[PATCH net] ipv6: Fix dangling pointer when ipv6 fragment

[PATCH net] ipv6: Fix dangling pointer when ipv6 fragment

[hujunwei]

From: Junwei Hu <hujunwei4@huawei.com>

At the beginning of ip6_fragment func, the prevhdr pointer is
obtained in the ip6_find_1stfragopt func.
However, all the pointers pointing into skb header may change
when calling skb_checksum_help func with
skb->ip_summed = CHECKSUM_PARTIAL condition.
The prevhdr pointe will be dangling if it is not reloaded after
calling __skb_linearize func in skb_checksum_help func.

Here, I add a variable, nexthdr_offset, to evaluate the offset,
which does not changes even after calling __skb_linearize func.

Fixes: 405c92f7a541 ("ipv6: add defensive check for CHECKSUM_PARTIAL skbs in ip_fragment")
Signed-off-by: Junwei Hu <hujunwei4@huawei.com>
Reported-by: Wenhao Zhang <zhangwenhao8@huawei.com>
Reviewed-by: Zhiqiang Liu <liuzhiqiang26@huawei.com>

---
 net/ipv6/ip6_output.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
index edbd12067170..6db3c60b3b66 100644
--- a/net/ipv6/ip6_output.c
+++ b/net/ipv6/ip6_output.c
@@ -606,12 +606,14 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
     __be32 frag_id;
     int ptr, offset = 0, err = 0;
     u8 *prevhdr, nexthdr = 0;
+    u8 nexthdr_offset;
 
     err = ip6_find_1stfragopt(skb, &prevhdr);
     if (err < 0)
         goto fail;
     hlen = err;
     nexthdr = *prevhdr;
+    nexthdr_offset = prevhdr - skb_network_header(skb);
 
     mtu = ip6_skb_dst_mtu(skb);
 
@@ -646,6 +648,8 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
         (err = skb_checksum_help(skb)))
         goto fail;
 
+    prevhdr = skb_network_header(skb) + nexthdr_offset;
+
     hroom = LL_RESERVED_SPACE(rt->dst.dev);
     if (skb_has_frag_list(skb)) {
         unsigned int first_len = skb_pagelen(skb);
-- 
2.19.1

Re: [PATCH net] ipv6: Fix dangling pointer when ipv6 fragment

[Eric Dumazet]

On 03/30/2019 12:29 AM, hujunwei wrote:
> From: Junwei Hu <hujunwei4@huawei.com>
> 
> At the beginning of ip6_fragment func, the prevhdr pointer is
> obtained in the ip6_find_1stfragopt func.
> However, all the pointers pointing into skb header may change
> when calling skb_checksum_help func with
> skb->ip_summed = CHECKSUM_PARTIAL condition.
> The prevhdr pointe will be dangling if it is not reloaded after
> calling __skb_linearize func in skb_checksum_help func.
> 
> Here, I add a variable, nexthdr_offset, to evaluate the offset,
> which does not changes even after calling __skb_linearize func.
> 
> Fixes: 405c92f7a541 ("ipv6: add defensive check for CHECKSUM_PARTIAL skbs in ip_fragment")
> Signed-off-by: Junwei Hu <hujunwei4@huawei.com>
> Reported-by: Wenhao Zhang <zhangwenhao8@huawei.com>
> Reviewed-by: Zhiqiang Liu <liuzhiqiang26@huawei.com>

Interesting.

We got a syzbot report yesterday about this issue. (email thread : BUG: unable to handle kernel paging request in ip6_fragment)

<quote>
syzbot found the following crash on:

HEAD commit:    8c838f53 dpaa2-eth: fix race condition with bql frame acco..
git tree:       net
console output: https://syzkaller.appspot.com/x/log.txt?x=12b83a9b200000
kernel config:  https://syzkaller.appspot.com/x/.config?x=f05902bca21d8935
dashboard link: https://syzkaller.appspot.com/bug?extid=e8ce541d095e486074fc
compiler:       gcc (GCC) 9.0.0 20181231 (experimental)

Unfortunately, I don't have any reproducer for this crash yet.

IMPORTANT: if you fix the bug, please add the following tag to the commit:
Reported-by: syzbot+e8ce541d095e486074fc@syzkaller.appspotmail.com
</quote>


