On 19-08-2010 15:41, Daniel P. Berrange wrote: >> + case VIR_DOMAIN_CHR_TYPE_PIPE: >> + /* XXX could open the pipe & just pass the FDs */ > Any reason not to let the code deal with PIPE too ? It seems > like the code should work equally well for both PIPE & FILE. > (well drop the O_CREATE|O_APPEND - assume the user has got > the pipe pre-created with 'mkfifo'. Not in particular. I was just in a hurry, wanted to share the patch sooner rather than later and then attack the pipes later on. I'll fix it up, but perhaps I won't get to it today. > I think this leaks file descriptors in libvirtd. There's no existing > code which ever closes the FDs in keepfd after spawning UML later > on in the function. True. I somehow thought virExecWhatever took care of that, but I see now that it doesn't, and I guess that would be troublesome. I'll handle this in the uml driver. -- Soren Hansen Ubuntu Developer http://www.ubuntu.com/ -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list