make-network-process's gethostbyname usage

make-network-process's gethostbyname usage

[Alexander Kotelnikov]

Hello.

make-network-process calles gethostbyname, but never calls
res_init. The problem, which appears is: if contents of
/etc/resolv.conf was changed old dns servers are unavailable,
gethostbyname still tries them, not new ones. So this bug is pretty
much the same as http://bugs.debian.org/438680 (which has a patch).

-- 
Alexander Kotelnikov
Saint-Petersburg, Russia

Re: make-network-process's gethostbyname usage

[Kim F. Storm]

Alexander Kotelnikov <sacha@myxomop.com> writes:

> Hello.
>
> make-network-process calles gethostbyname, but never calls
> res_init. The problem, which appears is: if contents of
> /etc/resolv.conf was changed old dns servers are unavailable,
> gethostbyname still tries them, not new ones. So this bug is pretty
> much the same as http://bugs.debian.org/438680 (which has a patch).

Is res_init available on all platforms which have gethostbyname ?

Otherwise, we need to arrange for configure to test for its
existence...

-- 
Kim F. Storm <storm@cua.dk> http://www.cua.dk

Re: make-network-process's gethostbyname usage

[Alexander Kotelnikov]

>>>>> On Tue, 21 Aug 2007 10:43:01 +0200
>>>>> "KFS" == Kim F Storm <storm@cua.dk> wrote:
KFS> 
KFS> Alexander Kotelnikov <sacha@myxomop.com> writes:
>> Hello.
>> 
>> make-network-process calles gethostbyname, but never calls
>> res_init. The problem, which appears is: if contents of
>> /etc/resolv.conf was changed old dns servers are unavailable,
>> gethostbyname still tries them, not new ones. So this bug is pretty
>> much the same as http://bugs.debian.org/438680 (which has a patch).
KFS> 
KFS> Is res_init available on all platforms which have gethostbyname ?

I am alsmost sure, no. BTW I think on my linux box getaddrinfo() is
used instead of gethostbyname(), but anyway, res_init is needed. And
its presence really helps: made a hack on emacs22-22.1+1 yesterday at
home and today I had not had to restart my emacs with gnus in the
office.

KFS> Otherwise, we need to arrange for configure to test for its
KFS> existence...

IIRC my fgrep res_init found some occurrences in autoconf/autoheader
emacs files.

-- 
Alexander Kotelnikov
Saint-Petersburg, Russia

Re: make-network-process's gethostbyname usage

[Richard Stallman]

    make-network-process calles gethostbyname, but never calls
    res_init. The problem, which appears is: if contents of
    /etc/resolv.conf was changed old dns servers are unavailable,
    gethostbyname still tries them, not new ones. So this bug is pretty
    much the same as http://bugs.debian.org/438680 (which has a patch).

Would someone please DTRT about this, then ack?
It needs to be fixed in Emacs 22.

Re: make-network-process's gethostbyname usage

[Kim F. Storm]

Richard Stallman <rms@gnu.org> writes:

>     make-network-process calles gethostbyname, but never calls
>     res_init. The problem, which appears is: if contents of
>     /etc/resolv.conf was changed old dns servers are unavailable,
>     gethostbyname still tries them, not new ones. So this bug is pretty
>     much the same as http://bugs.debian.org/438680 (which has a patch).
>
> Would someone please DTRT about this, then ack?
> It needs to be fixed in Emacs 22.

I don't know what to do about this.

configure already has a check for res_init, but on my system (GNU/Linux),
HAVE_RES_INIT is undefined, indicating that it doesn't exist.

However, I believe that it does exist, but only if -lresolv is specified.

So how can we fix configure to DTRT?

a) Include -lresolv in the check for HAVE_RES_INIT.

b) Include -lresolv in LIBS if and only if HAVE_RES_INIT is defined.


If that is in place, it should be trivial to add

#if HAVE_RES_INIT
    res_init();
#endif

before all calls to gethostbyname and getaddrinfo.


