Re: [Bug-tar] Small bug in tar-1.29

[Jack Howarth]

On Tue, Sep 06, 2016 at 06:48:23PM -0400, Jack Howarth wrote:
> On Tue, Sep 06, 2016 at 04:12:12PM -0400, Jack Howarth wrote:
> > On Tue, Sep 06, 2016 at 08:12:05PM +0100, Ralph Corderoy wrote:
> > > Hi Jack,
> > > 
> > > > > gnutar: Child died with signal 11
> > > ...
> > > > stop reason = signal SIGPIPE
> > > 
> > > SIGPIPE is 13.  We want 11, SIGSEGV.  Perhaps a "c" or two for continue
> > > will have the debugger carry on until the SEGV bites.
> > 
> > $ ulimit -c unlimited
> > $ gnutar -zcvf test_data.tar.gz test_data 
> > test_data/
> > test_data/.noe.tbl.swp
> > test_data/dihedral.tbl
> > test_data/file.nam
> > test_data/io.mc
> > test_data/noe.tbl
> > gnutar: test_data.tar.gz: Cannot write: Broken pipe
> > gnutar: Child died with signal 11
> > gnutar: Error is not recoverable: exiting now
> > $ lldb -c /cores/core.21601 `which gnutar`
> > (lldb) target create "/opt/local/bin/gnutar" --core "/cores/core.21601"
> > warning: (x86_64) /cores/core.21601 load command 84 LC_SEGMENT_64 has a 
> > fileoff + filesize (0x296fe000) that extends beyond the end of the file 
> > (0x296fd000), the segment will be truncated to match
> > Core file '/cores/core.21601' (x86_64) was loaded.
> > (lldb) bt
> > * thread #1: tid = 0x0000, 0x00007fff94ebb6d7 
> > libdispatch.dylib`_dispatch_queue_push_queue + 463, stop reason = signal 
> > SIGSTOP
> >   * frame #0: 0x00007fff94ebb6d7 
> > libdispatch.dylib`_dispatch_queue_push_queue + 463
> >     frame #1: 0x00007fff7478c4d0 libdispatch.dylib`_OS_dispatch_mach_vtable 
> > + 48
> >     frame #2: 0x00007fff94eb9b06 
> > libdispatch.dylib`_dispatch_queue_wakeup_with_qos_slow + 126
> >     frame #3: 0x00007fff94ec013f libdispatch.dylib`_dispatch_mach_msg_send 
> > + 1952
> >     frame #4: 0x00007fff94ebf8dc libdispatch.dylib`dispatch_mach_send + 262
> >     frame #5: 0x00007fff89448fc9 
> > libxpc.dylib`xpc_connection_send_message_with_reply + 131
> >     frame #6: 0x00007fff8e31ebdf 
> > CoreFoundation`__66-[CFPrefsSearchListSource 
> > generationCountFromListOfSources:count:]_block_invoke_2 + 143
> >     frame #7: 0x00007fff8e31ea0d 
> > CoreFoundation`_CFPrefsWithDaemonConnection + 381
> >     frame #8: 0x00007fff8e31db96 
> > CoreFoundation`__66-[CFPrefsSearchListSource 
> > generationCountFromListOfSources:count:]_block_invoke + 150
> >     frame #9: 0x00007fff8e31d933 CoreFoundation`-[CFPrefsSearchListSource 
> > generationCountFromListOfSources:count:] + 179
> >     frame #10: 0x00007fff8e31d214 CoreFoundation`-[CFPrefsSearchListSource 
> > alreadylocked_copyDictionary] + 324
> >     frame #11: 0x00007fff8e31ce5c CoreFoundation`-[CFPrefsSearchListSource 
> > alreadylocked_copyValueForKey:] + 60
> >     frame #12: 0x00007fff8e31cdec 
> > CoreFoundation`___CFPreferencesCopyAppValueWithContainer_block_invoke + 60
> >     frame #13: 0x00007fff8e314b10 CoreFoundation`+[CFPrefsSearchListSource 
> > withSearchListForIdentifier:container:perform:] + 608
> >     frame #14: 0x00007fff8e314867 
> > CoreFoundation`_CFPreferencesCopyAppValueWithContainer + 183
> >     frame #15: 0x00000001021a6538 
> > libintl.8.dylib`_nl_language_preferences_default + 70
> >     frame #16: 0x00000001021a462f libintl.8.dylib`libintl_dcigettext + 667
> >     frame #17: 0x00000001021491f2 gnutar`sys_child_open_for_compress + 125
> >     frame #18: 0x000000010213011f gnutar`open_archive + 603
> >     frame #19: 0x00000001021355ab gnutar`create_archive + 67
> >     frame #20: 0x000000010214b494 gnutar`main + 3510
> >     frame #21: 0x00007fff895605ad libdyld.dylib`start + 1
> >     frame #22: 0x00007fff895605ad libdyld.dylib`start + 1
> > (lldb)
> > 
> 
> Using the same approach that I took in http://savannah.gnu.org/bugs/?46261 to 
> determine that the threading in the CoreFoundation
> framework linked through libintl was causing breakage on darwin15 in make 
> 4.0/4,1...
> 
> rebuilding gettext with the additional configure options...
> 
> gt_cv_func_CFPreferencesCopyAppValue=no \ 
> gt_cv_func_CFLocaleCopyCurrent=no \ 
> gt_cv_func_CFPreferencesCopyAppValue=no \ 
> gt_cv_func_CFLocaleCopyCurrent=no 
> 
> to change the linkage from....
> 
> $ otool -L /opt/local/lib/libintl.8.dylib
> /opt/local/lib/libintl.8.dylib:
>       /opt/local/lib/libintl.8.dylib (compatibility version 10.0.0, current 
> version 10.5.0)
>       /opt/local/lib/libiconv.2.dylib (compatibility version 8.0.0, current 
> version 8.1.0)
>       /usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current 
> version 1226.10.1)
>       
> /System/Library/Frameworks/CoreFoundation.framework/Versions/A/CoreFoundation 
> (compatibility version 150.0.0, current version 1258.1.0)
> 
> $ otool -L /opt/local/lib/libintl.8.dylib
> /opt/local/lib/libintl.8.dylib:
>       /opt/local/lib/libintl.8.dylib (compatibility version 10.0.0, current 
> version 10.5.0)
>       /opt/local/lib/libiconv.2.dylib (compatibility version 8.0.0, current 
> version 8.1.0)
>       /usr/lib/libSystem.B.dylib (compatibility version 1.0.0, current 
> version 1226.10.1)
> 
> does indeed suppress the bug in tar 1.29 when LANG is unset. In the case of 
> make, this was fixed by
> the addition of pselect() calls to protect the threads before forking...
> 
> http://git.savannah.gnu.org/cgit/make.git/commit/?id=85c788572d054bc2c41b84007875edbd37ad3ed5
> 
> [SV 46261] Use pselect() for jobserver where supported.
> * Makefile.am, configure.ac: Check for pselect() and sys/select.h.
> * main.c (main): Block SIGCHLD if we have pselect() support.
> * posixos.c (jobserver_acquire): If we support pselect() then use
> it to query the jobserver pipe, while also listening for SIGCHLD.
> Also pselect() supports a timeout so avoid alarm() calls.
> 
> Perhaps something similar needs to be done for the xpipe() call in tar 1.29?

