include/net/eth.h

   1 /*
   2  * QEMU network structures definitions and helper functions
   3  *
   4  * Copyright (c) 2012 Ravello Systems LTD (http://ravellosystems.com)
   5  *
   6  * Developed by Daynix Computing LTD (http://www.daynix.com)
   7  *
   8  * Portions developed by Free Software Foundation, Inc
   9  * Copyright (C) 1991-1997, 2001, 2003, 2006 Free Software Foundation, Inc.
  10  * See netinet/ip6.h and netinet/in.h (GNU C Library)
  11  *
  12  * Portions developed by Igor Kovalenko
  13  * Copyright (c) 2006 Igor Kovalenko
  14  * See hw/rtl8139.c (QEMU)
  15  *
  16  * Authors:
  17  * Dmitry Fleytman <dmitry@daynix.com>
  18  * Tamir Shomer <tamirs@daynix.com>
  19  * Yan Vugenfirer <yan@daynix.com>
  20  *
  21  * This work is licensed under the terms of the GNU GPL, version 2 or later.
  22  * See the COPYING file in the top-level directory.
  23  *
  24  */
  25
  26 #ifndef QEMU_ETH_H
  27 #define QEMU_ETH_H
  28
  29 #include "qemu/bswap.h"
  30 #include "qemu/iov.h"
  31
  32 #define ETH_ALEN 6
  33 #define ETH_HLEN 14
  34 #define ETH_ZLEN 60     /* Min. octets in frame without FCS */
  35
  36 struct eth_header {
  37     uint8_t  h_dest[ETH_ALEN];   /* destination eth addr */
  38     uint8_t  h_source[ETH_ALEN]; /* source ether addr    */
  39     uint16_t h_proto;            /* packet type ID field */
  40 };
  41
  42 struct vlan_header {
  43     uint16_t  h_tci;     /* priority and VLAN ID  */
  44     uint16_t  h_proto;   /* encapsulated protocol */
  45 };
  46
  47 struct ip_header {
  48     uint8_t  ip_ver_len;     /* version and header length */
  49     uint8_t  ip_tos;         /* type of service */
  50     uint16_t ip_len;         /* total length */
  51     uint16_t ip_id;          /* identification */
  52     uint16_t ip_off;         /* fragment offset field */
  53     uint8_t  ip_ttl;         /* time to live */
  54     uint8_t  ip_p;           /* protocol */
  55     uint16_t ip_sum;         /* checksum */
  56     uint32_t ip_src, ip_dst; /* source and destination address */
  57 };
  58
  59 typedef struct tcp_header {
  60     uint16_t th_sport;          /* source port */
  61     uint16_t th_dport;          /* destination port */
  62     uint32_t th_seq;            /* sequence number */
  63     uint32_t th_ack;            /* acknowledgment number */
  64     uint16_t th_offset_flags;   /* data offset, reserved 6 bits, */
  65                                 /* TCP protocol flags */
  66     uint16_t th_win;            /* window */
  67     uint16_t th_sum;            /* checksum */
  68     uint16_t th_urp;            /* urgent pointer */
  69 } tcp_header;
  70
  71 #define TCP_FLAGS_ONLY(flags) ((flags) & 0x3f)
  72
  73 #define TCP_HEADER_FLAGS(tcp) \
  74     TCP_FLAGS_ONLY(be16_to_cpu((tcp)->th_offset_flags))
  75
  76 #define TCP_FLAG_ACK  0x10
  77
  78 #define TCP_HEADER_DATA_OFFSET(tcp) \
  79     (((be16_to_cpu((tcp)->th_offset_flags) >> 12) & 0xf) << 2)
  80
  81 typedef struct udp_header {
  82     uint16_t uh_sport; /* source port */
  83     uint16_t uh_dport; /* destination port */
  84     uint16_t uh_ulen;  /* udp length */
  85     uint16_t uh_sum;   /* udp checksum */
  86 } udp_header;
  87
