1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (c) 2009, Microsoft Corporation. 4 * 5 * Authors: 6 * Haiyang Zhang <haiyangz@microsoft.com> 7 * Hank Janssen <hjanssen@microsoft.com> 8 */ 9 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 10 11 #include <linux/init.h> 12 #include <linux/atomic.h> 13 #include <linux/module.h> 14 #include <linux/highmem.h> 15 #include <linux/device.h> 16 #include <linux/io.h> 17 #include <linux/delay.h> 18 #include <linux/netdevice.h> 19 #include <linux/inetdevice.h> 20 #include <linux/etherdevice.h> 21 #include <linux/pci.h> 22 #include <linux/skbuff.h> 23 #include <linux/if_vlan.h> 24 #include <linux/in.h> 25 #include <linux/slab.h> 26 #include <linux/rtnetlink.h> 27 #include <linux/netpoll.h> 28 29 #include <net/arp.h> 30 #include <net/route.h> 31 #include <net/sock.h> 32 #include <net/pkt_sched.h> 33 #include <net/checksum.h> 34 #include <net/ip6_checksum.h> 35 36 #include "hyperv_net.h" 37 38 #define RING_SIZE_MIN 64 39 #define RETRY_US_LO 5000 40 #define RETRY_US_HI 10000 41 #define RETRY_MAX 2000 /* >10 sec */ 42 43 #define LINKCHANGE_INT (2 * HZ) 44 #define VF_TAKEOVER_INT (HZ / 10) 45 46 static unsigned int ring_size __ro_after_init = 128; 47 module_param(ring_size, uint, 0444); 48 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)"); 49 unsigned int netvsc_ring_bytes __ro_after_init; 50 51 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE | 52 NETIF_MSG_LINK | NETIF_MSG_IFUP | 53 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR | 54 NETIF_MSG_TX_ERR; 55 56 static int debug = -1; 57 module_param(debug, int, 0444); 58 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)"); 59 60 static LIST_HEAD(netvsc_dev_list); 61 62 static void netvsc_change_rx_flags(struct net_device *net, int change) 63 { 64 struct net_device_context *ndev_ctx = netdev_priv(net); 65 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev); 66 int inc; 67 68 if (!vf_netdev) 69 return; 70 71 if (change & IFF_PROMISC) { 72 inc = (net->flags & IFF_PROMISC) ? 1 : -1; 73 dev_set_promiscuity(vf_netdev, inc); 74 } 75 76 if (change & IFF_ALLMULTI) { 77 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1; 78 dev_set_allmulti(vf_netdev, inc); 79 } 80 } 81 82 static void netvsc_set_rx_mode(struct net_device *net) 83 { 84 struct net_device_context *ndev_ctx = netdev_priv(net); 85 struct net_device *vf_netdev; 86 struct netvsc_device *nvdev; 87 88 rcu_read_lock(); 89 vf_netdev = rcu_dereference(ndev_ctx->vf_netdev); 90 if (vf_netdev) { 91 dev_uc_sync(vf_netdev, net); 92 dev_mc_sync(vf_netdev, net); 93 } 94 95 nvdev = rcu_dereference(ndev_ctx->nvdev); 96 if (nvdev) 97 rndis_filter_update(nvdev); 98 rcu_read_unlock(); 99 } 100 101 static void netvsc_tx_enable(struct netvsc_device *nvscdev, 102 struct net_device *ndev) 103 { 104 nvscdev->tx_disable = false; 105 virt_wmb(); /* ensure queue wake up mechanism is on */ 106 107 netif_tx_wake_all_queues(ndev); 108 } 109 110 static int netvsc_open(struct net_device *net) 111 { 112 struct net_device_context *ndev_ctx = netdev_priv(net); 113 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev); 114 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev); 115 struct rndis_device *rdev; 116 int ret = 0; 117 118 netif_carrier_off(net); 119 120 /* Open up the device */ 121 ret = rndis_filter_open(nvdev); 122 if (ret != 0) { 123 netdev_err(net, "unable to open device (ret %d).\n", ret); 124 return ret; 125 } 126 127 rdev = nvdev->extension; 128 if (!rdev->link_state) { 129 netif_carrier_on(net); 130 netvsc_tx_enable(nvdev, net); 131 } 132 133 if (vf_netdev) { 134 /* Setting synthetic device up transparently sets 135 * slave as up. If open fails, then slave will be 136 * still be offline (and not used). 137 */ 138 ret = dev_open(vf_netdev, NULL); 139 if (ret) 140 netdev_warn(net, 141 "unable to open slave: %s: %d\n", 142 vf_netdev->name, ret); 143 } 144 return 0; 145 } 146 147 static int netvsc_wait_until_empty(struct netvsc_device *nvdev) 148 { 149 unsigned int retry = 0; 150 int i; 151 152 /* Ensure pending bytes in ring are read */ 153 for (;;) { 154 u32 aread = 0; 155 156 for (i = 0; i < nvdev->num_chn; i++) { 157 struct vmbus_channel *chn 158 = nvdev->chan_table[i].channel; 159 160 if (!chn) 161 continue; 162 163 /* make sure receive not running now */ 164 napi_synchronize(&nvdev->chan_table[i].napi); 165 166 aread = hv_get_bytes_to_read(&chn->inbound); 167 if (aread) 168 break; 169 170 aread = hv_get_bytes_to_read(&chn->outbound); 171 if (aread) 172 break; 173 } 174 175 if (aread == 0) 176 return 0; 177 178 if (++retry > RETRY_MAX) 179 return -ETIMEDOUT; 180 181 usleep_range(RETRY_US_LO, RETRY_US_HI); 182 } 183 } 184 185 static void netvsc_tx_disable(struct netvsc_device *nvscdev, 186 struct net_device *ndev) 187 { 188 if (nvscdev) { 189 nvscdev->tx_disable = true; 190 virt_wmb(); /* ensure txq will not wake up after stop */ 191 } 192 193 netif_tx_disable(ndev); 194 } 195 196 static int netvsc_close(struct net_device *net) 197 { 198 struct net_device_context *net_device_ctx = netdev_priv(net); 199 struct net_device *vf_netdev 200 = rtnl_dereference(net_device_ctx->vf_netdev); 201 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev); 202 int ret; 203 204 netvsc_tx_disable(nvdev, net); 205 206 /* No need to close rndis filter if it is removed already */ 207 if (!nvdev) 208 return 0; 209 210 ret = rndis_filter_close(nvdev); 211 if (ret != 0) { 212 netdev_err(net, "unable to close device (ret %d).\n", ret); 213 return ret; 214 } 215 216 ret = netvsc_wait_until_empty(nvdev); 217 if (ret) 218 netdev_err(net, "Ring buffer not empty after closing rndis\n"); 219 220 if (vf_netdev) 221 dev_close(vf_netdev); 222 223 return ret; 224 } 225 226 static inline void *init_ppi_data(struct rndis_message *msg, 227 u32 ppi_size, u32 pkt_type) 228 { 229 struct rndis_packet *rndis_pkt = &msg->msg.pkt; 230 struct rndis_per_packet_info *ppi; 231 232 rndis_pkt->data_offset += ppi_size; 233 ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset 234 + rndis_pkt->per_pkt_info_len; 235 236 ppi->size = ppi_size; 237 ppi->type = pkt_type; 238 ppi->internal = 0; 239 ppi->ppi_offset = sizeof(struct rndis_per_packet_info); 240 241 rndis_pkt->per_pkt_info_len += ppi_size; 242 243 return ppi + 1; 244 } 245 246 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented 247 * packets. We can use ethtool to change UDP hash level when necessary. 248 */ 249 static inline u32 netvsc_get_hash( 250 struct sk_buff *skb, 251 const struct net_device_context *ndc) 252 { 253 struct flow_keys flow; 254 u32 hash, pkt_proto = 0; 255 static u32 hashrnd __read_mostly; 256 257 net_get_random_once(&hashrnd, sizeof(hashrnd)); 258 259 if (!skb_flow_dissect_flow_keys(skb, &flow, 0)) 260 return 0; 261 262 switch (flow.basic.ip_proto) { 263 case IPPROTO_TCP: 264 if (flow.basic.n_proto == htons(ETH_P_IP)) 265 pkt_proto = HV_TCP4_L4HASH; 266 else if (flow.basic.n_proto == htons(ETH_P_IPV6)) 267 pkt_proto = HV_TCP6_L4HASH; 268 269 break; 270 271 case IPPROTO_UDP: 272 if (flow.basic.n_proto == htons(ETH_P_IP)) 273 pkt_proto = HV_UDP4_L4HASH; 274 else if (flow.basic.n_proto == htons(ETH_P_IPV6)) 275 pkt_proto = HV_UDP6_L4HASH; 276 277 break; 278 } 279 280 if (pkt_proto & ndc->l4_hash) { 281 return skb_get_hash(skb); 282 } else { 283 if (flow.basic.n_proto == htons(ETH_P_IP)) 284 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd); 285 else if (flow.basic.n_proto == htons(ETH_P_IPV6)) 286 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd); 287 else 288 hash = 0; 289 290 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3); 291 } 292 293 return hash; 294 } 295 296 static inline int netvsc_get_tx_queue(struct net_device *ndev, 297 struct sk_buff *skb, int old_idx) 298 { 299 const struct net_device_context *ndc = netdev_priv(ndev); 300 struct sock *sk = skb->sk; 301 int q_idx; 302 303 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) & 304 (VRSS_SEND_TAB_SIZE - 1)]; 305 306 /* If queue index changed record the new value */ 307 if (q_idx != old_idx && 308 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache)) 309 sk_tx_queue_set(sk, q_idx); 310 311 return q_idx; 312 } 313 314 /* 315 * Select queue for transmit. 316 * 317 * If a valid queue has already been assigned, then use that. 318 * Otherwise compute tx queue based on hash and the send table. 319 * 320 * This is basically similar to default (netdev_pick_tx) with the added step 321 * of using the host send_table when no other queue has been assigned. 322 * 323 * TODO support XPS - but get_xps_queue not exported 324 */ 325 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb) 326 { 327 int q_idx = sk_tx_queue_get(skb->sk); 328 329 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) { 330 /* If forwarding a packet, we use the recorded queue when 331 * available for better cache locality. 