1 /* 2 * Copyright (c) 2009, Microsoft Corporation. 3 * 4 * This program is free software; you can redistribute it and/or modify it 5 * under the terms and conditions of the GNU General Public License, 6 * version 2, as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope it will be useful, but WITHOUT 9 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 10 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 11 * more details. 12 * 13 * You should have received a copy of the GNU General Public License along with 14 * this program; if not, see <http://www.gnu.org/licenses/>. 15 * 16 * Authors: 17 * Haiyang Zhang <haiyangz@microsoft.com> 18 * Hank Janssen <hjanssen@microsoft.com> 19 */ 20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 21 22 #include <linux/init.h> 23 #include <linux/atomic.h> 24 #include <linux/module.h> 25 #include <linux/highmem.h> 26 #include <linux/device.h> 27 #include <linux/io.h> 28 #include <linux/delay.h> 29 #include <linux/netdevice.h> 30 #include <linux/inetdevice.h> 31 #include <linux/etherdevice.h> 32 #include <linux/pci.h> 33 #include <linux/skbuff.h> 34 #include <linux/if_vlan.h> 35 #include <linux/in.h> 36 #include <linux/slab.h> 37 #include <linux/rtnetlink.h> 38 #include <linux/netpoll.h> 39 40 #include <net/arp.h> 41 #include <net/route.h> 42 #include <net/sock.h> 43 #include <net/pkt_sched.h> 44 #include <net/checksum.h> 45 #include <net/ip6_checksum.h> 46 47 #include "hyperv_net.h" 48 49 #define RING_SIZE_MIN 64 50 #define RETRY_US_LO 5000 51 #define RETRY_US_HI 10000 52 #define RETRY_MAX 2000 /* >10 sec */ 53 54 #define LINKCHANGE_INT (2 * HZ) 55 #define VF_TAKEOVER_INT (HZ / 10) 56 57 static unsigned int ring_size __ro_after_init = 128; 58 module_param(ring_size, uint, 0444); 59 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)"); 60 unsigned int netvsc_ring_bytes __ro_after_init; 61 62 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE | 63 NETIF_MSG_LINK | NETIF_MSG_IFUP | 64 NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR | 65 NETIF_MSG_TX_ERR; 66 67 static int debug = -1; 68 module_param(debug, int, 0444); 69 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)"); 70 71 static LIST_HEAD(netvsc_dev_list); 72 73 static void netvsc_change_rx_flags(struct net_device *net, int change) 74 { 75 struct net_device_context *ndev_ctx = netdev_priv(net); 76 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev); 77 int inc; 78 79 if (!vf_netdev) 80 return; 81 82 if (change & IFF_PROMISC) { 83 inc = (net->flags & IFF_PROMISC) ? 1 : -1; 84 dev_set_promiscuity(vf_netdev, inc); 85 } 86 87 if (change & IFF_ALLMULTI) { 88 inc = (net->flags & IFF_ALLMULTI) ? 1 : -1; 89 dev_set_allmulti(vf_netdev, inc); 90 } 91 } 92 93 static void netvsc_set_rx_mode(struct net_device *net) 94 { 95 struct net_device_context *ndev_ctx = netdev_priv(net); 96 struct net_device *vf_netdev; 97 struct netvsc_device *nvdev; 98 99 rcu_read_lock(); 100 vf_netdev = rcu_dereference(ndev_ctx->vf_netdev); 101 if (vf_netdev) { 102 dev_uc_sync(vf_netdev, net); 103 dev_mc_sync(vf_netdev, net); 104 } 105 106 nvdev = rcu_dereference(ndev_ctx->nvdev); 107 if (nvdev) 108 rndis_filter_update(nvdev); 109 rcu_read_unlock(); 110 } 111 112 static void netvsc_tx_enable(struct netvsc_device *nvscdev, 113 struct net_device *ndev) 114 { 115 nvscdev->tx_disable = false; 116 virt_wmb(); /* ensure queue wake up mechanism is on */ 117 118 netif_tx_wake_all_queues(ndev); 119 } 120 121 static int netvsc_open(struct net_device *net) 122 { 123 struct net_device_context *ndev_ctx = netdev_priv(net); 124 struct net_device *vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev); 125 struct netvsc_device *nvdev = rtnl_dereference(ndev_ctx->nvdev); 126 struct rndis_device *rdev; 127 int ret = 0; 128 129 netif_carrier_off(net); 130 131 /* Open up the device */ 132 ret = rndis_filter_open(nvdev); 133 if (ret != 0) { 134 netdev_err(net, "unable to open device (ret %d).\n", ret); 135 return ret; 136 } 137 138 rdev = nvdev->extension; 139 if (!rdev->link_state) { 140 netif_carrier_on(net); 141 netvsc_tx_enable(nvdev, net); 142 } 143 144 if (vf_netdev) { 145 /* Setting synthetic device up transparently sets 146 * slave as up. If open fails, then slave will be 147 * still be offline (and not used). 148 */ 149 ret = dev_open(vf_netdev, NULL); 150 if (ret) 151 netdev_warn(net, 152 "unable to open slave: %s: %d\n", 153 vf_netdev->name, ret); 154 } 155 return 0; 156 } 157 158 static int netvsc_wait_until_empty(struct netvsc_device *nvdev) 159 { 160 unsigned int retry = 0; 161 int i; 162 163 /* Ensure pending bytes in ring are read */ 164 for (;;) { 165 u32 aread = 0; 166 167 for (i = 0; i < nvdev->num_chn; i++) { 168 struct vmbus_channel *chn 169 = nvdev->chan_table[i].channel; 170 171 if (!chn) 172 continue; 173 174 /* make sure receive not running now */ 175 napi_synchronize(&nvdev->chan_table[i].napi); 176 177 aread = hv_get_bytes_to_read(&chn->inbound); 178 if (aread) 179 break; 180 181 aread = hv_get_bytes_to_read(&chn->outbound); 182 if (aread) 183 break; 184 } 185 186 if (aread == 0) 187 return 0; 188 189 if (++retry > RETRY_MAX) 190 return -ETIMEDOUT; 191 192 usleep_range(RETRY_US_LO, RETRY_US_HI); 193 } 194 } 195 196 static void netvsc_tx_disable(struct netvsc_device *nvscdev, 197 struct net_device *ndev) 198 { 199 if (nvscdev) { 200 nvscdev->tx_disable = true; 201 virt_wmb(); /* ensure txq will not wake up after stop */ 202 } 203 204 netif_tx_disable(ndev); 205 } 206 207 static int netvsc_close(struct net_device *net) 208 { 209 struct net_device_context *net_device_ctx = netdev_priv(net); 210 struct net_device *vf_netdev 211 = rtnl_dereference(net_device_ctx->vf_netdev); 212 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev); 213 int ret; 214 215 netvsc_tx_disable(nvdev, net); 216 217 /* No need to close rndis filter if it is removed already */ 218 if (!nvdev) 219 return 0; 220 221 ret = rndis_filter_close(nvdev); 222 if (ret != 0) { 223 netdev_err(net, "unable to close device (ret %d).