[PATCH v3 0/4] PR_SET_NO_NEW_PRIVS, unshare, and chroot

[Andy Lutomirski <luto@amacapital.net>]

This adds PR_{GET,SET}_NO_NEW_PRIVS.  As an example of its use, it
allows some unshare operations and (sometimes) chroot when no_new_privs
is set.  Another example is the experimental pam module here:

http://web.mit.edu/luto/www/linux/

After some impressively long mailing list threads, I still think that
blocking setresuid, setuid, and capset in no_new_privs mode is
unnecessary and overcomplicated.  Additionally, blocking those calls
will make my pam module either fail or become a giant security hole
(depending on how carefully the core pam stuff is written -- I haven't
checked).

Changes from v2:
 - Rebased onto a very recent -linus tree.
 - Changed prctl numbering.  (Needed because prctl 35 is now taken.)
 - Fixed a typo or two.
 - Removed explicit propagation of no_new_privs.  dup_task_struct is enough.
 - Reworked the chroot patch.  It now uses hopefully much more sane logic
   to decide whether the user is chrooted.  It also checks that fs is not
   shared (which was a big security hole in the earlier version).

For the git-inclined, this series is here:
https://git.kernel.org/?p=linux/kernel/git/luto/linux.git;a=shortlog;h=refs/heads/security/no_new_privs/patch_v3

Test it like this:

---- begin test case

#include <sys/prctl.h>
#include <stdio.h>
#include <unistd.h>
#include <errno.h>

#define PR_SET_NO_NEW_PRIVS 36
#define PR_GET_NO_NEW_PRIVS 37

int main()
{
  int nnp = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
  if (nnp == -EINVAL) {
    printf("Failed!\n");
    return 1;
  }

  printf("nnp was %d\n", nnp);

  if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0) != 0) {
    printf("Failed!\n");
    return 1;
  }

  nnp = prctl(PR_GET_NO_NEW_PRIVS, 0, 0, 0, 0);
  if (nnp == -EINVAL) {
    printf("Failed!\n");
    return 1;
  }

  printf("nnp is %d\n", nnp);

  printf("here goes...\n");
  execlp("bash", "bash", NULL);
  printf("Failed to exec bash\n");
  return 1;
}

---- end test case

Andy Lutomirski (3):
  Add PR_{GET,SET}_NO_NEW_PRIVS to prevent execve from granting privs
  Allow unprivileged CLONE_NEWUTS and CLONE_NEWIPC with no_new_privs
  Allow unprivileged chroot when safe

John Johansen (1):
  Fix apparmor for PR_{GET,SET}_NO_NEW_PRIVS

 fs/exec.c                  |   10 ++++++++-
 fs/open.c                  |   46 ++++++++++++++++++++++++++++++++++++++++++-
 include/linux/prctl.h      |   15 ++++++++++++++
 include/linux/sched.h      |    2 +
 include/linux/security.h   |    1 +
 kernel/nsproxy.c           |    8 ++++++-
 kernel/sys.c               |   10 +++++++++
 security/apparmor/domain.c |   35 +++++++++++++++++++++++++++++++++
 security/commoncap.c       |    7 ++++-
 security/selinux/hooks.c   |   10 ++++++++-
 10 files changed, 137 insertions(+), 7 deletions(-)

-- 
1.7.7.6