target/linux/brcm2708/patches-4.4/0223-bpf-add-skb_postpush_rcsum-and-fix-dev_forward_skb-o.patch

   1 From 7ce01cd3013b7c9f1b257a6033b6d1f67fed5b5c Mon Sep 17 00:00:00 2001
   2 From: Daniel Borkmann <daniel@iogearbox.net>
   3 Date: Thu, 7 Jan 2016 15:50:23 +0100
   4 Subject: [PATCH 223/232] bpf: add skb_postpush_rcsum and fix dev_forward_skb
   5  occasions
   6
   7 Add a small helper skb_postpush_rcsum() and fix up redirect locations
   8 that need CHECKSUM_COMPLETE fixups on ingress. dev_forward_skb() expects
   9 a proper csum that covers also Ethernet header, f.e. since 2c26d34bbcc0
  10 ("net/core: Handle csum for CHECKSUM_COMPLETE VXLAN forwarding"), we
  11 also do skb_postpull_rcsum() after pulling Ethernet header off via
  12 eth_type_trans().
  13
  14 When using eBPF in a netns setup f.e. with vxlan in collect metadata mode,
  15 I can trigger the following csum issue with an IPv6 setup:
  16
  17   [  505.144065] dummy1: hw csum failure
  18   [...]
  19   [  505.144108] Call Trace:
  20   [  505.144112]  <IRQ>  [<ffffffff81372f08>] dump_stack+0x44/0x5c
  21   [  505.144134]  [<ffffffff81607cea>] netdev_rx_csum_fault+0x3a/0x40
  22   [  505.144142]  [<ffffffff815fee3f>] __skb_checksum_complete+0xcf/0xe0
  23   [  505.144149]  [<ffffffff816f0902>] nf_ip6_checksum+0xb2/0x120
  24   [  505.144161]  [<ffffffffa08c0e0e>] icmpv6_error+0x17e/0x328 [nf_conntrack_ipv6]
  25   [  505.144170]  [<ffffffffa0898eca>] ? ip6t_do_table+0x2fa/0x645 [ip6_tables]
  26   [  505.144177]  [<ffffffffa08c0725>] ? ipv6_get_l4proto+0x65/0xd0 [nf_conntrack_ipv6]
  27   [  505.144189]  [<ffffffffa06c9a12>] nf_conntrack_in+0xc2/0x5a0 [nf_conntrack]
  28   [  505.144196]  [<ffffffffa08c039c>] ipv6_conntrack_in+0x1c/0x20 [nf_conntrack_ipv6]
  29   [  505.144204]  [<ffffffff8164385d>] nf_iterate+0x5d/0x70
  30   [  505.144210]  [<ffffffff816438d6>] nf_hook_slow+0x66/0xc0
  31   [  505.144218]  [<ffffffff816bd302>] ipv6_rcv+0x3f2/0x4f0
  32   [  505.144225]  [<ffffffff816bca40>] ? ip6_make_skb+0x1b0/0x1b0
  33   [  505.144232]  [<ffffffff8160b77b>] __netif_receive_skb_core+0x36b/0x9a0
  34   [  505.144239]  [<ffffffff8160bdc8>] ? __netif_receive_skb+0x18/0x60
  35   [  505.144245]  [<ffffffff8160bdc8>] __netif_receive_skb+0x18/0x60
  36   [  505.144252]  [<ffffffff8160ccff>] process_backlog+0x9f/0x140
  37   [  505.144259]  [<ffffffff8160c4a5>] net_rx_action+0x145/0x320
  38   [...]
  39
  40 What happens is that on ingress, we push Ethernet header back in, either
  41 from cls_bpf or right before skb_do_redirect(), but without updating csum.
  42 The "hw csum failure" can be fixed by using the new skb_postpush_rcsum()
  43 helper for the dev_forward_skb() case to correct the csum diff again.