\n", ret); 224 return ret; 225 } 226 227 ret = netvsc_wait_until_empty(nvdev); 228 if (ret) 229 netdev_err(net, "Ring buffer not empty after closing rndis\n"); 230 231 if (vf_netdev) 232 dev_close(vf_netdev); 233 234 return ret; 235 } 236 237 static inline void *init_ppi_data(struct rndis_message *msg, 238 u32 ppi_size, u32 pkt_type) 239 { 240 struct rndis_packet *rndis_pkt = &msg->msg.pkt; 241 struct rndis_per_packet_info *ppi; 242 243 rndis_pkt->data_offset += ppi_size; 244 ppi = (void *)rndis_pkt + rndis_pkt->per_pkt_info_offset 245 + rndis_pkt->per_pkt_info_len; 246 247 ppi->size = ppi_size; 248 ppi->type = pkt_type; 249 ppi->internal = 0; 250 ppi->ppi_offset = sizeof(struct rndis_per_packet_info); 251 252 rndis_pkt->per_pkt_info_len += ppi_size; 253 254 return ppi + 1; 255 } 256 257 /* Azure hosts don't support non-TCP port numbers in hashing for fragmented 258 * packets. We can use ethtool to change UDP hash level when necessary. 259 */ 260 static inline u32 netvsc_get_hash( 261 struct sk_buff *skb, 262 const struct net_device_context *ndc) 263 { 264 struct flow_keys flow; 265 u32 hash, pkt_proto = 0; 266 static u32 hashrnd __read_mostly; 267 268 net_get_random_once(&hashrnd, sizeof(hashrnd)); 269 270 if (!skb_flow_dissect_flow_keys(skb, &flow, 0)) 271 return 0; 272 273 switch (flow.basic.ip_proto) { 274 case IPPROTO_TCP: 275 if (flow.basic.n_proto == htons(ETH_P_IP)) 276 pkt_proto = HV_TCP4_L4HASH; 277 else if (flow.basic.n_proto == htons(ETH_P_IPV6)) 278 pkt_proto = HV_TCP6_L4HASH; 279 280 break; 281 282 case IPPROTO_UDP: 283 if (flow.basic.n_proto == htons(ETH_P_IP)) 284 pkt_proto = HV_UDP4_L4HASH; 285 else if (flow.basic.n_proto == htons(ETH_P_IPV6)) 286 pkt_proto = HV_UDP6_L4HASH; 287 288 break; 289 } 290 291 if (pkt_proto & ndc->l4_hash) { 292 return skb_get_hash(skb); 293 } else { 294 if (flow.basic.n_proto == htons(ETH_P_IP)) 295 hash = jhash2((u32 *)&flow.addrs.v4addrs, 2, hashrnd); 296 else if (flow.basic.n_proto == htons(ETH_P_IPV6)) 297 hash = jhash2((u32 *)&flow.addrs.v6addrs, 8, hashrnd); 298 else 299 hash = 0; 300 301 skb_set_hash(skb, hash, PKT_HASH_TYPE_L3); 302 } 303 304 return hash; 305 } 306 307 static inline int netvsc_get_tx_queue(struct net_device *ndev, 308 struct sk_buff *skb, int old_idx) 309 { 310 const struct net_device_context *ndc = netdev_priv(ndev); 311 struct sock *sk = skb->sk; 312 int q_idx; 313 314 q_idx = ndc->tx_table[netvsc_get_hash(skb, ndc) & 315 (VRSS_SEND_TAB_SIZE - 1)]; 316 317 /* If queue index changed record the new value */ 318 if (q_idx != old_idx && 319 sk && sk_fullsock(sk) && rcu_access_pointer(sk->sk_dst_cache)) 320 sk_tx_queue_set(sk, q_idx); 321 322 return q_idx; 323 } 324 325 /* 326 * Select queue for transmit. 327 * 328 * If a valid queue has already been assigned, then use that. 329 * Otherwise compute tx queue based on hash and the send table. 330 * 331 * This is basically similar to default (__netdev_pick_tx) with the added step 332 * of using the host send_table when no other queue has been assigned. 333 * 334 * TODO support XPS - but get_xps_queue not exported 335 */ 336 static u16 netvsc_pick_tx(struct net_device *ndev, struct sk_buff *skb) 337 { 338 int q_idx = sk_tx_queue_get(skb->sk); 339 340 if (q_idx < 0 || skb->ooo_okay || q_idx >= ndev->real_num_tx_queues) { 341 /* If forwarding a packet, we use the recorded queue when 342 * available for better cache locality. 343 */ 344 if (skb_rx_queue_recorded(skb)) 345 q_idx = skb_get_rx_queue(skb); 346 else 347 q_idx = netvsc_get_tx_queue(ndev, skb, q_idx); 348 } 349 350 return q_idx; 351 } 352 353 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb, 354 struct net_device *sb_dev, 355 select_queue_fallback_t fallback) 356 { 357 struct net_device_context *ndc = netdev_priv(ndev); 358 struct net_device *vf_netdev; 359 u16 txq; 360 361 rcu_read_lock(); 362 vf_netdev = rcu_dereference(ndc->vf_netdev); 363 if (vf_netdev) { 364 const struct net_device_ops *vf_ops = vf_netdev->netdev_ops; 365 366 if (vf_ops->ndo_select_queue) 367 txq = vf_ops->ndo_select_queue(vf_netdev, skb, 368 sb_dev, fallback); 369 else 370 txq = fallback(vf_netdev, skb, NULL); 371 372 /* Record the queue selected by VF so that it can be 373 * used for common case where VF has more queues than 374 * the synthetic device. 375 */ 376 qdisc_skb_cb(skb)->slave_dev_queue_mapping = txq; 377 } else { 378 txq = netvsc_pick_tx(ndev, skb); 379 } 380 rcu_read_unlock(); 381 382 while (unlikely(txq >= ndev->real_num_tx_queues)) 383 txq -= ndev->real_num_tx_queues; 384 385 return txq; 386 } 387 388 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len, 389 struct hv_page_buffer *pb) 390 { 391 int j = 0; 392 393 /* Deal with compound pages by ignoring unused part 394 * of the page. 395 */ 396 page += (offset >> PAGE_SHIFT); 397 offset &= ~PAGE_MASK; 398 399 while (len > 0) { 400 unsigned long bytes; 401 402 bytes = PAGE_SIZE - offset; 403 if (bytes > len) 404 bytes = len; 405 pb[j].pfn = page_to_pfn(page); 406 pb[j].offset = offset; 407 pb[j].len = bytes; 408 409 offset += bytes; 410 len -= bytes; 411 412 if (offset == PAGE_SIZE && len) { 413 page++; 414 offset = 0; 415 j++; 416 } 417 } 418 419 return j + 1; 420 } 421 422 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb, 423 struct hv_netvsc_packet *packet, 424 struct hv_page_buffer *pb) 425 { 426 u32 slots_used = 0; 427 char *data = skb->data; 428 int frags = skb_shinfo(skb)->nr_frags; 429 int i; 430 431 /* The packet is laid out thus: 432 * 1. hdr: RNDIS header and PPI 433 * 2. skb linear data 434 * 3. skb fragment data 435 */ 436 slots_used += fill_pg_buf(virt_to_page(hdr), 437 