> 
> ---
>  net/ipv6/ip6_output.c | 4 ++++
>  1 file changed, 4 insertions(+)
> 
> diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
> index edbd12067170..6db3c60b3b66 100644
> --- a/net/ipv6/ip6_output.c
> +++ b/net/ipv6/ip6_output.c
> @@ -606,12 +606,14 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
>      __be32 frag_id;
>      int ptr, offset = 0, err = 0;
>      u8 *prevhdr, nexthdr = 0;
> +    u8 nexthdr_offset;
>  
>      err = ip6_find_1stfragopt(skb, &prevhdr);
>      if (err < 0)
>          goto fail;
>      hlen = err;
>      nexthdr = *prevhdr;
> +    nexthdr_offset = prevhdr - skb_network_header(skb);
>  
>      mtu = ip6_skb_dst_mtu(skb);
>  
> @@ -646,6 +648,8 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
>          (err = skb_checksum_help(skb)))
>          goto fail;
>  
> +    prevhdr = skb_network_header(skb) + nexthdr_offset;
> +
>      hroom = LL_RESERVED_SPACE(rt->dst.dev);
>      if (skb_has_frag_list(skb)) {
>          unsigned int first_len = skb_pagelen(skb);
> 

Re: [PATCH net] ipv6: Fix dangling pointer when ipv6 fragment

[Eric Dumazet]

On 03/30/2019 12:48 AM, Eric Dumazet wrote:
> 
> 
> On 03/30/2019 12:29 AM, hujunwei wrote:
>> From: Junwei Hu <hujunwei4@huawei.com>
>>
>> At the beginning of ip6_fragment func, the prevhdr pointer is
>> obtained in the ip6_find_1stfragopt func.
>> However, all the pointers pointing into skb header may change
>> when calling skb_checksum_help func with
>> skb->ip_summed = CHECKSUM_PARTIAL condition.
>> The prevhdr pointe will be dangling if it is not reloaded after
>> calling __skb_linearize func in skb_checksum_help func.
>>
>> Here, I add a variable, nexthdr_offset, to evaluate the offset,
>> which does not changes even after calling __skb_linearize func.
>>

...

>> diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
>> index edbd12067170..6db3c60b3b66 100644
>> --- a/net/ipv6/ip6_output.c
>> +++ b/net/ipv6/ip6_output.c
>> @@ -606,12 +606,14 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
>>      __be32 frag_id;
>>      int ptr, offset = 0, err = 0;
>>      u8 *prevhdr, nexthdr = 0;
>> +    u8 nexthdr_offset;

Why u8 here ?

I would use "unsigned int" really.

>>  
>>      err = ip6_find_1stfragopt(skb, &prevhdr);
>>      if (err < 0)
>>          goto fail;
>>      hlen = err;
>>      nexthdr = *prevhdr;
>> +    nexthdr_offset = prevhdr - skb_network_header(skb);
>>  
>>      mtu = ip6_skb_dst_mtu(skb);
>>  
>> @@ -646,6 +648,8 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
>>          (err = skb_checksum_help(skb)))
>>          goto fail;
>>  
>> +    prevhdr = skb_network_header(skb) + nexthdr_offset;
>> +
>>      hroom = LL_RESERVED_SPACE(rt->dst.dev);
>>      if (skb_has_frag_list(skb)) {
>>          unsigned int first_len = skb_pagelen(skb);
>>

Re: [PATCH net] ipv6: Fix dangling pointer when ipv6 fragment

[hujunwei]

Hi Eri,

Thanks for your suggestion, u8 may not enough when the packet have a lot of exthdr.

I will update the patch in v2, by the way update Report-by tag.


On 2019/3/30 15:57, Eric Dumazet wrote:
>
> On 03/30/2019 12:48 AM, Eric Dumazet wrote:
>>
>> On 03/30/2019 12:29 AM, hujunwei wrote:
>>> From: Junwei Hu <hujunwei4@huawei.com>
>>>
>>> At the beginning of ip6_fragment func, the prevhdr pointer is
>>> obtained in the ip6_find_1stfragopt func.
>>> However, all the pointers pointing into skb header may change
>>> when calling skb_checksum_help func with
>>> skb->ip_summed = CHECKSUM_PARTIAL condition.
>>> The prevhdr pointe will be dangling if it is not reloaded after
>>> calling __skb_linearize func in skb_checksum_help func.
>>>
>>> Here, I add a variable, nexthdr_offset, to evaluate the offset,
>>> which does not changes even after calling __skb_linearize func.
>>>
> ...
>
>>> diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
>>> index edbd12067170..6db3c60b3b66 100644
>>> --- a/net/ipv6/ip6_output.c
>>> +++ b/net/ipv6/ip6_output.c
>>> @@ -606,12 +606,14 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
>>>      __be32 frag_id;
>>>      int ptr, offset = 0, err = 0;
>>>      u8 *prevhdr, nexthdr = 0;
>>> +    u8 nexthdr_offset;
> Why u8 here ?
>
> I would use "unsigned int" really.
>
>>>  
>>>      err = ip6_find_1stfragopt(skb, &prevhdr);
>>>      if (err < 0)
>>>          goto fail;
>>>      hlen = err;
>>>      nexthdr = *prevhdr;
>>> +    nexthdr_offset = prevhdr - skb_network_header(skb);
>>>  
>>>      mtu = ip6_skb_dst_mtu(skb);
>>>  
>>> @@ -646,6 +648,8 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
>>>          (err = skb_checksum_help(skb)))
>>>          goto fail;
>>>  
>>> +    prevhdr = skb_network_header(skb) + nexthdr_offset;
>>> +
>>>      hroom = LL_RESERVED_SPACE(rt->dst.dev);
>>>      if (skb_has_frag_list(skb)) {
>>>          unsigned int first_len = skb_pagelen(skb);
>>>
> .
>

