implement Solaris support for system-process-attributes and list-system-processes

implement Solaris support for system-process-attributes and list-system-processes

[Dan Nicolaescu]

These functions are needed so that M-x proced can work.  They need
system specific implementation.  At the moment they are only supported
on GNU/Linux and Windows.

There are a few warts in getting these 2 functions supported on Solaris,
and I would like some feedback.

1. First some include hackery:

#ifdef SOLARIS2
#if !defined (_LP64) && defined (_FILE_OFFSET_BITS) &&  (_FILE_OFFSET_BITS  ==  64)
#define PROCFS_FILE_OFFSET_BITS_HACK 1
#undef _FILE_OFFSET_BITS
#else
#define PROCFS_FILE_OFFSET_BITS_HACK 0
#endif
#include <procfs.h>
#if PROCFS_FILE_OFFSET_BITS_HACK ==  1
#define _FILE_OFFSET_BITS 64
#endif
#endif /* SOLARIS2 */

procfs.h is the header file that contains the proc data structures, but
it has an #error if compiled in 32 bit mode and _FILE_OFFSET_BITS is
64.
emacs/src/config.in will define _FILE_OFFSET_BITS to 64 when compiled on
a 32 bit solaris system.  Hence the above hackery.


2. process.c has a function `procfs_list_system_processes' that works on
Solaris (and probably on all systems that use a /proc).  
The problem is that procfs_list_system_processes is inside a #ifdef HAVE_PROCFS
that contains a few other functions (time_from_jiffies, get_up_time)
that are Linux specific.
Should I move those functions inside a #ifdef LINUX ?

3. An implementation for the Solaris version of
system-process-attributes is at the end of this message.

4. Flist_system_processes and Fsystem_process_attributes use the
LISTPROC and PROCATTR macros for function calls that need to be defined in src/s/*.h 
These macros are different from how other platform specific functions
are implemented in emacs, and they seem to be a bit redundant: the
functions to implement these are system specific anyway....

So what's the best way to integrate Solaris support?
Again the existing procfs_list_system_processes function can probably be
used by all systems using /proc, but procfs_system_process_attributes
is very likely to be OS specific. 

Thanks


static Lisp_Object
procfs_system_process_attributes (pid)
     Lisp_Object pid;
{
  char procfn[PATH_MAX], fn[PATH_MAX];
  struct stat st;
  struct passwd *pw;
  struct group *gr;
  char *procfn_end;
  struct psinfo pinfo;
  int fd;
  ssize_t nread;
  int proc_id, uid, gid;
  Lisp_Object attrs = Qnil;
  Lisp_Object decoded_cmd, tem;
  struct gcpro gcpro1, gcpro2;
  EMACS_INT uid_eint, gid_eint;

  CHECK_NUMBER_OR_FLOAT (pid);
  proc_id = FLOATP (pid) ? XFLOAT_DATA (pid) : XINT (pid);
  sprintf (procfn, "/proc/%lu", proc_id);
  if (stat (procfn, &st) < 0)
    return attrs;

  GCPRO2 (attrs, decoded_cmd);

  /* euid egid */
  uid = st.st_uid;
  /* Use of EMACS_INT stops GCC whining about limited range of data type.  */
  uid_eint = uid;
  attrs = Fcons (Fcons (Qeuid, make_fixnum_or_float (uid_eint)), attrs);
  BLOCK_INPUT;
  pw = getpwuid (uid);
  UNBLOCK_INPUT;
  if (pw)
    attrs = Fcons (Fcons (Quser, build_string (pw->pw_name)), attrs);

  gid = st.st_gid;
  gid_eint = gid;
  attrs = Fcons (Fcons (Qegid, make_fixnum_or_float (gid_eint)), attrs);
  BLOCK_INPUT;
  gr = getgrgid (gid);
  UNBLOCK_INPUT;
  if (gr)
    attrs = Fcons (Fcons (Qgroup, build_string (gr->gr_name)), attrs);

  strcpy (fn, procfn);
  procfn_end = fn + strlen (fn);
  strcpy (procfn_end, "/psinfo");
  fd = emacs_open (fn, O_RDONLY, 0);
  if (fd >= 0
      && (nread = read (fd, (char*)&pinfo, sizeof(struct psinfo)) > 0))
    {
          attrs = Fcons (Fcons (Qppid, make_fixnum_or_float (pinfo.pr_ppid)), attrs);
	  attrs = Fcons (Fcons (Qpgrp, make_fixnum_or_float (pinfo.pr_pgid)), attrs);
	  attrs = Fcons (Fcons (Qsess, make_fixnum_or_float (pinfo.pr_sid)), attrs);