332 */ 333 if (skb_rx_queue_recorded(skb)) 334 q_idx = skb_get_rx_queue(skb); 335 else 336 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx); 337 } 338 339 return q_idx; 340 } 341 342 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb, 343 struct net_device *sb_dev) 344 { 345 struct net_device_context *ndc = netdev_priv(ndev); 346 struct net_device *vf_netdev; 347 u16 txq; 348 349 rcu_read_lock(); 350 vf_netdev = rcu_dereference(ndc->vf_netdev); 351 if (vf_netdev) { 352 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops; 353 354 if (vf_ops->ndo_select_queue) 355 txq = vf_ops->ndo_select_queue(vf_netdev, skb, sb_dev); 356 else 357 txq = netdev_pick_tx(vf_netdev, skb, NULL); 358 359 /* Record the queue selected by VF so that it can be 360 * used for common case where VF has more queues than 361 * the synthetic device. 362 */ 363 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq; 364 } else { 365 txq = netvsc_pick_tx(ndev, skb); 366 } 367 rcu_read_unlock(); 368 369 while (unlikely(txq >= ndev->real_num_tx_queues)) 370 txq -= ndev->real_num_tx_queues; 371 372 return txq; 373 } 374 375 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len, 376 struct hv_page_buffer *pb) 377 { 378 int j = 0; 379 380 /* Deal with compound pages by ignoring unused part 381 * of the page. 382 */ 383 page += (offset >> PAGE_SHIFT); 384 offset &= ~PAGE_MASK; 385 386 while (len > 0) { 387 unsigned long bytes; 388 389 bytes = PAGE_SIZE - offset; 390 if (bytes > len) 391 bytes = len; 392 pb[j].pfn = page_to_pfn(page); 393 pb[j].offset = offset; 394 pb[j].len = bytes; 395 396 offset += bytes; 397 len -= bytes; 398 399 if (offset == PAGE_SIZE && len) { 400 page++; 401 offset = 0; 402 j++; 403 } 404 } 405 406 return j + 1; 407 } 408 409 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb, 410 struct hv_netvsc_packet *packet, 411 struct hv_page_buffer *pb) 412 { 413 u32 slots_used = 0; 414 char *data = skb->data; 415 int frags = skb_shinfo(skb)->nr_frags; 416 int i; 417 418 /* The packet is laid out thus: 419 * 1. hdr: RNDIS header and PPI 420 * 2. skb linear data 421 * 3. skb fragment data 422 */ 423 slots_used += fill_pg_buf(virt_to_page(hdr), 424 offset_in_page(hdr), 425 len, &pb[slots_used]); 426 427 packet->rmsg_size = len; 428 packet->rmsg_pgcnt = slots_used; 429 430 slots_used += fill_pg_buf(virt_to_page(data), 431 offset_in_page(data), 432 skb_headlen(skb), &pb[slots_used]); 433 434 for (i = 0; i < frags; i++) { 435 skb_frag_t *frag = skb_shinfo(skb)->frags + i; 436 437 slots_used += fill_pg_buf(skb_frag_page(frag), 438 frag->page_offset, 439 skb_frag_size(frag), &pb[slots_used]); 440 } 441 return slots_used; 442 } 443 444 static int count_skb_frag_slots(struct sk_buff *skb) 445 { 446 int i, frags = skb_shinfo(skb)->nr_frags; 447 int pages = 0; 448 449 for (i = 0; i < frags; i++) { 450 skb_frag_t *frag = skb_shinfo(skb)->frags + i; 451 unsigned long size = skb_frag_size(frag); 452 unsigned long offset = frag->page_offset; 453 454 /* Skip unused frames from start of page */ 455 offset &= ~PAGE_MASK; 456 pages += PFN_UP(offset + size); 457 } 458 return pages; 459 } 460 461 static int netvsc_get_slots(struct sk_buff *skb) 462 { 463 char *data = skb->data; 464 unsigned int offset = offset_in_page(data); 465 unsigned int len = skb_headlen(skb); 466 int slots; 467 int frag_slots; 468 469 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE); 470 frag_slots = count_skb_frag_slots(skb); 471 return slots + frag_slots; 472 } 473 474 static u32 net_checksum_info(struct sk_buff *skb) 475 { 476 if (skb->protocol == htons(ETH_P_IP)) { 477 struct iphdr *ip = ip_hdr(skb); 478 479 if (ip->protocol == IPPROTO_TCP) 480 return TRANSPORT_INFO_IPV4_TCP; 481 else if (ip->protocol == IPPROTO_UDP) 482 return TRANSPORT_INFO_IPV4_UDP; 483 } else { 484 struct ipv6hdr *ip6 = ipv6_hdr(skb); 485 486 if (ip6->nexthdr == IPPROTO_TCP) 487 return TRANSPORT_INFO_IPV6_TCP; 488 else if (ip6->nexthdr == IPPROTO_UDP) 489 return TRANSPORT_INFO_IPV6_UDP; 490 } 491 492 return TRANSPORT_INFO_NOT_IP; 493 } 494 495 /* Send skb on the slave VF device. */ 496 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev, 497 struct sk_buff *skb) 498 { 499 struct net_device_context *ndev_ctx = netdev_priv(net); 500 unsigned int len = skb->len; 501 int rc; 502 503 skb->dev = vf_netdev; 504 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping; 505 506 rc = dev_queue_xmit(skb); 507 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) { 508 struct netvsc_vf_pcpu_stats *pcpu_stats 509 = this_cpu_ptr(ndev_ctx->vf_stats); 510 511 u64_stats_update_begin(&pcpu_stats->syncp); 512 pcpu_stats->tx_packets++; 513 pcpu_stats->tx_bytes += len; 514 u64_stats_update_end(&pcpu_stats->syncp); 515 } else { 516 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped); 517 } 518 519 return rc; 520 } 521 522 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net) 523 { 524 struct net_device_context *net_device_ctx = netdev_priv(net); 525 struct hv_netvsc_packet *packet = NULL; 526 int ret; 527 unsigned int num_data_pgs; 528 struct rndis_message *rndis_msg; 529 struct net_device *vf_netdev; 530 u32 rndis_msg_size; 531 u32 hash; 532 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT]; 533 534 /* if VF is present and up then redirect packets 535 * already called with rcu_read_lock_bh 536 */ 537 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev); 538 if (vf_netdev && netif_running(vf_netdev) && 539 !netpoll_tx_running(net)) 540 return netvsc_vf_xmit(net, vf_netdev, skb); 541 542 /* We will atmost need two pages to describe the rndis 543 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number 544 * of pages in a single packet. If skb is scattered around 545 * more pages we try linearizing it. 546 */ 547 548 num_data_pgs = netvsc_get_slots(skb) + 2; 549 550 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) { 551 ++net_device_ctx->eth_stats.tx_scattered; 552 553 if (skb_linearize(skb)) 554 goto no_memory; 555 556 num_data_pgs = netvsc_get_slots(skb) + 2; 557 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) { 558 ++net_device_ctx->eth_stats.tx_too_big; 559 goto drop; 560 } 561 } 562 563 /* 564 * Place the rndis header in the skb head room and 565 * the skb->cb will be used for hv_netvsc_packet 566 * structure. 567 */ 568 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE); 569 if (ret) 570 goto no_memory; 571 572 /* Use the skb control buffer for building up the packet */ 573 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) > 574 FIELD_SIZEOF(struct sk_buff, cb)); 575 packet = (struct hv_netvsc_packet *)skb->cb; 576 577 packet->q_idx = skb_get_queue_mapping(skb); 578 579 packet->total_data_buflen = skb->len; 580 packet->total_bytes = skb->len; 581 packet->total_packets = 1; 582 583 rndis_msg = (struct rndis_message *)skb->head; 584 585 /* Add the rndis header */ 586 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET; 587 rndis_msg->msg_len = packet->total_data_buflen; 588 589 rndis_msg->msg.pkt = (struct rndis_packet) { 590 .data_offset = sizeof(struct rndis_packet), 591 .data_len = packet->total_data_buflen, 592 .per_pkt_info_offset = sizeof(struct rndis_packet), 593 }; 594 595 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet); 596 597 hash = skb_get_hash_raw(skb); 598 if (hash != 0 && net->real_num_tx_queues > 1) { 599 u32 *hash_info; 600 601 rndis_msg_size += NDIS_HASH_PPI_SIZE; 602 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE, 603 NBL_HASH_VALUE); 604 *hash_info = hash; 605 } 606 607 if (skb_vlan_tag_present(skb)) { 608 struct ndis_pkt_8021q_info *vlan; 609 610 rndis_msg_size += NDIS_VLAN_PPI_SIZE; 611 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE, 612 IEEE_8021Q_INFO); 613 614 vlan->value = 0; 615 vlan->vlanid = skb_vlan_tag_get_id(skb); 616 vlan->cfi = skb_vlan_tag_get_cfi(skb); 617 vlan->pri = skb_vlan_tag_get_prio(skb); 618 } 619 620 if (skb_is_gso(skb)) { 621 struct ndis_tcp_lso_info *lso_info; 622 623 rndis_msg_size += NDIS_LSO_PPI_SIZE; 624 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE, 625 TCP_LARGESEND_PKTINFO); 626 627 lso_info->value = 0; 628 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE; 629 if (skb->protocol == htons(ETH_P_IP)) { 630 lso_info->lso_v2_transmit.ip_version = 631 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4; 632 ip_hdr(skb)->tot_len = 0; 633 ip_hdr(skb)->check = 0; 634 tcp_hdr(skb)->check = 635 ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 636 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0); 637 } else { 638 lso_info->lso_v2_transmit.ip_version = 639 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6; 640 ipv6_hdr(skb)->payload_len = 0; 641 tcp_hdr(skb)->check = 642 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 643 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0); 644 } 645 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb); 646 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size; 647 