offset_in_page(hdr), 438 len, &pb[slots_used]); 439 440 packet->rmsg_size = len; 441 packet->rmsg_pgcnt = slots_used; 442 443 slots_used += fill_pg_buf(virt_to_page(data), 444 offset_in_page(data), 445 skb_headlen(skb), &pb[slots_used]); 446 447 for (i = 0; i < frags; i++) { 448 skb_frag_t *frag = skb_shinfo(skb)->frags + i; 449 450 slots_used += fill_pg_buf(skb_frag_page(frag), 451 frag->page_offset, 452 skb_frag_size(frag), &pb[slots_used]); 453 } 454 return slots_used; 455 } 456 457 static int count_skb_frag_slots(struct sk_buff *skb) 458 { 459 int i, frags = skb_shinfo(skb)->nr_frags; 460 int pages = 0; 461 462 for (i = 0; i < frags; i++) { 463 skb_frag_t *frag = skb_shinfo(skb)->frags + i; 464 unsigned long size = skb_frag_size(frag); 465 unsigned long offset = frag->page_offset; 466 467 /* Skip unused frames from start of page */ 468 offset &= ~PAGE_MASK; 469 pages += PFN_UP(offset + size); 470 } 471 return pages; 472 } 473 474 static int netvsc_get_slots(struct sk_buff *skb) 475 { 476 char *data = skb->data; 477 unsigned int offset = offset_in_page(data); 478 unsigned int len = skb_headlen(skb); 479 int slots; 480 int frag_slots; 481 482 slots = DIV_ROUND_UP(offset + len, PAGE_SIZE); 483 frag_slots = count_skb_frag_slots(skb); 484 return slots + frag_slots; 485 } 486 487 static u32 net_checksum_info(struct sk_buff *skb) 488 { 489 if (skb->protocol == htons(ETH_P_IP)) { 490 struct iphdr *ip = ip_hdr(skb); 491 492 if (ip->protocol == IPPROTO_TCP) 493 return TRANSPORT_INFO_IPV4_TCP; 494 else if (ip->protocol == IPPROTO_UDP) 495 return TRANSPORT_INFO_IPV4_UDP; 496 } else { 497 struct ipv6hdr *ip6 = ipv6_hdr(skb); 498 499 if (ip6->nexthdr == IPPROTO_TCP) 500 return TRANSPORT_INFO_IPV6_TCP; 501 else if (ip6->nexthdr == IPPROTO_UDP) 502 return TRANSPORT_INFO_IPV6_UDP; 503 } 504 505 return TRANSPORT_INFO_NOT_IP; 506 } 507 508 /* Send skb on the slave VF device. */ 509 static int netvsc_vf_xmit(struct net_device *net, struct net_device *vf_netdev, 510 struct sk_buff *skb) 511 { 512 struct net_device_context *ndev_ctx = netdev_priv(net); 513 unsigned int len = skb->len; 514 int rc; 515 516 skb->dev = vf_netdev; 517 skb->queue_mapping = qdisc_skb_cb(skb)->slave_dev_queue_mapping; 518 519 rc = dev_queue_xmit(skb); 520 if (likely(rc == NET_XMIT_SUCCESS || rc == NET_XMIT_CN)) { 521 struct netvsc_vf_pcpu_stats *pcpu_stats 522 = this_cpu_ptr(ndev_ctx->vf_stats); 523 524 u64_stats_update_begin(&pcpu_stats->syncp); 525 pcpu_stats->tx_packets++; 526 pcpu_stats->tx_bytes += len; 527 u64_stats_update_end(&pcpu_stats->syncp); 528 } else { 529 this_cpu_inc(ndev_ctx->vf_stats->tx_dropped); 530 } 531 532 return rc; 533 } 534 535 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net) 536 { 537 struct net_device_context *net_device_ctx = netdev_priv(net); 538 struct hv_netvsc_packet *packet = NULL; 539 int ret; 540 unsigned int num_data_pgs; 541 struct rndis_message *rndis_msg; 542 struct net_device *vf_netdev; 543 u32 rndis_msg_size; 544 u32 hash; 545 struct hv_page_buffer pb[MAX_PAGE_BUFFER_COUNT]; 546 547 /* if VF is present and up then redirect packets 548 * already called with rcu_read_lock_bh 549 */ 550 vf_netdev = rcu_dereference_bh(net_device_ctx->vf_netdev); 551 if (vf_netdev && netif_running(vf_netdev) && 552 !netpoll_tx_running(net)) 553 return netvsc_vf_xmit(net, vf_netdev, skb); 554 555 /* We will atmost need two pages to describe the rndis 556 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number 557 * of pages in a single packet. If skb is scattered around 558 * more pages we try linearizing it. 559 */ 560 561 num_data_pgs = netvsc_get_slots(skb) + 2; 562 563 if (unlikely(num_data_pgs > MAX_PAGE_BUFFER_COUNT)) { 564 ++net_device_ctx->eth_stats.tx_scattered; 565 566 if (skb_linearize(skb)) 567 goto no_memory; 568 569 num_data_pgs = netvsc_get_slots(skb) + 2; 570 if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) { 571 ++net_device_ctx->eth_stats.tx_too_big; 572 goto drop; 573 } 574 } 575 576 /* 577 * Place the rndis header in the skb head room and 578 * the skb->cb will be used for hv_netvsc_packet 579 * structure. 580 */ 581 ret = skb_cow_head(skb, RNDIS_AND_PPI_SIZE); 582 if (ret) 583 goto no_memory; 584 585 /* Use the skb control buffer for building up the packet */ 586 BUILD_BUG_ON(sizeof(struct hv_netvsc_packet) > 587 FIELD_SIZEOF(struct sk_buff, cb)); 588 packet = (struct hv_netvsc_packet *)skb->cb; 589 590 packet->q_idx = skb_get_queue_mapping(skb); 591 592 packet->total_data_buflen = skb->len; 593 packet->total_bytes = skb->len; 594 packet->total_packets = 1; 595 596 rndis_msg = (struct rndis_message *)skb->head; 597 598 /* Add the rndis header */ 599 rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET; 600 rndis_msg->msg_len = packet->total_data_buflen; 601 602 rndis_msg->msg.pkt = (struct rndis_packet) { 603 .data_offset = sizeof(struct rndis_packet), 604 .data_len = packet->total_data_buflen, 605 .per_pkt_info_offset = sizeof(struct rndis_packet), 606 }; 607 608 rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet); 609 610 hash = skb_get_hash_raw(skb); 611 if (hash != 0 && net->real_num_tx_queues > 1) { 612 u32 *hash_info; 613 614 rndis_msg_size += NDIS_HASH_PPI_SIZE; 615 hash_info = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE, 616 NBL_HASH_VALUE); 617 *hash_info = hash; 618 } 619 620 if (skb_vlan_tag_present(skb)) { 621 struct ndis_pkt_8021q_info *vlan; 622 623 rndis_msg_size += NDIS_VLAN_PPI_SIZE; 624 vlan = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE, 625 IEEE_8021Q_INFO); 626 627 vlan->value = 0; 628 vlan->vlanid = skb_vlan_tag_get_id(skb); 629 vlan->cfi = skb_vlan_tag_get_cfi(skb); 630 vlan->pri = skb_vlan_tag_get_prio(skb); 631 } 632 633 if (skb_is_gso(skb)) { 634 struct ndis_tcp_lso_info *lso_info; 635 636 rndis_msg_size += NDIS_LSO_PPI_SIZE; 