[PATCH v2] checkpatch.pl: New instances of ENOSYS are errors

[PATCH v2] checkpatch.pl: New instances of ENOSYS are errors

[Andy Lutomirski]

ENOSYS means that a nonexistent system call was called.  We have a
bad habit of using it for things like invalid operations on
otherwise valid syscalls.  We should avoid this in new code.

Signed-off-by: Andy Lutomirski <luto@amacapital.net>
---

Pervasive incorrect usage of ENOSYS came up at the kernel summit ABI
review discussion.  Let's see if checkpatch can help.

I'll submit a separate patch for include/uapi/asm-generic/errno.h.

Changes from v1:
 - Moved later so that it won't warn on context lines.
 - Use $herecur.
 - Improve regex pattern.

 scripts/checkpatch.pl | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
index 182be0f..0a68a29 100755
--- a/scripts/checkpatch.pl
+++ b/scripts/checkpatch.pl
@@ -2931,6 +2931,14 @@ sub process {
 			     "Prefer dev_$level(... to dev_printk(KERN_$orig, ...\n" . $herecurr);
 		}
 
+# ENOSYS means "bad syscall nr" and nothing else
+# (note that this doesn't run on assembly files, so entry*.S is okay)
+		if ($line =~ /\bENOSYS\b/) {
+			ERROR("ENOSYS",
+			      "ENOSYS means 'invalid syscall nr' and nothing else\n" .
+			      "       (ignore if this really is syscall entry code)\n" . $herecurr);
+		}
+
 # function brace can't be on same line, except for #defines of do while,
 # or if closed on same line
 		if (($line=~/$Type\s*$Ident\(.*\).*\s{/) and
-- 
1.9.3

Re: [PATCH v2] checkpatch.pl: New instances of ENOSYS are errors

[Joe Perches]

On Fri, 2014-08-22 at 09:26 -0700, Andy Lutomirski wrote:
> ENOSYS means that a nonexistent system call was called.  We have a
> bad habit of using it for things like invalid operations on
> otherwise valid syscalls.  We should avoid this in new code.
[]
> diff --git a/scripts/checkpatch.pl b/scripts/checkpatch.pl
[]
> @@ -2931,6 +2931,14 @@ sub process {
>  			     "Prefer dev_$level(... to dev_printk(KERN_$orig, ...\n" . $herecurr);
>  		}
>  
> +# ENOSYS means "bad syscall nr" and nothing else
> +# (note that this doesn't run on assembly files, so entry*.S is okay)
> +		if ($line =~ /\bENOSYS\b/) {
> +			ERROR("ENOSYS",
> +			      "ENOSYS means 'invalid syscall nr' and nothing else\n" .
> +			      "       (ignore if this really is syscall entry code)\n" . $herecurr);
> +		}
> +
>  # function brace can't be on same line, except for #defines of do while,
>  # or if closed on same line
>  		if (($line=~/$Type\s*$Ident\(.*\).*\s{/) and

Hi again Andy.

I think this is OK, but you seem to be making
this patch against an old version of checkpatch.

It applies with an offset of a few hundred lines
to Linus' tree and to -next.

Some trivial comments:

I think the "(note that" comment line isn't very useful.
All the tests after:
		next if ($realfile !~ /\.(h|c)$/);
have that attribute.

checkpatch doesn't use multi-line message descriptions.
The "(ignore if this" line should not be on a separate line
but simply use a longer single output line.

ERROR types should not have false positives.
I'd prefer WARN.

cheers, Joe

Re: [PATCH v2] checkpatch.pl: New instances of ENOSYS are errors

[Pavel Machek]

On Fri 2014-08-22 09:26:31, Andy Lutomirski wrote:
> ENOSYS means that a nonexistent system call was called.  We have a
> bad habit of using it for things like invalid operations on
> otherwise valid syscalls.  We should avoid this in new code.

Is it good idea? I mean, doing EINVAL for subcalls is pretty
unhelpful.
								Pavel

-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html

Re: [PATCH v2] checkpatch.pl: New instances of ENOSYS are errors

[Andy Lutomirski]

On Aug 25, 2014 3:29 AM, "Pavel Machek" <pavel@ucw.cz> wrote:
>
> On Fri 2014-08-22 09:26:31, Andy Lutomirski wrote:
> > ENOSYS means that a nonexistent system call was called.  We have a
> > bad habit of using it for things like invalid operations on
> > otherwise valid syscalls.  We should avoid this in new code.
>
> Is it good idea? I mean, doing EINVAL for subcalls is pretty
> unhelpful.

EOPNOTSUPP?

The problem is that user code wants to do:

bool foo_wrapper(int op)
{
  if (foo_not_supported)
    return false

  if (foo(op) != 0) {
    if (errno == ENOSYS)
      foo_not_supported = true;
    return false;
  }

  return true;
}

foo_wrapper(FOO_OP_A);
foo_wrapper(FOO_OP_B);

If FOO_OP_A returns -ENOSYS but FOO_OP_B does not, then this doesn't work.

--Andy