1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * drivers/net/veth.c 4 * 5 * Copyright (C) 2007 OpenVZ http://openvz.org, SWsoft Inc 6 * 7 * Author: Pavel Emelianov <xemul@openvz.org> 8 * Ethtool interface from: Eric W. Biederman <ebiederm@xmission.com> 9 * 10 */ 11 12 #include <linux/netdevice.h> 13 #include <linux/slab.h> 14 #include <linux/ethtool.h> 15 #include <linux/etherdevice.h> 16 #include <linux/u64_stats_sync.h> 17 18 #include <net/rtnetlink.h> 19 #include <net/dst.h> 20 #include <net/xfrm.h> 21 #include <net/xdp.h> 22 #include <linux/veth.h> 23 #include <linux/module.h> 24 #include <linux/bpf.h> 25 #include <linux/filter.h> 26 #include <linux/ptr_ring.h> 27 #include <linux/bpf_trace.h> 28 #include <linux/net_tstamp.h> 29 30 #define DRV_NAME "veth" 31 #define DRV_VERSION "1.0" 32 33 #define VETH_XDP_FLAG BIT(0) 34 #define VETH_RING_SIZE 256 35 #define VETH_XDP_HEADROOM (XDP_PACKET_HEADROOM + NET_IP_ALIGN) 36 37 /* Separating two types of XDP xmit */ 38 #define VETH_XDP_TX BIT(0) 39 #define VETH_XDP_REDIR BIT(1) 40 41 struct veth_rq_stats { 42 u64 xdp_packets; 43 u64 xdp_bytes; 44 u64 xdp_drops; 45 struct u64_stats_sync syncp; 46 }; 47 48 struct veth_rq { 49 struct napi_struct xdp_napi; 50 struct net_device *dev; 51 struct bpf_prog __rcu *xdp_prog; 52 struct xdp_mem_info xdp_mem; 53 struct veth_rq_stats stats; 54 bool rx_notify_masked; 55 struct ptr_ring xdp_ring; 56 struct xdp_rxq_info xdp_rxq; 57 }; 58 59 struct veth_priv { 60 struct net_device __rcu *peer; 61 atomic64_t dropped; 62 struct bpf_prog *_xdp_prog; 63 struct veth_rq *rq; 64 unsigned int requested_headroom; 65 }; 66 67 /* 68 * ethtool interface 69 */ 70 71 struct veth_q_stat_desc { 72 char desc[ETH_GSTRING_LEN]; 73 size_t offset; 74 }; 75 76 #define VETH_RQ_STAT(m) offsetof(struct veth_rq_stats, m) 77 78 static const struct veth_q_stat_desc veth_rq_stats_desc[] = { 79 { "xdp_packets", VETH_RQ_STAT(xdp_packets) }, 80 { "xdp_bytes", VETH_RQ_STAT(xdp_bytes) }, 81 { "xdp_drops", VETH_RQ_STAT(xdp_drops) }, 82 }; 83 84 #define VETH_RQ_STATS_LEN ARRAY_SIZE(veth_rq_stats_desc) 85 86 static struct { 87 const char string[ETH_GSTRING_LEN]; 88 } ethtool_stats_keys[] = { 89 { "peer_ifindex" }, 90 }; 91 92 static int veth_get_link_ksettings(struct net_device *dev, 93 struct ethtool_link_ksettings *cmd) 94 { 95 cmd->base.speed = SPEED_10000; 96 cmd->base.duplex = DUPLEX_FULL; 97 cmd->base.port = PORT_TP; 98 cmd->base.autoneg = AUTONEG_DISABLE; 99 return 0; 100 } 101 102 static void veth_get_drvinfo(struct net_device *dev, struct ethtool_drvinfo *info) 103 { 104 strlcpy(info->driver, DRV_NAME, sizeof(info->driver)); 105 strlcpy(info->version, DRV_VERSION, sizeof(info->version)); 106 } 107 108 static void veth_get_strings(struct net_device *dev, u32 stringset, u8 *buf) 109 { 110 char *p = (char *)buf; 111 int i, j; 112 113 switch(stringset) { 114 case ETH_SS_STATS: 115 memcpy(p, ðtool_stats_keys, sizeof(ethtool_stats_keys)); 116 p += sizeof(ethtool_stats_keys); 117 for (i = 0; i < dev->real_num_rx_queues; i++) { 118 for (j = 0; j < VETH_RQ_STATS_LEN; j++) { 119 snprintf(p, ETH_GSTRING_LEN, 120 "rx_queue_%u_%.11s", 121 i, veth_rq_stats_desc[j].desc); 122 p += ETH_GSTRING_LEN; 123 } 124 } 125 break; 126 } 127 } 128 129 static int veth_get_sset_count(struct net_device *dev, int sset) 130 { 131 switch (sset) { 132 case ETH_SS_STATS: 133 return ARRAY_SIZE(ethtool_stats_keys) + 134 VETH_RQ_STATS_LEN * dev->real_num_rx_queues; 135 default: 136 return -EOPNOTSUPP; 137 } 138 } 139 140 static void veth_get_ethtool_stats(struct net_device *dev, 141 struct ethtool_stats *stats, u64 *data) 142 { 143 struct veth_priv *priv = netdev_priv(dev); 144 struct net_device *peer = rtnl_dereference(priv->peer); 145 int i, j, idx; 146 147 data[0] = peer ? peer->ifindex : 0; 148 idx = 1; 149 for (i = 0; i < dev->real_num_rx_queues; i++) { 150 const struct veth_rq_stats *rq_stats = &priv->rq[i].stats; 151 const void *stats_base = (void *)rq_stats; 152 unsigned int start; 153 size_t offset; 154 155 do { 156 start = u64_stats_fetch_begin_irq(&rq_stats->syncp); 157 for (j = 0; j < VETH_RQ_STATS_LEN; j++) { 158 offset = veth_rq_stats_desc[j].offset; 159 data[idx + j] = *(u64 *)(stats_base + offset); 160 } 161 } while (u64_stats_fetch_retry_irq(&rq_stats->syncp, start)); 162 idx += VETH_RQ_STATS_LEN; 163 } 164 } 165 166 static const struct ethtool_ops veth_ethtool_ops = { 167 .get_drvinfo = veth_get_drvinfo, 168 .get_link = ethtool_op_get_link, 169 .get_strings = veth_get_strings, 170 .get_sset_count = veth_get_sset_count, 171 .get_ethtool_stats = veth_get_ethtool_stats, 172 .get_link_ksettings = veth_get_link_ksettings, 173 .get_ts_info = ethtool_op_get_ts_info, 174 }; 175 176 /* general routines */ 177 178 static bool veth_is_xdp_frame(void *ptr) 179 { 180 return (unsigned long)ptr & VETH_XDP_FLAG; 181 } 182 183 static void *veth_ptr_to_xdp(void *ptr) 184 { 185 return (void *)((unsigned long)ptr & ~VETH_XDP_FLAG); 186 } 187 188 static void *veth_xdp_to_ptr(void *ptr) 189 { 190 return (void *)((unsigned long)ptr | VETH_XDP_FLAG); 191 } 192 193 static void veth_ptr_free(void *ptr) 194 { 195 if (veth_is_xdp_frame(ptr)) 196 xdp_return_frame(veth_ptr_to_xdp(ptr)); 197 else 198 kfree_skb(ptr); 199 } 200 201 static void __veth_xdp_flush(struct veth_rq *rq) 202 { 203 /* Write ptr_ring before reading rx_notify_masked */ 204 smp_mb(); 205 if (!rq->rx_notify_masked) { 206 rq->rx_notify_masked = true; 207 napi_schedule(&rq->xdp_napi); 208 } 209 } 210 211 static int veth_xdp_rx(struct veth_rq *rq, struct sk_buff *skb) 212 { 213 if (unlikely(ptr_ring_produce(&rq->xdp_ring, skb))) { 214 dev_kfree_skb_any(skb); 215 return NET_RX_DROP; 216 } 217 218 return NET_RX_SUCCESS; 219 } 220 221 static int veth_forward_skb(struct net_device *dev, struct sk_buff *skb, 222 struct veth_rq *rq, bool xdp) 223 { 224 return __dev_forward_skb(dev, skb) ?: xdp ? 225 veth_xdp_rx(rq, skb) : 226 netif_rx(skb); 227 } 228 229 static netdev_tx_t veth_xmit(struct sk_buff *skb, struct net_device *dev) 230 { 231 struct veth_priv *rcv_priv, *priv = netdev_priv(dev); 232 struct veth_rq *rq = NULL; 233 struct net_device *rcv; 234 int length = skb->len; 235 bool rcv_xdp = false; 236 int rxq; 237 238 rcu_read_lock(); 239 rcv = rcu_dereference(priv->peer); 240 if (unlikely(!rcv)) { 241 kfree_skb(skb); 242 goto drop; 243 } 244 245 rcv_priv = netdev_priv(rcv); 246 rxq = skb_get_queue_mapping(skb); 247 if (rxq < rcv->real_num_rx_queues) { 248 rq = &rcv_priv->rq[rxq]; 249 rcv_xdp = rcu_access_pointer(rq->xdp_prog); 250 if (rcv_xdp) 251 skb_record_rx_queue(skb, rxq); 252 } 253 254 skb_tx_timestamp(skb); 255 if (likely(veth_forward_skb(rcv, skb, rq, rcv_xdp) == NET_RX_SUCCESS)) { 256 if (!rcv_xdp) { 257 struct pcpu_lstats *stats = this_cpu_ptr(dev->lstats); 258 259 u64_stats_update_begin(&stats->syncp); 260 stats->bytes += length; 261 stats->packets++; 262 u64_stats_update_end(&stats->syncp); 263 } 264 } else { 265 drop: 266 atomic64_inc(&priv->dropped); 267 } 268 269 if (rcv_xdp) 270 __veth_xdp_flush(rq); 271 272 rcu_read_unlock(); 273 274 return NETDEV_TX_OK; 275 } 276 277 static u64 veth_stats_tx(struct pcpu_lstats *result, struct net_device *dev) 278 { 279 struct veth_priv *priv = netdev_priv(dev); 280 int cpu; 281 282 result->packets = 0; 283 result->bytes = 0; 284 for_each_possible_cpu(cpu) { 285 struct pcpu_lstats *stats = per_cpu_ptr(dev->lstats, cpu); 286 u64 packets, bytes; 287 unsigned int start; 288 289 do { 290 start = u64_stats_fetch_begin_irq(&stats->syncp); 291 packets = stats->packets; 292 bytes = stats->bytes; 293 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 294 result->packets += packets; 295 result->bytes += bytes; 296 } 297 return atomic64_read(&priv->dropped); 298 } 299 300 static void veth_stats_rx(struct veth_rq_stats *result, struct net_device *dev) 301 { 302 struct veth_priv *priv = netdev_priv(dev); 303 int i; 304 305 result->xdp_packets = 0; 306 result->xdp_bytes = 0; 307 result->xdp_drops = 0; 308 for (i = 0; i < dev->num_rx_queues; i++) { 309 struct veth_rq_stats *stats = &priv->rq[i].stats; 310 u64 packets, bytes, drops; 311 unsigned int start; 312 313 do { 314 start = u64_stats_fetch_begin_irq(&stats->syncp); 315 packets = stats->xdp_packets; 316 bytes = stats->xdp_bytes; 317 drops = stats->xdp_drops; 318 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 319 result->xdp_packets += packets; 320 result->xdp_bytes += bytes; 321 result->xdp_drops += drops; 322 } 323 } 324 325 static void veth_get_stats64(struct net_device *dev, 326 struct rtnl_link_stats64 *tot) 327 { 328 struct veth_priv *priv = netdev_priv(dev); 329 struct net_device *peer; 330 struct veth_rq_stats rx; 331 struct pcpu_lstats tx; 332 333 tot->tx_dropped = veth_stats_tx(&tx, dev); 334 tot->tx_bytes = tx.bytes; 335 tot->tx_packets = tx.packets; 336 337 veth_stats_rx(&rx, dev); 338 tot->rx_dropped = rx.xdp_drops; 339 tot->rx_bytes = rx.xdp_bytes; 340 tot->rx_packets = rx.xdp_packets; 341 342 rcu_read_lock(); 343 peer = rcu_dereference(priv->peer); 344 if (peer) { 345 tot->rx_dropped += veth_stats_tx(&tx, peer); 346 tot->rx_bytes += tx.bytes; 347 tot->rx_packets += tx.packets; 348 349 veth_stats_rx(&rx, peer); 350 tot->tx_bytes += rx.xdp_bytes; 351 tot->tx_packets += rx.xdp_packets; 352 } 353 rcu_read_unlock(); 354 } 355 356 /* fake multicast ability */ 357 static void veth_set_multicast_list(struct net_device *dev) 358 { 359 } 360 361 static struct sk_buff *veth_build_skb(void *head, int headroom, int len, 362 int buflen) 363 { 364 struct sk_buff *skb; 365 366 if (!buflen) { 367 buflen = SKB_DATA_ALIGN(headroom + len) + 368 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 369 } 370 skb = build_skb(head, buflen); 371 if (!skb) 372 return NULL; 373 374 skb_reserve(skb, headroom); 375 skb_put(skb, len); 376 377 return skb; 378 } 379 380 static int veth_select_rxq(struct net_device *dev) 381 { 382 return smp_processor_id() % dev->real_num_rx_queues; 383 } 384 385 static int veth_xdp_xmit(struct net_device *dev, int n, 386 struct xdp_frame **frames, u32 flags) 387 { 388 struct veth_priv *rcv_priv, *priv = netdev_priv(dev); 389 struct net_device *rcv; 390 int i, ret, drops = n; 391 unsigned int max_len; 392 struct veth_rq *rq; 393 394 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) { 395 ret = -EINVAL; 396 goto drop; 397 } 398 399 rcv = rcu_dereference(priv->peer); 400 if (unlikely(!rcv)) { 401 ret = -ENXIO; 402 goto drop; 403 } 404 405 rcv_priv = netdev_priv(rcv); 406 rq = &rcv_priv->rq[veth_select_rxq(rcv)]; 407 /* Non-NULL xdp_prog ensures that xdp_ring is initialized on receive 408 * side. This means an XDP program is loaded on the peer and the peer 409 * device is up. 410 */ 411 if (!rcu_access_pointer(rq->xdp_prog)) { 412 ret = -ENXIO; 413 goto drop; 414 } 415 416 drops = 0; 417 max_len = rcv->mtu + rcv->hard_header_len + VLAN_HLEN; 418 419 spin_lock(&rq->xdp_ring.producer_lock); 420 for (i = 0; i < n; i++) { 421 struct xdp_frame *frame = frames[i]; 422 void *ptr = veth_xdp_to_ptr(frame); 423 424 if (unlikely(frame->len > max_len || 425 __ptr_ring_produce(&rq->xdp_ring, ptr))) { 426 xdp_return_frame_rx_napi(frame); 427 drops++; 428 } 429 } 430 spin_unlock(&rq->xdp_ring.producer_lock); 431 432 if (flags & XDP_XMIT_FLUSH) 433 __veth_xdp_flush(rq); 434 435 if (likely(!drops)) 436 return n; 437 438 ret = n - drops; 439 drop: 440 atomic64_add(drops, &priv->dropped); 441 442 return ret; 443 } 444 445 static void veth_xdp_flush(struct net_device *dev) 446 { 447 struct veth_priv *rcv_priv, *priv = netdev_priv(dev); 448 struct net_device *rcv; 449 struct veth_rq *rq; 450 451 rcu_read_lock(); 452 rcv = rcu_dereference(priv->peer); 453 if (unlikely(!rcv)) 454 goto out; 455 456 rcv_priv = netdev_priv(rcv); 457 rq = &rcv_priv->rq[veth_select_rxq(rcv)]; 458 /* xdp_ring is initialized on receive side? */ 459 if (unlikely(!rcu_access_pointer(rq->xdp_prog))) 460 goto out; 461 462 __veth_xdp_flush(rq); 463 out: 464 rcu_read_unlock(); 465 } 466 467 static int veth_xdp_tx(struct net_device *dev, struct xdp_buff *xdp) 468 { 469 struct xdp_frame *frame = convert_to_xdp_frame(xdp); 470 471 if (unlikely(!frame)) 472 return -EOVERFLOW; 473 474 return veth_xdp_xmit(dev, 1, &frame, 0); 475 } 476 477 static struct sk_buff *veth_xdp_rcv_one(struct veth_rq *rq, 478 struct xdp_frame *frame, 479 unsigned int *xdp_xmit) 480 { 481 void *hard_start = frame->data - frame->headroom; 482 void *head = hard_start - sizeof(struct xdp_frame); 483 int len = frame->len, delta = 0; 484 struct xdp_frame orig_frame; 485 struct bpf_prog *xdp_prog; 486 unsigned int headroom; 487 struct sk_buff *skb; 488 489 rcu_read_lock(); 490 xdp_prog = rcu_dereference(rq->xdp_prog); 491 if (likely(xdp_prog)) { 492 struct xdp_buff xdp; 493 u32 act; 494 495 xdp.data_hard_start = hard_start; 496 xdp.data = frame->data; 497 xdp.data_end = frame->data + frame->len; 498 xdp.data_meta = frame->data - frame->metasize; 499 xdp.rxq = &rq->xdp_rxq; 500 501 act = bpf_prog_run_xdp(xdp_prog, &xdp); 502 503 switch (act) { 504 case XDP_PASS: 505 delta = frame->data - xdp.data; 506 len = xdp.data_end - xdp.data; 507 break; 508 case XDP_TX: 509 orig_frame = *frame; 510 xdp.data_hard_start = head; 511 xdp.rxq->mem = frame->mem; 512 if (unlikely(veth_xdp_tx(rq->dev, &xdp) < 0)) { 513 trace_xdp_exception(rq->dev, xdp_prog, act); 