lists.openwall.net   lists  /  announce  owl-users  owl-dev  john-users  john-dev  passwdqc-users  yescrypt  popa3d-users  /  oss-security  kernel-hardening  musl  sabotage  tlsify  passwords  /  crypt-dev  xvendor  /  Bugtraq  Full-Disclosure  linux-kernel  linux-netdev  linux-ext4  PHC 
Open Source and information security mailing list archives
 
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date:   Tue, 30 Jun 2020 09:16:32 -0700
From:   Eric Dumazet <eric.dumazet@...il.com>
To:     Willem de Bruijn <willemdebruijn.kernel@...il.com>,
        Tom Herbert <tom@...bertland.com>
Cc:     Eric Dumazet <eric.dumazet@...il.com>,
        Network Development <netdev@...r.kernel.org>,
        David Miller <davem@...emloft.net>
Subject: Re: [PATCH net-next] icmp: support rfc 4884



On 6/30/20 6:57 AM, Willem de Bruijn wrote:
> On Mon, Jun 29, 2020 at 10:19 PM Willem de Bruijn
> <willemdebruijn.kernel@...il.com> wrote:
>>
>> On Mon, Jun 29, 2020 at 8:37 PM Tom Herbert <tom@...bertland.com> wrote:
>>>
>>> On Mon, Jun 29, 2020 at 4:07 PM Eric Dumazet <eric.dumazet@...il.com> wrote:
>>>>
>>>>
>>>>
>>>> On 6/29/20 2:30 PM, Willem de Bruijn wrote:
>>>>> On Mon, Jun 29, 2020 at 5:15 PM Eric Dumazet <eric.dumazet@...il.com> wrote:
>>>>>>
>>>>>>
>>>>>>
>>>>>> On 6/29/20 9:57 AM, Willem de Bruijn wrote:
>>>>>>> From: Willem de Bruijn <willemb@...gle.com>
>>>>>>>
>>>>>>> ICMP messages may include an extension structure after the original
>>>>>>> datagram. RFC 4884 standardized this behavior.
>>>>>>>
>>>>>>> It introduces an explicit original datagram length field in the ICMP
>>>>>>> header to delineate the original datagram from the extension struct.
>>>>>>>
>>>>>>> Return this field when reading an ICMP error from the error queue.
>>>>>>
>>>>>> RFC mentions a 'length' field of 8 bits, your patch chose to export the whole
>>>>>> second word of icmp header.
>>>>>>
>>>>>> Why is this field mapped to a prior one (icmp_hdr(skb)->un.gateway) ?
>>>>>>
>>>>>> Should we add an element in the union to make this a little bit more explicit/readable ?
>>>>>>
>>>>>> diff --git a/include/uapi/linux/icmp.h b/include/uapi/linux/icmp.h
>>>>>> index 5589eeb791ca580bb182e1dc38c05eab1c75adb9..427ed5a6765316a4c1e2fa06f3b6618447c01564 100644
>>>>>> --- a/include/uapi/linux/icmp.h
>>>>>> +++ b/include/uapi/linux/icmp.h
>>>>>> @@ -76,6 +76,7 @@ struct icmphdr {
>>>>>>                 __be16  sequence;
>>>>>>         } echo;
>>>>>>         __be32  gateway;
>>>>>> +       __be32  second_word; /* RFC 4884 4.[123] : <unused:8>,<length:8>,<mtu:16> */
>>>>>>         struct {
>>>>>>                 __be16  __unused;
>>>>>>                 __be16  mtu;
>>>>>
>>>>> Okay. How about a variant of the existing struct frag?
>>>>>
>>>>> @@ -80,6 +80,11 @@ struct icmphdr {
>>>>>                 __be16  __unused;
>>>>>                 __be16  mtu;
>>>>>         } frag;
>>>>> +       struct {
>>>>> +               __u8    __unused;
>>>>> +               __u8    length;
>>>>> +               __be16  mtu;
>>>>> +       } rfc_4884;
>>>>>         __u8    reserved[4];
>>>>>    } un;
>>>>>
>>>>
>>>> Sure, but my point was later in the code :
>>>>
>>>>>>> +     if (inet_sk(sk)->recverr_rfc4884)
>>>>>>> +             info = ntohl(icmp_hdr(skb)->un.gateway);
>>>>>>
>>>>>> ntohl(icmp_hdr(skb)->un.second_word);
>>>>
>>>> If you leave there "info = ntohl(icmp_hdr(skb)->un.gateway)" it is a bit hard for someone
>>>> reading linux kernel code to understand why we do this.
>>>>
>>> It's also potentially problematic. The other bits are Unused, which
>>> isn't the same thing as necessarily being zero. Userspace might assume
>>> that info is the length without checking its bounded.
>>
>> It shouldn't. The icmp type and code are passed in sock_extended_err
>> as ee_type and ee_code. So it can demultiplex the meaning of the rest
>> of the icmp header.
>>
>> It just needs access to the other 32-bits, which indeed are context
>> sensitive. It makes more sense to me to let userspace demultiplex this
>> in one place, rather than demultiplex in the kernel and define a new,
>> likely no simpler, data structure to share with userspace.
>>
>> Specific to RFC 4884, the 8-bit length field coexists with the
>> 16-bit mtu field in case of ICMP_FRAG_NEEDED, so we cannot just pass
>> the first as ee_info in RFC 4884 mode. sock_extended_err additionally
>> has ee_data, but after that we're out of fields, too, so this approach
>> is not very future proof to additional ICMP extensions.
>>
>> On your previous point, it might be useful to define struct rfc_4884
>> equivalent outside struct icmphdr, so that an application can easily
>> cast to that. RFC 4884 itself does not define any extension objects.
>> That is out of scope there, and in my opinion, here. Again, better
>> left to userspace. Especially because as it describes, it standardized
>> the behavior after observing non-compliant, but existing in the wild,
>> proprietary extension variants. Users may have to change how they
>> interpret the fields based on what they have deployed.
> 
> As this just shares the raw icmp header data, I should probably
> change the name to something less specific to RFC 4884.
> 
> Since it would also help with decoding other extensions, such as
> the one you mention in  draft-ietf-6man-icmp-limits-08.
> 
> Unfortunately I cannot simply reserve IP_RECVERR with integer 2.
> Perhaps IP_RECVERR_EXINFO.
> 

Perhaps let the icmp header as is, but provides the extra information
as an explicit ancillary message in ip_recv_error() ?

Really this is all about documentation and providing stable API.

Possible alternative would be to add an union over ee_pad

Legacy applications would always get 0 for ee_pad/ee_length, while
applications enabling IP_RECVERR_RFC4884 would access the wire value.


diff --git a/include/uapi/linux/errqueue.h b/include/uapi/linux/errqueue.h
index ca5cb3e3c6df745aa4c886ba7b4f88179fa22d86..509a5a6ccc555705ef867d7580553d289d559786 100644
--- a/include/uapi/linux/errqueue.h
+++ b/include/uapi/linux/errqueue.h
@@ -10,7 +10,10 @@ struct sock_extended_err {
        __u8    ee_origin;
        __u8    ee_type;
        __u8    ee_code;
-       __u8    ee_pad;
+       union {
+               __u8    ee_pad;
+               __u8    ee_length; /* RFC 4884 (see IP_RECVERR_RFC4884) */
+       };
        __u32   ee_info;
        __u32   ee_data;
 };

Powered by blists - more mailing lists