[libvirt] [Libguestfs] [PATCH libguestfs v3] lib: Handle slow USB devices more gracefully.

[Richard W.M. Jones]

On Tue, Jan 19, 2016 at 05:00:27PM +0000, Daniel P. Berrange wrote:
> On Tue, Jan 19, 2016 at 04:18:46PM +0000, Richard W.M. Jones wrote:
> > Libvirt has a fixed 15 second timeout for qemu to exit.  If qemu is
> > writing to a slow USB key, it can hang (in D state) for much longer
> > than this - many minutes usually.
> > 
> > The solution is to check specifically for the libvirt EBUSY error when
> > this happens, and retry the virDomainDestroyFlags operation
> > (indefinitely).
> > 
> > See also the description here:
> > https://www.redhat.com/archives/libvir-list/2016-January/msg00767.html
> > 
> > Similar to the following OpenStack Nova commit:
> > http://git.openstack.org/cgit/openstack/nova/commit/?id=3907867
> > 
> > Thanks: Kashyap Chamarthy and Daniel Berrange.
> > ---
> >  src/launch-libvirt.c | 45 +++++++++++++++++++++++++++++++++++----------
> >  1 file changed, 35 insertions(+), 10 deletions(-)
> > 
> > diff --git a/src/launch-libvirt.c b/src/launch-libvirt.c
> > index 8215e02..90b6c49 100644
> > --- a/src/launch-libvirt.c
> > +++ b/src/launch-libvirt.c
> > @@ -25,6 +25,7 @@
> >  #include <unistd.h>
> >  #include <fcntl.h>
> >  #include <grp.h>
> > +#include <errno.h>
> >  #include <sys/types.h>
> >  #include <sys/stat.h>
> >  #include <assert.h>
> > @@ -2015,6 +2016,8 @@ ignore_errors (void *ignore, virErrorPtr ignore2)
> >    /* empty */
> >  }
> >  
> > +static int destroy_domain (guestfs_h *g, virDomainPtr dom, int check_for_errors);
> > +
> >  static int
> >  shutdown_libvirt (guestfs_h *g, void *datav, int check_for_errors)
> >  {
> > @@ -2023,23 +2026,14 @@ shutdown_libvirt (guestfs_h *g, void *datav, int check_for_errors)
> >    virDomainPtr dom = data->dom;
> >    size_t i;
> >    int ret = 0;
> > -  int flags;
> >  
> >    /* Note that we can be called back very early in launch (specifically
> >     * from launch_libvirt itself), when conn and dom might be NULL.
> >     */
> > -
> >    if (dom != NULL) {
> > -    flags = check_for_errors ? VIR_DOMAIN_DESTROY_GRACEFUL : 0;
> > -    debug (g, "calling virDomainDestroy \"%s\" flags=%s",
> > -           data->name, check_for_errors ? "VIR_DOMAIN_DESTROY_GRACEFUL" : "0");
> > -    if (virDomainDestroyFlags (dom, flags) == -1) {
> > -      libvirt_error (g, _("could not destroy libvirt domain"));
> > -      ret = -1;
> > -    }
> > +    ret = destroy_domain (g, dom, check_for_errors);
> >      virDomainFree (dom);
> >    }
> > -
> >    if (conn != NULL)
> >      virConnectClose (conn);
> >  
> > @@ -2068,6 +2062,37 @@ shutdown_libvirt (guestfs_h *g, void *datav, int check_for_errors)
> >    return ret;
> >  }
> >  
> > +/* Wrapper around virDomainDestroy which handles errors and retries.. */
> > +static int
> > +destroy_domain (guestfs_h *g, virDomainPtr dom, int check_for_errors)
> > +{
> > +  const int flags = check_for_errors ? VIR_DOMAIN_DESTROY_GRACEFUL : 0;
> > +  virErrorPtr err;
> > +
> > + again:
> > +  debug (g, "calling virDomainDestroy flags=%s",
> > +         check_for_errors ? "VIR_DOMAIN_DESTROY_GRACEFUL" : "0");
> > +  if (virDomainDestroyFlags (dom, flags) == -1) {
> > +    err = virGetLastError ();
> > +
> > +    /* Second chance if we're just waiting for qemu to shut down.  See:
> > +     * https://www.redhat.com/archives/libvir-list/2016-January/msg00767.html
> > +     */
> > +    if ((flags & VIR_DOMAIN_DESTROY_GRACEFUL) &&
> > +        err && err->code == VIR_ERR_SYSTEM_ERROR && err->int1 == EBUSY)
> > +      goto again;
> 
> NB, you could get that error even if you don't specify
> VIR_DOMAIN_DESTROY_GRACEFUL, because even SIGKILL can
> fail to kill QEMU if it is in an uninterruptable sleep
> on a dead NFS server for example and hard,nointr is used
> as mount options.
> 
> So perhaps you don't want to check flags here ?

It's a bit confusing: As you say, testing `flags' is wrong, that part
was carried over from an earlier patch.

But I think I should test `check_for_errors' instead, since that flag
basically means did we come from the guestfs_shutdown method -- where
we want to be careful about data integrity -- or did we come from the
guestfs_close method -- where the user just wants to close the
connection without caring about data integrity.

> > +
> > +    /* "Domain not found" is not treated as an error. */
> > +    if (err && err->code == VIR_ERR_NO_DOMAIN)
> > +      return 0;
> > +
> > +    libvirt_error (g, _("could not destroy libvirt domain"));
> > +    return -1;
> > +  }
> > +
> > +  return 0;
> > +}
> 
> I don't know the use context from libguestfs POV, but is there any
> scenario in which a user would want a way to time out this, instead
> of making libguestfs wait forever ?

There's no good answer, but I'll see if the need arises before
making any further changes.

Thanks for the review!

Rich.

-- 
Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
Fedora Windows cross-compiler. Compile Windows programs, test, and
build Windows installers. Over 100 libraries supported.
http://fedoraproject.org/wiki/MinGW