1 /* 2 * Copyright (C) 2014 Vincenzo Maffione. All rights reserved. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. Redistributions in binary form must reproduce the above copyright 10 * notice, this list of conditions and the following disclaimer in the 11 * documentation and/or other materials provided with the distribution. 12 * 13 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 14 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 15 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 16 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 17 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 18 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 19 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 20 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 21 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 22 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 23 * SUCH DAMAGE. 24 */ 25 26 /* $FreeBSD$ */ 27 28 #if defined(__FreeBSD__) 29 #include <sys/cdefs.h> /* prerequisite */ 30 31 #include <sys/types.h> 32 #include <sys/errno.h> 33 #include <sys/param.h> /* defines used in kernel.h */ 34 #include <sys/kernel.h> /* types used in module initialization */ 35 #include <sys/sockio.h> 36 #include <sys/socketvar.h> /* struct socket */ 37 #include <sys/socket.h> /* sockaddrs */ 38 #include <net/if.h> 39 #include <net/if_var.h> 40 #include <machine/bus.h> /* bus_dmamap_* */ 41 #include <sys/endian.h> 42 43 #elif defined(linux) 44 45 #include "bsd_glue.h" 46 47 #elif defined(__APPLE__) 48 49 #warning OSX support is only partial 50 #include "osx_glue.h" 51 52 #else 53 54 #error Unsupported platform 55 56 #endif /* unsupported */ 57 58 #include <net/netmap.h> 59 #include <dev/netmap/netmap_kern.h> 60 61 62 63 /* This routine is called by bdg_mismatch_datapath() when it finishes 64 * accumulating bytes for a segment, in order to fix some fields in the 65 * segment headers (which still contain the same content as the header 66 * of the original GSO packet). 'buf' points to the beginning (e.g. 67 * the ethernet header) of the segment, and 'len' is its length. 68 */ 69 static void gso_fix_segment(uint8_t *buf, size_t len, u_int idx, 70 u_int segmented_bytes, u_int last_segment, 71 u_int tcp, u_int iphlen) 72 { 73 struct nm_iphdr *iph = (struct nm_iphdr *)(buf + 14); 74 struct nm_ipv6hdr *ip6h = (struct nm_ipv6hdr *)(buf + 14); 75 uint16_t *check = NULL; 76 uint8_t *check_data = NULL; 77 78 if (iphlen == 20) { 79 /* Set the IPv4 "Total Length" field. */ 80 iph->tot_len = htobe16(len-14); 81 ND("ip total length %u", be16toh(ip->tot_len)); 82 83 /* Set the IPv4 "Identification" field. */ 84 iph->id = htobe16(be16toh(iph->id) + idx); 85 ND("ip identification %u", be16toh(iph->id)); 86 87 /* Compute and insert the IPv4 header checksum. */ 88 iph->check = 0; 89 iph->check = nm_csum_ipv4(iph); 90 ND("IP csum %x", be16toh(iph->check)); 91 } else {/* if (iphlen == 40) */ 92 /* Set the IPv6 "Payload Len" field. */ 93 ip6h->payload_len = htobe16(len-14-iphlen); 94 } 95 96 if (tcp) { 97 struct nm_tcphdr *tcph = (struct nm_tcphdr *)(buf + 14 + iphlen); 98 99 /* Set the TCP sequence number. */ 100 tcph->seq = htobe32(be32toh(tcph->seq) + segmented_bytes); 101 ND("tcp seq %u", be32toh(tcph->seq)); 102 103 /* Zero the PSH and FIN TCP flags if this is not the last 104 segment. */ 105 if (!last_segment) 106 tcph->flags &= ~(0x8 | 0x1); 107 ND("last_segment %u", last_segment); 108 109 check = &tcph->check; 110 check_data = (uint8_t *)tcph; 111 } else { /* UDP */ 112 struct nm_udphdr *udph = (struct nm_udphdr *)(buf + 14 + iphlen); 113 114 /* Set the UDP 'Length' field. */ 115 udph->len = htobe16(len-14-iphlen); 116 117 check = &udph->check; 118 check_data = (uint8_t *)udph; 119 } 120 121 /* Compute and insert TCP/UDP checksum. */ 122 *check = 0; 123 if (iphlen == 20) 124 nm_csum_tcpudp_ipv4(iph, check_data, len-14-iphlen, check); 125 else 126 nm_csum_tcpudp_ipv6(ip6h, check_data, len-14-iphlen, check); 127 128 ND("TCP/UDP