-- 
Kim F. Storm <storm@cua.dk> http://www.cua.dk

Re: make-network-process's gethostbyname usage

[Jan Djärv]

Kim F. Storm skrev:
> Richard Stallman <rms@gnu.org> writes:
> 
>>     make-network-process calles gethostbyname, but never calls
>>     res_init. The problem, which appears is: if contents of
>>     /etc/resolv.conf was changed old dns servers are unavailable,
>>     gethostbyname still tries them, not new ones. So this bug is pretty
>>     much the same as http://bugs.debian.org/438680 (which has a patch).
>>
>> Would someone please DTRT about this, then ack?
>> It needs to be fixed in Emacs 22.
> 
> I don't know what to do about this.
> 
> configure already has a check for res_init, but on my system (GNU/Linux),
> HAVE_RES_INIT is undefined, indicating that it doesn't exist.
> 
> However, I believe that it does exist, but only if -lresolv is specified.
> 
> So how can we fix configure to DTRT?
> 
> a) Include -lresolv in the check for HAVE_RES_INIT.
> 
> b) Include -lresolv in LIBS if and only if HAVE_RES_INIT is defined.
> 
> 

Should we not first try if res_init exists without any special libraries, and 
if that fails, check for -lresolv and try to find res_init again?

But there seems to be some problem with that, comment in configure.in says:

dnl This causes -lresolv to get used in subsequent tests,
dnl which causes failures on some systems such as HPUX 9.
dnl AC_CHECK_LIB(resolv, gethostbyname)


But otherwise:

AC_CHECK_FUNCS(res_init)
if test "X$HAVE_RES_INIT" = X; then
   AC_SEARCH_LIBS(res_init, resolv, HAVE_RES_INIT=yes)
fi

Untested :-)


	Jan D.



> If that is in place, it should be trivial to add
> 
> #if HAVE_RES_INIT
>     res_init();
> #endif
> 
> before all calls to gethostbyname and getaddrinfo.
> 
> 

Re: make-network-process's gethostbyname usage

[Richard Stallman]

    So how can we fix configure to DTRT?

    a) Include -lresolv in the check for HAVE_RES_INIT.

We can certainly do that, if that gives the right results.
Do you know of any residual problem that would remain
if that is fixed?

    b) Include -lresolv in LIBS if and only if HAVE_RES_INIT is defined.

    If that is in place, it should be trivial to add

    #if HAVE_RES_INIT
	res_init();
    #endif

    before all calls to gethostbyname and getaddrinfo.

By all means.

Re: make-network-process's gethostbyname usage

[Chong Yidong]

storm@cua.dk (Kim F. Storm) writes:

> Richard Stallman <rms@gnu.org> writes:
>
>>     make-network-process calles gethostbyname, but never calls
>>     res_init. The problem, which appears is: if contents of
>>     /etc/resolv.conf was changed old dns servers are unavailable,
>>     gethostbyname still tries them, not new ones. So this bug is pretty
>>     much the same as http://bugs.debian.org/438680 (which has a patch).
>>
>> Would someone please DTRT about this, then ack?
>> It needs to be fixed in Emacs 22.
>
> I don't know what to do about this.
>
> configure already has a check for res_init, but on my system (GNU/Linux),
> HAVE_RES_INIT is undefined, indicating that it doesn't exist.
>
> However, I believe that it does exist, but only if -lresolv is specified.
>
> So how can we fix configure to DTRT?

See the attached patch.

Strangely enough, even though autoconf needs -lresolv to detect
res_init(), it doesn't seem as though we need to add -lresolv when
compiling Emacs itself.  I don't know why this should be the case.
Does anyone have an idea?

*** emacs/configure.in.~1.444.2.2.~	2007-08-22 17:50:31.000000000 -0400
--- emacs/configure.in	2007-08-25 14:34:43.000000000 -0400
***************
*** 2580,2586 ****
  
  AC_CHECK_FUNCS(gethostname getdomainname dup2 \
  rename closedir mkdir rmdir sysinfo getrusage get_current_dir_name \
! random lrand48 bcopy bcmp logb frexp fmod rint cbrt ftime res_init setsid \
  strerror fpathconf select mktime euidaccess getpagesize tzset setlocale \
  utimes setrlimit setpgid getcwd getwd shutdown getaddrinfo \
  __fpending mblen mbrlen mbsinit strsignal setitimer ualarm index rindex \
--- 2580,2586 ----
  
