On Thu, Nov 12, 2015 at 23:47:54 +0200, noxdafox wrote: > Greetings, > > I was investigating on an issue in which QEMU's dynamic ownership was > not properly working when calling qemuDomainCoreDumpWithFormat(). Could describe this issue you are investigating? > The core of the issue seems to be the qemuOpenFileAs() function which > does not handle the dynamic ownership. This might affect other libvirt's > features within as well. Because you are most likely looking at a wrong place; qemuOpenFileAs is a quite low level function which is just supposed to open/create a file accessible to a given user. It's up to the caller to decide what the user should be. ... > The issue is that all the functions within the qemu_driver.c module are > static. I could indeed include the module itself in my tests but I'm not > sure whether this is acceptable. We solve this kind of issues by removing "static" from the functions and adding a new header file (if it doesn't exist yet) called *priv.h (qemu_driverpriv.h in this case) with the prototypes for such functions. Only tests are allowed to include such header files. > Furthermore I'd like to have some clarification about the NFS related > code. It seems that some effort has been put in order to tackle > something I'm not aware of. Could someone briefly explain how to > reproduce NFS failing scenarios? The main problem with NFS which this ugly function is trying to handle is called "root-squash". This feature maps all access from UID 0 to an unprivileged UID. That is, libvirtd (even though it is running as root) will not be able to access the desired file. Jirka -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list