637 lso_info = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE, 638 TCP_LARGESEND_PKTINFO); 639 640 lso_info->value = 0; 641 lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE; 642 if (skb->protocol == htons(ETH_P_IP)) { 643 lso_info->lso_v2_transmit.ip_version = 644 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4; 645 ip_hdr(skb)->tot_len = 0; 646 ip_hdr(skb)->check = 0; 647 tcp_hdr(skb)->check = 648 ~csum_tcpudp_magic(ip_hdr(skb)->saddr, 649 ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0); 650 } else { 651 lso_info->lso_v2_transmit.ip_version = 652 NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6; 653 ipv6_hdr(skb)->payload_len = 0; 654 tcp_hdr(skb)->check = 655 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr, 656 &ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0); 657 } 658 lso_info->lso_v2_transmit.tcp_header_offset = skb_transport_offset(skb); 659 lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size; 660 } else if (skb->ip_summed == CHECKSUM_PARTIAL) { 661 if (net_checksum_info(skb) & net_device_ctx->tx_checksum_mask) { 662 struct ndis_tcp_ip_checksum_info *csum_info; 663 664 rndis_msg_size += NDIS_CSUM_PPI_SIZE; 665 csum_info = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE, 666 TCPIP_CHKSUM_PKTINFO); 667 668 csum_info->value = 0; 669 csum_info->transmit.tcp_header_offset = skb_transport_offset(skb); 670 671 if (skb->protocol == htons(ETH_P_IP)) { 672 csum_info->transmit.is_ipv4 = 1; 673 674 if (ip_hdr(skb)->protocol == IPPROTO_TCP) 675 csum_info->transmit.tcp_checksum = 1; 676 else 677 csum_info->transmit.udp_checksum = 1; 678 } else { 679 csum_info->transmit.is_ipv6 = 1; 680 681 if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP) 682 csum_info->transmit.tcp_checksum = 1; 683 else 684 csum_info->transmit.udp_checksum = 1; 685 } 686 } else { 687 /* Can't do offload of this type of checksum */ 688 if (skb_checksum_help(skb)) 689 goto drop; 690 } 691 } 692 693 /* Start filling in the page buffers with the rndis hdr */ 694 rndis_msg->msg_len += rndis_msg_size; 695 packet->total_data_buflen = rndis_msg->msg_len; 696 packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size, 697 skb, packet, pb); 698 699 /* timestamp packet in software */ 700 skb_tx_timestamp(skb); 701 702 ret = netvsc_send(net, packet, rndis_msg, pb, skb); 703 if (likely(ret == 0)) 704 return NETDEV_TX_OK; 705 706 if (ret == -EAGAIN) { 707 ++net_device_ctx->eth_stats.tx_busy; 708 return NETDEV_TX_BUSY; 709 } 710 711 if (ret == -ENOSPC) 712 ++net_device_ctx->eth_stats.tx_no_space; 713 714 drop: 715 dev_kfree_skb_any(skb); 716 net->stats.tx_dropped++; 717 718 return NETDEV_TX_OK; 719 720 no_memory: 721 ++net_device_ctx->eth_stats.tx_no_memory; 722 goto drop; 723 } 724 725 /* 726 * netvsc_linkstatus_callback - Link up/down notification 727 */ 728 void netvsc_linkstatus_callback(struct net_device *net, 729 struct rndis_message *resp) 730 { 731 struct rndis_indicate_status *indicate = &resp->msg.indicate_status; 732 struct net_device_context *ndev_ctx = netdev_priv(net); 733 struct netvsc_reconfig *event; 734 unsigned long flags; 735 736 /* Update the physical link speed when changing to another vSwitch */ 737 if (indicate->status == RNDIS_STATUS_LINK_SPEED_CHANGE) { 738 u32 speed; 739 740 speed = *(u32 *)((void *)indicate 741 + indicate->status_buf_offset) / 10000; 742 ndev_ctx->speed = speed; 743 return; 744 } 745 746 /* Handle these link change statuses below */ 747 if (indicate->status != RNDIS_STATUS_NETWORK_CHANGE && 748 indicate->status != RNDIS_STATUS_MEDIA_CONNECT && 749 indicate->status != RNDIS_STATUS_MEDIA_DISCONNECT) 750 return; 751 752 if (net->reg_state != NETREG_REGISTERED) 753 return; 754 755 event = kzalloc(sizeof(*event), GFP_ATOMIC); 756 if (!event) 757 return; 758 event->event = indicate->status; 759 760 spin_lock_irqsave(&ndev_ctx->lock, flags); 761 list_add_tail(&event->list, &ndev_ctx->reconfig_events); 762 spin_unlock_irqrestore(&ndev_ctx->lock, flags); 763 764 schedule_delayed_work(&ndev_ctx->dwork, 0); 765 } 766 767 static void netvsc_comp_ipcsum(struct sk_buff *skb) 768 { 769 struct iphdr *iph = (struct iphdr *)skb->data; 770 771 iph->check = 0; 772 iph->check = ip_fast_csum(iph, iph->ihl); 773 } 774 775 static struct sk_buff *netvsc_alloc_recv_skb(struct net_device *net, 776 struct netvsc_channel *nvchan) 777 { 778 struct napi_struct *napi = &nvchan->napi; 779 const struct ndis_pkt_8021q_info *vlan = nvchan->rsc.vlan; 780 const struct ndis_tcp_ip_checksum_info *csum_info = 781 nvchan->rsc.csum_info; 782 struct sk_buff *skb; 783 int i; 784 785 skb = napi_alloc_skb(napi, nvchan->rsc.pktlen); 786 if (!skb) 787 return skb; 788 789 /* 790 * Copy to skb. This copy is needed here since the memory pointed by 791 * hv_netvsc_packet cannot be deallocated 792 */ 793 for (i = 0; i < nvchan->rsc.cnt; i++) 794 skb_put_data(skb, nvchan->rsc.data[i], nvchan->rsc.len[i]); 795 796 skb->protocol = eth_type_trans(skb, net); 797 798 /* skb is already created with CHECKSUM_NONE */ 799 skb_checksum_none_assert(skb); 800 801 /* Incoming packets may have IP header checksum verified by the host. 802 * They may not have IP header checksum computed after coalescing. 803 * We compute it here if the flags are set, because on Linux, the IP 804 * checksum is always checked. 805 */ 806 if (csum_info && csum_info->receive.ip_checksum_value_invalid && 807 csum_info->receive.ip_checksum_succeeded && 808 skb->protocol == htons(ETH_P_IP)) 809 netvsc_comp_ipcsum(skb); 810 811 /* Do L4 checksum offload if enabled and present. 