  88 typedef struct ip_pseudo_header {
  89     uint32_t ip_src;
  90     uint32_t ip_dst;
  91     uint8_t  zeros;
  92     uint8_t  ip_proto;
  93     uint16_t ip_payload;
  94 } ip_pseudo_header;
  95
  96 /* IPv6 address */
  97 struct in6_address {
  98     union {
  99         uint8_t __u6_addr8[16];
 100     } __in6_u;
 101 };
 102
 103 struct ip6_header {
 104     union {
 105         struct ip6_hdrctl {
 106             uint32_t ip6_un1_flow; /* 4 bits version, 8 bits TC,
 107                                       20 bits flow-ID */
 108             uint16_t ip6_un1_plen; /* payload length */
 109             uint8_t  ip6_un1_nxt;  /* next header */
 110             uint8_t  ip6_un1_hlim; /* hop limit */
 111         } ip6_un1;
 112         uint8_t ip6_un2_vfc;       /* 4 bits version, top 4 bits tclass */
 113         struct ip6_ecn_access {
 114             uint8_t  ip6_un3_vfc;  /* 4 bits version, top 4 bits tclass */
 115             uint8_t  ip6_un3_ecn;  /* 2 bits ECN, top 6 bits payload length */
 116         } ip6_un3;
 117     } ip6_ctlun;
 118     struct in6_address ip6_src;    /* source address */
 119     struct in6_address ip6_dst;    /* destination address */
 120 };
 121
 122 typedef struct ip6_pseudo_header {
 123     struct in6_address ip6_src;
 124     struct in6_address ip6_dst;
 125     uint32_t           len;
 126     uint8_t            zero[3];
 127     uint8_t            next_hdr;
 128 } ip6_pseudo_header;
 129
 130 struct ip6_ext_hdr {
 131     uint8_t        ip6r_nxt;   /* next header */
 132     uint8_t        ip6r_len;   /* length in units of 8 octets */
 133 };
 134
 135 struct ip6_ext_hdr_routing {
 136     uint8_t     nxt;
 137     uint8_t     len;
 138     uint8_t     rtype;
 139     uint8_t     segleft;
 140     uint8_t     rsvd[4];
 141 };
 142
 143 struct ip6_option_hdr {
 144 #define IP6_OPT_PAD1   (0x00)
 145 #define IP6_OPT_HOME   (0xC9)
 146     uint8_t type;
 147     uint8_t len;
 148 };
 149
 150 struct udp_hdr {
 151   uint16_t uh_sport;           /* source port */
 152   uint16_t uh_dport;           /* destination port */
 153   uint16_t uh_ulen;            /* udp length */
 154   uint16_t uh_sum;             /* udp checksum */
 155 };
 156
 157 struct tcp_hdr {
 158     u_short     th_sport;   /* source port */
 159     u_short     th_dport;   /* destination port */
 160     uint32_t    th_seq;     /* sequence number */
 161     uint32_t    th_ack;     /* acknowledgment number */
 162 #if HOST_BIG_ENDIAN
 163     u_char  th_off : 4,     /* data offset */
 164         th_x2:4;            /* (unused) */
 165 #else
 166     u_char  th_x2 : 4,      /* (unused) */
 167         th_off:4;           /* data offset */
 168 #endif
 169
 170 #define TH_ELN  0x1 /* explicit loss notification */
 171 #define TH_ECN  0x2 /* explicit congestion notification */
 172 #define TH_FS   0x4 /* fast start */
 173
 174     u_char  th_flags;
 175 #define TH_FIN  0x01
 176 #define TH_SYN  0x02
 177 #define TH_RST  0x04
 178 #define TH_PUSH 0x08
 179 #define TH_ACK  0x10
 180 #define TH_URG  0x20
 181 #define TH_ECE  0x40
 182 #define TH_CWR  0x80
 183     u_short th_win;      /* window */
 184     u_short th_sum;      /* checksum */
 185     u_short th_urp;      /* urgent pointer */
