path: root/security/sandbox/linux/SandboxInfo.cpp

/* -*- Mode: C++; tab-width: 8; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/* vim: set ts=8 sts=2 et sw=2 tw=80: */
/* This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this file,
 * You can obtain one at http://mozilla.org/MPL/2.0/. */

#include "SandboxInfo.h"
#include "SandboxLogging.h"
#include "LinuxSched.h"

#include <errno.h>
#include <stdlib.h>
#include <sys/prctl.h>
#include <sys/stat.h>
#include <sys/syscall.h>
#include <sys/wait.h>
#include <unistd.h>

#include "base/posix/eintr_wrapper.h"
#include "mozilla/Assertions.h"
#include "mozilla/ArrayUtils.h"
#include "sandbox/linux/system_headers/linux_seccomp.h"
#include "sandbox/linux/system_headers/linux_syscalls.h"

#ifdef MOZ_VALGRIND
#include <valgrind/valgrind.h>
#endif


// A note about assertions: in general, the worst thing this module
// should be able to do is disable sandboxing features, so release
// asserts or MOZ_CRASH should be avoided, even for seeming
// impossibilities like an unimplemented syscall returning success
// (which has happened: https://crbug.com/439795 ).
//
// MOZ_DIAGNOSTIC_ASSERT (debug builds, plus Nightly/Aurora non-debug)
// is probably the best choice for conditions that shouldn't be able
// to fail without the help of bugs in the kernel or system libraries.
//
// Regardless of assertion type, whatever condition caused it to fail
// should generally also disable the corresponding feature on builds
// that omit the assertion.

namespace mozilla {

// Bug 1229136: this is copied from ../SandboxUtil.cpp to avoid
// complicated build issues; renamespaced to avoid the possibility of
// symbol conflict.
namespace {

static bool
IsSingleThreaded()
{
  // This detects the thread count indirectly.  /proc/<pid>/task has a
  // subdirectory for each thread in <pid>'s thread group, and the
  // link count on the "task" directory follows Unix expectations: the
  // link from its parent, the "." link from itself, and the ".." link
  // from each subdirectory; thus, 2+N links for N threads.
  struct stat sb;
  if (stat("/proc/self/task", &sb) < 0) {
    MOZ_DIAGNOSTIC_ASSERT(false, "Couldn't access /proc/self/task!");
    return false;
  }
  MOZ_DIAGNOSTIC_ASSERT(sb.st_nlink >= 3);
  return sb.st_nlink == 3;
}

} // anonymous namespace

static bool
HasSeccompBPF()
{
  // Allow simulating the absence of seccomp-bpf support, for testing.
  if (getenv("MOZ_FAKE_NO_SANDBOX")) {
    return false;
  }

  // Valgrind and the sandbox don't interact well, probably because Valgrind
  // does various system calls which aren't allowed, even if Firefox itself
  // is playing by the rules.
# if defined(MOZ_VALGRIND)
  if (RUNNING_ON_VALGRIND) {
    return false;
  }
# endif

  // Determine whether seccomp-bpf is supported by trying to
  // enable it with an invalid pointer for the filter.  This will
  // fail with EFAULT if supported and EINVAL if not, without
  // changing the process's state.

  int rv = prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, nullptr);
  MOZ_DIAGNOSTIC_ASSERT(rv == -1, "prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER,"
                        " nullptr) didn't fail");
  MOZ_DIAGNOSTIC_ASSERT(errno == EFAULT || errno == EINVAL);
  return rv == -1 && errno == EFAULT;
}

static bool
HasSeccompTSync()
{
  // Similar to above, but for thread-sync mode.  See also Chromium's
  // sandbox::SandboxBPF::SupportsSeccompThreadFilterSynchronization
  if (getenv("MOZ_FAKE_NO_SECCOMP_TSYNC")) {
    return false;
  }
  int rv = syscall(__NR_seccomp, SECCOMP_SET_MODE_FILTER,
                   SECCOMP_FILTER_FLAG_TSYNC, nullptr);
  MOZ_DIAGNOSTIC_ASSERT(rv == -1, "seccomp(..., SECCOMP_FILTER_FLAG_TSYNC,"
                        " nullptr) didn't fail");
  MOZ_DIAGNOSTIC_ASSERT(errno == EFAULT || errno == EINVAL || errno == ENOSYS);
  return rv == -1 && errno == EFAULT;
}

static bool
HasUserNamespaceSupport()
{
  // Note: the /proc/<pid>/ns/* files track setns(2) support, which in
  // some cases (e.g., pid) significantly postdates kernel support for
  // the namespace type, so in general this type of check could be a
  // false negative.  However, for user namespaces, any kernel new
  // enough for the feature to be usable for us has setns support
  // (v3.8), so this is okay.
  //
  // The non-user namespaces all default to "y" in init/Kconfig, but
  // check them explicitly in case someone has a weird custom config.
  static const char* const paths[] = {
    "/proc/self/ns/user",
    "/proc/self/ns/pid",
    "/proc/self/ns/net",
    "/proc/self/ns/ipc",
  };
  for (size_t i = 0; i < ArrayLength(paths); ++i) {
    if (access(paths[i], F_OK) == -1) {
      MOZ_ASSERT(errno == ENOENT);
      return false;
    }
  }
  return true;
}

