After an evaluation, GNOME has moved from Bugzilla to GitLab. Learn more about GitLab.
No new issues can be reported in GNOME Bugzilla anymore.
To report an issue in a GNOME project, go to GNOME GitLab.
Do not go to GNOME Gitlab for: Bluefish, Doxygen, GnuCash, GStreamer, java-gnome, LDTP, NetworkManager, Tomboy.
Bug 740923 - Crash in tracker-store at tracker_fts_shutdown_db
Crash in tracker-store at tracker_fts_shutdown_db
Status: RESOLVED FIXED
Product: tracker
Classification: Core
Component: FTS
unspecified
Other Windows
: Normal normal
: ---
Assigned To: tracker-general
Depends on:
Blocks:
 
 
Reported: 2014-11-30 14:07 UTC by Philip Van Hoof
Modified: 2014-12-27 16:39 UTC
See Also:
GNOME target: ---
GNOME version: ---

Attachments
Protect global variable against simul destruction (1.03 KB, patch)
2014-11-30 14:24 UTC, Philip Van Hoof 		rejected Details | Review

Description Philip Van Hoof 2014-11-30 14:07:48 UTC 



+
Trace
    234380






Thread 1
          (LWP 2028)


    

  • 
#0
__GI___libc_free

    
                at malloc.c
                  line
                  2970

    

    • 

  • 
#1
g_free

    
                at gmem.c
                  line
                  197

    

    • 

  • 
#2
g_strfreev

    
                at gstrfuncs.c
                  line
                  2424

    

    • 

  • 
#3
tracker_fts_shutdown_db

    
                at tracker-fts.c
                  line
                  268

    

    • 

  • 
#4
tracker_db_interface_sqlite_finalize

    
                at tracker-db-interface-sqlite.c
                  line
                  1338

    

    • 

  • 
#5
g_object_unref

    
                at gobject.c
                  line
                  3197

    

    • 

  • 
#6
__nptl_deallocate_tsd

    
                at pthread_create.c
                  line
                  156

    

    • 

  • 
#7
start_thread

    
                at pthread_create.c
                  line
                  315

    

    • 

  • 
#8
??

    
                from
                /lib/libc.so.6

    

    • 

  • 
#9
??

    
                from
                /lib/libc.so.6

    

    • 



Thread 2
          (LWP 2025)


    

  • 
#0
releaseMemArray

    
                at sqlite3.c
                  line
                  62951

    

    • 

  • 
#1
releaseMemArray

    
                at sqlite3.c
                  line
                  62940

    

    • 

  • 
#2
closeAllCursors

    
                at sqlite3.c
                  line
                  63520

    

    • 

  • 
#3
sqlite3VdbeHalt

    
                at sqlite3.c
                  line
                  64016

    

    • 

  • 
#4
sqlite3VdbeReset

    
                at sqlite3.c
                  line
                  64259

    

    • 

  • 
#5
sqlite3VdbeFinalize

    
                at sqlite3.c
                  line
                  64334

    

    • 

  • 
#6
sqlite3_finalize

    
                at sqlite3.c
                  line
                  65819

    

    • 

  • 
#7
tracker_db_statement_finalize

    
                at tracker-db-interface-sqlite.c
                  line
                  1813

    

    • 

  • 
#8
g_object_unref

    
                at gobject.c
                  line
                  3197

    

    • 

  • 
#9
g_hash_table_remove_all_nodes

    
                at ghash.c
                  line
                  500

    

    • 

  • 
#10
g_hash_table_unref

    
                at ghash.c
                  line
                  1026

    

    • 

  • 
#11
close_database

    
                at tracker-db-interface-sqlite.c
                  line
                  1098

    

    • 

  • 
#12
tracker_db_interface_sqlite_finalize

    
                at tracker-db-interface-sqlite.c
                  line
                  1341

    

    • 

  • 
#13
g_object_unref

    
                at gobject.c
                  line
                  3197

    

    • 

  • 
#14
__nptl_deallocate_tsd

    
                at pthread_create.c
                  line
                  156

    

    • 

  • 
#15
start_thread

    
                at pthread_create.c
                  line
                  315

    

    • 

  • 
#16
??

    
                from
                /lib/libc.so.6

    

    • 

  • 
#17
??

    
                from
                /lib/libc.so.6

    

    • 



Thread 4
          (LWP 2027)


    