 186 };
 187
 188 #define ip6_nxt      ip6_ctlun.ip6_un1.ip6_un1_nxt
 189 #define ip6_ecn_acc  ip6_ctlun.ip6_un3.ip6_un3_ecn
 190 #define ip6_plen     ip6_ctlun.ip6_un1.ip6_un1_plen
 191
 192 #define PKT_GET_ETH_HDR(p)        \
 193     ((struct eth_header *)(p))
 194 #define PKT_GET_VLAN_HDR(p)       \
 195     ((struct vlan_header *) (((uint8_t *)(p)) + sizeof(struct eth_header)))
 196 #define PKT_GET_DVLAN_HDR(p)       \
 197     (PKT_GET_VLAN_HDR(p) + 1)
 198 #define PKT_GET_IP_HDR(p)         \
 199     ((struct ip_header *)(((uint8_t *)(p)) + eth_get_l2_hdr_length(p)))
 200 #define IP_HDR_GET_LEN(p)         \
 201     ((ldub_p(p + offsetof(struct ip_header, ip_ver_len)) & 0x0F) << 2)
 202 #define IP_HDR_GET_P(p)                                           \
 203     (ldub_p(p + offsetof(struct ip_header, ip_p)))
 204 #define PKT_GET_IP_HDR_LEN(p)     \
 205     (IP_HDR_GET_LEN(PKT_GET_IP_HDR(p)))
 206 #define PKT_GET_IP6_HDR(p)        \
 207     ((struct ip6_header *) (((uint8_t *)(p)) + eth_get_l2_hdr_length(p)))
 208 #define IP_HEADER_VERSION(ip)     \
 209     (((ip)->ip_ver_len >> 4) & 0xf)
 210 #define IP4_IS_FRAGMENT(ip) \
 211     ((be16_to_cpu((ip)->ip_off) & (IP_OFFMASK | IP_MF)) != 0)
 212
 213 #define ETH_P_IP                  (0x0800)      /* Internet Protocol packet  */
 214 #define ETH_P_ARP                 (0x0806)      /* Address Resolution packet */
 215 #define ETH_P_IPV6                (0x86dd)
 216 #define ETH_P_VLAN                (0x8100)
 217 #define ETH_P_DVLAN               (0x88a8)
 218 #define ETH_P_NCSI                (0x88f8)
 219 #define ETH_P_UNKNOWN             (0xffff)
 220 #define VLAN_VID_MASK             0x0fff
 221 #define IP_HEADER_VERSION_4       (4)
 222 #define IP_HEADER_VERSION_6       (6)
 223 #define IP_PROTO_TCP              (6)
 224 #define IP_PROTO_UDP              (17)
 225 #define IPTOS_ECN_MASK            0x03
 226 #define IPTOS_ECN(x)              ((x) & IPTOS_ECN_MASK)
 227 #define IPTOS_ECN_CE              0x03
 228 #define IP6_ECN_MASK              0xC0
 229 #define IP6_ECN(x)                ((x) & IP6_ECN_MASK)
 230 #define IP6_ECN_CE                0xC0
 231 #define IP4_DONT_FRAGMENT_FLAG    (1 << 14)
 232
 233 #define IS_SPECIAL_VLAN_ID(x)     \
 234     (((x) == 0) || ((x) == 0xFFF))
 235
 236 #define ETH_MAX_L2_HDR_LEN  \
 237     (sizeof(struct eth_header) + 2 * sizeof(struct vlan_header))
 238
 239 #define ETH_MAX_IP4_HDR_LEN   (60)
 240 #define ETH_MAX_IP_DGRAM_LEN  (0xFFFF)
 241
 242 #define IP_FRAG_UNIT_SIZE     (8)
 243 #define IP_FRAG_ALIGN_SIZE(x) ((x) & ~0x7)
 244 #define IP_RF                 0x8000           /* reserved fragment flag */
 245 #define IP_DF                 0x4000           /* don't fragment flag */
 246 #define IP_MF                 0x2000           /* more fragments flag */
 247 #define IP_OFFMASK            0x1fff           /* mask for fragmenting bits */
 248
 249 #define IP6_EXT_GRANULARITY   (8)  /* Size granularity for
 250                                       IPv6 extension headers */
 251
 252 /* IP6 extension header types */
 253 #define IP6_HOP_BY_HOP        (0)
 254 #define IP6_ROUTING           (43)
 255 #define IP6_FRAGMENT          (44)