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 648 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) { 649 struct ndis_tcp_ip_checksum_info *csum_info; 650 651 rndis_msg_size += NDIS_CSUM_PPI_SIZE; 652 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE, 653 TCPIP_CHKSUM_PKTINFO); 654 655 csum_info->value = 0; 656 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb); 657 658 if (skb->protocol == htons(ETH_P_IP)) { 659 csum_info->transmit.is_ipv4 = 1; 660 661 if (ip_hdr(skb)->protocol == IPPROTO_TCP) 662 csum_info->transmit.tcp_checksum = 1; 663 else 664 csum_info->transmit.udp_checksum = 1; 665 } else { 666 csum_info->transmit.is_ipv6 = 1; 667 668 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP) 669 csum_info->transmit.tcp_checksum = 1; 670 else 671 csum_info->transmit.udp_checksum = 1; 672 } 673 } else { 674 /* Can't do offload of this type of checksum */ 675 if (skb_checksum_help(skb)) 676 goto drop; 677 } 678 } 679 680 /* Start filling in the page buffers with the rndis hdr */ 681 rndis_msg->msg_len += rndis_msg_size; 682 packet->total_data_buflen = rndis_msg->msg_len; 683 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size, 684 skb, packet, pb); 685 686 /* timestamp packet in software */ 687 skb_tx_timestamp(skb); 688 689 ret = netvsc_send(net, packet, rndis_msg, pb, skb); 690 if (likely(ret == 0)) 691 return NETDEV_TX_OK; 692 693 if (ret == -EAGAIN) { 694 ++net_device_ctx->eth_stats.tx_busy; 695 return NETDEV_TX_BUSY; 696 } 697 698 if (ret == -ENOSPC) 699 ++net_device_ctx->eth_stats.tx_no_space; 700 701 drop: 702 dev_kfree_skb_any(skb); 703 net->stats.tx_dropped++; 704 705 return NETDEV_TX_OK; 706 707 no_memory: 708 ++net_device_ctx->eth_stats.tx_no_memory; 709 goto drop; 710 } 711 712 /* 713 * netvsc_linkstatus_callback - Link up/down notification 714 */ 715 void netvsc_linkstatus_callback(struct net_device *net, 716 struct rndis_message *resp) 717 { 718 struct rndis_indicate_status *indicate = &resp->msg.indicate_status; 719 struct net_device_context *ndev_ctx = netdev_priv(net); 720 struct netvsc_reconfig *event; 721 unsigned long flags; 722 723 /* Update the physical link speed when changing to another vSwitch */ 724 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) { 725 u32 speed; 726 727 speed = *(u32 *)((void *)indicate 728 + indicate->status_buf_offset) / 10000; 729 ndev_ctx->speed = speed; 730 return; 731 } 732 733 /* Handle these link change statuses below */ 734 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE && 735 indicate->status != RNDIS_STATUS_MEDIA_CONNECT && 736 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT) 737 return; 738 739 if (net->reg_state != NETREG_REGISTERED) 740 return; 741 742 event = kzalloc(sizeof(*event), GFP_ATOMIC); 743 if (!event) 744 return; 745 event->event = indicate->status; 746 747 spin_lock_irqsave(&ndev_ctx->lock, flags); 748 list_add_tail(&event->list, &ndev_ctx->reconfig_events); 749 spin_unlock_irqrestore(&ndev_ctx->lock, flags); 750 751 schedule_delayed_work(&ndev_ctx->dwork, 0); 752 } 753 754 static void netvsc_comp_ipcsum(struct sk_buff *skb) 755 { 756 struct iphdr *iph = (struct iphdr *)skb->data; 757 758 iph->check = 0; 759 iph->check = ip_fast_csum(iph, iph->ihl); 760 } 761 762 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net, 763 struct netvsc_channel *nvchan) 764 { 765 struct napi_struct *napi = &nvchan->napi; 766 const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan; 767 const struct ndis_tcp_ip_checksum_info *csum_info = 768 nvchan->rsc.csum_info; 769 struct sk_buff *skb; 770 int i; 771 772 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen); 773 if (!skb) 774 return skb; 775 776 /* 777 * Copy to skb. This copy is needed here since the memory pointed by 778 * hv_netvsc_packet cannot be deallocated 779 */ 780 for (i = 0; i < nvchan->rsc.cnt; i++) 781 skb_put_data(skb, nvchan->rsc.data[i], nvchan->rsc.len[i]); 782 783 skb->protocol = eth_type_trans(skb, net); 784 785 /* skb is already created with CHECKSUM_NONE */ 786 skb_checksum_none_assert(skb); 787 788 /* Incoming packets may have IP header checksum verified by the host. 789 * They may not have IP header checksum computed after coalescing. 790 * We compute it here if the flags are set, because on Linux, the IP 791 * checksum is always checked. 792 */ 793 if (csum_info && csum_info->receive.ip_checksum_value_invalid && 794 csum_info->receive.ip_checksum_succeeded && 795 skb->protocol == htons(ETH_P_IP)) 796 netvsc_comp_ipcsum(skb); 797 798 /* Do L4 checksum offload if enabled and present. 799 */ 800 if (csum_info && (net->features & NETIF_F_RXCSUM)) { 801 if (csum_info->receive.tcp_checksum_succeeded || 802 csum_info->receive.udp_checksum_succeeded) 803 skb->ip_summed = CHECKSUM_UNNECESSARY; 804 } 805 806 if (vlan) { 807 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) | 808 (vlan->cfi ? VLAN_CFI_MASK : 0); 809 810 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), 811 vlan_tci); 812 } 813 814 return skb; 815 } 816 817 /* 818 * netvsc_recv_callback - Callback when we receive a packet from the 819 * "wire" on the specified device. 820 */ 821 int netvsc_recv_callback(struct net_device *net, 822 struct netvsc_device *net_device, 823 struct netvsc_channel *nvchan) 824 { 825 struct net_device_context *net_device_ctx = netdev_priv(net); 826 struct vmbus_channel *channel = nvchan->channel; 827 u16 q_idx = channel->offermsg.offer.sub_channel_index; 828 struct sk_buff *skb; 829 struct netvsc_stats *rx_stats; 830 831 if (net->reg_state != NETREG_REGISTERED) 832 return NVSP_STAT_FAIL; 833 834 /* Allocate a skb - TODO direct I/O to pages? */ 835 skb = netvsc_alloc_recv_skb(net, nvchan); 836 837 if (unlikely(!skb)) { 838 ++net_device_ctx->eth_stats.rx_no_memory; 839 rcu_read_unlock(); 840 return NVSP_STAT_FAIL; 841 } 842 843 skb_record_rx_queue(skb, q_idx); 844 845 /* 846 * Even if injecting the packet, record the statistics 847 * on the synthetic device because modifying the VF device 848 * statistics will not work correctly. 849 */ 850 rx_stats = &nvchan->rx_stats; 851 u64_stats_update_begin(&rx_stats->syncp); 852 rx_stats->packets++; 853 rx_stats->bytes += nvchan->rsc.pktlen; 854 855 if (skb->pkt_type == PACKET_BROADCAST) 856 ++rx_stats->broadcast; 857 else if (skb->pkt_type == PACKET_MULTICAST) 858 ++rx_stats->multicast; 859 u64_stats_update_end(&rx_stats->syncp); 860 861 napi_gro_receive(&nvchan->napi, skb); 862 return NVSP_STAT_SUCCESS; 863 } 864 865 static void netvsc_get_drvinfo(struct net_device *net, 866 struct ethtool_drvinfo *info) 867 { 868 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); 869 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version)); 870 } 871 872 static void netvsc_get_channels(struct net_device *net, 873 struct ethtool_channels *channel) 874 { 875 struct net_device_context *net_device_ctx = netdev_priv(net); 876 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev); 877 878 if (nvdev) { 879 channel->max_combined = nvdev->max_chn; 880 channel->combined_count = nvdev->num_chn; 881 } 882 } 883 884 /* Alloc struct netvsc_device_info, and initialize it from either existing 885 * struct netvsc_device, or from default values. 886 */ 887 static struct netvsc_device_info *netvsc_devinfo_get 888 (struct netvsc_device *nvdev) 889 { 890 struct netvsc_device_info *dev_info; 891 892 dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC); 893 894 if (!dev_info) 895 return NULL; 896 897 if (nvdev) { 898 dev_info->num_chn = nvdev->num_chn; 899 dev_info->send_sections = nvdev->send_section_cnt; 900 dev_info->send_section_size = nvdev->send_section_size; 901 dev_info->recv_sections = nvdev->recv_section_cnt; 902 dev_info->recv_section_size = nvdev->recv_section_size; 903 904 memcpy(dev_info->rss_key, nvdev->extension->rss_key, 905 NETVSC_HASH_KEYLEN); 906 } else { 907 dev_info->num_chn = VRSS_CHANNEL_DEFAULT; 908 dev_info->send_sections = NETVSC_DEFAULT_TX; 909 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE; 910 dev_info->recv_sections = NETVSC_DEFAULT_RX; 911 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE; 912 } 913 914 return dev_info; 915 } 916 917 static int netvsc_detach(struct net_device *ndev, 918 struct netvsc_device *nvdev) 919 { 920 struct net_device_context *ndev_ctx = netdev_priv(ndev); 921 struct hv_device *hdev = ndev_ctx->device_ctx; 922 int ret; 923 924 /* Don't try continuing to try and setup sub channels */ 925 if (cancel_work_sync(&nvdev->subchan_work)) 926 nvdev->num_chn = 1; 927 928 /* If device was up (receiving) then shutdown */ 929 if (netif_running(ndev)) { 930 netvsc_tx_disable(nvdev, ndev); 931 932 ret = rndis_filter_close(nvdev); 933 if (ret) { 934 netdev_err(ndev, 935 "unable to close device (ret %d).