514 frame = &orig_frame; 515 goto err_xdp; 516 } 517 *xdp_xmit |= VETH_XDP_TX; 518 rcu_read_unlock(); 519 goto xdp_xmit; 520 case XDP_REDIRECT: 521 orig_frame = *frame; 522 xdp.data_hard_start = head; 523 xdp.rxq->mem = frame->mem; 524 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) { 525 frame = &orig_frame; 526 goto err_xdp; 527 } 528 *xdp_xmit |= VETH_XDP_REDIR; 529 rcu_read_unlock(); 530 goto xdp_xmit; 531 default: 532 bpf_warn_invalid_xdp_action(act); 533 /* fall through */ 534 case XDP_ABORTED: 535 trace_xdp_exception(rq->dev, xdp_prog, act); 536 /* fall through */ 537 case XDP_DROP: 538 goto err_xdp; 539 } 540 } 541 rcu_read_unlock(); 542 543 headroom = sizeof(struct xdp_frame) + frame->headroom - delta; 544 skb = veth_build_skb(head, headroom, len, 0); 545 if (!skb) { 546 xdp_return_frame(frame); 547 goto err; 548 } 549 550 xdp_scrub_frame(frame); 551 skb->protocol = eth_type_trans(skb, rq->dev); 552 err: 553 return skb; 554 err_xdp: 555 rcu_read_unlock(); 556 xdp_return_frame(frame); 557 xdp_xmit: 558 return NULL; 559 } 560 561 static struct sk_buff *veth_xdp_rcv_skb(struct veth_rq *rq, struct sk_buff *skb, 562 unsigned int *xdp_xmit) 563 { 564 u32 pktlen, headroom, act, metalen; 565 void *orig_data, *orig_data_end; 566 struct bpf_prog *xdp_prog; 567 int mac_len, delta, off; 568 struct xdp_buff xdp; 569 570 skb_orphan(skb); 571 572 rcu_read_lock(); 573 xdp_prog = rcu_dereference(rq->xdp_prog); 574 if (unlikely(!xdp_prog)) { 575 rcu_read_unlock(); 576 goto out; 577 } 578 579 mac_len = skb->data - skb_mac_header(skb); 580 pktlen = skb->len + mac_len; 581 headroom = skb_headroom(skb) - mac_len; 582 583 if (skb_shared(skb) || skb_head_is_locked(skb) || 584 skb_is_nonlinear(skb) || headroom < XDP_PACKET_HEADROOM) { 585 struct sk_buff *nskb; 586 int size, head_off; 587 void *head, *start; 588 struct page *page; 589 590 size = SKB_DATA_ALIGN(VETH_XDP_HEADROOM + pktlen) + 591 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 592 if (size > PAGE_SIZE) 593 goto drop; 594 595 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN); 596 if (!page) 597 goto drop; 598 599 head = page_address(page); 600 start = head + VETH_XDP_HEADROOM; 601 if (skb_copy_bits(skb, -mac_len, start, pktlen)) { 602 page_frag_free(head); 603 goto drop; 604 } 605 606 nskb = veth_build_skb(head, 607 VETH_XDP_HEADROOM + mac_len, skb->len, 608 PAGE_SIZE); 609 if (!nskb) { 610 page_frag_free(head); 611 goto drop; 612 } 613 614 skb_copy_header(nskb, skb); 615 head_off = skb_headroom(nskb) - skb_headroom(skb); 616 skb_headers_offset_update(nskb, head_off); 617 consume_skb(skb); 618 skb = nskb; 619 } 620 621 xdp.data_hard_start = skb->head; 622 xdp.data = skb_mac_header(skb); 623 xdp.data_end = xdp.data + pktlen; 624 xdp.data_meta = xdp.data; 625 xdp.rxq = &rq->xdp_rxq; 626 orig_data = xdp.data; 627 orig_data_end = xdp.data_end; 628 629 act = bpf_prog_run_xdp(xdp_prog, &xdp); 630 631 switch (act) { 632 case XDP_PASS: 633 break; 634 case XDP_TX: 635 get_page(virt_to_page(xdp.data)); 636 consume_skb(skb); 637 xdp.rxq->mem = rq->xdp_mem; 638 if (unlikely(veth_xdp_tx(rq->dev, &xdp) < 0)) { 639 trace_xdp_exception(rq->dev, xdp_prog, act); 640 goto err_xdp; 641 } 642 *xdp_xmit |= VETH_XDP_TX; 643 rcu_read_unlock(); 644 goto xdp_xmit; 645 case XDP_REDIRECT: 646 get_page(virt_to_page(xdp.data)); 647 consume_skb(skb); 648 xdp.rxq->mem = rq->xdp_mem; 649 if (xdp_do_redirect(rq->dev, &xdp, xdp_prog)) 650 goto err_xdp; 651 *xdp_xmit |= VETH_XDP_REDIR; 652 rcu_read_unlock(); 653 goto xdp_xmit; 654 default: 655 bpf_warn_invalid_xdp_action(act); 656 /* fall through */ 657 case XDP_ABORTED: 658 trace_xdp_exception(rq->dev, xdp_prog, act); 659 /* fall through */ 660 case XDP_DROP: 661 goto drop; 662 } 663 rcu_read_unlock(); 664 665 delta = orig_data - xdp.data; 666 off = mac_len + delta; 667 if (off > 0) 668 __skb_push(skb, off); 669 else if (off < 0) 670 __skb_pull(skb, -off); 671 skb->mac_header -= delta; 672 off = xdp.data_end - orig_data_end; 673 if (off != 0) 674 __skb_put(skb, off); 675 skb->protocol = eth_type_trans(skb, rq->dev); 676 677 metalen = xdp.data - xdp.data_meta; 678 if (metalen) 679 skb_metadata_set(skb, metalen); 680 out: 681 return skb; 682 drop: 683 rcu_read_unlock(); 684 kfree_skb(skb); 685 return NULL; 686 err_xdp: 687 rcu_read_unlock(); 688 page_frag_free(xdp.data); 689 xdp_xmit: 690 return NULL; 691 } 692 693 static int veth_xdp_rcv(struct