  44
  45 Thanks to Hannes Frederic Sowa for the csum_partial() idea!
  46
  47 Fixes: 3896d655f4d4 ("bpf: introduce bpf_clone_redirect() helper")
  48 Fixes: 27b29f63058d ("bpf: add bpf_redirect() helper")
  49 Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
  50 Acked-by: Alexei Starovoitov <ast@kernel.org>
  51 Signed-off-by: David S. Miller <davem@davemloft.net>
  52 ---
  53  include/linux/skbuff.h | 17 +++++++++++++++++
  54  net/core/filter.c      | 17 +++++++++++++----
  55  2 files changed, 30 insertions(+), 4 deletions(-)
  56
  57 --- a/include/linux/skbuff.h
  58 +++ b/include/linux/skbuff.h
  59 @@ -2737,6 +2737,23 @@ static inline void skb_postpull_rcsum(st
  60
  61  unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
  62
  63 +static inline void skb_postpush_rcsum(struct sk_buff *skb,
  64 +                                     const void *start, unsigned int len)
  65 +{
  66 +       /* For performing the reverse operation to skb_postpull_rcsum(),
  67 +        * we can instead of ...
  68 +        *
  69 +        *   skb->csum = csum_add(skb->csum, csum_partial(start, len, 0));
  70 +        *
  71 +        * ... just use this equivalent version here to save a few
  72 +        * instructions. Feeding csum of 0 in csum_partial() and later
  73 +        * on adding skb->csum is equivalent to feed skb->csum in the
  74 +        * first place.
  75 +        */
  76 +       if (skb->ip_summed == CHECKSUM_COMPLETE)
  77 +               skb->csum = csum_partial(start, len, skb->csum);
  78 +}
  79 +
  80  /**
  81   *     pskb_trim_rcsum - trim received skb and update checksum
  82   *     @skb: buffer to trim
  83 --- a/net/core/filter.c
  84 +++ b/net/core/filter.c
  85 @@ -1288,8 +1288,9 @@ static u64 bpf_skb_store_bytes(u64 r1, u
  86                 /* skb_store_bits cannot return -EFAULT here */
  87                 skb_store_bits(skb, offset, ptr, len);
  88
  89 -       if (BPF_RECOMPUTE_CSUM(flags) && skb->ip_summed == CHECKSUM_COMPLETE)
  90 -               skb->csum = csum_add(skb->csum, csum_partial(ptr, len, 0));
  91 +       if (BPF_RECOMPUTE_CSUM(flags))
  92 +               skb_postpush_rcsum(skb, ptr, len);
  93 +
  94         return 0;
  95  }
  96
  97 @@ -1415,8 +1416,12 @@ static u64 bpf_clone_redirect(u64 r1, u6
  98         if (unlikely(!skb2))
  99                 return -ENOMEM;
 100
 101 -       if (BPF_IS_REDIRECT_INGRESS(flags))
 102 +       if (BPF_IS_REDIRECT_INGRESS(flags)) {
 103 +               if (skb_at_tc_ingress(skb2))
 104 +                       skb_postpush_rcsum(skb2, skb_mac_header(skb2),
 105 +                                          skb2->mac_len);
 106                 return dev_forward_skb(dev, skb2);
 107 +       }
 108
 109         skb2->dev = dev;
 110         skb_sender_cpu_clear(skb2);
 111 @@ -1459,8 +1464,12 @@ int skb_do_redirect(struct sk_buff *skb)
 112                 return -EINVAL;
 113         }
 114
 115 -       if (BPF_IS_REDIRECT_INGRESS(ri->flags))
 116 +       if (BPF_IS_REDIRECT_INGRESS(ri->flags)) {
 117 +               if (skb_at_tc_ingress(skb))
 118 +                       skb_postpush_rcsum(skb, skb_mac_header(skb),
 119 +                                          skb->mac_len);
 120                 return dev_forward_skb(dev, skb);
 121 +       }
 122
 123         skb->dev = dev;
 124         skb_sender_cpu_clear(skb);