lib/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_ether.h>
  10
  11 #include "ip_frag_common.h"
  12
  13 /* Fragment Offset */
  14 #define RTE_IPV4_HDR_DF_SHIFT                   14
  15 #define RTE_IPV4_HDR_MF_SHIFT                   13
  16 #define RTE_IPV4_HDR_FO_SHIFT                   3
  17
  18 #define IPV4_HDR_DF_MASK                        (1 << RTE_IPV4_HDR_DF_SHIFT)
  19 #define IPV4_HDR_MF_MASK                        (1 << RTE_IPV4_HDR_MF_SHIFT)
  20
  21 #define IPV4_HDR_FO_ALIGN                       (1 << RTE_IPV4_HDR_FO_SHIFT)
  22
  23 #define IPV4_HDR_MAX_LEN                        60
  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 static inline uint16_t __create_ipopt_frag_hdr(uint8_t *iph,
  45         uint16_t ipopt_len, uint8_t *ipopt_frag_hdr)
  46 {
  47         uint16_t len = ipopt_len;
  48         struct rte_ipv4_hdr *iph_opt = (struct rte_ipv4_hdr *)ipopt_frag_hdr;
  49
  50         ipopt_len = 0;
  51         rte_memcpy(ipopt_frag_hdr, iph, sizeof(struct rte_ipv4_hdr));
  52         ipopt_frag_hdr += sizeof(struct rte_ipv4_hdr);
  53
  54         uint8_t *p_opt = iph + sizeof(struct rte_ipv4_hdr);
  55
  56         while (len > 0) {
  57                 if (unlikely(*p_opt == RTE_IPV4_HDR_OPT_NOP)) {
  58                         len--;
  59                         p_opt++;
  60                         continue;
  61                 } else if (unlikely(*p_opt == RTE_IPV4_HDR_OPT_EOL))
  62                         break;
  63
  64                 if (unlikely(p_opt[1] < 2 || p_opt[1] > len))
  65                         break;
  66
  67                 if (RTE_IPV4_HDR_OPT_COPIED(*p_opt)) {
  68                         rte_memcpy(ipopt_frag_hdr + ipopt_len,
  69                                 p_opt, p_opt[1]);
  70                         ipopt_len += p_opt[1];
  71                 }
  72
  73                 len -= p_opt[1];
  74                 p_opt += p_opt[1];
  75         }
  76
  77         len = RTE_ALIGN_CEIL(ipopt_len, RTE_IPV4_IHL_MULTIPLIER);
  78         memset(ipopt_frag_hdr + ipopt_len,
  79                 RTE_IPV4_HDR_OPT_EOL, len - ipopt_len);
  80         ipopt_len = len;
  81         iph_opt->ihl = (sizeof(struct rte_ipv4_hdr) + ipopt_len) /
  82                 RTE_IPV4_IHL_MULTIPLIER;
  83
  84         return ipopt_len;
  85 }
  86
  87 /**
  88  * IPv4 fragmentation.
  89  *
  90  * This function implements the fragmentation of IPv4 packets.
  91  *
  92  * @param pkt_in
  93  *   The input packet.
  94  * @param pkts_out
  95  *   Array storing the output fragments.
  96  * @param mtu_size
  97  *   Size in bytes of the Maximum Transfer Unit (MTU) for the outgoing IPv4
  98  *   datagrams. This value includes the size of the IPv4 header.
  99  * @param pool_direct
 100  *   MBUF pool used for allocating direct buffers for the output fragments.
 101  * @param pool_indirect
 102  *   MBUF pool used for allocating indirect buffers for the output fragments.
 103  * @return
 104  *   Upon successful completion - number of output fragments placed
 105  *   in the pkts_out array.
 106  *   Otherwise - (-1) * <errno>.
 107  */
 108 int32_t
 109 rte_ipv4_fragment_packet(struct rte_mbuf *pkt_in,
 110         struct rte_mbuf **pkts_out,
 111         uint16_t nb_pkts_out,
 112         uint16_t mtu_size,
 113         struct rte_mempool *pool_direct,
 114         struct rte_mempool *pool_indirect)
 115 {
 116         struct rte_mbuf *in_seg = NULL;
 117         struct rte_ipv4_hdr *in_hdr;
 118         uint32_t out_pkt_pos, in_seg_data_pos;
 119         uint32_t more_in_segs;
 120         uint16_t fragment_offset, flag_offset, frag_size, header_len;
 121         uint16_t frag_bytes_remaining;
 122         uint8_t ipopt_frag_hdr[IPV4_HDR_MAX_LEN];
 123         uint16_t ipopt_len;
 124
 125         /*
 126          * Formal parameter checking.
 127          */
 128         if (unlikely(pkt_in == NULL) || unlikely(pkts_out == NULL) ||
 129             unlikely(nb_pkts_out == 0) ||
 130             unlikely(pool_direct == NULL) || unlikely(pool_indirect == NULL) ||
 131             unlikely(mtu_size < RTE_ETHER_MIN_MTU))
 132                 return -EINVAL;
 133
 134         in_hdr = rte_pktmbuf_mtod(pkt_in, struct rte_ipv4_hdr *);
 135         header_len = (in_hdr->version_ihl & RTE_IPV4_HDR_IHL_MASK) *
 136             RTE_IPV4_IHL_MULTIPLIER;
 137
 138         /* Check IP header length */
 139         if (unlikely(pkt_in->data_len < header_len) ||
 140             unlikely(mtu_size < header_len))
 141                 return -EINVAL;
 142
 143         /*
 144          * Ensure the IP payload length of all fragments is aligned to a
 145          * multiple of 8 bytes as per RFC791 section 2.3.
