Re: [PATCH v4 net] net: udp: fix Fast/frag0 UDP GRO

[Alexander Lobakin]

From: Willem de Bruijn <willemdebruijn.kernel@xxxxxxxxx>
Date: Tue, 10 Nov 2020 13:49:56 -0500

> On Mon, Nov 9, 2020 at 7:29 PM Alexander Lobakin <alobakin@xxxxx> wrote:
>>
>> From: Alexander Lobakin <alobakin@xxxxx>
>> Date: Tue, 10 Nov 2020 00:17:18 +0000
>>
>>> While testing UDP GSO fraglists forwarding through driver that uses
>>> Fast GRO (via napi_gro_frags()), I was observing lots of out-of-order
>>> iperf packets:
>>>
>>> [ ID] Interval           Transfer     Bitrate         Jitter
>>> [SUM]  0.0-40.0 sec  12106 datagrams received out-of-order
>>>
>>> Simple switch to napi_gro_receive() or any other method without frag0
>>> shortcut completely resolved them.
>>>
>>> I've found that UDP GRO uses udp_hdr(skb) in its .gro_receive()
>>> callback. While it's probably OK for non-frag0 paths (when all
>>> headers or even the entire frame are already in skb->data), this
>>> inline points to junk when using Fast GRO (napi_gro_frags() or
>>> napi_gro_receive() with only Ethernet header in skb->data and all
>>> the rest in shinfo->frags) and breaks GRO packet compilation and
>>> the packet flow itself.
>>> To support both modes, skb_gro_header_fast() + skb_gro_header_slow()
>>> are typically used. UDP even has an inline helper that makes use of
>>> them, udp_gro_udphdr(). Use that instead of troublemaking udp_hdr()
>>> to get rid of the out-of-order delivers.
>>>
>>> Present since the introduction of plain UDP GRO in 5.0-rc1.
>>>
>>> Since v3 [1]:
>>>  - restore the original {,__}udp{4,6}_lib_lookup_skb() and use
>>>    private versions of them inside GRO code (Willem).
>>
>> Note: this doesn't cover a support for nested tunnels as it's out of
>> the subject and requires more invasive changes. It will be handled
>> separately in net-next series.
>
> Thanks for looking into that.

Thank you (and Eric) for all your comments and reviews :)

> In that case, should the p->data + off change be deferred to that,
> too? It adds some risk unrelated to the bug fix.

Change to p->data + off is absolutely safe and even can prevent from
any other potentional problems with Fast/frag0 GRO of UDP fraglists.
I find them pretty fragile currently.

>>> Since v2 [2]:
>>>  - dropped redundant check introduced in v2 as it's performed right
>>>    before (thanks to Eric);
>>>  - udp_hdr() switched to data + off for skbs from list (also Eric);
>>>  - fixed possible malfunction of {,__}udp{4,6}_lib_lookup_skb() with
>>>    Fast/frag0 due to ip{,v6}_hdr() usage (Willem).
>>>
>>> Since v1 [3]:
>>>  - added a NULL pointer check for "uh" as suggested by Willem.
>>>
>>> [1] https://lore.kernel.org/netdev/MgZce9htmEtCtHg7pmWxXXfdhmQ6AHrnltXC41zOoo@xxxxxxxxxxxxxxxxxxxx
>>> [2] https://lore.kernel.org/netdev/0eaG8xtbtKY1dEKCTKUBubGiC9QawGgB3tVZtNqVdY@xxxxxxxxxxxxxxxxxxxx
>>> [3] https://lore.kernel.org/netdev/YazU6GEzBdpyZMDMwJirxDX7B4sualpDG68ADZYvJI@xxxxxxxxxxxxxxxxxxxx
>>>
>>> Fixes: e20cf8d3f1f7 ("udp: implement GRO for plain UDP sockets.")
>>> Cc: Eric Dumazet <edumazet@xxxxxxxxxx>
>>> Cc: Willem de Bruijn <willemb@xxxxxxxxxx>
>>> Signed-off-by: Alexander Lobakin <alobakin@xxxxx>
>>> ---
>>>  net/ipv4/udp_offload.c | 23 +++++++++++++++++++----
>>>  net/ipv6/udp_offload.c | 14 +++++++++++++-
>>>  2 files changed, 32 insertions(+), 5 deletions(-)
>>>
>>> diff --git a/net/ipv4/udp_offload.c b/net/ipv4/udp_offload.c
>>> index e67a66fbf27b..6064efe17cdb 100644
>>> --- a/net/ipv4/udp_offload.c
>>> +++ b/net/ipv4/udp_offload.c
>>> @@ -366,11 +366,11 @@ static struct sk_buff *udp4_ufo_fragment(struct sk_buff *skb,
>>>  static struct sk_buff *udp_gro_receive_segment(struct list_head *head,
>>>                                              struct sk_buff *skb)
>>>  {
>>> -     struct udphdr *uh = udp_hdr(skb);
>>> +     struct udphdr *uh = udp_gro_udphdr(skb);
>>>       struct sk_buff *pp = NULL;
>>>       struct udphdr *uh2;
>>>       struct sk_buff *p;
>>> -     unsigned int ulen;
>>> +     u32 ulen, off;
>
> a specific reason for changing type here?

