On Thu, Nov 11, 2021 at 08:50:33AM +0100, Johannes Berg wrote: > On Thu, 2021-11-11 at 06:43 +0000, syzbot wrote: > > > > console output: https://syzkaller.appspot.com/x/log.txt?x=15b45fb6b00000 > > So we see that fault injection is triggering a memory allocation failure > deep within the device_rename(): > > int __dev_change_net_namespace(struct net_device *dev, struct net *net, > const char *pat, int new_ifindex) > { > ... > /* Fixup kobjects */ > err = device_rename(&dev->dev, dev->name); > WARN_ON(err); > > > So we hit that WARN_ON(). > > I'm not really sure what to do about that though. Feels like we should > be able to cope with failures here, but clearly we don't, and it seems > like it would also be tricky to do after all the work already done at > this point. > > Perhaps device_rename() could grow an API to preallocate all the > memories, but that would also be fairly involved, I imagine? That would be a mess to unwind at times. For fault-injection stuff like this, that can not be hit in "real world operation", if the issue can not be easily handled, I don't think it is worth worrying about. We have some things like this in the tty layer at boot time, if a memory failure happens then, we have bigger overall problems in the system than trying to recover from minor stuff like this. thanks, greg k-h
