Re: [PATCH 9/12] ipv4: assign PDE->data before gluing PDE into /proc tree

[ebiederm@xmission.com (Eric W. Biederman)]

"Stoyan Gaydarov" <stoyboyker@gmail.com> writes:

> First off, sorry to bring such an old email back but I can seem to get
> a bad feeling when looking back over it.
>
> On Tue, Apr 29, 2008 at 6:13 AM, Denis V. Lunev <den@openvz.org> wrote:
>> The check for PDE->data != NULL becomes useless after the replacement
>> of proc_net_fops_create with proc_create_data.
>>
>> Signed-off-by: Denis V. Lunev <den@openvz.org>
>> Cc: Alexey Dobriyan <adobriyan@openvz.org>
>> Cc: Eric W. Biederman <ebiederm@xmission.com>
>> Cc: David S. Miller <davem@davemloft.net>
>> ---
>>  net/ipv4/tcp_ipv4.c |   10 +++-------
>>  net/ipv4/udp.c      |    7 +++----
>>  2 files changed, 6 insertions(+), 11 deletions(-)
>>
>> diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
>> index 7766151..4d97b28 100644
>> --- a/net/ipv4/tcp_ipv4.c
>> +++ b/net/ipv4/tcp_ipv4.c
>> @@ -2214,9 +2214,6 @@ static int tcp_seq_open(struct inode *inode, struct file
> *file)
>>        struct tcp_iter_state *s;
>>        int err;
>>
>> -       if (unlikely(afinfo == NULL))
>> -               return -EINVAL;
> I think that this check needs to stay in some form, reason below.
>> -
>>        err = seq_open_net(inode, file, &afinfo->seq_ops,
>>                          sizeof(struct tcp_iter_state));
>>        if (err < 0)
>> @@ -2241,10 +2238,9 @@ int tcp_proc_register(struct net *net, struct
> tcp_seq_afinfo *afinfo)
>>        afinfo->seq_ops.next            = tcp_seq_next;
>>        afinfo->seq_ops.stop            = tcp_seq_stop;
>>
>> - p = proc_net_fops_create(net, afinfo->name, S_IRUGO, &afinfo->seq_fops);
>> -       if (p)
>> -               p->data = afinfo;
>> -       else
>> +       p = proc_create_data(afinfo->name, S_IRUGO, net->proc_net,
>
> When you try to pass in afinfo->name (and also the seq_fops) you are
> assuming that afinfo is not null meaning in the unlikely(as shown
> above) even that it is null you get a very bad null pointer problem.
> If I am just way off do let me know because this just seams to me like
> a bad idea. This is also still present in 2.6.26-rc9.

It appears you are getting things confused.  The original window is that tcp_seq_open
(which is what get called when you open the proc file) had a small race that p->data
could be read before it was set.

With proc_create_data that race was closed.

You are saying that it is a problem for tcp_seq_open to be passed a NULL afinfo.
It is.  That has nothing to do with the original race (as that is a very
different part of the code).  Feel free to audit all of the callers if
you like.  That problem however is not subtle or racy.

So I see nothing wrong with this patch unless you can find a problem with
proc_create_data.

Eric