 146          */
 147         frag_size = RTE_ALIGN_FLOOR((mtu_size - header_len),
 148                                     IPV4_HDR_FO_ALIGN);
 149
 150         flag_offset = rte_cpu_to_be_16(in_hdr->fragment_offset);
 151
 152         /* If Don't Fragment flag is set */
 153         if (unlikely ((flag_offset & IPV4_HDR_DF_MASK) != 0))
 154                 return -ENOTSUP;
 155
 156         /* Check that pkts_out is big enough to hold all fragments */
 157         if (unlikely(frag_size * nb_pkts_out <
 158             (uint16_t)(pkt_in->pkt_len - header_len)))
 159                 return -EINVAL;
 160
 161         in_seg = pkt_in;
 162         in_seg_data_pos = header_len;
 163         out_pkt_pos = 0;
 164         fragment_offset = 0;
 165
 166         ipopt_len = header_len - sizeof(struct rte_ipv4_hdr);
 167         if (unlikely(ipopt_len > RTE_IPV4_HDR_OPT_MAX_LEN))
 168                 return -EINVAL;
 169
 170         more_in_segs = 1;
 171         while (likely(more_in_segs)) {
 172                 struct rte_mbuf *out_pkt = NULL, *out_seg_prev = NULL;
 173                 uint32_t more_out_segs;
 174                 struct rte_ipv4_hdr *out_hdr;
 175
 176                 /* Allocate direct buffer */
 177                 out_pkt = rte_pktmbuf_alloc(pool_direct);
 178                 if (unlikely(out_pkt == NULL)) {
 179                         __free_fragments(pkts_out, out_pkt_pos);
 180                         return -ENOMEM;
 181                 }
 182
 183                 /* Reserve space for the IP header that will be built later */
 184                 out_pkt->data_len = header_len;
 185                 out_pkt->pkt_len = header_len;
 186                 frag_bytes_remaining = frag_size;
 187
 188                 out_seg_prev = out_pkt;
 189                 more_out_segs = 1;
 190                 while (likely(more_out_segs && more_in_segs)) {
 191                         struct rte_mbuf *out_seg = NULL;
 192                         uint32_t len;
 193
 194                         /* Allocate indirect buffer */
 195                         out_seg = rte_pktmbuf_alloc(pool_indirect);
 196                         if (unlikely(out_seg == NULL)) {
 197                                 rte_pktmbuf_free(out_pkt);
 198                                 __free_fragments(pkts_out, out_pkt_pos);
 199                                 return -ENOMEM;
 200                         }
 201                         out_seg_prev->next = out_seg;
 202                         out_seg_prev = out_seg;
 203
 204                         /* Prepare indirect buffer */
 205                         rte_pktmbuf_attach(out_seg, in_seg);
 206                         len = frag_bytes_remaining;
 207                         if (len > (in_seg->data_len - in_seg_data_pos)) {
 208                                 len = in_seg->data_len - in_seg_data_pos;
 209                         }
 210                         out_seg->data_off = in_seg->data_off + in_seg_data_pos;
 211                         out_seg->data_len = (uint16_t)len;
 212                         out_pkt->pkt_len = (uint16_t)(len +
 213                             out_pkt->pkt_len);
 214                         out_pkt->nb_segs += 1;
 215                         in_seg_data_pos += len;
 216                         frag_bytes_remaining -= len;
 217
 218                         /* Current output packet (i.e. fragment) done ? */
 219                         if (unlikely(frag_bytes_remaining == 0))
 220                                 more_out_segs = 0;
 221
 222                         /* Current input segment done ? */
 223                         if (unlikely(in_seg_data_pos == in_seg->data_len)) {
 224                                 in_seg = in_seg->next;
 225                                 in_seg_data_pos = 0;
 226
 227                                 if (unlikely(in_seg == NULL))
 228                                         more_in_segs = 0;
 229                         }
 230                 }
 231
 232                 /* Build the IP header */
 233
 234                 out_hdr = rte_pktmbuf_mtod(out_pkt, struct rte_ipv4_hdr *);
 235
 236                 __fill_ipv4hdr_frag(out_hdr, in_hdr, header_len,
 237                     (uint16_t)out_pkt->pkt_len,
 238                     flag_offset, fragment_offset, more_in_segs);
 239
 240                 if (unlikely((fragment_offset == 0) && (ipopt_len) &&
 241                             ((flag_offset & RTE_IPV4_HDR_OFFSET_MASK) == 0))) {
 242                         ipopt_len = __create_ipopt_frag_hdr((uint8_t *)in_hdr,
 243                                 ipopt_len, ipopt_frag_hdr);
 244                         fragment_offset = (uint16_t)(fragment_offset +
 245                                 out_pkt->pkt_len - header_len);
 246                         out_pkt->l3_len = header_len;
 247
 248                         header_len = sizeof(struct rte_ipv4_hdr) + ipopt_len;
 249                         in_hdr = (struct rte_ipv4_hdr *)ipopt_frag_hdr;
 250                 } else {
 251                         fragment_offset = (uint16_t)(fragment_offset +
 252                                 out_pkt->pkt_len - header_len);
 253                         out_pkt->l3_len = header_len;
 254                 }
 255
 256                 /* Write the fragment to the output list */
 257                 pkts_out[out_pkt_pos] = out_pkt;
 258                 out_pkt_pos ++;
 259         }
 260
 261         return out_pkt_pos;
 262 }