	  {
	    char state_str[2];
	    state_str[0] =  pinfo.pr_lwp.pr_sname;
	    state_str[1] =  '\0';
	    tem =   build_string (state_str);
	    attrs =  Fcons (Fcons (Qstate,  tem),  attrs);
	  }
	    
	  /* FIXME: missing Qttyname. psinfo.pr_ttydev is a dev_t,
	     need to get a string from it. */
	  
	  /* FIXME: missing: Qtpgid */

	  /* FIXME: missing:
		Qminflt
		Qmajflt
		Qcminflt
		Qcmajflt

		Qstime
		Qcstime  
		Are they available? */

	  
	  attrs = Fcons (Fcons (Qutime,
	  			list3 (make_number (pinfo.pr_time.tv_sec >> 16),
	  			       make_number (pinfo.pr_time.tv_sec & 0xffff),
	  			       make_number (pinfo.pr_time.tv_nsec))),
	  		 attrs);

	  attrs = Fcons (Fcons (Qcutime,
	  			list3 (make_number (pinfo.pr_ctime.tv_sec >> 16),
	  			       make_number (pinfo.pr_ctime.tv_sec & 0xffff),
	  			       make_number (pinfo.pr_ctime.tv_nsec))),
	  		 attrs);

	  attrs = Fcons (Fcons (Qpri, make_number (pinfo.pr_lwp.pr_pri)), attrs);
	  attrs = Fcons (Fcons (Qnice, make_number (pinfo.pr_lwp.pr_nice)), attrs);
	  attrs = Fcons (Fcons (Qthcount, make_fixnum_or_float (pinfo.pr_nlwp)), attrs);

	  attrs = Fcons (Fcons (Qstart,
	  			list3 (make_number (pinfo.pr_start.tv_sec >> 16),
	  			       make_number (pinfo.pr_start.tv_sec & 0xffff),
	  			       make_number (pinfo.pr_start.tv_nsec))),
	  		 attrs);
	  attrs = Fcons (Fcons (Qvsize, make_fixnum_or_float (pinfo.pr_size)), attrs);
	  attrs = Fcons (Fcons (Qrss, make_fixnum_or_float (pinfo.pr_rssize)), attrs);

	  /* pr_pctcpu and pr_pctmem are encoded as a fixed point 16 bit number in  [0 ... 1]  */
	  attrs = Fcons (Fcons (Qpcpu, (pinfo.pr_pctcpu * 100.0) / (double)0x8000), attrs);
	  attrs = Fcons (Fcons (Qpmem, (pinfo.pr_pctmem * 100.0) / (double)0x8000), attrs);

	  decoded_cmd
	    =  code_convert_string_norecord (make_unibyte_string (pinfo.pr_fname,
								  strlen (pinfo.pr_fname)),
					     Vlocale_coding_system,  0);
	  attrs =  Fcons (Fcons (Qcomm,  decoded_cmd),  attrs);
	  decoded_cmd
	    =  code_convert_string_norecord (make_unibyte_string (pinfo.pr_psargs,
								  strlen (pinfo.pr_psargs)),
					     Vlocale_coding_system,  0);
	  attrs =  Fcons (Fcons (Qargs,  decoded_cmd),  attrs);
    }

  UNGCPRO;
  return attrs;
}

Re: implement Solaris support for system-process-attributes and list-system-processes

[Eli Zaretskii]

> Date: Thu, 18 Dec 2008 01:16:19 -0800 (PST)
> From: Dan Nicolaescu <dann@ics.uci.edu>
> 
> #ifdef SOLARIS2
> #if !defined (_LP64) && defined (_FILE_OFFSET_BITS) &&  (_FILE_OFFSET_BITS  ==  64)
> #define PROCFS_FILE_OFFSET_BITS_HACK 1
> #undef _FILE_OFFSET_BITS
> #else
> #define PROCFS_FILE_OFFSET_BITS_HACK 0
> #endif
> #include <procfs.h>
> #if PROCFS_FILE_OFFSET_BITS_HACK ==  1
> #define _FILE_OFFSET_BITS 64
> #endif
> #endif /* SOLARIS2 */
> 
> procfs.h is the header file that contains the proc data structures, but
> it has an #error if compiled in 32 bit mode and _FILE_OFFSET_BITS is
> 64.
> emacs/src/config.in will define _FILE_OFFSET_BITS to 64 when compiled on
> a 32 bit solaris system.  Hence the above hackery.

