Re: IPv4/IPv6 sysctl unregistration deadlock

[Patrick McHardy <kaber@trash.net>]

Patrick McHardy wrote:
> Stephen Hemminger wrote:
>>>> @@ -77,7 +77,9 @@ static ssize_t netdev_store(struct devic
>>>>      if (endp == buf)
>>>>          goto err;
>>>>  
>>>> -    rtnl_lock();
>>>> +    if (!rtnl_trylock())
>>>> +        return -ERESTARTSYS;
>>>> +
>>>>      if (dev_isalive(net)) {
>>>>          if ((ret = (*set)(net, new)) == 0)
>>>>              ret = len;
>>> I can test this to see if it fixes my problem.  Are the above lines the
>>> entirety of the patch?
>>
>> yes
> 
> We should be able to avoid the restart looping in most cases,
> we only need to do this while unregistration is in progress.

It doesn't seem to work. The idea was something like this:

static int addrconf_fixup_forwarding(struct ctl_table *table, int *p, 
int old)
{
	struct inet6_dev *idev;
	struct net *net;

	net = (struct net *)table->extra2;
	if (p == &net->ipv6.devconf_dflt->forwarding)
		return 0;

	/* Unregistration in progress */
	idev = table->extra1;
	if (idev->cnf.sysctl == NULL)
		return -ERESTARTSYS;

	rtnl_lock();
...

but the process might have entered this function while the rtnl
is already held, but the unregistration hasn't been triggered yet.
So it would still deadlock.