xfrm_device.rst 5.3 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151
  1. .. SPDX-License-Identifier: GPL-2.0
  2. ===============================================
  3. XFRM device - offloading the IPsec computations
  4. ===============================================
  5. Shannon Nelson <shannon.nelson@oracle.com>
  6. Overview
  7. ========
  8. IPsec is a useful feature for securing network traffic, but the
  9. computational cost is high: a 10Gbps link can easily be brought down
  10. to under 1Gbps, depending on the traffic and link configuration.
  11. Luckily, there are NICs that offer a hardware based IPsec offload which
  12. can radically increase throughput and decrease CPU utilization. The XFRM
  13. Device interface allows NIC drivers to offer to the stack access to the
  14. hardware offload.
  15. Userland access to the offload is typically through a system such as
  16. libreswan or KAME/raccoon, but the iproute2 'ip xfrm' command set can
  17. be handy when experimenting. An example command might look something
  18. like this::
  19. ip x s add proto esp dst 14.0.0.70 src 14.0.0.52 spi 0x07 mode transport \
  20. reqid 0x07 replay-window 32 \
  21. aead 'rfc4106(gcm(aes))' 0x44434241343332312423222114131211f4f3f2f1 128 \
  22. sel src 14.0.0.52/24 dst 14.0.0.70/24 proto tcp \
  23. offload dev eth4 dir in
  24. Yes, that's ugly, but that's what shell scripts and/or libreswan are for.
  25. Callbacks to implement
  26. ======================
  27. ::
  28. /* from include/linux/netdevice.h */
  29. struct xfrmdev_ops {
  30. int (*xdo_dev_state_add) (struct xfrm_state *x);
  31. void (*xdo_dev_state_delete) (struct xfrm_state *x);
  32. void (*xdo_dev_state_free) (struct xfrm_state *x);
  33. bool (*xdo_dev_offload_ok) (struct sk_buff *skb,
  34. struct xfrm_state *x);
  35. void (*xdo_dev_state_advance_esn) (struct xfrm_state *x);
  36. };
  37. The NIC driver offering ipsec offload will need to implement these
  38. callbacks to make the offload available to the network stack's
  39. XFRM subsytem. Additionally, the feature bits NETIF_F_HW_ESP and
  40. NETIF_F_HW_ESP_TX_CSUM will signal the availability of the offload.
  41. Flow
  42. ====
  43. At probe time and before the call to register_netdev(), the driver should
  44. set up local data structures and XFRM callbacks, and set the feature bits.
  45. The XFRM code's listener will finish the setup on NETDEV_REGISTER.
  46. ::
  47. adapter->netdev->xfrmdev_ops = &ixgbe_xfrmdev_ops;
  48. adapter->netdev->features |= NETIF_F_HW_ESP;
  49. adapter->netdev->hw_enc_features |= NETIF_F_HW_ESP;
  50. When new SAs are set up with a request for "offload" feature, the
  51. driver's xdo_dev_state_add() will be given the new SA to be offloaded
  52. and an indication of whether it is for Rx or Tx. The driver should
  53. - verify the algorithm is supported for offloads
  54. - store the SA information (key, salt, target-ip, protocol, etc)
  55. - enable the HW offload of the SA
  56. - return status value:
  57. =========== ===================================
  58. 0 success
  59. -EOPNETSUPP offload not supported, try SW IPsec
  60. other fail the request
  61. =========== ===================================
  62. The driver can also set an offload_handle in the SA, an opaque void pointer
  63. that can be used to convey context into the fast-path offload requests::
  64. xs->xso.offload_handle = context;
  65. When the network stack is preparing an IPsec packet for an SA that has
  66. been setup for offload, it first calls into xdo_dev_offload_ok() with
  67. the skb and the intended offload state to ask the driver if the offload
  68. will serviceable. This can check the packet information to be sure the
  69. offload can be supported (e.g. IPv4 or IPv6, no IPv4 options, etc) and
  70. return true of false to signify its support.
  71. When ready to send, the driver needs to inspect the Tx packet for the
  72. offload information, including the opaque context, and set up the packet
  73. send accordingly::
  74. xs = xfrm_input_state(skb);
  75. context = xs->xso.offload_handle;
  76. set up HW for send
  77. The stack has already inserted the appropriate IPsec headers in the
  78. packet data, the offload just needs to do the encryption and fix up the
  79. header values.
  80. When a packet is received and the HW has indicated that it offloaded a
  81. decryption, the driver needs to add a reference to the decoded SA into
  82. the packet's skb. At this point the data should be decrypted but the
  83. IPsec headers are still in the packet data; they are removed later up
  84. the stack in xfrm_input().
  85. find and hold the SA that was used to the Rx skb::
  86. get spi, protocol, and destination IP from packet headers
  87. xs = find xs from (spi, protocol, dest_IP)
  88. xfrm_state_hold(xs);
  89. store the state information into the skb::
  90. sp = secpath_set(skb);
  91. if (!sp) return;
  92. sp->xvec[sp->len++] = xs;
  93. sp->olen++;
  94. indicate the success and/or error status of the offload::
  95. xo = xfrm_offload(skb);
  96. xo->flags = CRYPTO_DONE;
  97. xo->status = crypto_status;
  98. hand the packet to napi_gro_receive() as usual
  99. In ESN mode, xdo_dev_state_advance_esn() is called from xfrm_replay_advance_esn().
  100. Driver will check packet seq number and update HW ESN state machine if needed.
  101. When the SA is removed by the user, the driver's xdo_dev_state_delete()
  102. is asked to disable the offload. Later, xdo_dev_state_free() is called
  103. from a garbage collection routine after all reference counts to the state
  104. have been removed and any remaining resources can be cleared for the
  105. offload state. How these are used by the driver will depend on specific
  106. hardware needs.
  107. As a netdev is set to DOWN the XFRM stack's netdev listener will call
  108. xdo_dev_state_delete() and xdo_dev_state_free() on any remaining offloaded
  109. states.