  • 
#0
tracker_fts_shutdown_db

    
                at tracker-fts.c
                  line
                  265

    

    • 

  • 
#1
tracker_db_interface_sqlite_finalize

    
                at tracker-db-interface-sqlite.c
                  line
                  1338

    

    • 

  • 
#2
g_object_unref

    
                at gobject.c
                  line
                  3197

    

    • 

  • 
#3
__nptl_deallocate_tsd

    
                at pthread_create.c
                  line
                  156

    

    • 

  • 
#4
start_thread

    
                at pthread_create.c
                  line
                  315

    

    • 

  • 
#5
??

    
                from
                /lib/libc.so.6

    

    • 

  • 
#6
??

    
                from
                /lib/libc.so.6

    

    • 



Thread 5
          (LWP 2007)


    

  • 
#0
tracker_fts_shutdown_db

    
                at tracker-fts.c
                  line
                  265

    

    • 

  • 
#1
tracker_db_interface_sqlite_finalize

    
                at tracker-db-interface-sqlite.c
                  line
                  1338

    

    • 

  • 
#2
g_object_unref

    
                at gobject.c
                  line
                  3197

    

    • 

  • 
#3
g_private_replace

    
                at gthread-posix.c
                  line
                  1067

    

    • 

  • 
#4
tracker_db_manager_shutdown

    
                at tracker-db-manager.c
                  line
                  1279

    

    • 

  • 
#5
tracker_data_manager_shutdown

    
                at tracker-data-manager.c
                  line
                  4612

    

    • 

  • 
#6
tracker_main_main

    
                at /usr/src/debug/tracker-1.1.4+git4/tracker/src/tracker-store/tracker-main.vala
                  line
                  329

    

    • 

  • 
#7
main

    
                at /usr/src/debug/tracker-1.1.4+git4/tracker/src/tracker-store/tracker-main.vala
                  line
                  158

    

    • 







Backtrace stopped: previous frame identical to this frame (corrupt stack?)






Comment 1


Philip Van Hoof





          2014-11-30 14:24:10 UTC
        



Created attachment 291831 [details] [review]
Protect global variable against simul destruction






Comment 2


Martyn Russell





          2014-12-02 09:56:17 UTC
        



Comment on attachment 291831 [details] [review]
Protect global variable against simul destruction

Hi Philip, I thought we had a fix for this already somewhere...

Anyway, I kind of knew about this bug but wasn't able to reproduce it very easily.

This fix looks right to me, but I wonder if we shouldn't also protect the creation of the property_names too with the/a mutex? I am thinking of re-initialisation or race condition cases.






Comment 3


Philip Van Hoof





          2014-12-02 10:29:12 UTC
        



(In reply to comment #2)
> (From update of attachment 291831 [details] [review])
> Hi Philip, I thought we had a fix for this already somewhere...
> 
> Anyway, I kind of knew about this bug but wasn't able to reproduce it very
> easily.
> 
> This fix looks right to me, but I wonder if we shouldn't also protect the
> creation of the property_names too with the/a mutex? I am thinking of
> re-initialisation or race condition cases.

To be completely correct we should protect each access to the variable, including read, create and destroy, with the same static mutex.

Or we do a thread local copy of the table for each thread. That's also a possibility (which I think might even be better).






Comment 4


Martyn Russell





          2014-12-02 10:36:45 UTC
        



(In reply to comment #3)
> To be completely correct we should protect each access to the variable,
> including read, create and destroy, with the same static mutex.

Indeed.
 
> Or we do a thread local copy of the table for each thread. That's also a
> possibility (which I think might even be better).

I don't think a thread local copy is the right approach because then there could be discrepancies between the table information between threads. It's unlikely but ...






Comment 5


Martyn Russell





          2014-12-27 16:17:18 UTC
        



Comment on attachment 291831 [details] [review]
Protect global variable against simul destruction

This patch is no longer needed, Carlos applied a patch upstream using GPrivate APIs.






Comment 6


Martyn Russell





          2014-12-27 16:39:18 UTC
        



Carlos fixed this commit:

commit a105aacd5531d0d5e7d2af7c12c156c3375135af
Author: Carlos Garnacho <carlosg@gnome.org>
Date:   Sun Nov 30 12:34:31 2014 +0100

    tracker-fts: Make the property list per-thread data
    
    This is global memory, and shared across threads, which doesn't play well
    when (un)initializing it, as multiple simultaneous accesses can occur.