static bool
CanCreateUserNamespace()
{
  // Unfortunately, the only way to verify that this process can
  // create a new user namespace is to actually create one; because
  // this process's namespaces shouldn't be side-effected (yet), it's
  // necessary to clone (and collect) a child process.  See also
  // Chromium's sandbox::Credentials::SupportsNewUserNS.
  //
  // This is somewhat more expensive than the other tests, so it's
  // cached in the environment to prevent child processes from having
  // to re-run the test.
  //
  // This is run at static initializer time, while single-threaded, so
  // locking isn't needed to access the environment.
  static const char kCacheEnvName[] = "MOZ_ASSUME_USER_NS";
  const char* cached = getenv(kCacheEnvName);
  if (cached) {
    return cached[0] > '0';
  }

  // Valgrind might allow the clone, but doesn't know what to do with
  // unshare.  Check for that by unsharing nothing.  (Valgrind will
  // probably need sandboxing disabled entirely, but no need to break
  // things worse than strictly necessary.)
  if (syscall(__NR_unshare, 0) != 0) {
#ifdef MOZ_VALGRIND
    MOZ_ASSERT(errno == ENOSYS);
#else
    // If something else can cause that call to fail, we's like to know
    // about it; the right way to handle it might not be the same.
    MOZ_ASSERT(false);
#endif
    return false;
  }

  pid_t pid = syscall(__NR_clone, SIGCHLD | CLONE_NEWUSER,
                      nullptr, nullptr, nullptr, nullptr);
  if (pid == 0) {
    // In the child.  Do as little as possible.
    _exit(0);
  }
  if (pid == -1) {
    // Failure.
    MOZ_ASSERT(errno == EINVAL || // unsupported
               errno == EPERM  || // root-only, or we're already chrooted
               errno == EUSERS);  // already at user namespace nesting limit
    setenv(kCacheEnvName, "0", 1);
    return false;
  }
  // Otherwise, in the parent and successful.
  bool waitpid_ok = HANDLE_EINTR(waitpid(pid, nullptr, 0)) == pid;
  MOZ_ASSERT(waitpid_ok);
  if (!waitpid_ok) {
    return false;
  }
  setenv(kCacheEnvName, "1", 1);
  return true;
}

/* static */
SandboxInfo SandboxInfo::sSingleton = SandboxInfo();

SandboxInfo::SandboxInfo() {
  int flags = 0;
  static_assert(sizeof(flags) >= sizeof(Flags), "enum Flags fits in an int");

  if (HasSeccompBPF()) {
    flags |= kHasSeccompBPF;
    if (HasSeccompTSync()) {
      flags |= kHasSeccompTSync;
    }
  }

  // Detect the threading-problem signal from the parent process.
  if (getenv("MOZ_SANDBOX_UNEXPECTED_THREADS")) {
    flags |= kUnexpectedThreads;
  } else {
    if (HasUserNamespaceSupport()) {
      flags |= kHasPrivilegedUserNamespaces;
      if (CanCreateUserNamespace()) {
        flags |= kHasUserNamespaces;
      }
    }
  }

  if (getenv("MOZ_SANDBOX_VERBOSE")) {
    flags |= kVerbose;
  }

  mFlags = static_cast<Flags>(flags);
}

/* static */ void
SandboxInfo::ThreadingCheck()
{
  // Allow MOZ_SANDBOX_UNEXPECTED_THREADS to be set manually for testing.
  if (IsSingleThreaded() &&
      !getenv("MOZ_SANDBOX_UNEXPECTED_THREADS")) {
    return;
  }
  SANDBOX_LOG_ERROR("unexpected multithreading found; this prevents using"
                    " namespace sandboxing.%s",
                    // getenv isn't thread-safe, but see below.
                    getenv("LD_PRELOAD") ? "  (If you're LD_PRELOAD'ing"
                    " nVidia GL: that's not necessary for Gecko.)" : "");

  // Propagate this information for use by child processes.  (setenv
  // isn't thread-safe, but other threads are from non-Gecko code so
  // they wouldn't be using NSPR; we have to hope for the best.)
  setenv("MOZ_SANDBOX_UNEXPECTED_THREADS", "1", 0);
  int flags = sSingleton.mFlags;
  flags |= kUnexpectedThreads;
  flags &= ~(kHasUserNamespaces | kHasPrivilegedUserNamespaces);
  sSingleton.mFlags = static_cast<Flags>(flags);
}

} // namespace mozilla