Message-ID: <b392a477-2ab5-1045-a18c-4df915f78001@huawei.com>
Date:   Thu, 23 Apr 2020 16:40:53 +0800
From:   Yuehaibing <yuehaibing@...wei.com>
To:     Xin Long <lucien.xin@...il.com>
CC:     Steffen Klassert <steffen.klassert@...unet.com>,
        Herbert Xu <herbert@...dor.apana.org.au>,
        davem <davem@...emloft.net>, <kuba@...nel.org>,
        network dev <netdev@...r.kernel.org>,
        LKML <linux-kernel@...r.kernel.org>,
        Jamal Hadi Salim <hadi@...erus.ca>
Subject: Re: [PATCH] xfrm: policy: Only use mark as policy lookup key

On 2020/4/23 14:37, Xin Long wrote:
> On Thu, Apr 23, 2020 at 10:26 AM Yuehaibing <yuehaibing@...wei.com> wrote:
>>
>> On 2020/4/22 23:41, Xin Long wrote:
>>> On Wed, Apr 22, 2020 at 8:18 PM Yuehaibing <yuehaibing@...wei.com> wrote:
>>>>
>>>> On 2020/4/22 17:33, Steffen Klassert wrote:
>>>>> On Tue, Apr 21, 2020 at 10:31:49PM +0800, YueHaibing wrote:
>>>>>> While update xfrm policy as follow:
>>>>>>
>>>>>> ip -6 xfrm policy update src fd00::1/128 dst fd00::2/128 dir in \
>>>>>>  priority 1 mark 0 mask 0x10
>>>>>> ip -6 xfrm policy update src fd00::1/128 dst fd00::2/128 dir in \
>>>>>>  priority 2 mark 0 mask 0x00
>>>>>> ip -6 xfrm policy update src fd00::1/128 dst fd00::2/128 dir in \
>>>>>>  priority 2 mark 0 mask 0x10
>>>>>>
>>>>>> We get this warning:
>>>>>>
>>>>>> WARNING: CPU: 0 PID: 4808 at net/xfrm/xfrm_policy.c:1548
>>>>>> Kernel panic - not syncing: panic_on_warn set ...
>>>>>> CPU: 0 PID: 4808 Comm: ip Not tainted 5.7.0-rc1+ #151
>>>>>> Call Trace:
>>>>>> RIP: 0010:xfrm_policy_insert_list+0x153/0x1e0
>>>>>>  xfrm_policy_inexact_insert+0x70/0x330
>>>>>>  xfrm_policy_insert+0x1df/0x250
>>>>>>  xfrm_add_policy+0xcc/0x190 [xfrm_user]
>>>>>>  xfrm_user_rcv_msg+0x1d1/0x1f0 [xfrm_user]
>>>>>>  netlink_rcv_skb+0x4c/0x120
>>>>>>  xfrm_netlink_rcv+0x32/0x40 [xfrm_user]
>>>>>>  netlink_unicast+0x1b3/0x270
>>>>>>  netlink_sendmsg+0x350/0x470
>>>>>>  sock_sendmsg+0x4f/0x60
>>>>>>
>>>>>> Policy C and policy A has the same mark.v and mark.m, so policy A is
>>>>>> matched in first round lookup while updating C. However policy C and
>>>>>> policy B has same mark and priority, which also leads to matched. So
>>>>>> the WARN_ON is triggered.
>>>>>>
>>>>>> xfrm policy lookup should only be matched when the found policy has the
>>>>>> same lookup keys (mark.v & mark.m) no matter priority.
>>>>>>
>>>>>> Fixes: 7cb8a93968e3 ("xfrm: Allow inserting policies with matching mark and different priorities")
>>>>>> Signed-off-by: YueHaibing <yuehaibing@...wei.com>
>>>>>> ---
>>>>>>  net/xfrm/xfrm_policy.c | 16 +++++-----------
>>>>>>  1 file changed, 5 insertions(+), 11 deletions(-)
>>>>>>
>>>>>> diff --git a/net/xfrm/xfrm_policy.c b/net/xfrm/xfrm_policy.c
>>>>>> index 297b2fd..67d0469 100644
>>>>>> --- a/net/xfrm/xfrm_policy.c
>>>>>> +++ b/net/xfrm/xfrm_policy.c
>>>>>> @@ -1436,13 +1436,7 @@ static void xfrm_policy_requeue(struct xfrm_policy *old,
>>>>>>  static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
>>>>>>                                 struct xfrm_policy *pol)
>>>>>>  {
>>>>>> -    u32 mark = policy->mark.v & policy->mark.m;
>>>>>> -
>>>>>> -    if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
>>>>>> -            return true;
>>>>>> -
>>>>>> -    if ((mark & pol->mark.m) == pol->mark.v &&
>>>>>> -        policy->priority == pol->priority)
>>>>>
>>>>> If you remove the priority check, you can't insert policies with matching
>>>>> mark and different priorities anymore. This brings us back the old bug.
>>>>
>>>> Yes, this is true.
>>>>
>>>>>
>>>>> I plan to apply the patch from Xin Long, this seems to be the right way
>>>>> to address this problem.
>>>>
>>>> That still brings an issue, update like this:
>>>>
>>>> policy A (mark.v = 1, mark.m = 0, priority = 1)
>>>> policy B (mark.v = 1, mark.m = 0, priority = 1)
>>>>
>>>> A and B will all in the list.
>>> I think this is another issue even before:
>>> 7cb8a93968e3 ("xfrm: Allow inserting policies with matching mark and
>>> different priorities")
>>>
>>>>
>>>> So should do this:
>>>>
>>>>  static bool xfrm_policy_mark_match(struct xfrm_policy *policy,
>>>>                                    struct xfrm_policy *pol)
>>>>  {
>>>> -       u32 mark = policy->mark.v & policy->mark.m;
>>>> -
>>>> -       if (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m)
>>>> -               return true;
>>>> -
>>>> -       if ((mark & pol->mark.m) == pol->mark.v &&
>>>> +       if ((policy->mark.v & policy->mark.m) == (pol->mark.v & pol->mark.m) &&
>>>>             policy->priority == pol->priority)
>>>>                 return true;
>>> "mark.v & mark.m" looks weird to me, it should be:
>>> ((something & mark.m) == mark.v)
>>>
>>> So why should we just do this here?:
>>> (policy->mark.v == pol->mark.v && policy->mark.m == pol->mark.m &&
>>>  policy->priority == pol->priority)
>>
>>
>> This leads to this issue:
>>
>>  ip -6 xfrm policy add src fd00::1/128 dst fd00::2/128 dir in mark 0x00000001 mask 0x00000005
>>  ip -6 xfrm policy add src fd00::1/128 dst fd00::2/128 dir in mark 0x00000001 mask 0x00000003
>>
>> the two policies will be in list, which should not be allowed.
> I think these are two different policies.
> For instance:
> mark = 0x1234567b will match the 1st one only.
> mark = 0x1234567d will match the 2st one only
> 
> So these should have been allowed, no?

