Implement kernel_execve

This set of changes implements kernel_execve to remove the need for
kernel threads to pass in pointers to in-kernel data structures
to functions that take __user pointers.   Which is part of the
greater removal of set_fs work.

This set of changes makes do_execve static and so I have updated the
comments.  This affects the comments in the x86 entry point code
and the comments in tomoyo.  I believe I have updated them correctly.
If not please let me know.

I have moved the calls of copy_strings before the call of
security_bprm_creds_for_exec.  Which might be of interest to the
security folks.  I can't see that it matters but I have copied the
security folks just to be certain.

By moving the initialization of the new stack that copy_strings does
earlier it becomes possible to copy all of the parameters to exec before
anything else is done which makes it possible to have one function
kernel_execve that uncondtionally handles copying parameters from kernel
space, and another function do_execveat_common which handles copying
parameters from userspace.

This work was inspired by Christoph Hellwig's similar patchset, which my
earlier work to remove the file parameter to do_execveat_common
conflicted with.
https://lore.kernel.org/linux-fsdevel/20200627072704.2447163-1-hch@lst.de/

I figured that after causing all of that trouble for the set_fs work
the least I could do is implement the change myself.

The big practical change from Christoph's work is that he did not
separate out the copying of parameters from the rest of the work of
exec, which did not help the maintainability of the code.

Eric W. Biederman (7):
      exec: Remove unnecessary spaces from binfmts.h
      exec: Factor out alloc_bprm
      exec: Move initialization of bprm->filename into alloc_bprm
      exec: Move bprm_mm_init into alloc_bprm
      exec: Factor bprm_execve out of do_execve_common
      exec: Factor bprm_stack_limits out of prepare_arg_pages
      exec: Implement kernel_execve

 arch/x86/entry/entry_32.S      |   2 +-
 arch/x86/entry/entry_64.S      |   2 +-
 arch/x86/kernel/unwind_frame.c |   2 +-
 fs/exec.c                      | 301 ++++++++++++++++++++++++++++-------------
 include/linux/binfmts.h        |  20 ++-
 init/main.c                    |   4 +-
 kernel/umh.c                   |   6 +-
 security/tomoyo/common.h       |   2 +-
 security/tomoyo/domain.c       |   4 +-
 security/tomoyo/tomoyo.c       |   4 +-
 10 files changed, 224 insertions(+), 123 deletions(-)

Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Link: https://lkml.kernel.org/r/871rle8bw2.fsf@x220.int.ebiederm.org
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>

arch/x86/entry/entry_32.S

@@ -854,7 +854,7 @@ SYM_CODE_START(ret_from_fork)
 	CALL_NOSPEC ebx
 	/*
 	 * A kernel thread is allowed to return here after successfully
-	 * calling do_execve().  Exit to userspace to complete the execve()
+	 * calling kernel_execve().  Exit to userspace to complete the execve()
 	 * syscall.
 	 */
 	movl	$0, PT_EAX(%esp)

arch/x86/entry/entry_64.S

@@ -293,7 +293,7 @@ SYM_CODE_START(ret_from_fork)
 	CALL_NOSPEC rbx
 	/*
 	 * A kernel thread is allowed to return here after successfully
-	 * calling do_execve().  Exit to userspace to complete the execve()
+	 * calling kernel_execve().  Exit to userspace to complete the execve()
 	 * syscall.
 	 */
 	movq	$0, RAX(%rsp)

arch/x86/kernel/unwind_frame.c

@@ -275,7 +275,7 @@ bool unwind_next_frame(struct unwind_state *state)
 		 * This user_mode() check is slightly broader than a PF_KTHREAD
 		 * check because it also catches the awkward situation where a
 		 * newly forked kthread transitions into a user task by calling
-		 * do_execve(), which eventually clears PF_KTHREAD.
+		 * kernel_execve(), which eventually clears PF_KTHREAD.
 		 */
 		if (!user_mode(regs))
 			goto the_end;