csum %x", be16toh(*check)); 129 } 130 131 132 /* The VALE mismatch datapath implementation. */ 133 void bdg_mismatch_datapath(struct netmap_vp_adapter *na, 134 struct netmap_vp_adapter *dst_na, 135 struct nm_bdg_fwd *ft_p, struct netmap_ring *ring, 136 u_int *j, u_int lim, u_int *howmany) 137 { 138 struct netmap_slot *slot = NULL; 139 struct nm_vnet_hdr *vh = NULL; 140 /* Number of source slots to process. */ 141 u_int frags = ft_p->ft_frags; 142 struct nm_bdg_fwd *ft_end = ft_p + frags; 143 144 /* Source and destination pointers. */ 145 uint8_t *dst, *src; 146 size_t src_len, dst_len; 147 148 u_int j_start = *j; 149 u_int dst_slots = 0; 150 151 /* If the source port uses the offloadings, while destination doesn't, 152 * we grab the source virtio-net header and do the offloadings here. 153 */ 154 if (na->virt_hdr_len && !dst_na->virt_hdr_len) { 155 vh = (struct nm_vnet_hdr *)ft_p->ft_buf; 156 } 157 158 /* Init source and dest pointers. */ 159 src = ft_p->ft_buf; 160 src_len = ft_p->ft_len; 161 slot = &ring->slot[*j]; 162 dst = BDG_NMB(&dst_na->up, slot); 163 dst_len = src_len; 164 165 /* We are processing the first input slot and there is a mismatch 166 * between source and destination virt_hdr_len (SHL and DHL). 167 * When the a client is using virtio-net headers, the header length 168 * can be: 169 * - 10: the header corresponds to the struct nm_vnet_hdr 170 * - 12: the first 10 bytes correspond to the struct 171 * virtio_net_hdr, and the last 2 bytes store the 172 * "mergeable buffers" info, which is an optional 173 * hint that can be zeroed for compability 174 * 175 * The destination header is therefore built according to the 176 * following table: 177 * 178 * SHL | DHL | destination header 179 * ----------------------------- 180 * 0 | 10 | zero 181 * 0 | 12 | zero 182 * 10 | 0 | doesn't exist 183 * 10 | 12 | first 10 bytes are copied from source header, last 2 are zero 184 * 12 | 0 | doesn't exist 185 * 12 | 10 | copied from the first 10 bytes of source header 186 */ 187 bzero(dst, dst_na->virt_hdr_len); 188 if (na->virt_hdr_len && dst_na->virt_hdr_len) 189 memcpy(dst, src, sizeof(struct nm_vnet_hdr)); 190 /* Skip the virtio-net headers. */ 191 src += na->virt_hdr_len; 192 src_len -= na->virt_hdr_len; 193 dst += dst_na->virt_hdr_len; 194 dst_len = dst_na->virt_hdr_len + src_len; 195 196 /* Here it could be dst_len == 0 (which implies src_len == 0), 197 * so we avoid passing a zero length fragment. 198 */ 199 if (dst_len == 0) { 200 ft_p++; 201 src = ft_p->ft_buf; 202 src_len = ft_p->ft_len; 203 dst_len = src_len; 204 } 205 206 if (vh && vh->gso_type != VIRTIO_NET_HDR_GSO_NONE) { 207 u_int gso_bytes = 0; 208 /* Length of the GSO packet header. */ 209 u_int gso_hdr_len = 0; 210 /* Pointer to the GSO packet header. Assume it is in a single fragment. */ 211 uint8_t *gso_hdr = NULL; 212 /* Index of the current segment. */ 213 u_int gso_idx = 0; 214 /* Payload data bytes segmented so far (e.g. TCP data bytes). */ 215 u_int segmented_bytes = 0; 216 /* Length of the IP header (20 if IPv4, 40 if IPv6). */ 217 u_int iphlen = 0; 218 /* Is this a TCP or an UDP GSO packet? */ 219 u_int tcp = ((vh->gso_type & ~VIRTIO_NET_HDR_GSO_ECN) 220 == VIRTIO_NET_HDR_GSO_UDP) ? 0 : 1; 221 222 /* Segment the GSO packet contained into the input slots (frags). */ 223 while (ft_p != ft_end) { 224 size_t copy; 225 226 /* Grab the GSO header if we don't have it. */ 227 if (!gso_hdr) { 228 uint16_t ethertype; 229 230 gso_hdr = src; 231 232 /* Look at the 'Ethertype' field to see if this packet 233 * is IPv4 or IPv6. 234 */ 235 ethertype = be16toh(*((uint16_t *)(gso_hdr + 12))); 236 if (ethertype == 0x0800) 237 iphlen = 20; 238 else /* if (ethertype == 0x86DD) */ 239 iphlen = 40; 240 ND(3, "type=%04x", ethertype); 241 242 /* Compute gso_hdr_len. For TCP we need to read the 243 * content of the 'Data Offset' field. 