Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx> writes:
> syzbot wrote:
>> > On Sun, Mar 4, 2018 at 6:57 AM, Tetsuo Handa
>> > <penguin-kernel@xxxxxxxxxxxxxxxxxxx> wrote:
>> >> Switching from mm to fsdevel, for this report says that put_net(net) in
>> >> rpc_kill_sb() made net->count < 0 when mount_ns() failed due to
>> >> register_shrinker() failure.
>>
>> >> Relevant commits will be
>> >> commit 9ee332d99e4d5a97 ("sget(): handle failures of
>> >> register_shrinker()") and
>> >> commit d91ee87d8d85a080 ("vfs: Pass data, ns, and ns->userns to
>> >> mount_ns.").
>>
>> >> When sget_userns() in mount_ns() failed, mount_ns() returns an error
>> >> code to
>> >> the caller without calling fill_super(). That is, get_net(sb->s_fs_info)
>> >> was
>> >> not called by rpc_fill_super() (via fill_super callback passed to
>> >> mount_ns())
>> >> but put_net(sb->s_fs_info) is called by rpc_kill_sb() (via fs->kill_sb()
>> >> from
>> >> deactivate_locked_super()).
>>
>> >> ----------
>> >> static struct dentry *
>> >> rpc_mount(struct file_system_type *fs_type,
>> >> int flags, const char *dev_name, void *data)
>> >> {
>> >> struct net *net = current->nsproxy->net_ns;
>> >> return mount_ns(fs_type, flags, data, net, net->user_ns,
>> >> rpc_fill_super);
>> >> }
>> >> ----------
>>
>> > Messed kernel output, this is definitely not in should_fail.
>>
>> > #syz dup: WARNING: refcount bug in sk_alloc
>>
>> Can't find the corresponding bug.
>>
> I don't think this is a dup of existing bug.
> We need to fix either 9ee332d99e4d5a97 or d91ee87d8d85a080.
Even if expanding mount_ns to more filesystems was magically fixed,
proc would still have this issue with the pid namespace rather than
the net namespace.
This is a mess. I will take a look and see if I can see a a fix.
Eric