\n", ret); 936 return ret; 937 } 938 939 ret = netvsc_wait_until_empty(nvdev); 940 if (ret) { 941 netdev_err(ndev, 942 "Ring buffer not empty after closing rndis\n"); 943 return ret; 944 } 945 } 946 947 netif_device_detach(ndev); 948 949 rndis_filter_device_remove(hdev, nvdev); 950 951 return 0; 952 } 953 954 static int netvsc_attach(struct net_device *ndev, 955 struct netvsc_device_info *dev_info) 956 { 957 struct net_device_context *ndev_ctx = netdev_priv(ndev); 958 struct hv_device *hdev = ndev_ctx->device_ctx; 959 struct netvsc_device *nvdev; 960 struct rndis_device *rdev; 961 int ret; 962 963 nvdev = rndis_filter_device_add(hdev, dev_info); 964 if (IS_ERR(nvdev)) 965 return PTR_ERR(nvdev); 966 967 if (nvdev->num_chn > 1) { 968 ret = rndis_set_subchannel(ndev, nvdev, dev_info); 969 970 /* if unavailable, just proceed with one queue */ 971 if (ret) { 972 nvdev->max_chn = 1; 973 nvdev->num_chn = 1; 974 } 975 } 976 977 /* In any case device is now ready */ 978 netif_device_attach(ndev); 979 980 /* Note: enable and attach happen when sub-channels setup */ 981 netif_carrier_off(ndev); 982 983 if (netif_running(ndev)) { 984 ret = rndis_filter_open(nvdev); 985 if (ret) 986 return ret; 987 988 rdev = nvdev->extension; 989 if (!rdev->link_state) 990 netif_carrier_on(ndev); 991 } 992 993 return 0; 994 } 995 996 static int netvsc_set_channels(struct net_device *net, 997 struct ethtool_channels *channels) 998 { 999 struct net_device_context *net_device_ctx = netdev_priv(net); 1000 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev); 1001 unsigned int orig, count = channels->combined_count; 1002 struct netvsc_device_info *device_info; 1003 int ret; 1004 1005 /* We do not support separate count for rx, tx, or other */ 1006 if (count == 0 || 1007 channels->rx_count || channels->tx_count || channels->other_count) 1008 return -EINVAL; 1009 1010 if (!nvdev || nvdev->destroy) 1011 return -ENODEV; 1012 1013 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) 1014 return -EINVAL; 1015 1016 if (count > nvdev->max_chn) 1017 return -EINVAL; 1018 1019 orig = nvdev->num_chn; 1020 1021 device_info = netvsc_devinfo_get(nvdev); 1022 1023 if (!device_info) 1024 return -ENOMEM; 1025 1026 device_info->num_chn = count; 1027 1028 ret = netvsc_detach(net, nvdev); 1029 if (ret) 1030 goto out; 1031 1032 ret = netvsc_attach(net, device_info); 1033 if (ret) { 1034 device_info->num_chn = orig; 1035 if (netvsc_attach(net, device_info)) 1036 netdev_err(net, "restoring channel setting failed\n"); 1037 } 1038 1039 out: 1040 kfree(device_info); 1041 return ret; 1042 } 1043 1044 static bool 1045 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd) 1046 { 1047 struct ethtool_link_ksettings diff1 = *cmd; 1048 struct ethtool_link_ksettings diff2 = {}; 1049 1050 diff1.base.speed = 0; 1051 diff1.base.duplex = 0; 1052 /* advertising and cmd are usually set */ 1053 ethtool_link_ksettings_zero_link_mode(&diff1, advertising); 1054 diff1.base.cmd = 0; 1055 /* We set port to PORT_OTHER */ 1056 diff2.base.port = PORT_OTHER; 1057 1058 return !memcmp(&diff1, &diff2, sizeof(diff1)); 1059 } 1060 1061 static void netvsc_init_settings(struct net_device *dev) 1062 { 1063 struct net_device_context *ndc = netdev_priv(dev); 1064 1065 ndc->l4_hash = HV_DEFAULT_L4HASH; 1066 1067 ndc->speed = SPEED_UNKNOWN; 1068 ndc->duplex = DUPLEX_FULL; 1069 1070 dev->features = NETIF_F_LRO; 1071 } 1072 1073 static int netvsc_get_link_ksettings(struct net_device *dev, 1074 struct ethtool_link_ksettings *cmd) 1075 { 1076 struct net_device_context *ndc = netdev_priv(dev); 1077 1078 cmd->base.speed = ndc->speed; 1079 cmd->base.duplex = ndc->duplex; 1080 cmd->base.port = PORT_OTHER; 1081 1082 return 0; 1083 } 1084 1085 static int netvsc_set_link_ksettings(struct net_device *dev, 1086 const struct ethtool_link_ksettings *cmd) 1087 { 1088 struct net_device_context *ndc = netdev_priv(dev); 1089 u32 speed; 1090 1091 speed = cmd->base.speed; 1092 if (!ethtool_validate_speed(speed) || 1093 !ethtool_validate_duplex(cmd->base.duplex) || 1094 !netvsc_validate_ethtool_ss_cmd(cmd)) 1095 return -EINVAL; 1096 1097 ndc->speed = speed; 1098 ndc->duplex = cmd->base.duplex; 1099 1100 return 0; 1101 } 1102 1103 static int netvsc_change_mtu(struct net_device *ndev, int mtu) 1104 { 1105 struct net_device_context *ndevctx = netdev_priv(ndev); 1106 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev); 1107 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev); 1108 int orig_mtu = ndev->mtu; 1109 struct netvsc_device_info *device_info; 1110 int ret = 0; 1111 1112 if (!nvdev || nvdev->destroy) 1113 return -ENODEV; 1114 1115 device_info = netvsc_devinfo_get(nvdev); 1116 1117 if (!device_info) 1118 return -ENOMEM; 1119 1120 /* Change MTU of underlying VF netdev first. */ 1121 if (vf_netdev) { 1122 ret = dev_set_mtu(vf_netdev, mtu); 1123 if (ret) 1124 goto out; 1125 } 1126 1127 ret = netvsc_detach(ndev, nvdev); 1128 if (ret) 1129 goto rollback_vf; 1130 1131 ndev->mtu = mtu; 1132 1133 ret = netvsc_attach(ndev, device_info); 1134 if (!ret) 1135 goto out; 1136 1137 /* Attempt rollback to original MTU */ 1138 ndev->mtu = orig_mtu; 1139 1140 if (netvsc_attach(ndev, device_info)) 1141 netdev_err(ndev, "restoring mtu failed\n"); 1142 rollback_vf: 1143 if (vf_netdev) 1144 dev_set_mtu(vf_netdev, orig_mtu); 1145 1146 out: 1147 kfree(device_info); 1148 return ret; 1149 } 1150 1151 static void netvsc_get_vf_stats(struct net_device *net, 1152 struct netvsc_vf_pcpu_stats *tot) 1153 { 1154 struct net_device_context *ndev_ctx = netdev_priv(net); 1155 int i; 1156 1157 memset(tot, 0, sizeof(*tot)); 1158 1159 for_each_possible_cpu(i) { 1160 const struct netvsc_vf_pcpu_stats *stats 1161 = per_cpu_ptr(ndev_ctx->vf_stats, i); 1162 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 1163 unsigned int start; 1164 1165 do { 1166 start = u64_stats_fetch_begin_irq(&stats->syncp); 1167 rx_packets = stats->rx_packets; 1168 tx_packets = stats->tx_packets; 1169 rx_bytes = stats->rx_bytes; 1170 tx_bytes = stats->tx_bytes; 1171 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1172 1173 tot->rx_packets += rx_packets; 1174 tot->tx_packets += tx_packets; 1175 tot->rx_bytes += rx_bytes; 1176 tot->tx_bytes += tx_bytes; 1177 tot->tx_dropped += stats->tx_dropped; 1178 } 1179 } 1180 1181 static void netvsc_get_pcpu_stats(struct net_device *net, 1182 struct netvsc_ethtool_pcpu_stats *pcpu_tot) 1183 { 1184 struct net_device_context *ndev_ctx = netdev_priv(net); 1185 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev); 1186 int i; 1187 1188 /* fetch percpu stats of vf */ 1189 for_each_possible_cpu(i) { 1190 const struct netvsc_vf_pcpu_stats *stats = 1191 per_cpu_ptr(ndev_ctx->vf_stats, i); 1192 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i]; 1193 unsigned int start; 1194 1195 do { 1196 start = u64_stats_fetch_begin_irq(&stats->syncp); 1197 this_tot->vf_rx_packets = stats->rx_packets; 1198 this_tot->vf_tx_packets = stats->tx_packets; 1199 this_tot->vf_rx_bytes = stats->rx_bytes; 1200 this_tot->vf_tx_bytes = stats->tx_bytes; 1201 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1202 this_tot->rx_packets = this_tot->vf_rx_packets; 1203 this_tot->tx_packets = this_tot->vf_tx_packets; 1204 this_tot->rx_bytes = this_tot->vf_rx_bytes; 1205 this_tot->tx_bytes = this_tot->vf_tx_bytes; 1206 } 1207 1208 /* fetch percpu stats of netvsc */ 1209 for (i = 0; i < nvdev->num_chn; i++) { 1210 const struct netvsc_channel *nvchan = &nvdev->chan_table[i]; 1211 const struct netvsc_stats *stats; 1212 struct netvsc_ethtool_pcpu_stats *this_tot = 1213 &pcpu_tot[nvchan->channel->target_cpu]; 1214 u64 packets, bytes; 1215 unsigned int start; 1216 1217 stats = &nvchan->tx_stats; 1218 do { 1219 start = u64_stats_fetch_begin_irq(&stats->syncp); 1220 packets = stats->packets; 1221 bytes = stats->bytes; 1222 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1223 1224 this_tot->tx_bytes += bytes; 1225 this_tot->tx_packets += packets; 1226 1227 stats = &nvchan->rx_stats; 1228 do { 1229 start = u64_stats_fetch_begin_irq(&stats->syncp); 1230 packets = stats->packets; 1231 bytes = stats->bytes; 1232 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1233 1234 this_tot->rx_bytes += bytes; 1235 this_tot->rx_packets += packets; 1236 } 1237 } 1238 1239 static void netvsc_get_stats64(struct net_device *net, 1240 struct rtnl_link_stats64 *t) 1241 { 1242 struct net_device_context *ndev_ctx = netdev_priv(net); 1243 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev); 1244 struct netvsc_vf_pcpu_stats vf_tot; 1245 int i; 1246 1247 if (!nvdev) 1248 return; 1249 1250 netdev_stats_to_stats64(t, &net->stats); 1251 1252 netvsc_get_vf_stats(net, &vf_tot); 1253 t->rx_packets += vf_tot.rx_packets; 1254 t->tx_packets += vf_tot.tx_packets; 1255 t->rx_bytes += vf_tot.rx_bytes; 1256 t->tx_bytes += vf_tot.tx_bytes; 1257 t->tx_dropped += vf_tot.tx_dropped; 1258 1259 for (i = 0; i < nvdev->num_chn; i++) { 1260 const struct netvsc_channel *nvchan = &nvdev->chan_table[i]; 1261 const struct netvsc_stats *stats; 1262 u64 packets, bytes, multicast; 1263 unsigned int start; 1264 