Can this hackery be moved to a Solaris-specific header file in src/s/ ?

> 2. process.c has a function `procfs_list_system_processes' that works on
> Solaris (and probably on all systems that use a /proc).  
> The problem is that procfs_list_system_processes is inside a #ifdef HAVE_PROCFS
> that contains a few other functions (time_from_jiffies, get_up_time)
> that are Linux specific.
> Should I move those functions inside a #ifdef LINUX ?

I guess GNU_LINUX would be better, and perhaps move them to sysdep.c
while at that.

> 4. Flist_system_processes and Fsystem_process_attributes use the
> LISTPROC and PROCATTR macros for function calls that need to be defined in src/s/*.h 
> These macros are different from how other platform specific functions
> are implemented in emacs, and they seem to be a bit redundant: the
> functions to implement these are system specific anyway....

I hoped that systems with procfs will have only minor differences,
that's why I opted for these macros.  In addition, having the macros
provides an easy way of switching between 2 different implementations
on the same platform.  But these are not very important reasons, so I
don't object to some different solution.

> So what's the best way to integrate Solaris support?
> Again the existing procfs_list_system_processes function can probably be
> used by all systems using /proc, but procfs_system_process_attributes
> is very likely to be OS specific. 

If /proc systems are different to the degree that splitting them into
a common part and OS-specific parts is impractical or unjustified, I
think the implementations of procfs_system_process_attributes should
be moved to sysdep.c, and there we could have a separate variant for
each OS.

Re: implement Solaris support for system-process-attributes and list-system-processes

[Dan Nicolaescu]

Eli Zaretskii <eliz@gnu.org> writes:

  > > Date: Thu, 18 Dec 2008 01:16:19 -0800 (PST)
  > > From: Dan Nicolaescu <dann@ics.uci.edu>
  > > 
  > > #ifdef SOLARIS2
  > > #if !defined (_LP64) && defined (_FILE_OFFSET_BITS) &&  (_FILE_OFFSET_BITS  ==  64)
  > > #define PROCFS_FILE_OFFSET_BITS_HACK 1
  > > #undef _FILE_OFFSET_BITS
  > > #else
  > > #define PROCFS_FILE_OFFSET_BITS_HACK 0
  > > #endif
  > > #include <procfs.h>
  > > #if PROCFS_FILE_OFFSET_BITS_HACK ==  1
  > > #define _FILE_OFFSET_BITS 64
  > > #endif
  > > #endif /* SOLARIS2 */
  > > 
  > > procfs.h is the header file that contains the proc data structures, but
  > > it has an #error if compiled in 32 bit mode and _FILE_OFFSET_BITS is
  > > 64.
  > > emacs/src/config.in will define _FILE_OFFSET_BITS to 64 when compiled on
  > > a 32 bit solaris system.  Hence the above hackery.
  > 
  > Can this hackery be moved to a Solaris-specific header file in src/s/ ?

That file is included everywhere, so it might not be a good idea at this point.

  > > 2. process.c has a function `procfs_list_system_processes' that works on
  > > Solaris (and probably on all systems that use a /proc).  
  > > The problem is that procfs_list_system_processes is inside a #ifdef HAVE_PROCFS
  > > that contains a few other functions (time_from_jiffies, get_up_time)
  > > that are Linux specific.
  > > Should I move those functions inside a #ifdef LINUX ?
  > 
  > I guess GNU_LINUX would be better, and perhaps move them to sysdep.c
  > while at that.

How does this sound: 
- move the procfs_list_system_processes and procfs_system_process_attributes to sysdep.c
- remove the procfs_ prefix 
- add the proper #defines  (HAVE_PROCFS can probably be used by a few OSes)
- the default implementations just return Qnil
- make Fsystem_process_attributes and Flist_system_processes just call 
list_system_processes and system_process_attributes
- rename the w32 versions to system_process_attributes and list_system_processes
- remove the PROCATTR and LISTPROC macros
?

