[Libvirt-cim] Test Run Summary (Jan 19 2009): KVM on Fedora release 9.90.1 (Rawhide) with sfcb

[Kaitlin Rupert]

  > sfcb log says a sigsev:
> -#- Virt_ComputerSystemIndicationProvider - 10624 provider exiting due 
> to a SIGSEGV signal
> 
> Debugging there, a backtrace shows me a null mi->ft in 
> stdi_enable_indications() yet same pointer in sfcb's stack is valid:
> (gdb) bt
> #0  stdi_enable_indications (mi=0xbfb3e778, ctx=0xb801f0b8) at 
> std_indication.c:270
> #1  0xb7fc137c in enableIndications (hdr=0x80625c0, info=0x805b0d0, 
> requestor=-75) at providerDrv.c:2192
> #2  0xb7fc2d80 in processProviderInvocationRequestsThread 
> (prms=0x805a8d0) at providerDrv.c:2585
> #3  0xb7fc3514 in processProviderInvocationRequests (name=0x8050720 
> "Virt_ComputerSystemIndicationProvider") at providerDrv.c:2700
> #4  0xb7fb523c in getProcess (info=0x80507a8, proc=0xbfb3ef14) at 
> providerDrv.c:602
> #5  0xb7fb5709 in forkProvider (info=0x80507a8, req=0x805abb0, msg=0x0) 
> at providerDrv.c:653
> #6  0xb7fadc9d in lookupProviderList (type=4, requestor=0xbfb3f018, 
> req=0x805abb0) at providerMgr.c:368
> #7  0xb7fae3eb in processIndProviderList (requestor=0xbfb3f018, 
> req=0x805abb0) at providerMgr.c:449
> #8  0xb7fb025e in processProviderMgrRequests () at providerMgr.c:857
> #9  0x0804aa14 in main (argc=3, argv=0xbfb3f154) at sfcBroker.c:791
> (gdb) p *mi
> $5 = {hdl = 0x0, ft = 0x0}
> (gdb) up
> #1  0xb7fc137c in enableIndications (hdr=0x80625c0, info=0x805b0d0, 
> requestor=-75) at providerDrv.c:2192
> 2192 info->indicationMI->ft->enableIndications(info->indicationMI,ctx);
> (gdb) p *(info->indicationMI)
> $6 = {hdl = 0xb801f0a0, ft = 0xb801f180}
> (gdb)

Yes, this is really odd. The info is correct in sfcb's stack:

(gdb) up
#1  0xb7ff437c in enableIndications (hdr=0x805f2b8, info=0x805a900,
     requestor=-75) at providerDrv.c:2192
2192			info->indicationMI->ft->enableIndications(info->indicationMI,ctx);
(gdb) p *(info->indicationMI)
$4 = {hdl = 0xb80530a0, ft = 0xb8053180}
(gdb) p info->indicationMI->ft
$5 = (CMPIIndicationMIFT *) 0xb8053180
(gdb) p info->indicationMI->ft->miName
$6 = 0xb8050b44 "IndicationVirt_ComputerSystemIndicationProvider"

I put a break statement on the following line in libcmpiutil in 
stdi_enable_indications():

_ctx = (struct std_indication_ctx *)mi->hdl;

The contents of mi were already blown at that point.  We're not doing 
any kind of manipulation on mi, so I suspect that sfcb is handing us a 
bogus mi pointer.  I'm not sure how that is happening though.

-- 
Kaitlin Rupert
IBM Linux Technology Center
kaitlin at linux.vnet.ibm.com