 256 #define IP6_ESP               (50)
 257 #define IP6_AUTHENTICATION    (51)
 258 #define IP6_NONE              (59)
 259 #define IP6_DESTINATON        (60)
 260 #define IP6_MOBILITY          (135)
 261
 262 static inline int is_multicast_ether_addr(const uint8_t *addr)
 263 {
 264     return 0x01 & addr[0];
 265 }
 266
 267 static inline int is_broadcast_ether_addr(const uint8_t *addr)
 268 {
 269     return (addr[0] & addr[1] & addr[2] & addr[3] & addr[4] & addr[5]) == 0xff;
 270 }
 271
 272 static inline int is_unicast_ether_addr(const uint8_t *addr)
 273 {
 274     return !is_multicast_ether_addr(addr);
 275 }
 276
 277 typedef enum {
 278     ETH_PKT_UCAST = 0xAABBCC00,
 279     ETH_PKT_BCAST,
 280     ETH_PKT_MCAST
 281 } eth_pkt_types_e;
 282
 283 static inline eth_pkt_types_e
 284 get_eth_packet_type(const struct eth_header *ehdr)
 285 {
 286     if (is_broadcast_ether_addr(ehdr->h_dest)) {
 287         return ETH_PKT_BCAST;
 288     } else if (is_multicast_ether_addr(ehdr->h_dest)) {
 289         return ETH_PKT_MCAST;
 290     } else { /* unicast */
 291         return ETH_PKT_UCAST;
 292     }
 293 }
 294
 295 static inline uint32_t
 296 eth_get_l2_hdr_length(const void *p)
 297 {
 298     uint16_t proto = be16_to_cpu(PKT_GET_ETH_HDR(p)->h_proto);
 299     struct vlan_header *hvlan = PKT_GET_VLAN_HDR(p);
 300     switch (proto) {
 301     case ETH_P_VLAN:
 302         return sizeof(struct eth_header) + sizeof(struct vlan_header);
 303     case ETH_P_DVLAN:
 304         if (be16_to_cpu(hvlan->h_proto) == ETH_P_VLAN) {
 305             return sizeof(struct eth_header) + 2 * sizeof(struct vlan_header);
 306         } else {
 307             return sizeof(struct eth_header) + sizeof(struct vlan_header);
 308         }
 309     default:
 310         return sizeof(struct eth_header);
 311     }
 312 }
 313
 314 static inline uint32_t
 315 eth_get_l2_hdr_length_iov(const struct iovec *iov, int iovcnt)
 316 {
 317     uint8_t p[sizeof(struct eth_header) + sizeof(struct vlan_header)];
 318     size_t copied = iov_to_buf(iov, iovcnt, 0, p, ARRAY_SIZE(p));
 319
 320     if (copied < ARRAY_SIZE(p)) {
 321         return copied;
 322     }
 323
 324     return eth_get_l2_hdr_length(p);
 325 }
 326
 327 static inline uint16_t
 328 eth_get_pkt_tci(const void *p)
 329 {
 330     uint16_t proto = be16_to_cpu(PKT_GET_ETH_HDR(p)->h_proto);
 331     struct vlan_header *hvlan = PKT_GET_VLAN_HDR(p);
 332     switch (proto) {
 333     case ETH_P_VLAN:
 334     case ETH_P_DVLAN:
 335         return be16_to_cpu(hvlan->h_tci);
 336     default:
 337         return 0;
 338     }
 339 }
 340
 341 size_t
 342 eth_strip_vlan(const struct iovec *iov, int iovcnt, size_t iovoff,
 343                uint8_t *new_ehdr_buf,
 344                uint16_t *payload_offset, uint16_t *tci);
 345
 346 size_t
 347 eth_strip_vlan_ex(const struct iovec *iov, int iovcnt, size_t iovoff,
 348                   uint16_t vet, uint8_t *new_ehdr_buf,
 349                   uint16_t *payload_offset, uint16_t *tci);
 350
 351 uint16_t
 352 eth_get_l3_proto(const struct iovec *l2hdr_iov, int iovcnt, size_t l2hdr_len);
 353
 354 void eth_setup_vlan_headers_ex(struct eth_header *ehdr, uint16_t vlan_tag,
 355     uint16_t vlan_ethtype, bool *is_new);
