From jens.schmidt35@arcor.de

Created by jens.schmidt35@arcor.de

Consider the following test case​:

---- snip ----
use strict;
use warnings;
use File​::Copy qw( copy );

$SIG{INT} = sub { die };

copy( $ARGV[0], $ARGV[1] );
---- snip ----

Then call as "perl test.pl <large file> <target file>" and hit Ctrl-C
while it runs. In 90% of the cases I get the following error message​:

---- snip ----
[jschmidt@​mstest tmp]$ perl test.pl yyy zzz
Died at test.pl line 5.
refcnt_dec​: fd 4​: 0 <= 0
Quit
---- snip ----

and Perl hangs. I took a stack trace of the hanging Perl​:

---- snip ----
[jschmidt@​unix sbp]$ pstack 5127
#0 0x0022b402 in __kernel_vsyscall ()
#1 0x003a392e in __lll_mutex_lock_wait () from /lib/libpthread.so.0
#2 0x0039f79c in _L_mutex_lock_85 () from /lib/libpthread.so.0
#3 0x0039f2dd in pthread_mutex_lock () from /lib/libpthread.so.0
#4 0x0813dc7a in PerlIOUnix_refcnt_dec ()
#5 0x0813dd78 in PerlIOUnix_close ()
#6 0x0813e777 in PerlIOBase_close ()
#7 0x0813e7b1 in PerlIOBuf_close ()
#8 0x0813fa2b in Perl_PerlIO_close ()
#9 0x08120dc6 in Perl_io_close ()
#10 0x080e2eb4 in Perl_sv_clear ()
#11 0x080e31ba in Perl_sv_free2 ()
#12 0x0807abcf in Perl_gp_free ()
#13 0x080e2fcb in Perl_sv_clear ()
#14 0x080e31ba in Perl_sv_free2 ()
#15 0x080fd89b in Perl_leave_scope ()
#16 0x0807188b in S_my_exit_jump ()
#17 0x0807194f in Perl_my_failure_exit ()
#18 0x080bcaa1 in Perl_vcroak ()
#19 0x080bcb25 in Perl_croak ()
#20 0x0813dd30 in PerlIOUnix_refcnt_dec ()
#21 0x0813dd78 in PerlIOUnix_close ()
#22 0x0813e777 in PerlIOBase_close ()
#23 0x0813e7b1 in PerlIOBuf_close ()
#24 0x0813fa2b in Perl_PerlIO_close ()
#25 0x08120dc6 in Perl_io_close ()
#26 0x080e2eb4 in Perl_sv_clear ()
#27 0x080e31ba in Perl_sv_free2 ()
#28 0x0807abcf in Perl_gp_free ()
#29 0x080e2fcb in Perl_sv_clear ()
#30 0x080e31ba in Perl_sv_free2 ()
#31 0x080fd89b in Perl_leave_scope ()
#32 0x0807188b in S_my_exit_jump ()
#33 0x0807194f in Perl_my_failure_exit ()
#34 0x081096df in Perl_die_where ()
#35 0x080bf996 in Perl_vdie ()
#36 0x080bfa45 in Perl_die ()
#37 0x080c075f in Perl_sighandler ()
#38 0x080c0437 in Perl_despatch_signals ()
#39 0x0813dd9c in PerlIOUnix_close ()
#40 0x0813e777 in PerlIOBase_close ()
#41 0x0813e7b1 in PerlIOBuf_close ()
#42 0x0813fa2b in Perl_PerlIO_close ()
#43 0x08120dc6 in Perl_io_close ()
#44 0x08120f73 in Perl_do_close ()
#45 0x0811411e in Perl_pp_close ()
#46 0x080cfde9 in Perl_runops_standard ()
#47 0x08074c79 in perl_run ()
#48 0x0805fe4d in main ()
---- snip ----

Here is a listing if the file descriptors at hang time (yyy being the
source file of the copy function)​:

---- snip ----
[jschmidt@​unix sbp]$ ls -al /proc/5127/fd/
total 0
dr-x------ 2 sbpadm oraic 0 Dec 20 09​:26 .
dr-xr-xr-x 5 sbpadm oraic 0 Dec 20 09​:26 ..
lrwx------ 1 sbpadm oraic 64 Dec 20 09​:27 0 -> /dev/pts/25
lrwx------ 1 sbpadm oraic 64 Dec 20 09​:27 1 -> /dev/pts/25
lrwx------ 1 sbpadm oraic 64 Dec 20 09​:26 2 -> /dev/pts/25
lr-x------ 1 sbpadm oraic 64 Dec 20 09​:27 3 -> /net/sapmnt.oraicall/sbpmstest/tmp/yyy
---- snip----

I tried to simulate what File​::Copy​::copy does in my test program as well, using
various different methods (open, sysopen) to access the files, but the error
occurred in all cases.

Flags:
    category=core
    severity=medium

Site configuration information for perl 5.10.1:

Configured by oraic at Mon Nov 15 13:29:10 CET 2010.

