On 08/23/2018 07:27 AM, Daniel P. Berrangé wrote: > On Wed, Aug 22, 2018 at 05:43:21PM -0400, John Ferlan wrote: >> https://bugzilla.redhat.com/show_bug.cgi?id=1607202 >> >> It's stated that if the admin wants to shoot themselves in >> the foot by removing the nwfilter binding while the domain > > So based on your explanation in the other reply, this message > is what was misleading me. s/nwfilter binding/nwfilter/ > Actually perhaps it's more by "first removing the nwfilter binding and subsequently undefining the nwfilter that is/was in use for the running guest..." If just the nwfilter binding was removed, then libvirtd restart would have been fine because it would have recreated the nwfilter binding. Of course that may not be expected either... >> is running we will certainly allow that. However, in doing >> so we also run the risk that a libvirtd restart will cause >> the domain to be shutdown, which isn't a good thing. >> >> So add another boolean to virDomainConfNWFilterInstantiate >> which allows us to recover somewhat gracefully in the event >> the virNWFilterBindingCreateXML fails when we come from >> qemuProcessReconnect and we determine that the filter has >> been deleted. It was there at some point (it had to be), but >> if it's missing, then we don't want to cause the guest to >> stop running, so issue a warning and continue on. >> >> Signed-off-by: John Ferlan <jferlan@xxxxxxxxxx> >> --- >> src/conf/domain_nwfilter.c | 33 ++++++++++++++++++++++++++++----- >> src/conf/domain_nwfilter.h | 3 ++- >> src/lxc/lxc_process.c | 3 ++- >> src/qemu/qemu_hotplug.c | 7 ++++--- >> src/qemu/qemu_interface.c | 6 ++++-- >> src/qemu/qemu_process.c | 10 +++++++--- >> src/uml/uml_conf.c | 3 ++- >> 7 files changed, 49 insertions(+), 16 deletions(-) > > [snip] > >> static int >> -qemuProcessFiltersInstantiate(virDomainDefPtr def, bool ignoreExists) >> +qemuProcessFiltersInstantiate(virDomainDefPtr def, >> + bool ignoreExists, >> + bool ignoreDeleted) >> { >> size_t i; >> >> for (i = 0; i < def->nnets; i++) { >> virDomainNetDefPtr net = def->nets[i]; >> if ((net->filter) && (net->ifname)) { >> - if (virDomainConfNWFilterInstantiate(def->name, def->uuid, net, ignoreExists) < 0) >> + if (virDomainConfNWFilterInstantiate(def->name, def->uuid, net, >> + ignoreExists, >> + ignoreDeleted) < 0) >> return 1; >> } > > Rather than this extra "ignoreDeleted" arg, why can't we just do > > if (virDomainConfNWFilterInstantiate(def->name, def->uuid, net, > ignoreExists) < 0 && > ignoreDeleted) > return 1; > > This ensures that all things which can cause a nwfilter binding failure > on startup will be handled by avoiding tearing down the running guest. > Did you mean !ignoreDeleted? There's only one caller to qemuProcessFiltersInstantiate which does: if (qemuProcessFiltersInstantiate(obj->def, true)) goto error; Of course what's the purpose of distinguishing between ignoreExists and ignoreDeleted then? Essentially what you're indicating is we wouldn't care about any error because virDomainConfNWFilterInstantiate wouldn't be distinguishing between any error (because there's only one caller to qemuProcessFiltersInstantiate). I could change the last argument to virDomainConfNWFilterInstantiate to be a flag instead of a bool so that if we have future errors we care to ignore we don't keep adding bool arguments, but that's just a implementation detail. Again, I could go back and figure out a way for the *Delete path to remove the live net->filter based on the path taken so that when we get here we wouldn't find the net->filter and thus wouldn't call the instantiate function. It's just a different approach, but I'm hesitant mainly because then one wouldn't be able to use the *CreateXML API to reinstate the filter. I'm not as clear about all the (wacko) use cases. John -- libvir-list mailing list libvir-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/libvir-list