812 */ 813 if (csum_info && (net->features & NETIF_F_RXCSUM)) { 814 if (csum_info->receive.tcp_checksum_succeeded || 815 csum_info->receive.udp_checksum_succeeded) 816 skb->ip_summed = CHECKSUM_UNNECESSARY; 817 } 818 819 if (vlan) { 820 u16 vlan_tci = vlan->vlanid | (vlan->pri << VLAN_PRIO_SHIFT) | 821 (vlan->cfi ? VLAN_CFI_MASK : 0); 822 823 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), 824 vlan_tci); 825 } 826 827 return skb; 828 } 829 830 /* 831 * netvsc_recv_callback - Callback when we receive a packet from the 832 * "wire" on the specified device. 833 */ 834 int netvsc_recv_callback(struct net_device *net, 835 struct netvsc_device *net_device, 836 struct netvsc_channel *nvchan) 837 { 838 struct net_device_context *net_device_ctx = netdev_priv(net); 839 struct vmbus_channel *channel = nvchan->channel; 840 u16 q_idx = channel->offermsg.offer.sub_channel_index; 841 struct sk_buff *skb; 842 struct netvsc_stats *rx_stats; 843 844 if (net->reg_state != NETREG_REGISTERED) 845 return NVSP_STAT_FAIL; 846 847 /* Allocate a skb - TODO direct I/O to pages? */ 848 skb = netvsc_alloc_recv_skb(net, nvchan); 849 850 if (unlikely(!skb)) { 851 ++net_device_ctx->eth_stats.rx_no_memory; 852 rcu_read_unlock(); 853 return NVSP_STAT_FAIL; 854 } 855 856 skb_record_rx_queue(skb, q_idx); 857 858 /* 859 * Even if injecting the packet, record the statistics 860 * on the synthetic device because modifying the VF device 861 * statistics will not work correctly. 862 */ 863 rx_stats = &nvchan->rx_stats; 864 u64_stats_update_begin(&rx_stats->syncp); 865 rx_stats->packets++; 866 rx_stats->bytes += nvchan->rsc.pktlen; 867 868 if (skb->pkt_type == PACKET_BROADCAST) 869 ++rx_stats->broadcast; 870 else if (skb->pkt_type == PACKET_MULTICAST) 871 ++rx_stats->multicast; 872 u64_stats_update_end(&rx_stats->syncp); 873 874 napi_gro_receive(&nvchan->napi, skb); 875 return NVSP_STAT_SUCCESS; 876 } 877 878 static void netvsc_get_drvinfo(struct net_device *net, 879 struct ethtool_drvinfo *info) 880 { 881 strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver)); 882 strlcpy(info->fw_version, "N/A", sizeof(info->fw_version)); 883 } 884 885 static void netvsc_get_channels(struct net_device *net, 886 struct ethtool_channels *channel) 887 { 888 struct net_device_context *net_device_ctx = netdev_priv(net); 889 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev); 890 891 if (nvdev) { 892 channel->max_combined = nvdev->max_chn; 893 channel->combined_count = nvdev->num_chn; 894 } 895 } 896 897 /* Alloc struct netvsc_device_info, and initialize it from either existing 898 * struct netvsc_device, or from default values. 899 */ 900 static struct netvsc_device_info *netvsc_devinfo_get 901 (struct netvsc_device *nvdev) 902 { 903 struct netvsc_device_info *dev_info; 904 905 dev_info = kzalloc(sizeof(*dev_info), GFP_ATOMIC); 906 907 if (!dev_info) 908 return NULL; 909 910 if (nvdev) { 911 dev_info->num_chn = nvdev->num_chn; 912 dev_info->send_sections = nvdev->send_section_cnt; 913 dev_info->send_section_size = nvdev->send_section_size; 914 dev_info->recv_sections = nvdev->recv_section_cnt; 915 dev_info->recv_section_size = nvdev->recv_section_size; 916 917 memcpy(dev_info->rss_key, nvdev->extension->rss_key, 918 NETVSC_HASH_KEYLEN); 919 } else { 920 dev_info->num_chn = VRSS_CHANNEL_DEFAULT; 921 dev_info->send_sections = NETVSC_DEFAULT_TX; 922 dev_info->send_section_size = NETVSC_SEND_SECTION_SIZE; 923 dev_info->recv_sections = NETVSC_DEFAULT_RX; 924 dev_info->recv_section_size = NETVSC_RECV_SECTION_SIZE; 925 } 926 927 return dev_info; 928 } 929 930 static int netvsc_detach(struct net_device *ndev, 931 struct netvsc_device *nvdev) 932 { 933 struct net_device_context *ndev_ctx = netdev_priv(ndev); 934 struct hv_device *hdev = ndev_ctx->device_ctx; 935 int ret; 936 937 /* Don't try continuing to try and setup sub channels */ 938 if (cancel_work_sync(&nvdev->subchan_work)) 939 nvdev->num_chn = 1; 940 941 /* If device was up (receiving) then shutdown */ 942 if (netif_running(ndev)) { 943 netvsc_tx_disable(nvdev, ndev); 944 945 ret = rndis_filter_close(nvdev); 946 if (ret) { 947 netdev_err(ndev, 948 "unable to close device (ret %d).\n", ret); 949 return ret; 950 } 951 952 ret = netvsc_wait_until_empty(nvdev); 953 if (ret) { 954 netdev_err(ndev, 955 "Ring buffer not empty after closing rndis\n"); 956 return ret; 957 } 958 } 959 960 netif_device_detach(ndev); 961 962 rndis_filter_device_remove(hdev, nvdev); 963 964 return 0; 965 } 966 967 static int netvsc_attach(struct net_device *ndev, 968 struct netvsc_device_info *dev_info) 969 { 970 struct net_device_context *ndev_ctx = netdev_priv(ndev); 971 struct hv_device *hdev = ndev_ctx->device_ctx; 972 struct netvsc_device *nvdev; 973 struct rndis_device *rdev; 974 int ret; 975 976 nvdev = rndis_filter_device_add(hdev, dev_info); 977 if (IS_ERR(nvdev)) 978 return PTR_ERR(nvdev); 979 980 if (nvdev->num_chn > 1) { 981 ret = rndis_set_subchannel(ndev, nvdev, dev_info); 982 983 /* if unavailable, just proceed with one queue */ 984 if (ret) { 985 nvdev->max_chn = 1; 986 nvdev->num_chn = 1; 987 } 988 } 989 990 /* In any case device is now ready */ 991 netif_device_attach(ndev); 992 993 /* Note: enable and attach happen when sub-channels setup */ 994 netif_carrier_off(ndev); 995 996 if (netif_running(ndev)) { 997 ret = rndis_filter_open(nvdev); 998 if (ret) 999 return ret; 1000 1001 rdev = nvdev->extension; 1002 if (!rdev->link_state) 1003 netif_carrier_on(ndev); 1004 } 1005 1006 return 0; 1007 } 1008 1009 static int netvsc_set_channels(struct net_device *net, 1010 struct ethtool_channels *channels) 1011 { 1012 struct net_device_context *net_device_ctx = netdev_priv(net); 1013 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev); 1014 unsigned int orig, count = channels->combined_count; 1015 struct netvsc_device_info *device_info; 1016 int ret; 1017 1018 /* We do not support separate count for rx, tx, or other */ 1019 if (count == 0 || 1020 channels->rx_count || channels->tx_count || channels->other_count) 1021 return -EINVAL; 1022 1023 if (!nvdev || nvdev->destroy) 1024 return -ENODEV; 1025 1026 if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) 1027 return -EINVAL; 1028 1029 if (count > nvdev->max_chn) 1030 return -EINVAL; 1031 1032 orig = nvdev->num_chn; 1033 1034 device_info = netvsc_devinfo_get(nvdev); 1035 1036 if (!device_info) 1037 return -ENOMEM; 1038 1039 device_info->num_chn = count; 1040 1041 ret = netvsc_detach(net, nvdev); 1042 if (ret) 1043 goto out; 1044 1045 ret = netvsc_attach(net, device_info); 1046 if (ret) { 1047 device_info->num_chn = orig; 1048 if (netvsc_attach(net, device_info)) 1049 netdev_err(net, "restoring channel setting failed\n"); 1050 } 1051 1052 out: 1053 kfree(device_info); 1054 return ret; 1055 } 1056 1057 static bool 1058 netvsc_validate_ethtool_ss_cmd(const struct ethtool_link_ksettings *cmd) 1059 { 1060 struct ethtool_link_ksettings diff1 = *cmd; 1061 struct ethtool_link_ksettings diff2 = {}; 1062 1063 diff1.base.speed = 0; 1064 diff1.base.duplex = 0; 1065 /* advertising and cmd are usually set */ 1066 ethtool_link_ksettings_zero_link_mode(&diff1, advertising); 1067 diff1.base.cmd = 0; 1068 /* We set port to PORT_OTHER */ 1069 diff2.base.port = PORT_OTHER; 1070 1071 return !memcmp(&diff1, &diff2, sizeof(diff1)); 1072 } 1073 1074 static void netvsc_init_settings(struct net_device *dev) 1075 { 1076 struct net_device_context *ndc = netdev_priv(dev); 1077 1078 ndc->l4_hash = HV_DEFAULT_L4HASH; 1079 1080 ndc->speed = SPEED_UNKNOWN; 1081 ndc->duplex = DUPLEX_FULL; 1082 1083 dev->features = NETIF_F_LRO; 1084 } 1085 1086 static int netvsc_get_link_ksettings(struct net_device *dev, 1087 struct ethtool_link_ksettings *cmd) 1088 { 1089 struct net_device_context *ndc = netdev_priv(dev); 1090 1091 cmd->base.speed = ndc->speed; 1092 cmd->base.duplex = ndc->duplex; 1093 cmd->base.port = PORT_OTHER; 1094 1095 return 0; 1096 } 1097 1098 static int netvsc_set_link_ksettings(struct net_device *dev, 1099 const struct ethtool_link_ksettings *cmd) 1100 { 1101 struct net_device_context *ndc = netdev_priv(dev); 1102 u32 speed; 1103 1104 speed = cmd->base.speed; 1105 if (!ethtool_validate_speed(speed) || 1106 !ethtool_validate_duplex(cmd->base.duplex) || 1107 !netvsc_validate_ethtool_ss_cmd(cmd)) 1108 return -EINVAL; 1109 1110 ndc->speed = speed; 1111 ndc->duplex = cmd->base.duplex; 1112 1113 return 0; 1114 } 1115 1116 static int netvsc_change_mtu(struct net_device *ndev, int mtu) 1117 { 1118 struct net_device_context *ndevctx = netdev_priv(ndev); 1119 struct net_device *vf_netdev = rtnl_dereference(ndevctx->vf_netdev); 1120 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev); 1121 int orig_mtu = ndev->mtu; 1122 struct netvsc_device_info *device_info; 1123 int ret = 0; 1124 1125 if (!nvdev || nvdev->destroy) 1126 return -ENODEV; 1127 1128 device_info = netvsc_devinfo_get(nvdev); 1129 1130 if (!device_info) 1131 return -ENOMEM; 1132 1133 /* Change MTU of underlying VF netdev first. */ 1134 if (vf_netdev) { 1135 ret = dev_set_mtu(vf_netdev, mtu); 1136 if (ret) 1137 goto out; 1138 } 1139 1140 ret = netvsc_detach(ndev, nvdev); 1141 if (ret) 1142 goto rollback_vf; 1143 1144 ndev->mtu = mtu; 1145 1146 ret = netvsc_attach(ndev, device_info); 1147 if (!ret) 1148 goto out; 1149 1150 /* Attempt rollback to original MTU */ 1151 ndev->mtu = orig_mtu; 1152 1153 if (netvsc_attach(ndev, device_info)) 1154 netdev_err(ndev, "restoring mtu failed\n"); 1155 rollback_vf: 1156 if (vf_netdev) 1157 dev_set_mtu(vf_netdev, orig_mtu); 1158 1159 out: 1160 kfree(device_info); 1161 return ret; 1162 } 1163 1164 static void netvsc_get_vf_stats(struct net_device *net, 1165 struct netvsc_vf_pcpu_stats *tot) 1166 { 1167 struct net_device_context *ndev_ctx = netdev_priv(net); 1168 int i; 1169 1170 memset(tot, 0, sizeof(*tot)); 1171 1172 for_each_possible_cpu(i) { 1173 const struct netvsc_vf_pcpu_stats *stats 1174 = per_cpu_ptr(ndev_ctx->vf_stats, i); 1175 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 1176 unsigned int start; 1177 1178 do { 1179 start = u64_stats_fetch_begin_irq(&stats->syncp); 1180 rx_packets = stats->rx_packets; 1181 tx_packets = stats->tx_packets; 1182 rx_bytes = stats->rx_bytes; 1183 tx_bytes = stats->tx_bytes; 1184 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1185 1186 tot->rx_packets += rx_packets; 1187 tot->tx_packets += tx_packets; 1188 tot->rx_bytes += rx_bytes; 1189 tot->tx_bytes += tx_bytes; 1190 tot->tx_dropped += stats->tx_dropped; 1191 } 1192 } 1193 1194 static void netvsc_get_pcpu_stats(struct net_device *net, 1195 struct netvsc_ethtool_pcpu_stats *pcpu_tot) 1196 { 1197 struct net_device_context *ndev_ctx = netdev_priv(net); 1198 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev); 1199 int i; 1200 1201 /* fetch percpu stats of vf */ 1202 for_each_possible_cpu(i) { 1203 const struct netvsc_vf_pcpu_stats *stats = 1204 per_cpu_ptr(ndev_ctx->vf_stats, i); 1205 struct netvsc_ethtool_pcpu_stats *this_tot = &pcpu_tot[i]; 1206 unsigned int start; 1207 1208 do { 1209 start = u64_stats_fetch_begin_irq(&stats->syncp); 1210 this_tot->vf_rx_packets = stats->rx_packets; 1211 this_tot->vf_tx_packets = stats->tx_packets; 1212 this_tot->vf_rx_bytes = stats->rx_bytes; 1213 this_tot->vf_tx_bytes = stats->tx_bytes; 1214 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1215 this_tot->rx_packets = this_tot->vf_rx_packets; 1216 this_tot->tx_packets = this_tot->vf_tx_packets; 1217 this_tot->rx_bytes = this_tot->vf_rx_bytes; 1218 this_tot->tx_bytes = this_tot->vf_tx_bytes; 1219 } 1220 1221 /* fetch percpu stats of netvsc */ 1222 for (i = 0; i < nvdev->num_chn; i++) { 1223 const struct netvsc_channel *nvchan = &nvdev->chan_table[i]; 1224 const struct netvsc_stats *stats; 1225 struct netvsc_ethtool_pcpu_stats *this_tot = 1226 &pcpu_tot[nvchan->channel->target_cpu]; 1227 u64 packets, bytes; 1228 unsigned int start; 1229 