sch_frag.c 3.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152
  1. // SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
  2. #include <linux/if_vlan.h>
  3. #include <net/netlink.h>
  4. #include <net/sch_generic.h>
  5. #include <net/pkt_sched.h>
  6. #include <net/dst.h>
  7. #include <net/ip.h>
  8. #include <net/ip6_fib.h>
  9. struct sch_frag_data {
  10. unsigned long dst;
  11. struct qdisc_skb_cb cb;
  12. __be16 inner_protocol;
  13. u16 vlan_tci;
  14. __be16 vlan_proto;
  15. unsigned int l2_len;
  16. u8 l2_data[VLAN_ETH_HLEN];
  17. int (*xmit)(struct sk_buff *skb);
  18. };
  19. static DEFINE_PER_CPU(struct sch_frag_data, sch_frag_data_storage);
  20. static int sch_frag_xmit(struct net *net, struct sock *sk, struct sk_buff *skb)
  21. {
  22. struct sch_frag_data *data = this_cpu_ptr(&sch_frag_data_storage);
  23. if (skb_cow_head(skb, data->l2_len) < 0) {
  24. kfree_skb(skb);
  25. return -ENOMEM;
  26. }
  27. __skb_dst_copy(skb, data->dst);
  28. *qdisc_skb_cb(skb) = data->cb;
  29. skb->inner_protocol = data->inner_protocol;
  30. if (data->vlan_tci & VLAN_CFI_MASK)
  31. __vlan_hwaccel_put_tag(skb, data->vlan_proto,
  32. data->vlan_tci & ~VLAN_CFI_MASK);
  33. else
  34. __vlan_hwaccel_clear_tag(skb);
  35. /* Reconstruct the MAC header. */
  36. skb_push(skb, data->l2_len);
  37. memcpy(skb->data, &data->l2_data, data->l2_len);
  38. skb_postpush_rcsum(skb, skb->data, data->l2_len);
  39. skb_reset_mac_header(skb);
  40. return data->xmit(skb);
  41. }
  42. static void sch_frag_prepare_frag(struct sk_buff *skb,
  43. int (*xmit)(struct sk_buff *skb))
  44. {
  45. unsigned int hlen = skb_network_offset(skb);
  46. struct sch_frag_data *data;
  47. data = this_cpu_ptr(&sch_frag_data_storage);
  48. data->dst = skb->_skb_refdst;
  49. data->cb = *qdisc_skb_cb(skb);
  50. data->xmit = xmit;
  51. data->inner_protocol = skb->inner_protocol;
  52. if (skb_vlan_tag_present(skb))
  53. data->vlan_tci = skb_vlan_tag_get(skb) | VLAN_CFI_MASK;
  54. else
  55. data->vlan_tci = 0;
  56. data->vlan_proto = skb->vlan_proto;
  57. data->l2_len = hlen;
  58. memcpy(&data->l2_data, skb->data, hlen);
  59. memset(IPCB(skb), 0, sizeof(struct inet_skb_parm));
  60. skb_pull(skb, hlen);
  61. }
  62. static unsigned int
  63. sch_frag_dst_get_mtu(const struct dst_entry *dst)
  64. {
  65. return dst->dev->mtu;
  66. }
  67. static struct dst_ops sch_frag_dst_ops = {
  68. .family = AF_UNSPEC,
  69. .mtu = sch_frag_dst_get_mtu,
  70. };
  71. static int sch_fragment(struct net *net, struct sk_buff *skb,
  72. u16 mru, int (*xmit)(struct sk_buff *skb))
  73. {
  74. int ret = -1;
  75. if (skb_network_offset(skb) > VLAN_ETH_HLEN) {
  76. net_warn_ratelimited("L2 header too long to fragment\n");
  77. goto err;
  78. }
  79. if (skb_protocol(skb, true) == htons(ETH_P_IP)) {
  80. struct rtable sch_frag_rt = { 0 };
  81. unsigned long orig_dst;
  82. sch_frag_prepare_frag(skb, xmit);
  83. dst_init(&sch_frag_rt.dst, &sch_frag_dst_ops, NULL, 1,
  84. DST_OBSOLETE_NONE, DST_NOCOUNT);
  85. sch_frag_rt.dst.dev = skb->dev;
  86. orig_dst = skb->_skb_refdst;
  87. skb_dst_set_noref(skb, &sch_frag_rt.dst);
  88. IPCB(skb)->frag_max_size = mru;
  89. ret = ip_do_fragment(net, skb->sk, skb, sch_frag_xmit);
  90. refdst_drop(orig_dst);
  91. } else if (skb_protocol(skb, true) == htons(ETH_P_IPV6)) {
  92. unsigned long orig_dst;
  93. struct rt6_info sch_frag_rt;
  94. sch_frag_prepare_frag(skb, xmit);
  95. memset(&sch_frag_rt, 0, sizeof(sch_frag_rt));
  96. dst_init(&sch_frag_rt.dst, &sch_frag_dst_ops, NULL, 1,
  97. DST_OBSOLETE_NONE, DST_NOCOUNT);
  98. sch_frag_rt.dst.dev = skb->dev;
  99. orig_dst = skb->_skb_refdst;
  100. skb_dst_set_noref(skb, &sch_frag_rt.dst);
  101. IP6CB(skb)->frag_max_size = mru;
  102. ret = ipv6_stub->ipv6_fragment(net, skb->sk, skb,
  103. sch_frag_xmit);
  104. refdst_drop(orig_dst);
  105. } else {
  106. net_warn_ratelimited("Fail frag %s: eth=%x, MRU=%d, MTU=%d\n",
  107. netdev_name(skb->dev),
  108. ntohs(skb_protocol(skb, true)), mru,
  109. skb->dev->mtu);
  110. goto err;
  111. }
  112. return ret;
  113. err:
  114. kfree_skb(skb);
  115. return ret;
  116. }
  117. int sch_frag_xmit_hook(struct sk_buff *skb, int (*xmit)(struct sk_buff *skb))
  118. {
  119. u16 mru = tc_skb_cb(skb)->mru;
  120. int err;
  121. if (mru && skb->len > mru + skb->dev->hard_header_len)
  122. err = sch_fragment(dev_net(skb->dev), skb, mru, xmit);
  123. else
  124. err = xmit(skb);
  125. return err;
  126. }
  127. EXPORT_SYMBOL_GPL(sch_frag_xmit_hook);