veth_rq *rq, int budget, unsigned int *xdp_xmit) 694 { 695 int i, done = 0, drops = 0, bytes = 0; 696 697 for (i = 0; i < budget; i++) { 698 void *ptr = __ptr_ring_consume(&rq->xdp_ring); 699 unsigned int xdp_xmit_one = 0; 700 struct sk_buff *skb; 701 702 if (!ptr) 703 break; 704 705 if (veth_is_xdp_frame(ptr)) { 706 struct xdp_frame *frame = veth_ptr_to_xdp(ptr); 707 708 bytes += frame->len; 709 skb = veth_xdp_rcv_one(rq, frame, &xdp_xmit_one); 710 } else { 711 skb = ptr; 712 bytes += skb->len; 713 skb = veth_xdp_rcv_skb(rq, skb, &xdp_xmit_one); 714 } 715 *xdp_xmit |= xdp_xmit_one; 716 717 if (skb) 718 napi_gro_receive(&rq->xdp_napi, skb); 719 else if (!xdp_xmit_one) 720 drops++; 721 722 done++; 723 } 724 725 u64_stats_update_begin(&rq->stats.syncp); 726 rq->stats.xdp_packets += done; 727 rq->stats.xdp_bytes += bytes; 728 rq->stats.xdp_drops += drops; 729 u64_stats_update_end(&rq->stats.syncp); 730 731 return done; 732 } 733 734 static int veth_poll(struct napi_struct *napi, int budget) 735 { 736 struct veth_rq *rq = 737 container_of(napi, struct veth_rq, xdp_napi); 738 unsigned int xdp_xmit = 0; 739 int done; 740 741 xdp_set_return_frame_no_direct(); 742 done = veth_xdp_rcv(rq, budget, &xdp_xmit); 743 744 if (done < budget && napi_complete_done(napi, done)) { 745 /* Write rx_notify_masked before reading ptr_ring */ 746 smp_store_mb(rq->rx_notify_masked, false); 747 if (unlikely(!__ptr_ring_empty(&rq->xdp_ring))) { 748 rq->rx_notify_masked = true; 749 napi_schedule(&rq->xdp_napi); 750 } 751 } 752 753 if (xdp_xmit & VETH_XDP_TX) 754 veth_xdp_flush(rq->dev); 755 if (xdp_xmit & VETH_XDP_REDIR) 756 xdp_do_flush_map(); 757 xdp_clear_return_frame_no_direct(); 758 759 return done; 760 } 761 762 static int veth_napi_add(struct net_device *dev) 763 { 764 struct veth_priv *priv = netdev_priv(dev); 765 int err, i; 766 767 for (i = 0; i < dev->real_num_rx_queues; i++) { 768 struct veth_rq *rq = &priv->rq[i]; 769 770 err = ptr_ring_init(&rq->xdp_ring, VETH_RING_SIZE, GFP_KERNEL); 771 if (err) 772 goto err_xdp_ring; 773 } 774 775 for (i = 0; i < dev->real_num_rx_queues; i++) { 776 struct veth_rq *rq = &priv->rq[i]; 777 778 netif_napi_add(dev, &rq->xdp_napi, veth_poll, NAPI_POLL_WEIGHT); 779 napi_enable(&rq->xdp_napi); 780 } 781 782 return 0; 783 err_xdp_ring: 784 for (i--; i >= 0; i--) 785 ptr_ring_cleanup(&priv->rq[i].xdp_ring, veth_ptr_free); 786 787 return err; 788 } 789 790 static void veth_napi_del(struct net_device *dev) 791 { 792 struct veth_priv *priv = netdev_priv(dev); 793 int i; 794 795 for (i = 0; i < dev->real_num_rx_queues; i++) { 796 struct veth_rq *rq = &priv->rq[i]; 797 798 napi_disable(&rq->xdp_napi); 799 napi_hash_del(&rq->xdp_napi); 800 } 801 synchronize_net(); 802 803 for (i = 0; i < dev->real_num_rx_queues; i++) { 804 struct veth_rq *rq = &priv->rq[i]; 805 806 netif_napi_del(&rq->xdp_napi); 807 rq->rx_notify_masked = false; 808 ptr_ring_cleanup(&rq->xdp_ring, veth_ptr_free); 809 } 810 } 811 812 static int veth_enable_xdp(struct net_device *dev) 813 { 814 struct veth_priv *priv = netdev_priv(dev); 815 int err, i; 816 817 if (!xdp_rxq_info_is_reg(&priv->rq[0].xdp_rxq)) { 818 for (i = 0; i < dev->real_num_rx_queues; i++) { 819 struct veth_rq *rq = &priv->rq[i]; 820 821 err = xdp_rxq_info_reg(&rq->xdp_rxq, dev, i); 822 if (err < 0) 823 goto err_rxq_reg; 824 825 err = xdp_rxq_info_reg_mem_model(&rq->xdp_rxq, 826 MEM_TYPE_PAGE_SHARED, 827 NULL); 828 if (err < 0) 829 goto err_reg_mem; 830 831 /* Save original mem info as it can be overwritten */ 832 rq->xdp_mem = rq->xdp_rxq.mem; 833 } 834 835 err = veth_napi_add(dev); 836 if (err) 837 goto err_rxq_reg; 838 } 839 840 for (i = 0; i < dev->real_num_rx_queues; i++) 841 rcu_assign_pointer(priv->rq[i].xdp_prog, priv->_xdp_prog); 842 843 return 0; 844 err_reg_mem: 845 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq); 846 err_rxq_reg: 847 for (i--; i >= 0; i--) 848 xdp_rxq_info_unreg(&priv->rq[i].xdp_rxq); 849 850 return err; 851 } 852 853 static void veth_disable_xdp(struct net_device *dev) 854 { 855 struct veth_priv *priv = netdev_priv(dev); 856 int i; 857 858 for (i = 0; i < dev->real_num_rx_queues; i++) 859 rcu_assign_pointer(priv->rq[i].xdp_prog, NULL); 860 veth_napi_del(dev); 861 for (i = 0; i < dev->real_num_rx_queues; i++) { 862 struct veth_rq *rq = &priv->rq[i]; 863 864 rq->xdp_rxq.mem = rq->xdp_mem; 865 xdp_rxq_info_unreg(&rq->xdp_rxq); 866 } 867 } 868 869 static int veth_open(struct