ip_frag/rte_ipv4_fragmentation.c

   1 /* SPDX-License-Identifier: BSD-3-Clause
   2  * Copyright(c) 2010-2014 Intel Corporation
   3  */
   4
   5 #include <stddef.h>
   6 #include <errno.h>
   7
   8 #include <rte_memcpy.h>
   9 #include <rte_mempool.h>
  10 #include <rte_debug.h>
  11 #include <rte_ether.h>
  12
  13 #include "ip_frag_common.h"
  14
  15 /* Fragment Offset */
  16 #define RTE_IPV4_HDR_DF_SHIFT                   14
  17 #define RTE_IPV4_HDR_MF_SHIFT                   13
  18 #define RTE_IPV4_HDR_FO_SHIFT                   3
  19
  20 #define IPV4_HDR_DF_MASK                        (1 << RTE_IPV4_HDR_DF_SHIFT)
  21 #define IPV4_HDR_MF_MASK                        (1 << RTE_IPV4_HDR_MF_SHIFT)
  22
  23 #define IPV4_HDR_FO_ALIGN                       (1 << RTE_IPV4_HDR_FO_SHIFT)
  24
  25 static inline void __fill_ipv4hdr_frag(struct rte_ipv4_hdr *dst,
  26                 const struct rte_ipv4_hdr *src, uint16_t header_len,
  27                 uint16_t len, uint16_t fofs, uint16_t dofs, uint32_t mf)
  28 {
  29         rte_memcpy(dst, src, header_len);
  30         fofs = (uint16_t)(fofs + (dofs >> RTE_IPV4_HDR_FO_SHIFT));
  31         fofs = (uint16_t)(fofs | mf << RTE_IPV4_HDR_MF_SHIFT);
  32         dst->fragment_offset = rte_cpu_to_be_16(fofs);
  33         dst->total_length = rte_cpu_to_be_16(len);
  34         dst->hdr_checksum = 0;
  35 }
  36
  37 static inline void __free_fragments(struct rte_mbuf *mb[], uint32_t num)
  38 {
  39         uint32_t i;
  40         for (i = 0; i != num; i++)
  41                 rte_pktmbuf_free(mb[i]);
  42 }
  43
  44 /**
  45  * IPv4 fragmentation.
  46  *
  47  * This function implements the fragmentation of IPv4 packets.
  48  *
  49  * @param pkt_in
  50  *   The input packet.
  51  * @param pkts_out
  52  *   Array storing the output fragments.
  53  * @param mtu_size
  54  *   Size in bytes of the Maximum Transfer Unit (MTU) for the outgoing IPv4
  55  *   datagrams. This value includes the size of the IPv4 header.
  56  * @param pool_direct
  57  *   MBUF pool used for allocating direct buffers for the output fragments.
  58  * @param pool_indirect
  59  *   MBUF pool used for allocating indirect buffers for the output fragments.
  60  * @return
  61  *   Upon successful completion - number of output fragments placed
  62  *   in the pkts_out array.
  63  *   Otherwise - (-1) * <errno>.
  64  */
  65 int32_t
  66 rte_ipv4_fragment_packet(struct rte_mbuf *pkt_in,
  67         struct rte_mbuf **pkts_out,
  68         uint16_t nb_pkts_out,
  69         uint16_t mtu_size,
  70         struct rte_mempool *pool_direct,
  71         struct rte_mempool *pool_indirect)
  72 {
  73         struct rte_mbuf *in_seg = NULL;
  74         struct rte_ipv4_hdr *in_hdr;
  75         uint32_t out_pkt_pos, in_seg_data_pos;
  76         uint32_t more_in_segs;
  77         uint16_t fragment_offset, flag_offset, frag_size, header_len;
  78         uint16_t frag_bytes_remaining;
  79
  80         /*
  81          * Formal parameter checking.
  82          */
  83         if (unlikely(pkt_in == NULL) || unlikely(pkts_out == NULL) ||
  84             unlikely(nb_pkts_out == 0) ||
  85             unlikely(pool_direct == NULL) || unlikely(pool_indirect == NULL) ||
  86             unlikely(mtu_size < RTE_ETHER_MIN_MTU))
  87                 return -EINVAL;
  88
  89         in_hdr = rte_pktmbuf_mtod(pkt_in, struct rte_ipv4_hdr *);
  90         header_len = (in_hdr->version_ihl & RTE_IPV4_HDR_IHL_MASK) *
  91             RTE_IPV4_IHL_MULTIPLIER;
  92
  93         /* Check IP header length */
  94         if (unlikely(pkt_in->data_len < header_len) ||
  95             unlikely(mtu_size < header_len))
  96                 return -EINVAL;
  97
  98         /*
  99          * Ensure the IP payload length of all fragments is aligned to a
 100          * multiple of 8 bytes as per RFC791 section 2.3.