Re: implement Solaris support for system-process-attributes and list-system-processes

[Eli Zaretskii]

> Date: Thu, 18 Dec 2008 23:45:32 -0800 (PST)
> From: Dan Nicolaescu <dann@ics.uci.edu>
> Cc: emacs-devel@gnu.org
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> 
>   > > Date: Thu, 18 Dec 2008 01:16:19 -0800 (PST)
>   > > From: Dan Nicolaescu <dann@ics.uci.edu>
>   > > 
>   > > #ifdef SOLARIS2
>   > > #if !defined (_LP64) && defined (_FILE_OFFSET_BITS) &&  (_FILE_OFFSET_BITS  ==  64)
>   > > #define PROCFS_FILE_OFFSET_BITS_HACK 1
>   > > #undef _FILE_OFFSET_BITS
>   > > #else
>   > > #define PROCFS_FILE_OFFSET_BITS_HACK 0
>   > > #endif
>   > > #include <procfs.h>
>   > > #if PROCFS_FILE_OFFSET_BITS_HACK ==  1
>   > > #define _FILE_OFFSET_BITS 64
>   > > #endif
>   > > #endif /* SOLARIS2 */
>   > > 
>   > > procfs.h is the header file that contains the proc data structures, but
>   > > it has an #error if compiled in 32 bit mode and _FILE_OFFSET_BITS is
>   > > 64.
>   > > emacs/src/config.in will define _FILE_OFFSET_BITS to 64 when compiled on
>   > > a 32 bit solaris system.  Hence the above hackery.
>   > 
>   > Can this hackery be moved to a Solaris-specific header file in src/s/ ?
> 
> That file is included everywhere, so it might not be a good idea at this point.

It's just too ugly, so if there's a good place to hide it out of
sight, I'd be happier.

> How does this sound: 
> - move the procfs_list_system_processes and procfs_system_process_attributes to sysdep.c
> - remove the procfs_ prefix 
> - add the proper #defines  (HAVE_PROCFS can probably be used by a few OSes)
> - the default implementations just return Qnil
> - make Fsystem_process_attributes and Flist_system_processes just call 
> list_system_processes and system_process_attributes
> - rename the w32 versions to system_process_attributes and list_system_processes
> - remove the PROCATTR and LISTPROC macros
> ?

Sounds good to me.

Btw, it strikes me that the default implementation could return the
Emacs process instead of Qnil, but that can wait for another rainy
day.

Re: implement Solaris support for system-process-attributes and list-system-processes

[Miles Bader]

Dan Nicolaescu <dann@ics.uci.edu> writes:
> How does this sound: 
> - move the procfs_list_system_processes and procfs_system_process_attributes to sysdep.c
> - remove the procfs_ prefix 
> - add the proper #defines  (HAVE_PROCFS can probably be used by a few OSes)
> - the default implementations just return Qnil
> - make Fsystem_process_attributes and Flist_system_processes just call 
> list_system_processes and system_process_attributes
> - rename the w32 versions to system_process_attributes and list_system_processes
> - remove the PROCATTR and LISTPROC macros
> ?

... and while you're at it, rename Fsystem_process_attributes to
"Fprocess_attributes" and make it also accept emacs process objects as
an argument, ala signal-process.

-Miles


-- 
"Suppose He doesn't give a shit?  Suppose there is a God but He
just doesn't give a shit?"  [George Carlin]

Re: implement Solaris support for system-process-attributes and list-system-processes

[Dan Nicolaescu]

Eli Zaretskii <eliz@gnu.org> writes:

  > > Date: Thu, 18 Dec 2008 23:45:32 -0800 (PST)
  > > From: Dan Nicolaescu <dann@ics.uci.edu>
  > > Cc: emacs-devel@gnu.org
  > > 
  > > Eli Zaretskii <eliz@gnu.org> writes:
  > > 
  > >   > > Date: Thu, 18 Dec 2008 01:16:19 -0800 (PST)
  > >   > > From: Dan Nicolaescu <dann@ics.uci.edu>
  > >   > > 
  > >   > > #ifdef SOLARIS2
  > >   > > #if !defined (_LP64) && defined (_FILE_OFFSET_BITS) &&  (_FILE_OFFSET_BITS  ==  64)
  > >   > > #define PROCFS_FILE_OFFSET_BITS_HACK 1
  > >   > > #undef _FILE_OFFSET_BITS
  > >   > > #else
  > >   > > #define PROCFS_FILE_OFFSET_BITS_HACK 0
  > >   > > #endif
  > >   > > #include <procfs.h>
  > >   > > #if PROCFS_FILE_OFFSET_BITS_HACK ==  1
  > >   > > #define _FILE_OFFSET_BITS 64
  > >   > > #endif
  > >   > > #endif /* SOLARIS2 */
  > >   > > 
  > >   > > procfs.h is the header file that contains the proc data structures, but
  > >   > > it has an #error if compiled in 32 bit mode and _FILE_OFFSET_BITS is
  > >   > > 64.
  > >   > > emacs/src/config.in will define _FILE_OFFSET_BITS to 64 when compiled on
  > >   > > a 32 bit solaris system.  Hence the above hackery.
  > >   > 
  > >   > Can this hackery be moved to a Solaris-specific header file in src/s/ ?
  > > 
  > > That file is included everywhere, so it might not be a good idea at this point.
  > 
  > It's just too ugly, so if there's a good place to hide it out of
  > sight, I'd be happier.