net_device *dev) 870 { 871 struct veth_priv *priv = netdev_priv(dev); 872 struct net_device *peer = rtnl_dereference(priv->peer); 873 int err; 874 875 if (!peer) 876 return -ENOTCONN; 877 878 if (priv->_xdp_prog) { 879 err = veth_enable_xdp(dev); 880 if (err) 881 return err; 882 } 883 884 if (peer->flags & IFF_UP) { 885 netif_carrier_on(dev); 886 netif_carrier_on(peer); 887 } 888 889 return 0; 890 } 891 892 static int veth_close(struct net_device *dev) 893 { 894 struct veth_priv *priv = netdev_priv(dev); 895 struct net_device *peer = rtnl_dereference(priv->peer); 896 897 netif_carrier_off(dev); 898 if (peer) 899 netif_carrier_off(peer); 900 901 if (priv->_xdp_prog) 902 veth_disable_xdp(dev); 903 904 return 0; 905 } 906 907 static int is_valid_veth_mtu(int mtu) 908 { 909 return mtu >= ETH_MIN_MTU && mtu <= ETH_MAX_MTU; 910 } 911 912 static int veth_alloc_queues(struct net_device *dev) 913 { 914 struct veth_priv *priv = netdev_priv(dev); 915 int i; 916 917 priv->rq = kcalloc(dev->num_rx_queues, sizeof(*priv->rq), GFP_KERNEL); 918 if (!priv->rq) 919 return -ENOMEM; 920 921 for (i = 0; i < dev->num_rx_queues; i++) { 922 priv->rq[i].dev = dev; 923 u64_stats_init(&priv->rq[i].stats.syncp); 924 } 925 926 return 0; 927 } 928 929 static void veth_free_queues(struct net_device *dev) 930 { 931 struct veth_priv *priv = netdev_priv(dev); 932 933 kfree(priv->rq); 934 } 935 936 static int veth_dev_init(struct net_device *dev) 937 { 938 int err; 939 940 dev->lstats = netdev_alloc_pcpu_stats(struct pcpu_lstats); 941 if (!dev->lstats) 942 return -ENOMEM; 943 944 err = veth_alloc_queues(dev); 945 if (err) { 946 free_percpu(dev->lstats); 947 return err; 948 } 949 950 return 0; 951 } 952 953 static void veth_dev_free(struct net_device *dev) 954 { 955 veth_free_queues(dev); 956 free_percpu(dev->lstats); 957 } 958 959 #ifdef CONFIG_NET_POLL_CONTROLLER 960 static void veth_poll_controller(struct net_device *dev) 961 { 962 /* veth only receives frames when its peer sends one 963 * Since it has nothing to do with disabling irqs, we are guaranteed 964 * never to have pending data when we poll for it so 965 * there is nothing to do here. 966 * 967 * We need this though so netpoll recognizes us as an interface that 968 * supports polling, which enables bridge devices in virt setups to 969 * still use netconsole 970 */ 971 } 972 #endif /* CONFIG_NET_POLL_CONTROLLER */ 973 974 static int veth_get_iflink(const struct net_device *dev) 975 { 976 struct veth_priv *priv = netdev_priv(dev); 977 struct net_device *peer; 978 int iflink; 979 980 rcu_read_lock(); 981 peer = rcu_dereference(priv->peer); 982 iflink = peer ? peer->ifindex : 0; 983 rcu_read_unlock(); 984 985 return iflink; 986 } 987 988 static netdev_features_t veth_fix_features(struct net_device *dev, 989 netdev_features_t features) 990 { 991 struct veth_priv *priv = netdev_priv(dev); 992 struct net_device *peer; 993 994 peer = rtnl_dereference(priv->peer); 995 if (peer) { 996 struct veth_priv *peer_priv = netdev_priv(peer); 997 998 if (peer_priv->_xdp_prog) 999 features &= ~NETIF_F_GSO_SOFTWARE; 1000 } 1001 1002 return features; 1003 } 1004 1005 static void veth_set_rx_headroom(struct net_device *dev, int new_hr) 1006 { 1007 struct veth_priv *peer_priv, *priv = netdev_priv(dev); 1008 struct net_device *peer; 1009 1010 if (new_hr < 0) 1011 new_hr = 0; 1012 1013 rcu_read_lock(); 1014 peer = rcu_dereference(priv->peer); 1015 if (unlikely(!peer)) 1016 goto out; 1017 1018 peer_priv = netdev_priv(peer); 1019 priv->requested_headroom = new_hr; 1020 new_hr = max(priv->requested_headroom, peer_priv->requested_headroom); 1021 dev->needed_headroom = new_hr; 1022 peer->needed_headroom = new_hr; 1023 1024 out: 1025 rcu_read_unlock(); 1026 } 1027 1028 static int veth_xdp_set(struct net_device *dev, struct bpf_prog *prog, 1029 struct netlink_ext_ack *extack) 1030 { 1031 struct veth_priv *priv = netdev_priv(dev); 1032 struct bpf_prog *old_prog; 1033 struct net_device *peer; 1034 unsigned int max_mtu; 1035 int err; 1036 1037 old_prog = priv->_xdp_prog; 1038 priv->_xdp_prog = prog; 1039 peer = rtnl_dereference(priv->peer); 1040 1041 if (prog) { 1042 if (!peer) { 1043 NL_SET_ERR_MSG_MOD(extack, "Cannot set XDP when peer is detached"); 1044 err = -ENOTCONN; 1045 goto err; 1046 } 1047 1048 max_mtu = PAGE_SIZE - VETH_XDP_HEADROOM - 1049 peer->hard_header_len - 1050 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)); 1051 if (peer->mtu > max_mtu) { 1052 NL_SET_ERR_MSG_MOD(extack, "Peer MTU is too large to set XDP"); 1053 err = -ERANGE; 1054 goto err; 1055 } 1056 1057 if (dev->real_num_rx_queues < peer->real_num_tx_queues) { 1058 NL_SET_ERR_MSG_MOD(extack, "XDP expects number of rx queues not less than peer tx queues"); 1059 err = -ENOSPC; 1060 goto err; 1061 } 1062 1063 if (dev->flags & IFF_UP) { 1064 err = veth_enable_xdp(dev); 1065 if (err) { 1066 NL_SET_ERR_MSG_MOD(extack, "Setup for XDP failed"); 1067 goto err; 1068 } 1069 } 1070 1071 if (!old_prog) { 1072 peer->hw_features &= ~NETIF_F_GSO_SOFTWARE; 1073 peer->max_mtu = max_mtu; 1074 } 1075 } 1076 1077 if (old_prog) { 1078 if (!prog) { 1079 if (dev->flags & IFF_UP) 1080 veth_disable_xdp(dev); 1081 1082 if (peer) { 1083 peer->hw_features |= NETIF_F_GSO_SOFTWARE; 1084 peer->max_mtu = ETH_MAX_MTU; 1085 } 1086 } 1087 bpf_prog_put(old_prog); 1088 } 1089 1090 if ((!!old_prog ^ !!prog) && peer) 1091 netdev_update_features(peer); 1092 1093 return 0; 1094 err: 1095 priv->_xdp_prog = old_prog; 1096 1097 return err; 1098 } 1099 1100 static u32 veth_xdp_query(struct net_device *dev) 1101 { 1102 struct veth_priv *priv = netdev_priv(dev); 1103 const struct bpf_prog *xdp_prog; 1104 1105 xdp_prog = priv->_xdp_prog; 1106 if (xdp_prog) 1107 return xdp_prog->aux->id; 1108 1109 return 0; 1110 } 1111 1112 static int veth_xdp(struct net_device *dev, struct netdev_bpf *xdp) 1113 { 1114 switch (xdp->command) { 1115 case XDP_SETUP_PROG: 1116 return veth_xdp_set(dev, xdp->prog, xdp->extack); 1117 case XDP_QUERY_PROG: 1118 xdp->prog_id = veth_xdp_query(dev); 1119 return 0; 1120 default: 1121 return -EINVAL; 1122 } 1123 } 1124 1125 static const struct net_device_ops veth_netdev_ops = { 1126 .ndo_init = veth_dev_init, 1127 .ndo_open = veth_open, 1128 .ndo_stop = veth_close, 1129 .ndo_start_xmit = veth_xmit, 1130 .ndo_get_stats64 = veth_get_stats64, 1131 .ndo_set_rx_mode = veth_set_multicast_list, 1132 .ndo_set_mac_address = eth_mac_addr, 1133 #ifdef CONFIG_NET_POLL_CONTROLLER 1134 .ndo_poll_controller = veth_poll_controller, 1135 #endif 1136 .ndo_get_iflink = veth_get_iflink, 1137 .ndo_fix_features = veth_fix_features, 1138 .ndo_features_check = passthru_features_check, 1139 .ndo_set_rx_headroom = veth_set_rx_headroom, 1140 .ndo_bpf = veth_xdp, 1141 .ndo_xdp_xmit = veth_xdp_xmit, 1142 }; 1143 1144 #define VETH_FEATURES (NETIF_F_SG | NETIF_F_FRAGLIST | NETIF_F_HW_CSUM | \ 1145 NETIF_F_RXCSUM | NETIF_F_SCTP_CRC | NETIF_F_HIGHDMA | \ 1146 NETIF_F_GSO_SOFTWARE | NETIF_F_GSO_ENCAP_ALL | \ 1147 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX | \ 1148 NETIF_F_HW_VLAN_STAG_TX | NETIF_F_HW_VLAN_STAG_RX ) 1149 1150 static void veth_setup(struct net_device *dev) 1151 { 1152 ether_setup(dev); 1153 1154 dev->priv_flags &= ~IFF_TX_SKB_SHARING; 1155 dev->priv_flags |= IFF_LIVE_ADDR_CHANGE; 1156 dev->priv_flags |= IFF_NO_QUEUE; 1157 dev->priv_flags |= IFF_PHONY_HEADROOM; 1158 1159 dev->netdev_ops = &veth_netdev_ops; 1160 dev->ethtool_ops = &veth_ethtool_ops; 1161 dev->features |= NETIF_F_LLTX; 1162 dev->features |= VETH_FEATURES; 1163 dev->vlan_features = dev->features & 1164 ~(NETIF_F_HW_VLAN_CTAG_TX | 1165 NETIF_F_HW_VLAN_STAG_TX | 1166 NETIF_F_HW_VLAN_CTAG_RX | 1167 NETIF_F_HW_VLAN_STAG_RX); 1168 dev->needs_free_netdev = true; 1169 dev->priv_destructor = veth_dev_free; 1170 dev->max_mtu = ETH_MAX_MTU; 1171 1172 dev->hw_features = VETH_FEATURES; 1173 dev->hw_enc_features = VETH_FEATURES; 1174 dev->mpls_features = NETIF_F_HW_CSUM | NETIF_F_GSO_SOFTWARE; 1175 } 1176 1177 /* 1178 * netlink interface 1179 */ 1180 1181 static int veth_validate(struct nlattr *tb[], struct nlattr *data[], 1182 struct netlink_ext_ack *extack) 1183 { 1184 if (tb[IFLA_ADDRESS]) { 1185 if (nla_len(tb[IFLA_ADDRESS]) != ETH_ALEN) 1186 return -EINVAL; 1187 if (!is_valid_ether_addr(nla_data(tb[IFLA_ADDRESS]))) 1188 return -EADDRNOTAVAIL; 1189 } 1190 if (tb[IFLA_MTU]) { 1191 if (!is_valid_veth_mtu(nla_get_u32(tb[IFLA_MTU]))) 1192 return -EINVAL; 1193 } 1194 return 0; 1195 } 1196 1197 static struct rtnl_link_ops veth_link_ops; 1198 1199 static int veth_newlink(struct net *src_net, struct net_device *dev, 1200 struct nlattr *tb[], struct nlattr *data[], 1201 struct netlink_ext_ack *extack) 1202 { 1203 int err; 1204 struct net_device *peer; 1205 struct veth_priv *priv; 1206 char ifname[IFNAMSIZ]; 1207 struct nlattr *peer_tb[IFLA_MAX + 1], **tbp; 1208 unsigned char name_assign_type; 1209 