1265 stats = &nvchan->tx_stats; 1266 do { 1267 start = u64_stats_fetch_begin_irq(&stats->syncp); 1268 packets = stats->packets; 1269 bytes = stats->bytes; 1270 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1271 1272 t->tx_bytes += bytes; 1273 t->tx_packets += packets; 1274 1275 stats = &nvchan->rx_stats; 1276 do { 1277 start = u64_stats_fetch_begin_irq(&stats->syncp); 1278 packets = stats->packets; 1279 bytes = stats->bytes; 1280 multicast = stats->multicast + stats->broadcast; 1281 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1282 1283 t->rx_bytes += bytes; 1284 t->rx_packets += packets; 1285 t->multicast += multicast; 1286 } 1287 } 1288 1289 static int netvsc_set_mac_addr(struct net_device *ndev, void *p) 1290 { 1291 struct net_device_context *ndc = netdev_priv(ndev); 1292 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev); 1293 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); 1294 struct sockaddr *addr = p; 1295 int err; 1296 1297 err = eth_prepare_mac_addr_change(ndev, p); 1298 if (err) 1299 return err; 1300 1301 if (!nvdev) 1302 return -ENODEV; 1303 1304 if (vf_netdev) { 1305 err = dev_set_mac_address(vf_netdev, addr, NULL); 1306 if (err) 1307 return err; 1308 } 1309 1310 err = rndis_filter_set_device_mac(nvdev, addr->sa_data); 1311 if (!err) { 1312 eth_commit_mac_addr_change(ndev, p); 1313 } else if (vf_netdev) { 1314 /* rollback change on VF */ 1315 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN); 1316 dev_set_mac_address(vf_netdev, addr, NULL); 1317 } 1318 1319 return err; 1320 } 1321 1322 static const struct { 1323 char name[ETH_GSTRING_LEN]; 1324 u16 offset; 1325 } netvsc_stats[] = { 1326 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) }, 1327 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) }, 1328 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) }, 1329 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) }, 1330 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) }, 1331 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) }, 1332 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) }, 1333 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) }, 1334 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) }, 1335 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) }, 1336 }, pcpu_stats[] = { 1337 { "cpu%u_rx_packets", 1338 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) }, 1339 { "cpu%u_rx_bytes", 1340 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) }, 1341 { "cpu%u_tx_packets", 1342 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) }, 1343 { "cpu%u_tx_bytes", 1344 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) }, 1345 { "cpu%u_vf_rx_packets", 1346 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) }, 1347 { "cpu%u_vf_rx_bytes", 1348 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) }, 1349 { "cpu%u_vf_tx_packets", 1350 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) }, 1351 { "cpu%u_vf_tx_bytes", 1352 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) }, 1353 }, vf_stats[] = { 1354 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) }, 1355 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) }, 1356 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) }, 1357 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) }, 1358 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) }, 1359 }; 1360 1361 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats) 1362 #define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats) 1363 1364 /* statistics per queue (rx/tx packets/bytes) */ 1365 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats)) 1366 1367 /* 4 statistics per queue (rx/tx packets/bytes) */ 1368 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4) 1369 1370 static int netvsc_get_sset_count(struct net_device *dev, int string_set) 1371 { 1372 struct net_device_context *ndc = netdev_priv(dev); 1373 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); 1374 1375 if (!nvdev) 1376 return -ENODEV; 1377 1378 switch (string_set) { 1379 case ETH_SS_STATS: 1380 return NETVSC_GLOBAL_STATS_LEN 1381 + NETVSC_VF_STATS_LEN 1382 + NETVSC_QUEUE_STATS_LEN(nvdev) 1383 + NETVSC_PCPU_STATS_LEN; 1384 default: 1385 return -EINVAL; 1386 } 1387 } 1388 1389 static void netvsc_get_ethtool_stats(struct net_device *dev, 1390 struct ethtool_stats *stats, u64 *data) 1391 { 1392 struct net_device_context *ndc = netdev_priv(dev); 1393 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); 1394 const void *nds = &ndc->eth_stats; 1395 const struct netvsc_stats *qstats; 1396 struct netvsc_vf_pcpu_stats sum; 1397 struct netvsc_ethtool_pcpu_stats *pcpu_sum; 1398 unsigned int start; 1399 u64 packets, bytes; 1400 int i, j, cpu; 1401 1402 if (!nvdev) 1403 return; 1404 1405 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++) 1406 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset); 1407 1408 netvsc_get_vf_stats(dev, &sum); 1409 for (j = 0; j < NETVSC_VF_STATS_LEN; j++) 1410 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset); 1411 1412 for (j = 0; j < nvdev->num_chn; j++) { 1413 qstats = &nvdev->chan_table[j].tx_stats; 1414 1415 do { 1416 start = u64_stats_fetch_begin_irq(&qstats->syncp); 1417 packets = qstats->packets; 1418 bytes = qstats->bytes; 1419 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start)); 1420 data[i++] = packets; 1421 data[i++] = bytes; 1422 1423 qstats = &nvdev->chan_table[j].rx_stats; 1424 do { 1425 start = u64_stats_fetch_begin_irq(&qstats->syncp); 1426 packets = qstats->packets; 1427 bytes = qstats->bytes; 1428 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start)); 1429 data[i++] = packets; 1430 data[i++] = bytes; 1431 } 1432 1433 pcpu_sum = kvmalloc_array(num_possible_cpus(), 1434 sizeof(struct netvsc_ethtool_pcpu_stats), 1435 GFP_KERNEL); 1436 netvsc_get_pcpu_stats(dev, pcpu_sum); 1437 for_each_present_cpu(cpu) { 1438 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu]; 1439 1440 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++) 1441 data[i++] = *(u64 *)((void *)this_sum 1442 + pcpu_stats[j].offset); 1443 } 1444 kvfree(pcpu_sum); 1445 } 1446 1447 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data) 1448 { 1449 struct net_device_context *ndc = netdev_priv(dev); 1450 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); 1451 u8 *p = data; 1452 int i, cpu; 1453 1454 if (!nvdev) 1455 return; 1456 1457 switch (stringset) { 1458 case ETH_SS_STATS: 1459 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) { 1460 memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN); 1461 p += ETH_GSTRING_LEN; 1462 } 1463 1464 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) { 1465 memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN); 1466 p += ETH_GSTRING_LEN; 1467 } 1468 1469 for (i = 0; i < nvdev->num_chn; i++) { 1470 sprintf(p, "tx_queue_%u_packets", i); 1471 p += ETH_GSTRING_LEN; 1472 sprintf(p, "tx_queue_%u_bytes", i); 1473 p += ETH_GSTRING_LEN; 1474 sprintf(p, "rx_queue_%u_packets", i); 1475 p += ETH_GSTRING_LEN; 1476 sprintf(p, "rx_queue_%u_bytes", i); 1477 p += ETH_GSTRING_LEN; 1478 } 1479 1480 for_each_present_cpu(cpu) { 1481 for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) { 1482 sprintf(p, pcpu_stats[i].name, cpu); 1483 p += ETH_GSTRING_LEN; 1484 } 1485 } 1486 1487 break; 1488 } 1489 } 1490 1491 static int 1492 netvsc_get_rss_hash_opts(struct net_device_context *ndc, 1493 struct ethtool_rxnfc *info) 1494 { 1495 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3; 1496 1497 info->data = RXH_IP_SRC | RXH_IP_DST; 1498 1499 switch (info->flow_type) { 1500 case TCP_V4_FLOW: 1501 if (ndc->l4_hash & HV_TCP4_L4HASH) 1502 info->data |= l4_flag; 1503 1504 break; 1505 1506 case TCP_V6_FLOW: 1507 if (ndc->l4_hash & HV_TCP6_L4HASH) 1508 info->data |= l4_flag; 1509 1510 break; 1511 1512 case UDP_V4_FLOW: 1513 if (ndc->l4_hash & HV_UDP4_L4HASH) 1514 info->data |= l4_flag; 1515 1516 break; 1517 1518 case UDP_V6_FLOW: 1519 if (ndc->l4_hash & HV_UDP6_L4HASH) 1520 info->data |= l4_flag; 1521 1522 break; 1523 1524 case IPV4_FLOW: 1525 case IPV6_FLOW: 1526 break; 1527 default: 1528 info->data = 0; 1529 break; 1530 } 1531 1532 return 0; 1533 } 1534 1535 static int 1536 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info, 1537 u32 *rules) 1538 { 1539 struct net_device_context *ndc = netdev_priv(dev); 1540 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); 1541 1542 if (!nvdev) 1543 return -ENODEV; 1544 1545 switch (info->cmd) { 1546 case ETHTOOL_GRXRINGS: 1547 info->data = nvdev->num_chn; 1548 return 0; 1549 1550 case ETHTOOL_GRXFH: 1551 return netvsc_get_rss_hash_opts(ndc, info); 1552 } 1553 return -EOPNOTSUPP; 1554 } 1555 1556 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc, 1557 struct ethtool_rxnfc *info) 1558 { 1559 if (info->data == (RXH_IP_SRC | RXH_IP_DST | 1560 RXH_L4_B_0_1 | RXH_L4_B_2_3)) { 1561 switch (info->flow_type) { 