "" -f /root/rootless.key
     mkdir -m 0700 -p /home/rootless/.ssh
+    cp /root/rootless.key /home/rootless/.ssh/id_ecdsa
     cat /root/rootless.key.pub >> /home/rootless/.ssh/authorized_keys
     chown -R rootless.rootless /home/rootless
 

libcontainer/rootfs_linux.go

@@ -1065,7 +1065,22 @@ func remount(m *configs.Mount, rootfs string, mountFd *int) error {
 	}
 
 	return utils.WithProcfd(rootfs, m.Destination, func(procfd string) error {
-		return mount(source, m.Destination, procfd, m.Device, uintptr(m.Flags|unix.MS_REMOUNT), "")
+		flags := uintptr(m.Flags | unix.MS_REMOUNT)
+		err := mount(source, m.Destination, procfd, m.Device, flags, "")
+		if err == nil {
+			return nil
+		}
+		// Check if the source has ro flag...
+		var s unix.Statfs_t
+		if err := unix.Statfs(source, &s); err != nil {
+			return &os.PathError{Op: "statfs", Path: source, Err: err}
+		}
+		if s.Flags&unix.MS_RDONLY != unix.MS_RDONLY {
+			return err
+		}
+		// ... and retry the mount with ro flag set.
+		flags |= unix.MS_RDONLY
+		return mount(source, m.Destination, procfd, m.Device, flags, "")
 	})
 }
 

tests/integration/mounts_sshfs.bats

@@ -0,0 +1,40 @@
+#!/usr/bin/env bats
+
+load helpers
+
+function setup() {
+	# Create a ro fuse-sshfs mount; skip the test if it's not working.
+	local sshfs="sshfs
+		-o UserKnownHostsFile=/dev/null
+		-o StrictHostKeyChecking=no
+		-o PasswordAuthentication=no"
+
+	DIR="$BATS_RUN_TMPDIR/fuse-sshfs"
+	mkdir -p "$DIR"
+
+	if ! $sshfs -o ro rootless@localhost: "$DIR"; then
+		skip "test requires working sshfs mounts"
+	fi
+
+	setup_hello
+}
+
+function teardown() {
+	# New distros (Fedora 35) do not have fusermount installed
+	# as a dependency of fuse-sshfs, and good ol' umount works.
+	fusermount -u "$DIR" || umount "$DIR"
+
+	teardown_bundle
+}
+
+@test "runc run [rw bind mount of a ro fuse sshfs mount]" {
+	update_config '	  .mounts += [{
+					type: "bind",
+					source: "'"$DIR"'",
+					destination: "/mnt",
+					options: ["rw", "rprivate", "nosuid", "nodev", "rbind"]
+				}]'
+
+	runc run test_busybox
+	[ "$status" -eq 0 ]
+}