On 01/18/2013 01:18 PM, Eric Blake wrote:
On 01/18/2013 09:02 AM, Stefan Berger wrote:
Enable the passing of a file descriptor via fd=<..> to access the host's
TPM device using the TPM passthrough driver.
Do we still need this, or is it sufficient to use path=/dev/fdset/nnn,
so that we are reusing common fd passing mechanisms without inventing
yet more variants?
Well, it's similar to -netdev tap,fd=27,id=xyz
+++ b/qapi-schema.json
@@ -3033,11 +3033,13 @@
#
# @cancel_path: #optional Path to TPM backend device's cancel sysfs entry
#
+# @fd: #optional File descriptor for the TPM backend device
+#
# Since: 1.5.0
##
{ 'type': 'TPMInfo',
'data': {'model': 'str', 'id': 'str', 'type': 'str', '*path': 'str',
- '*cancel_path': 'str' } }
+ '*cancel_path': 'str', '*fd' : 'int' } }
Besides, what integer value would you use for fd? Older commands that
support fd passing did so via 'int' on the command line, but via a 'str'
via QMP (the name associated with the fd when using 'getfd'), since QMP
does not have a way to expose _which_ fd is the right number from qemu's
When libvirt forks, the child process inherits the file descriptors,
among them those of the taps and /dev/tpm0. The subsequent execve keeps
the file descriptor open. QEMU then reads the TPM file descriptor from
the command line into above TPMInfo->fd. This also works with 'exec
100<>/dev/tpm0' via command line.
Similar to the SELinux labeling of all the other file descriptors I also
use the one for the TPM device for SELinux labeling.
perspective (it's not necessarily the same fd as in the management process).
Hm.
I think this patch should just be dropped.
Regards,
Stefan
