1The execve system call can grant a newly-started program privileges that
2its parent did not have.  The most obvious examples are setuid/setgid
3programs and file capabilities.  To prevent the parent program from
4gaining these privileges as well, the kernel and user code must be
5careful to prevent the parent from doing anything that could subvert the
6child.  For example:
7
8 - The dynamic loader handles LD_* environment variables differently if
9   a program is setuid.
10
11 - chroot is disallowed to unprivileged processes, since it would allow
12   /etc/passwd to be replaced from the point of view of a process that
13   inherited chroot.
14
15 - The exec code has special handling for ptrace.
16
17These are all ad-hoc fixes.  The no_new_privs bit (since Linux 3.5) is a
18new, generic mechanism to make it safe for a process to modify its
19execution environment in a manner that persists across execve.  Any task
20can set no_new_privs.  Once the bit is set, it is inherited across fork,
21clone, and execve and cannot be unset.  With no_new_privs set, execve
22promises not to grant the privilege to do anything that could not have
23been done without the execve call.  For example, the setuid and setgid
24bits will no longer change the uid or gid; file capabilities will not
25add to the permitted set, and LSMs will not relax constraints after
26execve.
27
28To set no_new_privs, use prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0).
29
30Be careful, though: LSMs might also not tighten constraints on exec
31in no_new_privs mode.  (This means that setting up a general-purpose
32service launcher to set no_new_privs before execing daemons may
33interfere with LSM-based sandboxing.)
34
35Note that no_new_privs does not prevent privilege changes that do not
36involve execve.  An appropriately privileged task can still call
37setuid(2) and receive SCM_RIGHTS datagrams.
38
39There are two main use cases for no_new_privs so far:
40
41 - Filters installed for the seccomp mode 2 sandbox persist across
42   execve and can change the behavior of newly-executed programs.
43   Unprivileged users are therefore only allowed to install such filters
44   if no_new_privs is set.
45
46 - By itself, no_new_privs can be used to reduce the attack surface
47   available to an unprivileged user.  If everything running with a
48   given uid has no_new_privs set, then that uid will be unable to
49   escalate its privileges by directly attacking setuid, setgid, and
50   fcap-using binaries; it will need to compromise something without the
51   no_new_privs bit set first.
52
53In the future, other potentially dangerous kernel features could become
54available to unprivileged tasks if no_new_privs is set.  In principle,
55several options to unshare(2) and clone(2) would be safe when
56no_new_privs is set, and no_new_privs + chroot is considerable less
57dangerous than chroot by itself.
58