 101          */
 102         frag_size = RTE_ALIGN_FLOOR((mtu_size - header_len),
 103                                     IPV4_HDR_FO_ALIGN);
 104
 105         flag_offset = rte_cpu_to_be_16(in_hdr->fragment_offset);
 106
 107         /* If Don't Fragment flag is set */
 108         if (unlikely ((flag_offset & IPV4_HDR_DF_MASK) != 0))
 109                 return -ENOTSUP;
 110
 111         /* Check that pkts_out is big enough to hold all fragments */
 112         if (unlikely(frag_size * nb_pkts_out <
 113             (uint16_t)(pkt_in->pkt_len - header_len)))
 114                 return -EINVAL;
 115
 116         in_seg = pkt_in;
 117         in_seg_data_pos = header_len;
 118         out_pkt_pos = 0;
 119         fragment_offset = 0;
 120
 121         more_in_segs = 1;
 122         while (likely(more_in_segs)) {
 123                 struct rte_mbuf *out_pkt = NULL, *out_seg_prev = NULL;
 124                 uint32_t more_out_segs;
 125                 struct rte_ipv4_hdr *out_hdr;
 126
 127                 /* Allocate direct buffer */
 128                 out_pkt = rte_pktmbuf_alloc(pool_direct);
 129                 if (unlikely(out_pkt == NULL)) {
 130                         __free_fragments(pkts_out, out_pkt_pos);
 131                         return -ENOMEM;
 132                 }
 133
 134                 /* Reserve space for the IP header that will be built later */
 135                 out_pkt->data_len = header_len;
 136                 out_pkt->pkt_len = header_len;
 137                 frag_bytes_remaining = frag_size;
 138
 139                 out_seg_prev = out_pkt;
 140                 more_out_segs = 1;
 141                 while (likely(more_out_segs && more_in_segs)) {
 142                         struct rte_mbuf *out_seg = NULL;
 143                         uint32_t len;
 144
 145                         /* Allocate indirect buffer */
 146                         out_seg = rte_pktmbuf_alloc(pool_indirect);
 147                         if (unlikely(out_seg == NULL)) {
 148                                 rte_pktmbuf_free(out_pkt);
 149                                 __free_fragments(pkts_out, out_pkt_pos);
 150                                 return -ENOMEM;
 151                         }
 152                         out_seg_prev->next = out_seg;
 153                         out_seg_prev = out_seg;
 154
 155                         /* Prepare indirect buffer */
 156                         rte_pktmbuf_attach(out_seg, in_seg);
 157                         len = frag_bytes_remaining;
 158                         if (len > (in_seg->data_len - in_seg_data_pos)) {
 159                                 len = in_seg->data_len - in_seg_data_pos;
 160                         }
 161                         out_seg->data_off = in_seg->data_off + in_seg_data_pos;
 162                         out_seg->data_len = (uint16_t)len;
 163                         out_pkt->pkt_len = (uint16_t)(len +
 164                             out_pkt->pkt_len);
 165                         out_pkt->nb_segs += 1;
 166                         in_seg_data_pos += len;
 167                         frag_bytes_remaining -= len;
 168
 169                         /* Current output packet (i.e. fragment) done ? */
 170                         if (unlikely(frag_bytes_remaining == 0))
 171                                 more_out_segs = 0;
 172
 173                         /* Current input segment done ? */
 174                         if (unlikely(in_seg_data_pos == in_seg->data_len)) {
 175                                 in_seg = in_seg->next;
 176                                 in_seg_data_pos = 0;
 177
 178                                 if (unlikely(in_seg == NULL))
 179                                         more_in_segs = 0;
 180                         }
 181                 }
 182
 183                 /* Build the IP header */
 184
 185                 out_hdr = rte_pktmbuf_mtod(out_pkt, struct rte_ipv4_hdr *);
 186
 187                 __fill_ipv4hdr_frag(out_hdr, in_hdr, header_len,
 188                     (uint16_t)out_pkt->pkt_len,
 189                     flag_offset, fragment_offset, more_in_segs);
 190
 191                 fragment_offset = (uint16_t)(fragment_offset +
 192                     out_pkt->pkt_len - header_len);
 193
 194                 out_pkt->l3_len = header_len;
 195
 196                 /* Write the fragment to the output list */
 197                 pkts_out[out_pkt_pos] = out_pkt;
 198                 out_pkt_pos ++;
 199         }
 200
 201         return out_pkt_pos;
 202 }