Yep. unsigned int == u32, I had to add another variable, and the
easiest way to do this without breaking reverse christmas tree or
adding new lines was this.
Pure cosmetics, I can change this if somebody doesn't like that one.

>>>       int ret = 0;
>>>
>>>       /* requires non zero csum, for symmetry with GSO */
>>> @@ -385,6 +385,9 @@ static struct sk_buff *udp_gro_receive_segment(struct list_head *head,
>>>               NAPI_GRO_CB(skb)->flush = 1;
>>>               return NULL;
>>>       }
>>> +
>>> +     off = skb_gro_offset(skb);
>>> +
>>>       /* pull encapsulating udp header */
>>>       skb_gro_pull(skb, sizeof(struct udphdr));
>>>
>>> @@ -392,7 +395,7 @@ static struct sk_buff *udp_gro_receive_segment(struct list_head *head,
>>>               if (!NAPI_GRO_CB(p)->same_flow)
>>>                       continue;
>>>
>>> -             uh2 = udp_hdr(p);
>>> +             uh2 = (void *)p->data + off;
>>>
>>>               /* Match ports only, as csum is always non zero */
>>>               if ((*(u32 *)&uh->source != *(u32 *)&uh2->source)) {
>>> @@ -500,6 +503,16 @@ struct sk_buff *udp_gro_receive(struct list_head *head, struct sk_buff *skb,
>>>  }
>>>  EXPORT_SYMBOL(udp_gro_receive);
>>>
>>> +static struct sock *udp4_gro_lookup_skb(struct sk_buff *skb, __be16 sport,
>>> +                                     __be16 dport)
>>> +{
>>> +     const struct iphdr *iph = skb_gro_network_header(skb);
>>> +
>>> +     return __udp4_lib_lookup(dev_net(skb->dev), iph->saddr, sport,
>>> +                              iph->daddr, dport, inet_iif(skb),
>>> +                              inet_sdif(skb), &udp_table, NULL);
>>> +}
>>> +
>>>  INDIRECT_CALLABLE_SCOPE
>>>  struct sk_buff *udp4_gro_receive(struct list_head *head, struct sk_buff *skb)
>>>  {
>>> @@ -523,7 +536,9 @@ struct sk_buff *udp4_gro_receive(struct list_head *head, struct sk_buff *skb)
>>>  skip:
>>>       NAPI_GRO_CB(skb)->is_ipv6 = 0;
>>>       rcu_read_lock();
>>> -     sk = static_branch_unlikely(&udp_encap_needed_key) ? udp4_lib_lookup_skb(skb, uh->source, uh->dest) : NULL;
>>> +     sk = static_branch_unlikely(&udp_encap_needed_key) ?
>>> +          udp4_gro_lookup_skb(skb, uh->source, uh->dest) :
>>> +          NULL;
>
> Does this indentation pass checkpatch?

Sure, I always check my changes with checkpatch --strict.

> Else, the line limit is no longer strict,a and this only shortens the
> line, so a single line is fine.

These single lines is about 120 chars, don't find them eye-pleasant.
But, as with "u32" above, they're pure cosmetics and can be changed.

Thanks,
Al