1230 stats = &nvchan->tx_stats; 1231 do { 1232 start = u64_stats_fetch_begin_irq(&stats->syncp); 1233 packets = stats->packets; 1234 bytes = stats->bytes; 1235 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1236 1237 this_tot->tx_bytes += bytes; 1238 this_tot->tx_packets += packets; 1239 1240 stats = &nvchan->rx_stats; 1241 do { 1242 start = u64_stats_fetch_begin_irq(&stats->syncp); 1243 packets = stats->packets; 1244 bytes = stats->bytes; 1245 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1246 1247 this_tot->rx_bytes += bytes; 1248 this_tot->rx_packets += packets; 1249 } 1250 } 1251 1252 static void netvsc_get_stats64(struct net_device *net, 1253 struct rtnl_link_stats64 *t) 1254 { 1255 struct net_device_context *ndev_ctx = netdev_priv(net); 1256 struct netvsc_device *nvdev = rcu_dereference_rtnl(ndev_ctx->nvdev); 1257 struct netvsc_vf_pcpu_stats vf_tot; 1258 int i; 1259 1260 if (!nvdev) 1261 return; 1262 1263 netdev_stats_to_stats64(t, &net->stats); 1264 1265 netvsc_get_vf_stats(net, &vf_tot); 1266 t->rx_packets += vf_tot.rx_packets; 1267 t->tx_packets += vf_tot.tx_packets; 1268 t->rx_bytes += vf_tot.rx_bytes; 1269 t->tx_bytes += vf_tot.tx_bytes; 1270 t->tx_dropped += vf_tot.tx_dropped; 1271 1272 for (i = 0; i < nvdev->num_chn; i++) { 1273 const struct netvsc_channel *nvchan = &nvdev->chan_table[i]; 1274 const struct netvsc_stats *stats; 1275 u64 packets, bytes, multicast; 1276 unsigned int start; 1277 1278 stats = &nvchan->tx_stats; 1279 do { 1280 start = u64_stats_fetch_begin_irq(&stats->syncp); 1281 packets = stats->packets; 1282 bytes = stats->bytes; 1283 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1284 1285 t->tx_bytes += bytes; 1286 t->tx_packets += packets; 1287 1288 stats = &nvchan->rx_stats; 1289 do { 1290 start = u64_stats_fetch_begin_irq(&stats->syncp); 1291 packets = stats->packets; 1292 bytes = stats->bytes; 1293 multicast = stats->multicast + stats->broadcast; 1294 } while (u64_stats_fetch_retry_irq(&stats->syncp, start)); 1295 1296 t->rx_bytes += bytes; 1297 t->rx_packets += packets; 1298 t->multicast += multicast; 1299 } 1300 } 1301 1302 static int netvsc_set_mac_addr(struct net_device *ndev, void *p) 1303 { 1304 struct net_device_context *ndc = netdev_priv(ndev); 1305 struct net_device *vf_netdev = rtnl_dereference(ndc->vf_netdev); 1306 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); 1307 struct sockaddr *addr = p; 1308 int err; 1309 1310 err = eth_prepare_mac_addr_change(ndev, p); 1311 if (err) 1312 return err; 1313 1314 if (!nvdev) 1315 return -ENODEV; 1316 1317 if (vf_netdev) { 1318 err = dev_set_mac_address(vf_netdev, addr, NULL); 1319 if (err) 1320 return err; 1321 } 1322 1323 err = rndis_filter_set_device_mac(nvdev, addr->sa_data); 1324 if (!err) { 1325 eth_commit_mac_addr_change(ndev, p); 1326 } else if (vf_netdev) { 1327 /* rollback change on VF */ 1328 memcpy(addr->sa_data, ndev->dev_addr, ETH_ALEN); 1329 dev_set_mac_address(vf_netdev, addr, NULL); 1330 } 1331 1332 return err; 1333 } 1334 1335 static const struct { 1336 char name[ETH_GSTRING_LEN]; 1337 u16 offset; 1338 } netvsc_stats[] = { 1339 { "tx_scattered", offsetof(struct netvsc_ethtool_stats, tx_scattered) }, 1340 { "tx_no_memory", offsetof(struct netvsc_ethtool_stats, tx_no_memory) }, 1341 { "tx_no_space", offsetof(struct netvsc_ethtool_stats, tx_no_space) }, 1342 { "tx_too_big", offsetof(struct netvsc_ethtool_stats, tx_too_big) }, 1343 { "tx_busy", offsetof(struct netvsc_ethtool_stats, tx_busy) }, 1344 { "tx_send_full", offsetof(struct netvsc_ethtool_stats, tx_send_full) }, 1345 { "rx_comp_busy", offsetof(struct netvsc_ethtool_stats, rx_comp_busy) }, 1346 { "rx_no_memory", offsetof(struct netvsc_ethtool_stats, rx_no_memory) }, 1347 { "stop_queue", offsetof(struct netvsc_ethtool_stats, stop_queue) }, 1348 { "wake_queue", offsetof(struct netvsc_ethtool_stats, wake_queue) }, 1349 }, pcpu_stats[] = { 1350 { "cpu%u_rx_packets", 1351 offsetof(struct netvsc_ethtool_pcpu_stats, rx_packets) }, 1352 { "cpu%u_rx_bytes", 1353 offsetof(struct netvsc_ethtool_pcpu_stats, rx_bytes) }, 1354 { "cpu%u_tx_packets", 1355 offsetof(struct netvsc_ethtool_pcpu_stats, tx_packets) }, 1356 { "cpu%u_tx_bytes", 1357 offsetof(struct netvsc_ethtool_pcpu_stats, tx_bytes) }, 1358 { "cpu%u_vf_rx_packets", 1359 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_packets) }, 1360 { "cpu%u_vf_rx_bytes", 1361 offsetof(struct netvsc_ethtool_pcpu_stats, vf_rx_bytes) }, 1362 { "cpu%u_vf_tx_packets", 1363 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_packets) }, 1364 { "cpu%u_vf_tx_bytes", 1365 offsetof(struct netvsc_ethtool_pcpu_stats, vf_tx_bytes) }, 1366 }, vf_stats[] = { 1367 { "vf_rx_packets", offsetof(struct netvsc_vf_pcpu_stats, rx_packets) }, 1368 { "vf_rx_bytes", offsetof(struct netvsc_vf_pcpu_stats, rx_bytes) }, 1369 { "vf_tx_packets", offsetof(struct netvsc_vf_pcpu_stats, tx_packets) }, 1370 { "vf_tx_bytes", offsetof(struct netvsc_vf_pcpu_stats, tx_bytes) }, 1371 { "vf_tx_dropped", offsetof(struct netvsc_vf_pcpu_stats, tx_dropped) }, 1372 }; 1373 1374 #define NETVSC_GLOBAL_STATS_LEN ARRAY_SIZE(netvsc_stats) 1375 #define NETVSC_VF_STATS_LEN ARRAY_SIZE(vf_stats) 1376 1377 /* statistics per queue (rx/tx packets/bytes) */ 1378 #define NETVSC_PCPU_STATS_LEN (num_present_cpus() * ARRAY_SIZE(pcpu_stats)) 1379 1380 /* 4 statistics per queue (rx/tx packets/bytes) */ 1381 #define NETVSC_QUEUE_STATS_LEN(dev) ((dev)->num_chn * 4) 1382 1383 static int netvsc_get_sset_count(struct net_device *dev, int string_set) 1384 { 1385 struct net_device_context *ndc = netdev_priv(dev); 1386 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); 1387 1388 if (!nvdev) 1389 return -ENODEV; 1390 1391 switch (string_set) { 1392 case ETH_SS_STATS: 1393 return NETVSC_GLOBAL_STATS_LEN 1394 + NETVSC_VF_STATS_LEN 1395 + NETVSC_QUEUE_STATS_LEN(nvdev) 