[libvirt] [PATCH] Fix crash in SELinuxSecurityVerify

[Laine Stump]

On 01/13/2011 08:46 AM, Daniel P. Berrange wrote:
> On Thu, Jan 13, 2011 at 12:30:30AM -0500, Laine Stump wrote:
>> When attempting to edit a domain, libvirtd segfaulted in
>> SELinuxSecurityVerify() on this line:
>>
>>    if (!STREQ(virSecurityManagerGetModel(mgr), secdef->model)) {
>>
>> because secdef->model was NULL. Although I'm too tired to investigate
>> in depth, I noticed that all the other functions in that file that do
>> the same STREQ() will first check that def->seclabel.label is
>> non-NULL, but this function doesn't. I also noticed that label *is*
>> NULL in my case, so I tried adding that check to
>> SELinuxSecurityVerify(), and the crash goes away.
>>
>> I have no idea if this is the correct fix, but it allowed me to
>> continue my testing of a new (unrelated) feature.
>> ---
>>   src/security/security_selinux.c |    4 ++++
>>   1 files changed, 4 insertions(+), 0 deletions(-)
>>
>> diff --git a/src/security/security_selinux.c b/src/security/security_selinux.c
>> index d06afde..b97ca4c 100644
>> --- a/src/security/security_selinux.c
>> +++ b/src/security/security_selinux.c
>> @@ -871,6 +871,10 @@ SELinuxSecurityVerify(virSecurityManagerPtr mgr ATTRIBUTE_UNUSED,
>>                         virDomainDefPtr def)
>>   {
>>       const virSecurityLabelDefPtr secdef =&def->seclabel;
>> +
>> +    if (def->seclabel.label == NULL)
>> +        return 0;
>> +
>>       if (!STREQ(virSecurityManagerGetModel(mgr), secdef->model)) {
>>           virSecurityReportError(VIR_ERR_INTERNAL_ERROR,
>>                                  _("security label driver mismatch: "
> We don't want to skip a NULL label, but rather a NULL model.

Yeah, it didn't really make sense at the time, but all the other 
functions were doing it, it stopped the crash, and I was too tired to 
spend brain cells understanding the code :-)

> So I think you actually need to add a check
>
>    if (def->seclabel.model == NULL)
>      return 0;
>
> but in the Verify method in src/security/security_manager.c so
> that all drivers are protected instead of just SELinux.

Okay. Is that needed for the other methods that end up comparing 
secdef->model, too? Or are you guaranteed a non-null model by the time 
you get into any of those? (eg virSecurityManagerSetProcessLabel(), 
virSecurityManagerSetSecuritySocketLabel(),etc)