If mark = 0x12345671, it may match different policy depends on the order of inserting,

ip xfrm policy update src 172.16.2.0/24 dst 172.16.1.0/24 dir in ptype main \
tmpl src 192.168.2.10 dst 192.168.1.20 proto esp mode tunnel mark 0x00000001 mask 0x00000005

ip xfrm policy update src 172.16.2.0/24 dst 172.16.1.0/24 dir in ptype main \
tmpl src 192.168.2.100 dst 192.168.1.100 proto esp mode beet mark 0x00000001 mask 0x00000003

In fact, your case should use different priority to match.

> 
> I'm actually confused now.
> does the mask work against its own value, or the other value?
> as 'A == (mark.v&mark.m)' and '(A & mark.m) == mark.v' are different things.
> 
> This can date back to Jamal's xfrm by MARK:
> 
> https://lwn.net/Articles/375829/
> 
> where it does 'm->v & m->m' in xfrm_mark_get() and
> 'policy->mark.v & policy->mark.m' in xfrm_policy_insert() while
> it does '(A & pol->mark.m) == pol->mark.v' in other places.
> 
> Now I'm thinking 'm->v & m->m' is meaningless, by which if we get
> a value != m->v, it means this mark can never be matched by any.
> 
>   policy A (mark.v = 1, mark.m = 0, priority = 1)
>   policy B (mark.v = 1, mark.m = 0, priority = 1)
> 
> So probably we should avoid this case by check m->v == (m->v & m->m)
> when adding a new policy.
> 
> wdyt?
> 

Powered by blists - more mailing lists