Compile of mod.c fails with --enable-static-modules about pdsh

[GoogleCodeExporter]

--enable-static-modules was used in an effort to eliminate the need to set the 
LD_LIBRARY_PATH variable when using pdsh 2.25.  Unless LD_LIBRARY_PATH is set 
pdsh cannot load genders.so, evidently because it is linked with 
-nodefaultrpath.

Is there a way we can change/override -nodefaultrpath for the genders module so 
it is linked with an rpath so we don't need to set LD_LIBRARY_PATH? 

Original comment by [email protected] on 25 Apr 2011 at 3:01

[GoogleCodeExporter]

Sorry about --enable-static-modules, static modules probably have not been 
testing
in awhile. I'll work on getting that basically working again.

I'm not sure how to override -nodefaultrpath. It appears that libtool is setting
this, but I don't see a way to override it, see:

 http://www.mail-archive.com/[email protected]/msg01946.html

for what appears to be the original reason this is done on HP-UX.

Reading about the HP-UX linker, it seems that you can embed search paths
into binaries with the +b linker option. See

 http://docs.hp.com/en/B2355-90968/linkertasks.htm#MOVINGLIBRARIES-B

maybe you could try adding

 "-Wl,+b /path/to/libgenders"

to CFLAGS or just

 "+b /path/to/libgenders"

to LDFLAGS?

Original comment by [email protected] on 25 Apr 2011 at 4:18

• Changed state: Accepted

[GoogleCodeExporter]

For the build with --enable-static-modules, can you verify that the attached 
patch fixes the build time issue? There are some other fixes needed in the 
testuite, but
since the testsuite doesn't work on hp-ux either...

Original comment by [email protected] on 25 Apr 2011 at 5:18

Attachments:

• 0001-fix-compilation-of-pdsh-with-static-modules.patch

[GoogleCodeExporter]

This issue was updated by revision r1323.


Move prototypes in mod.c so they are not ifdef'd out when building
with --enable-static-modules. This should fix the static module
build.

Original comment by [email protected] on 25 Apr 2011 at 10:38

[GoogleCodeExporter]

This issue was updated by revision r1324.


Fixes for testsuite and --enable-static-modules

Original comment by [email protected] on 25 Apr 2011 at 10:38

[GoogleCodeExporter]

I built svn rev 1325 with static modules and everything looks good except for 
pdsh not having any embedded paths to load libraries from so it still needs to 
have LD_LIBRARY_PATH set.  I'm going to play with compiler/linker options to 
try to fix that.

I attached the "make check" output - there are a few errors you may want to 
check.

We really appreciate your fast response and patches to all of our issues!  We 
use pdsh about a million times/day and the genders support is going to be great 
to use.  Looking forward to replacing our old 2.7 soon. 

Original comment by [email protected] on 26 Apr 2011 at 2:15

[GoogleCodeExporter]

Original comment by [email protected] on 26 Apr 2011 at 2:15

Attachments:

• check.0426110902

[GoogleCodeExporter]

I attached the make check output to issue 22, in which I'm tracking testsuite
failures on HP-UX. It looks like HP-UX doesn't have seq(1). Will have to
encode one in a shell function I guess -- or is it perhaps in another path?

Original comment by [email protected] on 26 Apr 2011 at 2:36

[GoogleCodeExporter]

I also noticed you are not building the rcmd/exec module. This is going to
allow a lot of the tests to work, and is actually a very useful module to have
built (It allows you to run any command in parallel with pdsh, e.g. grep)

Original comment by [email protected] on 26 Apr 2011 at 2:40

[GoogleCodeExporter]

Right - HP-UX does not appear to have a seq command and a rudimentary search 
doesn't turn one up.

We use only ssh transport with pdsh - all other remote shells (e.g., remsh, 
rexec, rlogin) are disabled on all our systems due to security requirements.  
So we run any command we want in parallel on up to 300+ hosts using pdsh with 
ssh.

wilma:/home/root/genders-1.17# pdsh -V
pdsh-2.25 (+static-modules+readline)
rcmd modules: ssh
misc modules: genders

pdsh 2.25+1325 is in production use now.  Among other things we use pdsh in 
scripts to manage user accounts across 300+ Linux and hpux systems.  Here's a 
pdsh command from one of our scripts:

<snip>
pdsh -f $FANOUT -w $HOSTS '
  if (('$DEBUG')) ; then
    set -x
  fi

  LOCALDATE=$(date -u +%Y%j)
  if [[ $(uname) = Linux ]] ; then

    grep -q "^'$USER':"  /etc/shadow || exit 0
    grep -q "^'$USER':!" /etc/shadow && LOCKS="ADMIN"