  AC_CHECK_FUNCS(gethostname getdomainname dup2 \
  rename closedir mkdir rmdir sysinfo getrusage get_current_dir_name \
! random lrand48 bcopy bcmp logb frexp fmod rint cbrt ftime setsid \
  strerror fpathconf select mktime euidaccess getpagesize tzset setlocale \
  utimes setrlimit setpgid getcwd getwd shutdown getaddrinfo \
  __fpending mblen mbrlen mbsinit strsignal setitimer ualarm index rindex \
***************
*** 2624,2640 ****
  # than to expect to find it in ncurses.
  AC_CHECK_LIB(ncurses, tparm)
  
  # Do we need the Hesiod library to provide the support routines?
  if test "$with_hesiod" = yes ; then
    # Don't set $LIBS here -- see comments above.
-   resolv=no
    AC_CHECK_FUNC(res_send, , [AC_CHECK_FUNC(__res_send, ,
       [AC_CHECK_LIB(resolv, res_send, resolv=yes,
  		  [AC_CHECK_LIB(resolv, __res_send, resolv=yes)])])])
    if test "$resolv" = yes ; then
      RESOLVLIB=-lresolv
-     AC_DEFINE(HAVE_LIBRESOLV, 1,
- 	      [Define to 1 if you have the resolv library (-lresolv).])
    else
      RESOLVLIB=
    fi
--- 2624,2662 ----
  # than to expect to find it in ncurses.
  AC_CHECK_LIB(ncurses, tparm)
  
+ resolv=no
+ 
+ # Do we have res_init, for detecting changes in /etc/resolv.conf?
+ 
+ AC_CHECK_FUNC(res_init, have_res_init=yes, have_res_init=no)
+ if test "$have_res_init" = no; then
+   OLIBS="$LIBS"
+   LIBS="$LIBS -lresolv"
+   AC_MSG_CHECKING(for res_init with -lresolv)
+   AC_TRY_LINK([#include <netinet/in.h>
+ #include <arpa/nameser.h>
+ #include <resolv.h> ],
+     [res_init();],
+     have_res_init=yes, have_res_init=no)
+   AC_MSG_RESULT($have_res_init)
+   if test "$have_res_init" = yes ; then
+     resolv=yes
+   fi
+   LIBS="$OLIBS"
+ fi
+ 
+ if test "$have_res_init" = yes; then
+   AC_DEFINE(HAVE_RES_INIT, 1, [Define to 1 if res_init is available.])
+ fi
+ 
  # Do we need the Hesiod library to provide the support routines?
  if test "$with_hesiod" = yes ; then
    # Don't set $LIBS here -- see comments above.
    AC_CHECK_FUNC(res_send, , [AC_CHECK_FUNC(__res_send, ,
       [AC_CHECK_LIB(resolv, res_send, resolv=yes,
  		  [AC_CHECK_LIB(resolv, __res_send, resolv=yes)])])])
    if test "$resolv" = yes ; then
      RESOLVLIB=-lresolv
    else
      RESOLVLIB=
    fi
***************
*** 2644,2649 ****
--- 2666,2677 ----
  	:, $RESOLVLIB)])
  fi
  
+ # Do we need libresolv (due to res_init or Hesiod)?
+ if test "$resolv" = yes ; then
+   AC_DEFINE(HAVE_LIBRESOLV, 1,
+             [Define to 1 if you have the resolv library (-lresolv).])
+ fi
+ 
  # These tell us which Kerberos-related libraries to use.
  if test "${with_kerberos+set}" = set; then
    AC_CHECK_LIB(com_err, com_err)
*** emacs/src/process.c.~1.512.2.4.~	2007-08-08 14:07:26.000000000 -0400
--- emacs/src/process.c	2007-08-25 14:56:59.000000000 -0400
***************
*** 129,134 ****
--- 129,140 ----
  # endif
  #endif
  