I think that 
http://git.savannah.gnu.org/cgit/tar.git/commit/?id=589ba77faf60e4632771d80e2b25d9e3778d4a3f

Catch compressor execution errors.
* src/misc.c (write_fatal_details): Move to buffer.c
* src/buffer.c (write_fatal_details): Close the archive and wait for
the compressor program to terminate in order to catch eventual execution
errors.
* src/system.c (sys_child_open_for_compress): Ignore SIGPIPE so tar will
not silently terminate when unable to write to the compressor.
* tests/comperr.at: New file.
* tests/Makefile.am: Add comperr.at
* tests/testsuite.at: Include comperr.at

seems a likely candidate for code change causing this threading conflict with 
CoreFoundation.
In particular, the change...

diff --git a/src/system.c b/src/system.c
index 9414233..dc0b2b0 100644
--- a/src/system.c
+++ b/src/system.c
@@ -330,6 +330,7 @@ sys_child_open_for_compress (void)
   pid_t grandchild_pid;
   pid_t child_pid;
 
+  signal (SIGPIPE, SIG_IGN);
   xpipe (parent_pipe);
   child_pid = xfork ();
 
Perhaps pselect() could be used to fix this on darwin?

> > 
> > > 
> > > -- 
> > > Cheers, Ralph.
> > > https://plus.google.com/+RalphCorderoy