utmp symbol conflict

utmp symbol conflict

[Thomas Petazzoni]

Hello,

The shellinabox project
(https://github.com/shellinabox/shellinabox) fails to build with musl
due to the following define in musl's utmp.h:

#define utmp utmpx

Indeed shellinabox/launcher.c has the following piece of code
(https://github.com/shellinabox/shellinabox/blob/master/shellinabox/launcher.c#L1572) :

    struct utmpx utmpx            = utmp->utmpx;
    if (service->useLogin || service->authUser) {
      utmpx.ut_type               = LOGIN_PROCESS;
      memset(utmpx.ut_host, 0, sizeof(utmpx.ut_host));
    }

Where "utmp" is of type "struct Utmp *", which is an internal structure
type defined as:

struct Utmp {
  const char   pid[32];
  int          pty;
  int          useLogin;
#ifdef HAVE_UTMPX_H
  struct utmpx utmpx;
#endif
};

Unfortunately, with musl's definition #define utmp utmpx, the line:

	struct utmpx utmpx            = utmp->utmpx;

gets turned after pre-processing to:

	struct utmpx utmpx            = utmpx->utmpx;

which obviously causes a build failure:

shellinabox/launcher.c: In function ‘childProcess’:
shellinabox/launcher.c:1576:41: error: invalid type argument of ‘->’ (have ‘struct utmpx’)
     struct utmpx utmpx            = utmp->utmpx;
                                         ^
I don't think the symbol "utmp" as a variable name is reserved, so is
the #define utmp utmpx done by musl really a legal thing to do?

Best regards,

Thomas
-- 
Thomas Petazzoni, CTO, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com

Re: utmp symbol conflict

[Rich Felker]

On Thu, Aug 04, 2016 at 10:28:52PM +0200, Thomas Petazzoni wrote:
> Hello,
> 
> The shellinabox project
> (https://github.com/shellinabox/shellinabox) fails to build with musl
> due to the following define in musl's utmp.h:
> 
> #define utmp utmpx
> 
> Indeed shellinabox/launcher.c has the following piece of code
> (https://github.com/shellinabox/shellinabox/blob/master/shellinabox/launcher.c#L1572) :
> 
>     struct utmpx utmpx            = utmp->utmpx;
>     if (service->useLogin || service->authUser) {
>       utmpx.ut_type               = LOGIN_PROCESS;
>       memset(utmpx.ut_host, 0, sizeof(utmpx.ut_host));
>     }
> 
> Where "utmp" is of type "struct Utmp *", which is an internal structure
> type defined as:
> 
> struct Utmp {
>   const char   pid[32];
>   int          pty;
>   int          useLogin;
> #ifdef HAVE_UTMPX_H
>   struct utmpx utmpx;
> #endif
> };
> 
> Unfortunately, with musl's definition #define utmp utmpx, the line:
> 
> 	struct utmpx utmpx            = utmp->utmpx;
> 
> gets turned after pre-processing to:
> 
> 	struct utmpx utmpx            = utmpx->utmpx;
> 
> which obviously causes a build failure:
> 
> shellinabox/launcher.c: In function ‘childProcess’:
> shellinabox/launcher.c:1576:41: error: invalid type argument of ‘->’ (have ‘struct utmpx’)
>      struct utmpx utmpx            = utmp->utmpx;
>                                          ^
> I don't think the symbol "utmp" as a variable name is reserved, so is
> the #define utmp utmpx done by musl really a legal thing to do?

The header <utmp.h> is not a standard header so there are no namespace
constraints on it. An application using utmpx (the standard api)
should be including just <utmpx.h> not <utmp.h>.

Aside from that, utmp is not supported at all on musl (it's all stubs)
so it would probably make the most sense to compile with an option to
disable utmp support if possible. This might help avoid runtime
warnings about not being able to set/find utmp entries.

Rich