+ #ifdef HAVE_RES_INIT
+ #include <netinet/in.h>
+ #include <arpa/nameser.h>
+ #include <resolv.h>
+ #endif
+ 
  #include "lisp.h"
  #include "systime.h"
  #include "systty.h"
***************
*** 3082,3087 ****
--- 3088,3098 ----
        hints.ai_family = family;
        hints.ai_socktype = socktype;
        hints.ai_protocol = 0;
+ 
+ #ifdef HAVE_RES_INIT
+       res_init ();
+ #endif
+ 
        ret = getaddrinfo (SDATA (host), portstring, &hints, &res);
        if (ret)
  #ifdef HAVE_GAI_STRERROR
***************
*** 3127,3132 ****
--- 3138,3148 ----
  	 as it may `hang' Emacs for a very long time.  */
        immediate_quit = 1;
        QUIT;
+ 
+ #ifdef HAVE_RES_INIT
+       res_init ();
+ #endif
+ 
        host_info_ptr = gethostbyname (SDATA (host));
        immediate_quit = 0;

Re: make-network-process's gethostbyname usage

[Jan Djärv]

Chong Yidong skrev:
> storm@cua.dk (Kim F. Storm) writes:
> 
>> Richard Stallman <rms@gnu.org> writes:
>>
>>>     make-network-process calles gethostbyname, but never calls
>>>     res_init. The problem, which appears is: if contents of
>>>     /etc/resolv.conf was changed old dns servers are unavailable,
>>>     gethostbyname still tries them, not new ones. So this bug is pretty
>>>     much the same as http://bugs.debian.org/438680 (which has a patch).
>>>
>>> Would someone please DTRT about this, then ack?
>>> It needs to be fixed in Emacs 22.
>> I don't know what to do about this.
>>
>> configure already has a check for res_init, but on my system (GNU/Linux),
>> HAVE_RES_INIT is undefined, indicating that it doesn't exist.
>>
>> However, I believe that it does exist, but only if -lresolv is specified.
>>
>> So how can we fix configure to DTRT?
> 
> See the attached patch.
> 
> Strangely enough, even though autoconf needs -lresolv to detect
> res_init(), it doesn't seem as though we need to add -lresolv when
> compiling Emacs itself.  I don't know why this should be the case.
> Does anyone have an idea?

Probably some magic in GNU libc, I would not rely on this working on other
systems that don't use GNU libc and/or linker.

You should use the predefined macros instead of using AC_TRY_LINK, in this
case AC_CHECK_LIB.

	Jan D.