Summary of my perl5 (revision 5 version 10 subversion 1) configuration:
   
  Platform:
    osname=linux, osvers=2.6.18-8.el5, archname=i686-linux-thread-multi
    uname='linux ls3094 2.6.18-8.el5 #1 smp fri jan 26 14:15:21 est 2007 i686 i686 i386 gnulinux '
    config_args='-d -Dusethreads=yes -Dusevendorprefix=no -Dprefix=/sapmnt/oraic/tools/perl-5.10'
    hint=recommended, useposix=true, d_sigaction=define
    useithreads=define, usemultiplicity=define
    useperlio=define, d_sfio=undef, uselargefiles=define, usesocks=undef
    use64bitint=undef, use64bitall=undef, uselongdouble=undef
    usemymalloc=n, bincompat5005=undef
  Compiler:
    cc='cc', ccflags ='-D_REENTRANT -D_GNU_SOURCE -fno-strict-aliasing -pipe -fstack-protector -I/usr/local/include -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64',
    optimize='-O2',
    cppflags='-D_REENTRANT -D_GNU_SOURCE -fno-strict-aliasing -pipe -fstack-protector -I/usr/local/include'
    ccversion='', gccversion='4.1.1 20070105 (Red Hat 4.1.1-52)', gccosandvers=''
    intsize=4, longsize=4, ptrsize=4, doublesize=8, byteorder=1234
    d_longlong=define, longlongsize=8, d_longdbl=define, longdblsize=12
    ivtype='long', ivsize=4, nvtype='double', nvsize=8, Off_t='off_t', lseeksize=8
    alignbytes=4, prototype=define
  Linker and Libraries:
    ld='cc', ldflags =' -fstack-protector -L/usr/local/lib'
    libpth=/usr/local/lib /lib /usr/lib
    libs=-lnsl -lgdbm -ldb -ldl -lm -lcrypt -lutil -lpthread -lc
    perllibs=-lnsl -ldl -lm -lcrypt -lutil -lpthread -lc
    libc=/lib/libc-2.5.so, so=so, useshrplib=false, libperl=libperl.a
    gnulibc_version='2.5'
  Dynamic Linking:
    dlsrc=dl_dlopen.xs, dlext=so, d_dlsymun=undef, ccdlflags='-Wl,-E'
    cccdlflags='-fPIC', lddlflags='-shared -O2 -L/usr/local/lib -fstack-protector'

Locally applied patches:
    


@INC for perl 5.10.1:
    /sapmnt/oraic/tools/perl-5.10/lib/5.10.1/i686-linux-thread-multi
    /sapmnt/oraic/tools/perl-5.10/lib/5.10.1
    /sapmnt/oraic/tools/perl-5.10/lib/site_perl/5.10.1/i686-linux-thread-multi
    /sapmnt/oraic/tools/perl-5.10/lib/site_perl/5.10.1
    .


Environment for perl 5.10.1:
    HOME=/home/sbpadm
    LANG=C
    LANGUAGE (unset)
    LD_LIBRARY_PATH (unset)
    LOGDIR (unset)
    PATH=/usr/lib/qt-3.3/bin:/usr/kerberos/bin:/usr/local/bin:/bin:/usr/bin
    PERL_BADLANG (unset)
    SHELL=/bin/bash

From jens.schmidt35@arcor.de

Reproduces in perl 5.12.2 as well (same host as above).

jens.schmidt35@arcor.de - Status changed from 'new' to 'open'

jens.schmidt35@arcor.de - Status changed from 'open' to 'new'

From @iabyn

On Mon, Dec 20, 2010 at 12​:44​:33AM -0800, jens.schmidt35@​arcor.de wrote​:

Consider the following test case​:

---- snip ----
use strict;
use warnings;
use File​::Copy qw( copy );

$SIG{INT} = sub { die };

copy( $ARGV[0], $ARGV[1] );
---- snip ----

Then call as "perl test.pl <large file> <target file>" and hit Ctrl-C
while it runs. In 90% of the cases I get the following error message​:

---- snip ----
[jschmidt@​mstest tmp]$ perl test.pl yyy zzz
Died at test.pl line 5.
refcnt_dec​: fd 4​: 0 <= 0
Quit
---- snip ----

and Perl hangs. I took a stack trace of the hanging Perl​:

I suspect this bug (recursively calling into PerlIO via a signal handler)
was fixed by commit abf9167, which is
included in perl-5.14.0-RC1.

However, I was unable to reproduce this with 5.10.1 or any other perl,
so I can't confirm that its been fixed.

--
Spock (or Data) is fired from his high-ranking position for not being able
to understand the most basic nuances of about one in three sentences that
anyone says to him.
  -- Things That Never Happen in "Star Trek" #19

The RT System itself - Status changed from 'new' to 'open'

From jens.schmidt35@arcor.de

Thanks for looking into this.

To a pity, the problem still reproduces with perl-5.14.0-RC1.

I did some more tests (both with perl 5.10 and 5.14), and the problem
reproduces easiest (80-90%) if both source and target file of the copy
are located on an NFS-mounted volume. It reproduces less easier
(10-20%) if only the target file is on an NFS-mounted volume. (NetApp
Filer over a 1GBit network.)

It does not reproduce at all if both source and target are on a local disk.

Hope that helps.

From jens.schmidt35@arcor.de

On Tue Apr 26 06​:21​:18 2011, davem wrote​:

On Mon, Dec 20, 2010 at 12​:44​:33AM -0800, jens.schmidt35@​arcor.de wrote​:
I suspect this bug (recursively calling into PerlIO via a signal handler)
was fixed by commit abf9167, which is
included in perl-5.14.0-RC1.

Ups, I just now noticed the "Reply" button in perlbug, sorry. I've sort
of replied to your suspicion in the bug report, but not via that button.
So I added this reply to ensure that my previous update doesn't get lost.

Sorry for the inconvenience.

From jens.schmidt35@arcor.de

This seems to be even NFS-server dependent. We got a new NFS server a
year ago and now the problem does not reproduce any longer. I'll check
wehther I can find a server where the problem reproduces.