On Thu, Nov 6, 2014 at 8:07 AM, David Drysdale [off-list ref] wrote:

Here's another pass at this. Some things to discuss in particular:

1) The current approach for interpreted execs (i.e. mostly "#!" scripts)
   gives them an argv[1] filename like "/dev/fd/<fd>/<path>".  This
   means that script execution in a /proc-less system isn't going to
   work, at least until interpreters get smart enough to spot and
   special-case the leading "/dev/fd/<fd>", or until there's something
   to use in place of /dev/fd -> /proc/self/fd (e.g. Al's dupfs
   suggestion, https://lkml.org/lkml/2014/10/19/141).

   So is an execveat(2) that (currently) only works for non-interpreted
   programs still useful?

I think it is.  I would make sure to return a distinguishable error
code in the event that the failure happens because of one of the
unsupported cases.

2) I don't like having to add a new LOOKUP_EMPTY_NOPATH flag
   just to prevent O_PATH fds from being fexecve()ed -- alternative
   suggestions welcomed.  (More generally, I don't have a great
   feel for what O_PATH is for; how bad would it be to just allow
   them to be fexecve()ed?)

If you fexecve an O_PATH fd, does it at least check that you have
execute permission on the inode?  If so, it seems okay to allow it.

--Andy

.........

This patch set adds execveat(2) for x86, and is derived from Meredydd
Luff's patch from Sept 2012 (https://lkml.org/lkml/2012/9/11/528).

The primary aim of adding an execveat syscall is to allow an
implementation of fexecve(3) that does not rely on the /proc
filesystem, at least for executables (rather than scripts).  The
current glibc version of fexecve(3) is implemented via /proc, which
causes problems in sandboxed or otherwise restricted environments.

Given the desire for a /proc-free fexecve() implementation, HPA
suggested (https://lkml.org/lkml/2006/7/11/556) that an execveat(2)
syscall would be an appropriate generalization.

Also, having a new syscall means that it can take a flags argument
without back-compatibility concerns.  The current implementation just
defines the AT_EMPTY_PATH and AT_SYMLINK_NOFOLLOW flags, but other
flags could be added in future -- for example, flags for new namespaces
(as suggested at https://lkml.org/lkml/2006/7/11/474).

Related history:
 - https://lkml.org/lkml/2006/12/27/123 is an example of someone
   realizing that fexecve() is likely to fail in a chroot environment.
 - http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=514043 covered
   documenting the /proc requirement of fexecve(3) in its manpage, to
   "prevent other people from wasting their time".
 - https://bugzilla.kernel.org/show_bug.cgi?id=74481 documented that
   it's not possible to fexecve() a file descriptor for a script with
   close-on-exec set (which is possible with the implementation here).

Confused.  How does it work for a close-on-exec script?  I understand
how it works for a close-on-exec ELF binary.

--Andy

 - https://bugzilla.redhat.com/show_bug.cgi?id=241609 described a
   problem where a process that did setuid() could not fexecve()
   because it no longer had access to /proc/self/fd; this has since
   been fixed.


Changes since v5:
 - Set new flag in bprm->interp_flags for O_CLOEXEC fds, so that binfmts
   that invoke an interpreter fail the exec (as they will not be able
   to access the invoked file). [Andy Lutomirski]
 - Don't truncate long paths. [Andy Lutomirski]
 - Commonize code to open the executed file. [Eric W. Biederman]
 - Mark O_PATH file descriptors so they cannot be fexecve()ed.
 - Make self-test more helpful, and add additional cases:
     - file offset non-zero
     - binary file without execute bit
     - O_CLOEXEC fds

Changes since v4, suggested by Eric W. Biederman:
 - Use empty filename with AT_EMPTY_PATH flag rather than NULL
   pathname to request fexecve-like behaviour.
 - Build pathname as "/dev/fd/<fd>/<filename>" (or "/dev/fd/<fd>")
   rather than using d_path().
 - Patch against v3.17 (bfe01a5ba249)

Changes since Meredydd's v3 patch:
 - Added a selftest.
 - Added a man page.
 - Left open_exec() signature untouched to reduce patch impact
   elsewhere (as suggested by Al Viro).
 - Filled in bprm->filename with d_path() into a buffer, to avoid use
   of potentially-ephemeral dentry->d_name.
 - Patch against v3.14 (455c6fdbd21916).


David Drysdale (2):
  syscalls,x86: implement execveat() system call
  syscalls,x86: add selftest for execveat(2)

 arch/x86/ia32/audit.c                   |   1 +
 arch/x86/ia32/ia32entry.S               |   1 +
 arch/x86/kernel/audit_64.c              |   1 +
 arch/x86/kernel/entry_64.S              |  28 +++
 arch/x86/syscalls/syscall_32.tbl        |   1 +
 arch/x86/syscalls/syscall_64.tbl        |   2 +
 arch/x86/um/sys_call_table_64.c         |   1 +
 fs/binfmt_em86.c                        |   4 +
 fs/binfmt_misc.c                        |   4 +
 fs/binfmt_script.c                      |  10 +
 fs/exec.c                               | 115 ++++++++++--
 fs/namei.c                              |   8 +-
 include/linux/binfmts.h                 |   4 +
 include/linux/compat.h                  |   3 +
 include/linux/fs.h                      |   1 +
 include/linux/namei.h                   |   1 +
 include/linux/sched.h                   |   4 +
 include/linux/syscalls.h                |   4 +
 include/uapi/asm-generic/unistd.h       |   4 +-
 kernel/sys_ni.c                         |   3 +
 lib/audit.c                             |   3 +
 tools/testing/selftests/Makefile        |   1 +
 tools/testing/selftests/exec/.gitignore |   7 +
 tools/testing/selftests/exec/Makefile   |  25 +++
 tools/testing/selftests/exec/execveat.c | 321 ++++++++++++++++++++++++++++++++
 25 files changed, 542 insertions(+), 15 deletions(-)
 create mode 100644 tools/testing/selftests/exec/.gitignore
 create mode 100644 tools/testing/selftests/exec/Makefile
 create mode 100644 tools/testing/selftests/exec/execveat.c

--
2.1.0.rc2.206.gedb03e5

-- 
Andy Lutomirski
AMA Capital Management, LLC