[PATCH v2 net] ipv6: Fix dangling pointer when ipv6 fragment

[hujunwei]

From: Junwei Hu <hujunwei4@huawei.com>

At the beginning of ip6_fragment func, the prevhdr pointer is
obtained in the ip6_find_1stfragopt func.
However, all the pointers pointing into skb header may change
when calling skb_checksum_help func with
skb->ip_summed = CHECKSUM_PARTIAL condition.
The prevhdr pointe will be dangling if it is not reloaded after
calling __skb_linearize func in skb_checksum_help func.

Here, I add a variable, nexthdr_offset, to evaluate the offset,
which does not changes even after calling __skb_linearize func.

Fixes: 405c92f7a541 ("ipv6: add defensive check for CHECKSUM_PARTIAL skbs in ip_fragment")
Signed-off-by: Junwei Hu <hujunwei4@huawei.com>
Reported-by: Wenhao Zhang <zhangwenhao8@huawei.com>
Reported-by: syzbot+e8ce541d095e486074fc@syzkaller.appspotmail.com
Reviewed-by: Zhiqiang Liu <liuzhiqiang26@huawei.com>
---
V1->V2:
- change nexthdr_offset to unsigned int
- add Reported-by: syzbot+e8ce541d095e486074fc@syzkaller.appspotmail.com

 net/ipv6/ip6_output.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
index edbd12067170..e51f3c648b09 100644
--- a/net/ipv6/ip6_output.c
+++ b/net/ipv6/ip6_output.c
@@ -601,7 +601,7 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
                 inet6_sk(skb->sk) : NULL;
     struct ipv6hdr *tmp_hdr;
     struct frag_hdr *fh;
-    unsigned int mtu, hlen, left, len;
+    unsigned int mtu, hlen, left, len, nexthdr_offset;
     int hroom, troom;
     __be32 frag_id;
     int ptr, offset = 0, err = 0;
@@ -612,6 +612,7 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
         goto fail;
     hlen = err;
     nexthdr = *prevhdr;
+    nexthdr_offset = prevhdr - skb_network_header(skb);
 
     mtu = ip6_skb_dst_mtu(skb);
 
@@ -646,6 +647,7 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
         (err = skb_checksum_help(skb)))
         goto fail;
 
+    prevhdr = skb_network_header(skb) + nexthdr_offset;
     hroom = LL_RESERVED_SPACE(rt->dst.dev);
     if (skb_has_frag_list(skb)) {
         unsigned int first_len = skb_pagelen(skb);
-- 
2.19.1

Re: [PATCH v2 net] ipv6: Fix dangling pointer when ipv6 fragment

[Martin Lau]

On Sun, Mar 31, 2019 at 05:04:29PM +0800, hujunwei wrote:
> From: Junwei Hu <hujunwei4@huawei.com>
> 
> At the beginning of ip6_fragment func, the prevhdr pointer is
> obtained in the ip6_find_1stfragopt func.
> However, all the pointers pointing into skb header may change
> when calling skb_checksum_help func with
> skb->ip_summed = CHECKSUM_PARTIAL condition.
> The prevhdr pointe will be dangling if it is not reloaded after
> calling __skb_linearize func in skb_checksum_help func.
> 
> Here, I add a variable, nexthdr_offset, to evaluate the offset,
> which does not changes even after calling __skb_linearize func.
Acked-by: Martin KaFai Lau <kafai@fb.com>

Re: [PATCH v2 net] ipv6: Fix dangling pointer when ipv6 fragment

[David Miller]

From: hujunwei <hujunwei4@huawei.com>
Date: Sun, 31 Mar 2019 17:04:29 +0800

> diff --git a/net/ipv6/ip6_output.c b/net/ipv6/ip6_output.c
> index edbd12067170..e51f3c648b09 100644
> --- a/net/ipv6/ip6_output.c
> +++ b/net/ipv6/ip6_output.c
> @@ -601,7 +601,7 @@ int ip6_fragment(struct net *net, struct sock *sk, struct sk_buff *skb,
>                  inet6_sk(skb->sk) : NULL;
>      struct ipv6hdr *tmp_hdr;
>      struct frag_hdr *fh;
> -    unsigned int mtu, hlen, left, len;

Your patch was corrupted by your email client.

Please fix this, email a test patch to yourself, and only when you can
successfully apply that test patch should you repost this to the mailing
list.

Thank you.