1562 case TCP_V4_FLOW: 1563 ndc->l4_hash |= HV_TCP4_L4HASH; 1564 break; 1565 1566 case TCP_V6_FLOW: 1567 ndc->l4_hash |= HV_TCP6_L4HASH; 1568 break; 1569 1570 case UDP_V4_FLOW: 1571 ndc->l4_hash |= HV_UDP4_L4HASH; 1572 break; 1573 1574 case UDP_V6_FLOW: 1575 ndc->l4_hash |= HV_UDP6_L4HASH; 1576 break; 1577 1578 default: 1579 return -EOPNOTSUPP; 1580 } 1581 1582 return 0; 1583 } 1584 1585 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) { 1586 switch (info->flow_type) { 1587 case TCP_V4_FLOW: 1588 ndc->l4_hash &= ~HV_TCP4_L4HASH; 1589 break; 1590 1591 case TCP_V6_FLOW: 1592 ndc->l4_hash &= ~HV_TCP6_L4HASH; 1593 break; 1594 1595 case UDP_V4_FLOW: 1596 ndc->l4_hash &= ~HV_UDP4_L4HASH; 1597 break; 1598 1599 case UDP_V6_FLOW: 1600 ndc->l4_hash &= ~HV_UDP6_L4HASH; 1601 break; 1602 1603 default: 1604 return -EOPNOTSUPP; 1605 } 1606 1607 return 0; 1608 } 1609 1610 return -EOPNOTSUPP; 1611 } 1612 1613 static int 1614 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info) 1615 { 1616 struct net_device_context *ndc = netdev_priv(ndev); 1617 1618 if (info->cmd == ETHTOOL_SRXFH) 1619 return netvsc_set_rss_hash_opts(ndc, info); 1620 1621 return -EOPNOTSUPP; 1622 } 1623 1624 static u32 netvsc_get_rxfh_key_size(struct net_device *dev) 1625 { 1626 return NETVSC_HASH_KEYLEN; 1627 } 1628 1629 static u32 netvsc_rss_indir_size(struct net_device *dev) 1630 { 1631 return ITAB_NUM; 1632 } 1633 1634 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key, 1635 u8 *hfunc) 1636 { 1637 struct net_device_context *ndc = netdev_priv(dev); 1638 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev); 1639 struct rndis_device *rndis_dev; 1640 int i; 1641 1642 if (!ndev) 1643 return -ENODEV; 1644 1645 if (hfunc) 1646 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */ 1647 1648 rndis_dev = ndev->extension; 1649 if (indir) { 1650 for (i = 0; i < ITAB_NUM; i++) 1651 indir[i] = rndis_dev->rx_table[i]; 1652 } 1653 1654 if (key) 1655 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN); 1656 1657 return 0; 1658 } 1659 1660 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir, 1661 const u8 *key, const u8 hfunc) 1662 { 1663 struct net_device_context *ndc = netdev_priv(dev); 1664 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev); 1665 struct rndis_device *rndis_dev; 1666 int i; 1667 1668 if (!ndev) 1669 return -ENODEV; 1670 1671 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP) 1672 return -EOPNOTSUPP; 1673 1674 rndis_dev = ndev->extension; 1675 if (indir) { 1676 for (i = 0; i < ITAB_NUM; i++) 1677 if (indir[i] >= ndev->num_chn) 1678 return -EINVAL; 1679 1680 for (i = 0; i < ITAB_NUM; i++) 1681 rndis_dev->rx_table[i] = indir[i]; 1682 } 1683 1684 if (!key) { 1685 if (!indir) 1686 return 0; 1687 1688 key = rndis_dev->rss_key; 1689 } 1690 1691 return rndis_filter_set_rss_param(rndis_dev, key); 1692 } 1693 1694 /* Hyper-V RNDIS protocol does not have ring in the HW sense. 1695 * It does have pre-allocated receive area which is divided into sections. 1696 */ 1697 static void __netvsc_get_ringparam(struct netvsc_device *nvdev, 1698 struct ethtool_ringparam *ring) 1699 { 1700 u32 max_buf_size; 1701 1702 ring->rx_pending = nvdev->recv_section_cnt; 1703 ring->tx_pending = nvdev->send_section_cnt; 1704 1705 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2) 1706 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY; 1707 else 1708 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE; 1709 1710 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size; 1711 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE 1712 / nvdev->send_section_size; 1713 } 1714 1715 static void netvsc_get_ringparam(struct net_device *ndev, 1716 struct ethtool_ringparam *ring) 1717 { 1718 struct net_device_context *ndevctx = netdev_priv(ndev); 1719 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev); 1720 1721 if (!nvdev) 1722 return; 1723 1724 __netvsc_get_ringparam(nvdev, ring); 1725 } 1726 1727 static int netvsc_set_ringparam(struct net_device *ndev, 1728 struct ethtool_ringparam *ring) 1729 { 1730 struct net_device_context *ndevctx = netdev_priv(ndev); 1731 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev); 1732 struct netvsc_device_info *device_info; 1733 struct ethtool_ringparam orig; 1734 u32 new_tx, new_rx; 1735 int ret = 0; 1736 1737 if (!nvdev || nvdev->destroy) 1738 return -ENODEV; 1739 1740 memset(&orig, 0, sizeof(orig)); 1741 __netvsc_get_ringparam(nvdev, &orig); 1742 1743 new_tx = clamp_t(u32, ring->tx_pending, 1744 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending); 1745 new_rx = clamp_t(u32, ring->rx_pending, 1746 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending); 1747 1748 if (new_tx == orig.tx_pending && 1749 new_rx == orig.rx_pending) 1750 return 0; /* no change */ 1751 1752 device_info = netvsc_devinfo_get(nvdev); 1753 1754 if (!device_info) 1755 return -ENOMEM; 1756 1757 device_info->send_sections = new_tx; 1758 device_info->recv_sections = new_rx; 1759 1760 ret = netvsc_detach(ndev, nvdev); 1761 if (ret) 1762 goto out; 1763 1764 ret = netvsc_attach(ndev, device_info); 1765 if (ret) { 1766 device_info->send_sections = orig.tx_pending; 1767 device_info->recv_sections = orig.rx_pending; 1768 1769 if (netvsc_attach(ndev, device_info)) 1770 netdev_err(ndev, "restoring ringparam failed"); 1771 } 1772 1773 out: 1774 kfree(device_info); 1775 return ret; 1776 } 1777 1778 static int netvsc_set_features(struct net_device *ndev, 1779 netdev_features_t features) 1780 { 1781 netdev_features_t change = features ^ ndev->features; 1782 struct net_device_context *ndevctx = netdev_priv(ndev); 1783 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev); 1784 struct ndis_offload_params offloads; 1785 1786 if (!nvdev || nvdev->destroy) 1787 return -ENODEV; 1788 1789 if (!(change & NETIF_F_LRO)) 1790 return 0; 1791 1792 memset(&offloads, 0, sizeof(struct ndis_offload_params)); 1793 1794 if (features & NETIF_F_LRO) { 1795 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED; 1796 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED; 1797 } else { 1798 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED; 1799 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED; 1800 } 1801 1802 return rndis_filter_set_offload_params(ndev, nvdev, &offloads); 1803 } 1804 1805 static u32 netvsc_get_msglevel(struct net_device *ndev) 1806 { 1807 struct net_device_context *ndev_ctx = netdev_priv(ndev); 1808 1809 return ndev_ctx->msg_enable; 1810 } 1811 1812 static void netvsc_set_msglevel(struct net_device *ndev, u32 val) 1813 { 1814 struct net_device_context *ndev_ctx = netdev_priv(ndev); 1815 1816 ndev_ctx->msg_enable = val; 1817 } 1818 1819 static const struct ethtool_ops ethtool_ops = { 1820 .get_drvinfo = netvsc_get_drvinfo, 1821 .get_msglevel = netvsc_get_msglevel, 1822 .set_msglevel = netvsc_set_msglevel, 1823 .get_link = ethtool_op_get_link, 1824 .get_ethtool_stats = netvsc_get_ethtool_stats, 1825 .get_sset_count = netvsc_get_sset_count, 1826 .get_strings = netvsc_get_strings, 1827 .get_channels = netvsc_get_channels, 1828 .set_channels = netvsc_set_channels, 1829 .get_ts_info = ethtool_op_get_ts_info, 1830 .get_rxnfc = netvsc_get_rxnfc, 1831 .set_rxnfc = netvsc_set_rxnfc, 1832 .get_rxfh_key_size = netvsc_get_rxfh_key_size, 1833 .get_rxfh_indir_size = netvsc_rss_indir_size, 1834 .get_rxfh = netvsc_get_rxfh, 1835 .set_rxfh = netvsc_set_rxfh, 1836 .get_link_ksettings = netvsc_get_link_ksettings, 1837 .set_link_ksettings = netvsc_set_link_ksettings, 1838 .get_ringparam = netvsc_get_ringparam, 1839 .set_ringparam = netvsc_set_ringparam, 1840 }; 1841 1842 static const struct net_device_ops device_ops = { 1843 .ndo_open = netvsc_open, 1844 .ndo_stop = netvsc_close, 1845 .ndo_start_xmit = netvsc_start_xmit, 1846 .ndo_change_rx_flags = netvsc_change_rx_flags, 1847 .ndo_set_rx_mode = netvsc_set_rx_mode, 1848 .ndo_set_features = netvsc_set_features, 1849 .ndo_change_mtu = netvsc_change_mtu, 1850 .ndo_validate_addr = eth_validate_addr, 1851 .ndo_set_mac_address = netvsc_set_mac_addr, 1852 .ndo_select_queue = netvsc_select_queue, 1853 .ndo_get_stats64 = netvsc_get_stats64, 1854 }; 1855 1856 /* 1857 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link 1858 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is 1859 * present send GARP packet to network peers with netif_notify_peers(). 1860 */ 1861 static void netvsc_link_change(struct work_struct *w) 1862 { 1863 struct net_device_context *ndev_ctx = 1864 container_of(w, struct net_device_context, dwork.work); 1865 struct hv_device *device_obj = ndev_ctx->device_ctx; 1866 struct net_device *net = hv_get_drvdata(device_obj); 1867 struct netvsc_device *net_device; 1868 struct rndis_device *rdev; 1869 struct netvsc_reconfig *event = NULL; 1870 bool notify = false, reschedule = false; 1871 unsigned long flags, next_reconfig, delay; 1872 1873 /* if changes are happening, comeback later */ 1874 if (!rtnl_trylock()) { 1875 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT); 1876 return; 1877 } 1878 1879 net_device = rtnl_dereference(ndev_ctx->nvdev); 1880 if (!net_device) 1881 goto out_unlock; 1882 1883 rdev = net_device->extension; 1884 1885 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT; 1886 if (time_is_after_jiffies(next_reconfig)) { 1887 /* link_watch only sends one notification with current state 1888 * per second, avoid doing reconfig more frequently. Handle 1889 * wrap around. 