mythwelcome hanging (MythSystemUnix::Fork child QString use leads to possible deadlock)

[naf@…]

I noticed mythwelcome sporadically hanging when calling mythstatus. Logs ended like:

2011-04-07 20:08:19.377 Locking input devices
2011-04-07 20:08:19.386 Launching: /usr/local/bin/mythshutdown --status 0
2011-04-07 20:08:19.388 Managed child (PID: 27773) has started! * command=/usr/local/bin/mythshutdown --status 0, timeout=0

Backtrace of the hung process looked like:

#0  0x00007fe8f25d81ac in __lll_lock_wait () from /lib64/libpthread.so.0
#1  0x00007fe8f25d34bf in _L_lock_1058 () from /lib64/libpthread.so.0
#2  0x00007fe8f25d331b in pthread_mutex_lock () from /lib64/libpthread.so.0
#3  0x0000003c2d2978a9 in ?? () from /usr/lib64/libGL.so.1
#4  0x0000003c2d29a8b9 in ?? () from /usr/lib64/libGL.so.1
#5  0x0000003c2d29ace8 in ?? () from /usr/lib64/libGL.so.1
#6  0x0000003c2da0098d in ?? () from /usr/lib64/tls/libnvidia-tls.so.260.19.29
#7  0x00000031338b6623 in QString::fromLatin1_helper(char const*, int) () from /usr/lib64/libQtCore.so.4
#8  0x00000031338b92bd in QString::fromAscii_helper(char const*, int) () from /usr/lib64/libQtCore.so.4
#9  0x00007fe8f445902d in QString (this=0x1ffb5b0, timeout=0) at /usr/include/QtCore/qstring.h:426
#10 GetSetting (this=0x1ffb5b0, timeout=0) at mythsystem.h:86
#11 GetSetting (this=0x1ffb5b0, timeout=0) at mythsystem.h:147
#12 MythSystemUnix::Fork (this=0x1ffb5b0, timeout=0) at system-unix.cpp:766
#13 0x00007fe8f43d1732 in MythSystem::Run (this=0x201e6b0, timeout=0) at mythsystem.cpp:152
#14 0x00007fe8f43d58c6 in myth_system (command=..., flags=<value optimized out>, timeout=0) at mythsystem.cpp:368
#15 0x0000000000404ba1 in isRunning (program=0x40f770 "mythtranscode") at main.cpp:196
#16 0x00000000004052a6 in getStatus (bWantRecStatus=false) at main.cpp:231
#17 0x000000000040e8f1 in main (argc=3, argv=0x7fffeb559068) at main.cpp:932

So the deadlock was happening in MythSystemUnix::Fork() doing QString stuff during a call to GetSetting?() between fork() and execl().

It appears that bac42c09 fixed possible child deadlocks by removing all QString use between fork() and execl() [and added comments referencing their danger!], but another QString was reintroduced in 7bad51a1.

Attached patch moves the GetSetting?() call (and thus the use of QString) to before the fork(), which solved my intermittent mythwelcome freezes.

[naf]

Fix QString in forked child in mythsystem

Fixes #9714. Patch from the ticket. Thanks.

Signed-off-by: Gavin Hurlbut <ghurlbut@…>

Changeset: e355fe8471024aa026f6194a451c9b767d88bc32