struct ifinfomsg *ifmp; 1210 struct net *net; 1211 1212 /* 1213 * create and register peer first 1214 */ 1215 if (data != NULL && data[VETH_INFO_PEER] != NULL) { 1216 struct nlattr *nla_peer; 1217 1218 nla_peer = data[VETH_INFO_PEER]; 1219 ifmp = nla_data(nla_peer); 1220 err = rtnl_nla_parse_ifla(peer_tb, 1221 nla_data(nla_peer) + sizeof(struct ifinfomsg), 1222 nla_len(nla_peer) - sizeof(struct ifinfomsg), 1223 NULL); 1224 if (err < 0) 1225 return err; 1226 1227 err = veth_validate(peer_tb, NULL, extack); 1228 if (err < 0) 1229 return err; 1230 1231 tbp = peer_tb; 1232 } else { 1233 ifmp = NULL; 1234 tbp = tb; 1235 } 1236 1237 if (ifmp && tbp[IFLA_IFNAME]) { 1238 nla_strlcpy(ifname, tbp[IFLA_IFNAME], IFNAMSIZ); 1239 name_assign_type = NET_NAME_USER; 1240 } else { 1241 snprintf(ifname, IFNAMSIZ, DRV_NAME "%%d"); 1242 name_assign_type = NET_NAME_ENUM; 1243 } 1244 1245 net = rtnl_link_get_net(src_net, tbp); 1246 if (IS_ERR(net)) 1247 return PTR_ERR(net); 1248 1249 peer = rtnl_create_link(net, ifname, name_assign_type, 1250 &veth_link_ops, tbp, extack); 1251 if (IS_ERR(peer)) { 1252 put_net(net); 1253 return PTR_ERR(peer); 1254 } 1255 1256 if (!ifmp || !tbp[IFLA_ADDRESS]) 1257 eth_hw_addr_random(peer); 1258 1259 if (ifmp && (dev->ifindex != 0)) 1260 peer->ifindex = ifmp->ifi_index; 1261 1262 peer->gso_max_size = dev->gso_max_size; 1263 peer->gso_max_segs = dev->gso_max_segs; 1264 1265 err = register_netdevice(peer); 1266 put_net(net); 1267 net = NULL; 1268 if (err < 0) 1269 goto err_register_peer; 1270 1271 netif_carrier_off(peer); 1272 1273 err = rtnl_configure_link(peer, ifmp); 1274 if (err < 0) 1275 goto err_configure_peer; 1276 1277 /* 1278 * register dev last 1279 * 1280 * note, that since we've registered new device the dev's name 1281 * should be re-allocated 1282 */ 1283 1284 if (tb[IFLA_ADDRESS] == NULL) 1285 eth_hw_addr_random(dev); 1286 1287 if (tb[IFLA_IFNAME]) 1288 nla_strlcpy(dev->name, tb[IFLA_IFNAME], IFNAMSIZ); 1289 else 1290 snprintf(dev->name, IFNAMSIZ, DRV_NAME "%%d"); 1291 1292 err = register_netdevice(dev); 1293 if (err < 0) 1294 goto err_register_dev; 1295 1296 netif_carrier_off(dev); 1297 1298 /* 1299 * tie the deviced together 1300 */ 1301 1302 priv = netdev_priv(dev); 1303 rcu_assign_pointer(priv->peer, peer); 1304 1305 priv = netdev_priv(peer); 1306 rcu_assign_pointer(priv->peer, dev); 1307 1308 return 0; 1309 1310 err_register_dev: 1311 /* nothing to do */ 1312 err_configure_peer: 1313 unregister_netdevice(peer); 1314 return err; 1315 1316 err_register_peer: 1317 free_netdev(peer); 1318 return err; 1319 } 1320 1321 static void veth_dellink(struct net_device *dev, struct list_head *head) 1322 { 1323 struct veth_priv *priv; 1324 struct net_device *peer; 1325 1326 priv = netdev_priv(dev); 1327 peer = rtnl_dereference(priv->peer); 1328 1329 /* Note : dellink() is called from default_device_exit_batch(), 1330 * before a rcu_synchronize() point. The devices are guaranteed 1331 * not being freed before one RCU grace period. 1332 */ 1333 RCU_INIT_POINTER(priv->peer, NULL); 1334 unregister_netdevice_queue(dev, head); 1335 1336 if (peer) { 1337 priv = netdev_priv(peer); 1338 RCU_INIT_POINTER(priv->peer, NULL); 1339 unregister_netdevice_queue(peer, head); 1340 } 1341 } 1342 1343 static const struct nla_policy veth_policy[VETH_INFO_MAX + 1] = { 1344 [VETH_INFO_PEER] = { .len = sizeof(struct ifinfomsg) }, 1345 }; 1346 1347 static struct net *veth_get_link_net(const struct net_device *dev) 1348 { 1349 struct veth_priv *priv = netdev_priv(dev); 1350 struct net_device *peer = rtnl_dereference(priv->peer); 1351 1352 return peer ? dev_net(peer) : dev_net(dev); 1353 } 1354 1355 static struct rtnl_link_ops veth_link_ops = { 1356 .kind = DRV_NAME, 1357 .priv_size = sizeof(struct veth_priv), 1358 .setup = veth_setup, 1359 .validate = veth_validate, 1360 .newlink = veth_newlink, 1361 .dellink = veth_dellink, 1362 .policy = veth_policy, 1363 .maxtype = VETH_INFO_MAX, 1364 .get_link_net = veth_get_link_net, 1365 }; 1366 1367 /* 1368 * init/fini 1369 */ 1370 1371 static __init int veth_init(void) 1372 { 1373 return rtnl_link_register(&veth_link_ops); 1374 } 1375 1376 static __exit void veth_exit(void) 1377 { 1378 rtnl_link_unregister(&veth_link_ops); 1379 } 1380 1381 module_init(veth_init); 1382 module_exit(veth_exit); 1383 1384 MODULE_DESCRIPTION("Virtual Ethernet Tunnel"); 1385 MODULE_LICENSE("GPL v2"); 1386 MODULE_ALIAS_RTNL_LINK(DRV_NAME); 1387