1890 */ 1891 delay = next_reconfig - jiffies; 1892 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT; 1893 schedule_delayed_work(&ndev_ctx->dwork, delay); 1894 goto out_unlock; 1895 } 1896 ndev_ctx->last_reconfig = jiffies; 1897 1898 spin_lock_irqsave(&ndev_ctx->lock, flags); 1899 if (!list_empty(&ndev_ctx->reconfig_events)) { 1900 event = list_first_entry(&ndev_ctx->reconfig_events, 1901 struct netvsc_reconfig, list); 1902 list_del(&event->list); 1903 reschedule = !list_empty(&ndev_ctx->reconfig_events); 1904 } 1905 spin_unlock_irqrestore(&ndev_ctx->lock, flags); 1906 1907 if (!event) 1908 goto out_unlock; 1909 1910 switch (event->event) { 1911 /* Only the following events are possible due to the check in 1912 * netvsc_linkstatus_callback() 1913 */ 1914 case RNDIS_STATUS_MEDIA_CONNECT: 1915 if (rdev->link_state) { 1916 rdev->link_state = false; 1917 netif_carrier_on(net); 1918 netvsc_tx_enable(net_device, net); 1919 } else { 1920 notify = true; 1921 } 1922 kfree(event); 1923 break; 1924 case RNDIS_STATUS_MEDIA_DISCONNECT: 1925 if (!rdev->link_state) { 1926 rdev->link_state = true; 1927 netif_carrier_off(net); 1928 netvsc_tx_disable(net_device, net); 1929 } 1930 kfree(event); 1931 break; 1932 case RNDIS_STATUS_NETWORK_CHANGE: 1933 /* Only makes sense if carrier is present */ 1934 if (!rdev->link_state) { 1935 rdev->link_state = true; 1936 netif_carrier_off(net); 1937 netvsc_tx_disable(net_device, net); 1938 event->event = RNDIS_STATUS_MEDIA_CONNECT; 1939 spin_lock_irqsave(&ndev_ctx->lock, flags); 1940 list_add(&event->list, &ndev_ctx->reconfig_events); 1941 spin_unlock_irqrestore(&ndev_ctx->lock, flags); 1942 reschedule = true; 1943 } 1944 break; 1945 } 1946 1947 rtnl_unlock(); 1948 1949 if (notify) 1950 netdev_notify_peers(net); 1951 1952 /* link_watch only sends one notification with current state per 1953 * second, handle next reconfig event in 2 seconds. 1954 */ 1955 if (reschedule) 1956 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT); 1957 1958 return; 1959 1960 out_unlock: 1961 rtnl_unlock(); 1962 } 1963 1964 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev) 1965 { 1966 struct net_device_context *net_device_ctx; 1967 struct net_device *dev; 1968 1969 dev = netdev_master_upper_dev_get(vf_netdev); 1970 if (!dev || dev->netdev_ops != &device_ops) 1971 return NULL; /* not a netvsc device */ 1972 1973 net_device_ctx = netdev_priv(dev); 1974 if (!rtnl_dereference(net_device_ctx->nvdev)) 1975 return NULL; /* device is removed */ 1976 1977 return dev; 1978 } 1979 1980 /* Called when VF is injecting data into network stack. 1981 * Change the associated network device from VF to netvsc. 1982 * note: already called with rcu_read_lock 1983 */ 1984 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb) 1985 { 1986 struct sk_buff *skb = *pskb; 1987 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data); 1988 struct net_device_context *ndev_ctx = netdev_priv(ndev); 1989 struct netvsc_vf_pcpu_stats *pcpu_stats 1990 = this_cpu_ptr(ndev_ctx->vf_stats); 1991 1992 skb = skb_share_check(skb, GFP_ATOMIC); 1993 if (unlikely(!skb)) 1994 return RX_HANDLER_CONSUMED; 1995 1996 *pskb = skb; 1997 1998 skb->dev = ndev; 1999 2000 u64_stats_update_begin(&pcpu_stats->syncp); 2001 pcpu_stats->rx_packets++; 2002 pcpu_stats->rx_bytes += skb->len; 2003 u64_stats_update_end(&pcpu_stats->syncp); 2004 2005 return RX_HANDLER_ANOTHER; 2006 } 2007 2008 static int netvsc_vf_join(struct net_device *vf_netdev, 2009 struct net_device *ndev) 2010 { 2011 struct net_device_context *ndev_ctx = netdev_priv(ndev); 2012 int ret; 2013 2014 ret = netdev_rx_handler_register(vf_netdev, 2015 netvsc_vf_handle_frame, ndev); 2016 if (ret != 0) { 2017 netdev_err(vf_netdev, 2018 "can not register netvsc VF receive handler (err = %d)\n", 2019 ret); 2020 goto rx_handler_failed; 2021 } 2022 2023 ret = netdev_master_upper_dev_link(vf_netdev, ndev, 2024 NULL, NULL, NULL); 2025 if (ret != 0) { 2026 netdev_err(vf_netdev, 2027 "can not set master device %s (err = %d)\n", 2028 ndev->name, ret); 2029 goto upper_link_failed; 2030 } 2031 2032 /* set slave flag before open to prevent IPv6 addrconf */ 2033 vf_netdev->flags |= IFF_SLAVE; 2034 2035 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT); 2036 2037 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev); 2038 2039 netdev_info(vf_netdev, "joined to %s\n", ndev->name); 2040 return 0; 2041 2042 upper_link_failed: 2043 netdev_rx_handler_unregister(vf_netdev); 2044 rx_handler_failed: 2045 return ret; 2046 } 2047 2048 static void __netvsc_vf_setup(struct net_device *ndev, 2049 struct net_device *vf_netdev) 2050 { 2051 int ret; 2052 2053 /* Align MTU of VF with master */ 2054 ret = dev_set_mtu(vf_netdev, ndev->mtu); 2055 if (ret) 2056 netdev_warn(vf_netdev, 2057 "unable to change mtu to %u\n", ndev->mtu); 2058 2059 /* set multicast etc flags on VF */ 2060 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL); 2061 2062 /* sync address list from ndev to VF */ 2063 netif_addr_lock_bh(ndev); 2064 dev_uc_sync(vf_netdev, ndev); 2065 dev_mc_sync(vf_netdev, ndev); 2066 netif_addr_unlock_bh(ndev); 2067 2068 if (netif_running(ndev)) { 2069 ret = dev_open(vf_netdev, NULL); 2070 if (ret) 2071 netdev_warn(vf_netdev, 2072 "unable to open: %d\n", ret); 2073 } 2074 } 2075 2076 /* Setup VF as slave of the synthetic device. 2077 * Runs in workqueue to avoid recursion in netlink callbacks. 2078 */ 2079 static void netvsc_vf_setup(struct work_struct *w) 2080 { 2081 struct net_device_context *ndev_ctx 2082 = container_of(w, struct net_device_context, vf_takeover.work); 2083 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx); 2084 struct net_device *vf_netdev; 2085 2086 if (!rtnl_trylock()) { 2087 schedule_delayed_work(&ndev_ctx->vf_takeover, 0); 2088 return; 2089 } 2090 2091 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev); 2092 if (vf_netdev) 2093 __netvsc_vf_setup(ndev, vf_netdev); 2094 2095 rtnl_unlock(); 2096 } 2097 2098 /* Find netvsc by VF serial number. 2099 * The PCI hyperv controller records the serial number as the slot kobj name. 2100 */ 2101 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev) 2102 { 2103 struct device *parent = vf_netdev->dev.parent; 2104 struct net_device_context *ndev_ctx; 2105 struct pci_dev *pdev; 2106 u32 serial; 2107 2108 if (!parent || !dev_is_pci(parent)) 2109 return NULL; /* not a PCI device */ 2110 2111 pdev = to_pci_dev(parent); 2112 if (!pdev->slot) { 2113 netdev_notice(vf_netdev, "no PCI slot information\n"); 2114 return NULL; 2115 } 2116 2117 if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) { 2118 netdev_notice(vf_netdev, "Invalid vf serial:%s\n", 2119 pci_slot_name(pdev->slot)); 2120 return NULL; 2121 } 2122 2123 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) { 2124 if (!ndev_ctx->vf_alloc) 2125 continue; 2126 2127 if (ndev_ctx->vf_serial == serial) 2128 return hv_get_drvdata(ndev_ctx->device_ctx); 2129 } 2130 2131 netdev_notice(vf_netdev, 2132 "no netdev found for vf serial:%u\n", serial); 2133 return NULL; 2134 } 2135 2136 static int netvsc_register_vf(struct net_device *vf_netdev) 2137 { 2138 struct net_device_context *net_device_ctx; 2139 struct netvsc_device *netvsc_dev; 2140 struct net_device *ndev; 2141 int ret; 2142 2143 if (vf_netdev->addr_len != ETH_ALEN) 2144 return NOTIFY_DONE; 2145 2146 ndev = get_netvsc_byslot(vf_netdev); 2147 if (!ndev) 2148 return NOTIFY_DONE; 2149 2150 net_device_ctx = netdev_priv(ndev); 2151 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev); 2152 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev)) 2153 return NOTIFY_DONE; 2154 2155 /* if synthetic interface is a different namespace, 2156 * then move the VF to that namespace; join will be 2157 * done again in that context. 