244 */ 245 if (tcp) { 246 struct nm_tcphdr *tcph = 247 (struct nm_tcphdr *)&gso_hdr[14+iphlen]; 248 249 gso_hdr_len = 14 + iphlen + 4*(tcph->doff >> 4); 250 } else 251 gso_hdr_len = 14 + iphlen + 8; /* UDP */ 252 253 ND(3, "gso_hdr_len %u gso_mtu %d", gso_hdr_len, 254 dst_na->mfs); 255 256 /* Advance source pointers. */ 257 src += gso_hdr_len; 258 src_len -= gso_hdr_len; 259 if (src_len == 0) { 260 ft_p++; 261 if (ft_p == ft_end) 262 break; 263 src = ft_p->ft_buf; 264 src_len = ft_p->ft_len; 265 continue; 266 } 267 } 268 269 /* Fill in the header of the current segment. */ 270 if (gso_bytes == 0) { 271 memcpy(dst, gso_hdr, gso_hdr_len); 272 gso_bytes = gso_hdr_len; 273 } 274 275 /* Fill in data and update source and dest pointers. */ 276 copy = src_len; 277 if (gso_bytes + copy > dst_na->mfs) 278 copy = dst_na->mfs - gso_bytes; 279 memcpy(dst + gso_bytes, src, copy); 280 gso_bytes += copy; 281 src += copy; 282 src_len -= copy; 283 284 /* A segment is complete or we have processed all the 285 the GSO payload bytes. */ 286 if (gso_bytes >= dst_na->mfs || 287 (src_len == 0 && ft_p + 1 == ft_end)) { 288 /* After raw segmentation, we must fix some header 289 * fields and compute checksums, in a protocol dependent 290 * way. */ 291 gso_fix_segment(dst, gso_bytes, gso_idx, 292 segmented_bytes, 293 src_len == 0 && ft_p + 1 == ft_end, 294 tcp, iphlen); 295 296 ND("frame %u completed with %d bytes", gso_idx, (int)gso_bytes); 297 slot->len = gso_bytes; 298 slot->flags = 0; 299 segmented_bytes += gso_bytes - gso_hdr_len; 300 301 dst_slots++; 302 303 /* Next destination slot. */ 304 *j = nm_next(*j, lim); 305 slot = &ring->slot[*j]; 306 dst = BDG_NMB(&dst_na->up, slot); 307 308 gso_bytes = 0; 309 gso_idx++; 310 } 311 312 /* Next input slot. */ 313 if (src_len == 0) { 314 ft_p++; 315 if (ft_p == ft_end) 316 break; 317 src = ft_p->ft_buf; 318 src_len = ft_p->ft_len; 319 } 320 } 321 ND(3, "%d bytes segmented", segmented_bytes); 322 323 } else { 324 /* Address of a checksum field into a destination slot. */ 325 uint16_t *check = NULL; 326 /* Accumulator for an unfolded checksum. */ 327 rawsum_t csum = 0; 328 329 /* Process a non-GSO packet. */ 330 331 /* Init 'check' if necessary. */ 332 if (vh && (vh->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)) { 333 if (unlikely(vh->csum_offset + vh->csum_start > src_len)) 334 D("invalid checksum request"); 335 else 336 check = (uint16_t *)(dst + vh->csum_start + 337 vh->csum_offset); 338 } 339 340 while (ft_p != ft_end) { 341 /* Init/update the packet checksum if needed. */ 342 if (vh && (vh->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)) { 343 if (!dst_slots) 344 csum = nm_csum_raw(src + vh->csum_start, 345 src_len - vh->csum_start, 0); 346 else 347 csum = nm_csum_raw(src, src_len, csum); 348 } 349 350 /* Round to a multiple of 64 */ 351 src_len = (src_len + 63) & ~63; 352 353 if (ft_p->ft_flags & NS_INDIRECT) { 354 if (copyin(src, dst, src_len)) { 355 /* Invalid user pointer, pretend len is 0. */ 356 dst_len = 0; 357 } 358 } else { 359 memcpy(dst, src, (int)src_len); 360 } 361 slot->len = dst_len; 362 363 dst_slots++; 364 365 /* Next destination slot. */ 366 *j = nm_next(*j, lim); 367 slot = &ring->slot[*j]; 368 dst = BDG_NMB(&dst_na->up, slot); 369 370 /* Next source slot. */ 371 ft_p++; 372 src = ft_p->ft_buf; 373 dst_len = src_len = ft_p->ft_len; 374 375 } 376 377 /* Finalize (fold) the checksum if needed. */ 378 if (check && vh && (vh->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM)) { 379 *check = nm_csum_fold(csum); 380 } 381 ND(3, "using %u dst_slots", dst_slots); 382 383 /* A second pass on the desitations slots to set the slot flags, 384 * using the right number of destination slots. 385 */ 386 while (j_start != *j) { 387 slot = &ring->slot[j_start]; 388 slot->flags = (dst_slots << 8)| NS_MOREFRAG; 389 j_start = nm_next(j_start, lim); 390 } 391 /* Clear NS_MOREFRAG flag on last entry. */ 392 slot->flags = (dst_slots << 8); 393 } 394 395 /* Update howmany. */ 396 if (unlikely(dst_slots > *howmany)) { 397 dst_slots = *howmany; 398 D("Slot allocation error: Should never happen"); 399 } 400 *howmany -= dst_slots; 401 } 402