I'll put it in sysdep.c just before the Solaris specific
system_process_attributes implementation.

  > > How does this sound: 
  > > - move the procfs_list_system_processes and procfs_system_process_attributes to sysdep.c
  > > - remove the procfs_ prefix 
  > > - add the proper #defines  (HAVE_PROCFS can probably be used by a few OSes)
  > > - the default implementations just return Qnil
  > > - make Fsystem_process_attributes and Flist_system_processes just call 
  > > list_system_processes and system_process_attributes
  > > - rename the w32 versions to system_process_attributes and list_system_processes
  > > - remove the PROCATTR and LISTPROC macros
  > > ?
  > 
  > Sounds good to me.

Thanks.  Done.  I can't test on MSWindows, but hopefully didn't break anything.

I'll add the Solaris changes in the next step.

Re: implement Solaris support for system-process-attributes and list-system-processes

[Dan Nicolaescu]

Miles Bader <miles@gnu.org> writes:

  > Dan Nicolaescu <dann@ics.uci.edu> writes:
  > > How does this sound: 
  > > - move the procfs_list_system_processes and procfs_system_process_attributes to sysdep.c
  > > - remove the procfs_ prefix 
  > > - add the proper #defines  (HAVE_PROCFS can probably be used by a few OSes)
  > > - the default implementations just return Qnil
  > > - make Fsystem_process_attributes and Flist_system_processes just call 
  > > list_system_processes and system_process_attributes
  > > - rename the w32 versions to system_process_attributes and list_system_processes
  > > - remove the PROCATTR and LISTPROC macros
  > > ?
  > 
  > ... and while you're at it, rename Fsystem_process_attributes to
  > "Fprocess_attributes" and make it also accept emacs process objects as
  > an argument, ala signal-process.

Sounds good to me, but I'd rather let someone else do those changes.

Re: implement Solaris support for system-process-attributes and list-system-processes

[Miles Bader]

Dan Nicolaescu <dann@ics.uci.edu> writes:
>   > ... and while you're at it, rename Fsystem_process_attributes to
>   > "Fprocess_attributes" and make it also accept emacs process objects as
>   > an argument, ala signal-process.
>
> Sounds good to me, but I'd rather let someone else do those changes.

I actually started to make such a patch, but I think I'll wait until
you're finished with your current changes, as the code obviously
overlaps.

-miles

-- 
Innards, n. pl. The stomach, heart, soul, and other bowels.

Re: implement Solaris support for system-process-attributes and list-system-processes

[Dan Nicolaescu]

Miles Bader <miles@gnu.org> writes:

  > Dan Nicolaescu <dann@ics.uci.edu> writes:
  > >   > ... and while you're at it, rename Fsystem_process_attributes to
  > >   > "Fprocess_attributes" and make it also accept emacs process objects as
  > >   > an argument, ala signal-process.
  > >
  > > Sounds good to me, but I'd rather let someone else do those changes.
  > 
  > I actually started to make such a patch, but I think I'll wait until
  > you're finished with your current changes, as the code obviously
  > overlaps.

Go ahead, I'm done.

Re: implement Solaris support for system-process-attributes and list-system-processes

[Eli Zaretskii]

> Date: Fri, 19 Dec 2008 11:51:09 -0800 (PST)
> From: Dan Nicolaescu <dann@ics.uci.edu>
> Cc: emacs-devel@gnu.org
> 
> Thanks.  Done.  I can't test on MSWindows, but hopefully didn't break anything.

Works for me on MS-Windows, thanks.