2158 */ 2159 if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) { 2160 ret = dev_change_net_namespace(vf_netdev, 2161 dev_net(ndev), "eth%d"); 2162 if (ret) 2163 netdev_err(vf_netdev, 2164 "could not move to same namespace as %s: %d\n", 2165 ndev->name, ret); 2166 else 2167 netdev_info(vf_netdev, 2168 "VF moved to namespace with: %s\n", 2169 ndev->name); 2170 return NOTIFY_DONE; 2171 } 2172 2173 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name); 2174 2175 if (netvsc_vf_join(vf_netdev, ndev) != 0) 2176 return NOTIFY_DONE; 2177 2178 dev_hold(vf_netdev); 2179 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev); 2180 return NOTIFY_OK; 2181 } 2182 2183 /* VF up/down change detected, schedule to change data path */ 2184 static int netvsc_vf_changed(struct net_device *vf_netdev) 2185 { 2186 struct net_device_context *net_device_ctx; 2187 struct netvsc_device *netvsc_dev; 2188 struct net_device *ndev; 2189 bool vf_is_up = netif_running(vf_netdev); 2190 2191 ndev = get_netvsc_byref(vf_netdev); 2192 if (!ndev) 2193 return NOTIFY_DONE; 2194 2195 net_device_ctx = netdev_priv(ndev); 2196 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev); 2197 if (!netvsc_dev) 2198 return NOTIFY_DONE; 2199 2200 netvsc_switch_datapath(ndev, vf_is_up); 2201 netdev_info(ndev, "Data path switched %s VF: %s\n", 2202 vf_is_up ? "to" : "from", vf_netdev->name); 2203 2204 return NOTIFY_OK; 2205 } 2206 2207 static int netvsc_unregister_vf(struct net_device *vf_netdev) 2208 { 2209 struct net_device *ndev; 2210 struct net_device_context *net_device_ctx; 2211 2212 ndev = get_netvsc_byref(vf_netdev); 2213 if (!ndev) 2214 return NOTIFY_DONE; 2215 2216 net_device_ctx = netdev_priv(ndev); 2217 cancel_delayed_work_sync(&net_device_ctx->vf_takeover); 2218 2219 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name); 2220 2221 netdev_rx_handler_unregister(vf_netdev); 2222 netdev_upper_dev_unlink(vf_netdev, ndev); 2223 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL); 2224 dev_put(vf_netdev); 2225 2226 return NOTIFY_OK; 2227 } 2228 2229 static int netvsc_probe(struct hv_device *dev, 2230 const struct hv_vmbus_device_id *dev_id) 2231 { 2232 struct net_device *net = NULL; 2233 struct net_device_context *net_device_ctx; 2234 struct netvsc_device_info *device_info = NULL; 2235 struct netvsc_device *nvdev; 2236 int ret = -ENOMEM; 2237 2238 net = alloc_etherdev_mq(sizeof(struct net_device_context), 2239 VRSS_CHANNEL_MAX); 2240 if (!net) 2241 goto no_net; 2242 2243 netif_carrier_off(net); 2244 2245 netvsc_init_settings(net); 2246 2247 net_device_ctx = netdev_priv(net); 2248 net_device_ctx->device_ctx = dev; 2249 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg); 2250 if (netif_msg_probe(net_device_ctx)) 2251 netdev_dbg(net, "netvsc msg_enable: %d\n", 2252 net_device_ctx->msg_enable); 2253 2254 hv_set_drvdata(dev, net); 2255 2256 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change); 2257 2258 spin_lock_init(&net_device_ctx->lock); 2259 INIT_LIST_HEAD(&net_device_ctx->reconfig_events); 2260 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup); 2261 2262 net_device_ctx->vf_stats 2263 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats); 2264 if (!net_device_ctx->vf_stats) 2265 goto no_stats; 2266 2267 net->netdev_ops = &device_ops; 2268 net->ethtool_ops = ðtool_ops; 2269 SET_NETDEV_DEV(net, &dev->device); 2270 2271 /* We always need headroom for rndis header */ 2272 net->needed_headroom = RNDIS_AND_PPI_SIZE; 2273 2274 /* Initialize the number of queues to be 1, we may change it if more 2275 * channels are offered later. 2276 */ 2277 netif_set_real_num_tx_queues(net, 1); 2278 netif_set_real_num_rx_queues(net, 1); 2279 2280 /* Notify the netvsc driver of the new device */ 2281 device_info = netvsc_devinfo_get(NULL); 2282 2283 if (!device_info) { 2284 ret = -ENOMEM; 2285 goto devinfo_failed; 2286 } 2287 2288 nvdev = rndis_filter_device_add(dev, device_info); 2289 if (IS_ERR(nvdev)) { 2290 ret = PTR_ERR(nvdev); 2291 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret); 2292 goto rndis_failed; 2293 } 2294 2295 memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN); 2296 2297 /* We must get rtnl lock before scheduling nvdev->subchan_work, 2298 * otherwise netvsc_subchan_work() can get rtnl lock first and wait 2299 * all subchannels to show up, but that may not happen because 2300 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer() 2301 * -> ... -> device_add() -> ... -> __device_attach() can't get 2302 * the device lock, so all the subchannels can't be processed -- 2303 * finally netvsc_subchan_work() hangs forever. 2304 */ 2305 rtnl_lock(); 2306 2307 if (nvdev->num_chn > 1) 2308 schedule_work(&nvdev->subchan_work); 2309 2310 /* hw_features computed in rndis_netdev_set_hwcaps() */ 2311 net->features = net->hw_features | 2312 NETIF_F_HIGHDMA | NETIF_F_SG | 2313 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX; 2314 net->vlan_features = net->features; 2315 2316 netdev_lockdep_set_classes(net); 2317 2318 /* MTU range: 68 - 1500 or 65521 */ 2319 net->min_mtu = NETVSC_MTU_MIN; 2320 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2) 2321 net->max_mtu = NETVSC_MTU - ETH_HLEN; 2322 else 2323 net->max_mtu = ETH_DATA_LEN; 2324 2325 ret = register_netdevice(net); 2326 if (ret != 0) { 2327 pr_err("Unable to register netdev.\n"); 2328 goto register_failed; 2329 } 2330 2331 list_add(&net_device_ctx->list, &netvsc_dev_list); 2332 rtnl_unlock(); 2333 2334 kfree(device_info); 2335 return 0; 2336 2337 register_failed: 2338 rtnl_unlock(); 2339 rndis_filter_device_remove(dev, nvdev); 2340 rndis_failed: 2341 kfree(device_info); 2342 devinfo_failed: 2343 free_percpu(net_device_ctx->vf_stats); 2344 no_stats: 2345 hv_set_drvdata(dev, NULL); 2346 free_netdev(net); 2347 no_net: 2348 return ret; 2349 } 2350 2351 static int netvsc_remove(struct hv_device *dev) 2352 { 2353 struct net_device_context *ndev_ctx; 2354 struct net_device *vf_netdev, *net; 2355 struct netvsc_device *nvdev; 2356 2357 net = hv_get_drvdata(dev); 2358 if (net == NULL) { 2359 dev_err(&dev->device, "No net device to remove\n"); 2360 return 0; 2361 } 2362 2363 ndev_ctx = netdev_priv(net); 2364 2365 cancel_delayed_work_sync(&ndev_ctx->dwork); 2366 2367 rtnl_lock(); 2368 nvdev = rtnl_dereference(ndev_ctx->nvdev); 2369 if (nvdev) 2370 cancel_work_sync(&nvdev->subchan_work); 2371 2372 /* 2373 * Call to the vsc driver to let it know that the device is being 2374 * removed. Also blocks mtu and channel changes. 2375 */ 2376 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev); 2377 if (vf_netdev) 2378 netvsc_unregister_vf(vf_netdev); 2379 2380 if (nvdev) 2381 rndis_filter_device_remove(dev, nvdev); 2382 2383 unregister_netdevice(net); 2384 list_del(&ndev_ctx->list); 2385 2386 rtnl_unlock(); 2387 2388 hv_set_drvdata(dev, NULL); 2389 2390 free_percpu(ndev_ctx->vf_stats); 2391 free_netdev(net); 2392 return 0; 2393 } 2394 2395 static const struct hv_vmbus_device_id id_table[] = { 2396 /* Network guid */ 2397 { HV_NIC_GUID, }, 2398 { }, 2399 }; 2400 2401 MODULE_DEVICE_TABLE(vmbus, id_table); 2402 2403 /* The one and only one */ 2404 static struct hv_driver netvsc_drv = { 2405 .name = KBUILD_MODNAME, 2406 .id_table = id_table, 2407 .probe = netvsc_probe, 2408 .remove = netvsc_remove, 2409 .driver = { 2410 .probe_type = PROBE_PREFER_ASYNCHRONOUS, 2411 }, 2412 }; 2413 2414 /* 2415 * On Hyper-V, every VF interface is matched with a corresponding 2416 * synthetic interface. The synthetic interface is presented first 2417 * to the guest. When the corresponding VF instance is registered, 2418 * we will take care of switching the data path. 2419 */ 2420 static int netvsc_netdev_event(struct notifier_block *this, 2421 unsigned long event, void *ptr) 2422 { 2423 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 2424 2425 /* Skip our own events */ 2426 if (event_dev->netdev_ops == &device_ops) 2427 return NOTIFY_DONE; 2428 2429 /* Avoid non-Ethernet type devices */ 2430 if (event_dev->type != ARPHRD_ETHER) 2431 return NOTIFY_DONE; 2432 2433 /* Avoid Vlan dev with same MAC registering as VF */ 2434 if (is_vlan_dev(event_dev)) 2435 return NOTIFY_DONE; 2436 2437 /* Avoid Bonding master dev with same MAC registering as VF */ 2438 if ((event_dev->priv_flags & IFF_BONDING) && 2439 (event_dev->flags & IFF_MASTER)) 2440 return NOTIFY_DONE; 2441 2442 switch (event) { 2443 case NETDEV_REGISTER: 2444 return netvsc_register_vf(event_dev); 2445 case NETDEV_UNREGISTER: 2446 return netvsc_unregister_vf(event_dev); 2447 case NETDEV_UP: 2448 case NETDEV_DOWN: 2449 return netvsc_vf_changed(event_dev); 2450 default: 2451 return NOTIFY_DONE; 2452 } 2453 } 2454 2455 static struct notifier_block netvsc_netdev_notifier = { 2456 .notifier_call = netvsc_netdev_event, 2457 }; 2458 2459 static void __exit netvsc_drv_exit(void) 2460 { 2461 unregister_netdevice_notifier(&netvsc_netdev_notifier); 2462 vmbus_driver_unregister(&netvsc_drv); 2463 } 2464 2465 static int __init netvsc_drv_init(void) 2466 { 2467 int ret; 2468 2469 if (ring_size < RING_SIZE_MIN) { 2470 ring_size = RING_SIZE_MIN; 2471 pr_info("Increased ring_size to %u (min allowed)\n", 2472 ring_size); 2473 } 2474 netvsc_ring_bytes = ring_size * PAGE_SIZE; 2475 2476 ret = vmbus_driver_register(&netvsc_drv); 2477 if (ret) 2478 return ret; 2479 2480 register_netdevice_notifier(&netvsc_netdev_notifier); 2481 return 0; 2482 } 2483 2484 MODULE_LICENSE("GPL"); 2485 MODULE_DESCRIPTION("Microsoft Hyper-V network driver"); 2486 2487 module_init(netvsc_drv_init); 2488 module_exit(netvsc_drv_exit); 2489