 356
 357 static inline void
 358 eth_setup_vlan_headers(struct eth_header *ehdr, uint16_t vlan_tag,
 359     bool *is_new)
 360 {
 361     eth_setup_vlan_headers_ex(ehdr, vlan_tag, ETH_P_VLAN, is_new);
 362 }
 363
 364
 365 uint8_t eth_get_gso_type(uint16_t l3_proto, uint8_t *l3_hdr, uint8_t l4proto);
 366
 367 typedef struct eth_ip6_hdr_info_st {
 368     uint8_t l4proto;
 369     size_t  full_hdr_len;
 370     struct  ip6_header ip6_hdr;
 371     bool    has_ext_hdrs;
 372     bool    rss_ex_src_valid;
 373     struct  in6_address rss_ex_src;
 374     bool    rss_ex_dst_valid;
 375     struct  in6_address rss_ex_dst;
 376     bool    fragment;
 377 } eth_ip6_hdr_info;
 378
 379 typedef struct eth_ip4_hdr_info_st {
 380     struct ip_header ip4_hdr;
 381     bool   fragment;
 382 } eth_ip4_hdr_info;
 383
 384 typedef struct eth_l4_hdr_info_st {
 385     union {
 386         struct tcp_header tcp;
 387         struct udp_header udp;
 388     } hdr;
 389
 390     bool has_tcp_data;
 391 } eth_l4_hdr_info;
 392
 393 void eth_get_protocols(const struct iovec *iov, int iovcnt,
 394                        bool *isip4, bool *isip6,
 395                        bool *isudp, bool *istcp,
 396                        size_t *l3hdr_off,
 397                        size_t *l4hdr_off,
 398                        size_t *l5hdr_off,
 399                        eth_ip6_hdr_info *ip6hdr_info,
 400                        eth_ip4_hdr_info *ip4hdr_info,
 401                        eth_l4_hdr_info  *l4hdr_info);
 402
 403 void eth_setup_ip4_fragmentation(const void *l2hdr, size_t l2hdr_len,
 404                                  void *l3hdr, size_t l3hdr_len,
 405                                  size_t l3payload_len,
 406                                  size_t frag_offset, bool more_frags);
 407
 408 void
 409 eth_fix_ip4_checksum(void *l3hdr, size_t l3hdr_len);
 410
 411 uint32_t
 412 eth_calc_ip4_pseudo_hdr_csum(struct ip_header *iphdr,
 413                              uint16_t csl,
 414                              uint32_t *cso);
 415
 416 uint32_t
 417 eth_calc_ip6_pseudo_hdr_csum(struct ip6_header *iphdr,
 418                              uint16_t csl,
 419                              uint8_t l4_proto,
 420                              uint32_t *cso);
 421
 422 bool
 423 eth_parse_ipv6_hdr(const struct iovec *pkt, int pkt_frags,
 424                    size_t ip6hdr_off, eth_ip6_hdr_info *info);
 425
 426 /**
 427  * eth_pad_short_frame - pad a short frame to the minimum Ethernet frame length
 428  *
 429  * If the Ethernet frame size is shorter than 60 bytes, it will be padded to
 430  * 60 bytes at the address @padded_pkt.
 431  *
 432  * @padded_pkt: buffer address to hold the padded frame
 433  * @padded_buflen: pointer holding length of @padded_pkt. If the frame is
 434  *                 padded, the length will be updated to the padded one.
 435  * @pkt: address to hold the original Ethernet frame
 436  * @pkt_size: size of the original Ethernet frame
 437  * @return true if the frame is padded, otherwise false
 438  */
 439 bool eth_pad_short_frame(uint8_t *padded_pkt, size_t *padded_buflen,
 440                          const void *pkt, size_t pkt_size);
 441
 442 #endif