> 
> *** emacs/configure.in.~1.444.2.2.~	2007-08-22 17:50:31.000000000 -0400
> --- emacs/configure.in	2007-08-25 14:34:43.000000000 -0400
> ***************
> *** 2580,2586 ****
>   
>   AC_CHECK_FUNCS(gethostname getdomainname dup2 \
>   rename closedir mkdir rmdir sysinfo getrusage get_current_dir_name \
> ! random lrand48 bcopy bcmp logb frexp fmod rint cbrt ftime res_init setsid \
>   strerror fpathconf select mktime euidaccess getpagesize tzset setlocale \
>   utimes setrlimit setpgid getcwd getwd shutdown getaddrinfo \
>   __fpending mblen mbrlen mbsinit strsignal setitimer ualarm index rindex \
> --- 2580,2586 ----
>   
>   AC_CHECK_FUNCS(gethostname getdomainname dup2 \
>   rename closedir mkdir rmdir sysinfo getrusage get_current_dir_name \
> ! random lrand48 bcopy bcmp logb frexp fmod rint cbrt ftime setsid \
>   strerror fpathconf select mktime euidaccess getpagesize tzset setlocale \
>   utimes setrlimit setpgid getcwd getwd shutdown getaddrinfo \
>   __fpending mblen mbrlen mbsinit strsignal setitimer ualarm index rindex \
> ***************
> *** 2624,2640 ****
>   # than to expect to find it in ncurses.
>   AC_CHECK_LIB(ncurses, tparm)
>   
>   # Do we need the Hesiod library to provide the support routines?
>   if test "$with_hesiod" = yes ; then
>     # Don't set $LIBS here -- see comments above.
> -   resolv=no
>     AC_CHECK_FUNC(res_send, , [AC_CHECK_FUNC(__res_send, ,
>        [AC_CHECK_LIB(resolv, res_send, resolv=yes,
>   		  [AC_CHECK_LIB(resolv, __res_send, resolv=yes)])])])
>     if test "$resolv" = yes ; then
>       RESOLVLIB=-lresolv
> -     AC_DEFINE(HAVE_LIBRESOLV, 1,
> - 	      [Define to 1 if you have the resolv library (-lresolv).])
>     else
>       RESOLVLIB=
>     fi
> --- 2624,2662 ----
>   # than to expect to find it in ncurses.
>   AC_CHECK_LIB(ncurses, tparm)
>   
> + resolv=no
> + 
> + # Do we have res_init, for detecting changes in /etc/resolv.conf?
> + 
> + AC_CHECK_FUNC(res_init, have_res_init=yes, have_res_init=no)
> + if test "$have_res_init" = no; then
> +   OLIBS="$LIBS"
> +   LIBS="$LIBS -lresolv"
> +   AC_MSG_CHECKING(for res_init with -lresolv)
> +   AC_TRY_LINK([#include <netinet/in.h>
> + #include <arpa/nameser.h>
> + #include <resolv.h> ],
> +     [res_init();],
> +     have_res_init=yes, have_res_init=no)
> +   AC_MSG_RESULT($have_res_init)
> +   if test "$have_res_init" = yes ; then
> +     resolv=yes
> +   fi
> +   LIBS="$OLIBS"
> + fi
> + 
> + if test "$have_res_init" = yes; then
> +   AC_DEFINE(HAVE_RES_INIT, 1, [Define to 1 if res_init is available.])
> + fi
> + 
>   # Do we need the Hesiod library to provide the support routines?
>   if test "$with_hesiod" = yes ; then
>     # Don't set $LIBS here -- see comments above.
>     AC_CHECK_FUNC(res_send, , [AC_CHECK_FUNC(__res_send, ,
>        [AC_CHECK_LIB(resolv, res_send, resolv=yes,
>   		  [AC_CHECK_LIB(resolv, __res_send, resolv=yes)])])])
>     if test "$resolv" = yes ; then
>       RESOLVLIB=-lresolv
>     else
>       RESOLVLIB=
>     fi
> ***************
> *** 2644,2649 ****
> --- 2666,2677 ----
>   	:, $RESOLVLIB)])
>   fi
>   
> + # Do we need libresolv (due to res_init or Hesiod)?
> + if test "$resolv" = yes ; then
> +   AC_DEFINE(HAVE_LIBRESOLV, 1,
> +             [Define to 1 if you have the resolv library (-lresolv).])
> + fi
> + 
>   # These tell us which Kerberos-related libraries to use.
>   if test "${with_kerberos+set}" = set; then
>     AC_CHECK_LIB(com_err, com_err)
> *** emacs/src/process.c.~1.512.2.4.~	2007-08-08 14:07:26.000000000 -0400
> --- emacs/src/process.c	2007-08-25 14:56:59.000000000 -0400
> ***************
> *** 129,134 ****
> --- 129,140 ----
>   # endif
>   #endif
>   
> + #ifdef HAVE_RES_INIT
> + #include <netinet/in.h>
> + #include <arpa/nameser.h>
> + #include <resolv.h>
> + #endif
> + 
>   #include "lisp.h"
>   #include "systime.h"
>   #include "systty.h"
> ***************
> *** 3082,3087 ****
> --- 3088,3098 ----
>         hints.ai_family = family;
>         hints.ai_socktype = socktype;
>         hints.ai_protocol = 0;
> + 
> + #ifdef HAVE_RES_INIT
> +       res_init ();
> + #endif
> + 
>         ret = getaddrinfo (SDATA (host), portstring, &hints, &res);
>         if (ret)
>   #ifdef HAVE_GAI_STRERROR
> ***************
> *** 3127,3132 ****
> --- 3138,3148 ----
>   	 as it may `hang' Emacs for a very long time.  */
>         immediate_quit = 1;
>         QUIT;
> + 
> + #ifdef HAVE_RES_INIT
> +       res_init ();
> + #endif
> + 
>         host_info_ptr = gethostbyname (SDATA (host));
>         immediate_quit = 0;
>   
> 
> 
> _______________________________________________
> Emacs-devel mailing list
> Emacs-devel@gnu.org
> http://lists.gnu.org/mailman/listinfo/emacs-devel