    # convert seconds to days for linux password date
    LDATE=$(('$DATE'/86400))

    # update the password and date
    /usr/sbin/usermod -p '\'$PWD\''    '$USER'
    [[ '$SYSTEMDATE' = $LOCALDATE ]] \
      && /usr/bin/chage    -d $LDATE        '$USER'

    # remove lock
    /usr/bin/passwd   -u '$USER' >/dev/null

    # reset failed login count
    DENY=$(grep -Eo "deny=[0-9]" '$SYSTEMAUTH' |awk -F= '\'{print \$NF}\'')
    TALLY=$(/sbin/pam_tally --user '$USER' --reset |awk '\'{print \$NF}\'')
    [[ ! -z $DENY ]]            \
      && (($TALLY >= $DENY))    \
      && LOCKS="${LOCKS}${LOCKS:+, }MAXTRIES"

  else

    # format the lockout flags for display
    lockout=$(/usr/lbin/getprpw -m lockout '$USER' 2>/dev/null)
    [[ ($? > 0) ]] && exit 0

    #   NOTE: bash requires the parentheses in a test statement be escaped.
    lockout=$(echo $lockout |cut -f2 -d=)
    [[ $lockout = @\(1??????\) ]] && LOCKS="PWDEXPIRED"
    [[ $lockout = @\(?1?????\) ]] && LOCKS="${LOCKS}${LOCKS:+, }INACTIVE"
    [[ $lockout = @\(??1????\) ]] && LOCKS="${LOCKS}${LOCKS:+, }EXPIRED"
    [[ $lockout = @\(???1???\) ]] && LOCKS="${LOCKS}${LOCKS:+, }MAXTRIES"
    [[ $lockout = @\(????1??\) ]] && LOCKS="${LOCKS}${LOCKS:+, }NULLPWD"
    [[ $lockout = @\(?????1?\) ]] && LOCKS="${LOCKS}${LOCKS:+, }ADMIN"
    [[ $lockout = @\(??????1\) ]] && LOCKS="${LOCKS}${LOCKS:+, }PWDISA*"

    # remove all locks
    /usr/lbin/modprpw -k '$USER'

    # ensure the password date field exists
    [[ $(grep -c u_succhg '$UTCB') = 0 ]]       \
    && { UTCBVAR=$(< '$UTCB')
      echo "$UTCBVAR"                           \
      |sed -e "5s/:/:u_succhg#0:/"              \
      >'$UTCB'
      }

    # update the password and (if the local and remote
    # system dates match) date 
    [[ '$SYSTEMDATE' = $LOCALDATE ]]            \
      && EDTDATE='\'-e "s/u_succhg#[^:]*:/u_succhg#$DATE:/"\''
    UTCBVAR=$(< '$UTCB')
    echo "$UTCBVAR"                             \
    |sed $EDTDATE                               \
      -e "s#u_pwd=[^:]*:#u_pwd='$PWD':#"        \
      >'$UTCB'
  fi

  LOCKS=${LOCKS:-NONE}

  echo "'$USER' password updated, locks reset=$LOCKS"
'
</snip>

Original comment by [email protected] on 26 Apr 2011 at 7:36

[GoogleCodeExporter]

Nice, were you able then to get around the LD_LIBRARY_PATH issue?

Original comment by [email protected] on 26 Apr 2011 at 7:38

[GoogleCodeExporter]

Yes, using 

                LDFLAGS="-L/usr/local/lib/hpux32 -L/usr/lib/hpux32      \
                +b/usr/lib/hpux32:/usr/local/lib/hpux32"                \

fixed the LD_LIBRARY_PATH issue for pdsh.

But unfortunately that doesn't work for nodeattr in the genders package; 
LD_LIBRARY_PATH still needs to be set to use it.  I've recompiled it about 18 
times today trying different options but it always gets linked with NO embedded 
path.  Of course we don't use nodeattr with pdsh but we are testing it for use 
in some other things. 

So, pdsh with the genders module is working great and I'll be modifying some of 
our pdsh scripts to take advantage of genders.  genders is just what we needed 
and is even better that the proposed "include" enhancement to dshgroups that 
Sean (my coworker) submitted last week.

Original comment by [email protected] on 26 Apr 2011 at 8:40

[GoogleCodeExporter]

Original comment by [email protected] on 27 Apr 2011 at 6:15

• Changed state: Fixed