On Mon, Apr 02, 2018 at 07:44:37AM +0100, Al Viro wrote: > On Sun, Apr 01, 2018 at 02:48:54PM -0700, Eric Biggers wrote: > > [+Cc linux-nfs] > > > > > > [ 42.965515] net/sunrpc/rpc_pipe.c: __rpc_create_common failed to allocate inode for dentry blocklayout > > > [ 42.967234] net/sunrpc/rpc_pipe.c: rpc_mkpipe_dentry() failed to create pipe nfs/blocklayout (errno = -12) > > AFAICS, there's nothing to zero nn->bl_device_pipe->dentry after > nfs4blocklayout_unregister_sb(), is there? If nothing else, what's > going to happen after mount/umount/mount with failing > nfs4blocklayout_register_sb()? AFAICS, we'll have stale pointer to > dentry sitting in nn->bl_device_pipe->dentry, and call rpc_unlink() > on it while cleaning up after the failing mount. > > I don't think that's all there is to it, but it does smell like > a bug. That's not all. Making nfs4blocklayout_register_sb() immediately fail (without doing anything) leads to that oops on the very first attempt to mount rpc_pipefs. Matter of fact, rpc_gssd_dummy_depopulate() is garbage. I don't know how it had been tested, but it will do an extra dput() of gssd_dentry whenever it's called. _Any_ failure that sends us to err_depopulate: (== any failure in rpc_pipefs_notifier_list callbacks) means an oops there on attempt to dput() an already freed dentry. Hell, turn that if (err) goto err_depopulate; into if (err || !strcmp(current->comm, "bugger")) goto err_depopulate; cp /bin/mount /root/bugger, then boot with init=/bin/sh and cd /root; ./bugger -t rpc_pipefs none /mnt will trigger just that. -- To unsubscribe from this list: send the line "unsubscribe linux-nfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