1396 + NETVSC_PCPU_STATS_LEN; 1397 default: 1398 return -EINVAL; 1399 } 1400 } 1401 1402 static void netvsc_get_ethtool_stats(struct net_device *dev, 1403 struct ethtool_stats *stats, u64 *data) 1404 { 1405 struct net_device_context *ndc = netdev_priv(dev); 1406 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); 1407 const void *nds = &ndc->eth_stats; 1408 const struct netvsc_stats *qstats; 1409 struct netvsc_vf_pcpu_stats sum; 1410 struct netvsc_ethtool_pcpu_stats *pcpu_sum; 1411 unsigned int start; 1412 u64 packets, bytes; 1413 int i, j, cpu; 1414 1415 if (!nvdev) 1416 return; 1417 1418 for (i = 0; i < NETVSC_GLOBAL_STATS_LEN; i++) 1419 data[i] = *(unsigned long *)(nds + netvsc_stats[i].offset); 1420 1421 netvsc_get_vf_stats(dev, &sum); 1422 for (j = 0; j < NETVSC_VF_STATS_LEN; j++) 1423 data[i++] = *(u64 *)((void *)&sum + vf_stats[j].offset); 1424 1425 for (j = 0; j < nvdev->num_chn; j++) { 1426 qstats = &nvdev->chan_table[j].tx_stats; 1427 1428 do { 1429 start = u64_stats_fetch_begin_irq(&qstats->syncp); 1430 packets = qstats->packets; 1431 bytes = qstats->bytes; 1432 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start)); 1433 data[i++] = packets; 1434 data[i++] = bytes; 1435 1436 qstats = &nvdev->chan_table[j].rx_stats; 1437 do { 1438 start = u64_stats_fetch_begin_irq(&qstats->syncp); 1439 packets = qstats->packets; 1440 bytes = qstats->bytes; 1441 } while (u64_stats_fetch_retry_irq(&qstats->syncp, start)); 1442 data[i++] = packets; 1443 data[i++] = bytes; 1444 } 1445 1446 pcpu_sum = kvmalloc_array(num_possible_cpus(), 1447 sizeof(struct netvsc_ethtool_pcpu_stats), 1448 GFP_KERNEL); 1449 netvsc_get_pcpu_stats(dev, pcpu_sum); 1450 for_each_present_cpu(cpu) { 1451 struct netvsc_ethtool_pcpu_stats *this_sum = &pcpu_sum[cpu]; 1452 1453 for (j = 0; j < ARRAY_SIZE(pcpu_stats); j++) 1454 data[i++] = *(u64 *)((void *)this_sum 1455 + pcpu_stats[j].offset); 1456 } 1457 kvfree(pcpu_sum); 1458 } 1459 1460 static void netvsc_get_strings(struct net_device *dev, u32 stringset, u8 *data) 1461 { 1462 struct net_device_context *ndc = netdev_priv(dev); 1463 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); 1464 u8 *p = data; 1465 int i, cpu; 1466 1467 if (!nvdev) 1468 return; 1469 1470 switch (stringset) { 1471 case ETH_SS_STATS: 1472 for (i = 0; i < ARRAY_SIZE(netvsc_stats); i++) { 1473 memcpy(p, netvsc_stats[i].name, ETH_GSTRING_LEN); 1474 p += ETH_GSTRING_LEN; 1475 } 1476 1477 for (i = 0; i < ARRAY_SIZE(vf_stats); i++) { 1478 memcpy(p, vf_stats[i].name, ETH_GSTRING_LEN); 1479 p += ETH_GSTRING_LEN; 1480 } 1481 1482 for (i = 0; i < nvdev->num_chn; i++) { 1483 sprintf(p, "tx_queue_%u_packets", i); 1484 p += ETH_GSTRING_LEN; 1485 sprintf(p, "tx_queue_%u_bytes", i); 1486 p += ETH_GSTRING_LEN; 1487 sprintf(p, "rx_queue_%u_packets", i); 1488 p += ETH_GSTRING_LEN; 1489 sprintf(p, "rx_queue_%u_bytes", i); 1490 p += ETH_GSTRING_LEN; 1491 } 1492 1493 for_each_present_cpu(cpu) { 1494 for (i = 0; i < ARRAY_SIZE(pcpu_stats); i++) { 1495 sprintf(p, pcpu_stats[i].name, cpu); 1496 p += ETH_GSTRING_LEN; 1497 } 1498 } 1499 1500 break; 1501 } 1502 } 1503 1504 static int 1505 netvsc_get_rss_hash_opts(struct net_device_context *ndc, 1506 struct ethtool_rxnfc *info) 1507 { 1508 const u32 l4_flag = RXH_L4_B_0_1 | RXH_L4_B_2_3; 1509 1510 info->data = RXH_IP_SRC | RXH_IP_DST; 1511 1512 switch (info->flow_type) { 1513 case TCP_V4_FLOW: 1514 if (ndc->l4_hash & HV_TCP4_L4HASH) 1515 info->data |= l4_flag; 1516 1517 break; 1518 1519 case TCP_V6_FLOW: 1520 if (ndc->l4_hash & HV_TCP6_L4HASH) 1521 info->data |= l4_flag; 1522 1523 break; 1524 1525 case UDP_V4_FLOW: 1526 if (ndc->l4_hash & HV_UDP4_L4HASH) 1527 info->data |= l4_flag; 1528 1529 break; 1530 1531 case UDP_V6_FLOW: 1532 if (ndc->l4_hash & HV_UDP6_L4HASH) 1533 info->data |= l4_flag; 1534 1535 break; 1536 1537 case IPV4_FLOW: 1538 case IPV6_FLOW: 1539 break; 1540 default: 1541 info->data = 0; 1542 break; 1543 } 1544 1545 return 0; 1546 } 1547 1548 static int 1549 netvsc_get_rxnfc(struct net_device *dev, struct ethtool_rxnfc *info, 1550 u32 *rules) 1551 { 1552 struct net_device_context *ndc = netdev_priv(dev); 1553 struct netvsc_device *nvdev = rtnl_dereference(ndc->nvdev); 1554 1555 if (!nvdev) 1556 return -ENODEV; 1557 1558 switch (info->cmd) { 1559 case ETHTOOL_GRXRINGS: 1560 info->data = nvdev->num_chn; 1561 return 0; 1562 1563 case ETHTOOL_GRXFH: 1564 return netvsc_get_rss_hash_opts(ndc, info); 1565 } 1566 return -EOPNOTSUPP; 1567 } 1568 1569 static int netvsc_set_rss_hash_opts(struct net_device_context *ndc, 1570 struct ethtool_rxnfc *info) 1571 { 1572 if (info->data == (RXH_IP_SRC | RXH_IP_DST | 1573 RXH_L4_B_0_1 | RXH_L4_B_2_3)) { 1574 switch (info->flow_type) { 1575 case TCP_V4_FLOW: 1576 ndc->l4_hash |= HV_TCP4_L4HASH; 1577 break; 1578 1579 case TCP_V6_FLOW: 1580 ndc->l4_hash |= HV_TCP6_L4HASH; 1581 break; 1582 1583 case UDP_V4_FLOW: 1584 ndc->l4_hash |= HV_UDP4_L4HASH; 1585 break; 1586 1587 case UDP_V6_FLOW: 1588 ndc->l4_hash |= HV_UDP6_L4HASH; 1589 break; 1590 1591 default: 1592 return -EOPNOTSUPP; 1593 } 1594 1595 return 0; 1596 } 1597 1598 if (info->data == (RXH_IP_SRC | RXH_IP_DST)) { 1599 switch (info->flow_type) { 1600 case TCP_V4_FLOW: 1601 ndc->l4_hash &= ~HV_TCP4_L4HASH; 1602 break; 1603 1604 case TCP_V6_FLOW: 1605 ndc->l4_hash &= ~HV_TCP6_L4HASH; 1606 break; 1607 1608 case UDP_V4_FLOW: 1609 ndc->l4_hash &= ~HV_UDP4_L4HASH; 1610 break; 1611 1612 case UDP_V6_FLOW: 1613 ndc->l4_hash &= ~HV_UDP6_L4HASH; 1614 break; 1615 1616 default: 1617 return -EOPNOTSUPP; 1618 } 1619 1620 return 0; 1621 } 1622 1623 return -EOPNOTSUPP; 1624 } 1625 1626 static int 1627 netvsc_set_rxnfc(struct net_device *ndev, struct ethtool_rxnfc *info) 1628 { 1629 struct net_device_context *ndc = netdev_priv(ndev); 1630 1631 if (info->cmd == ETHTOOL_SRXFH) 1632 return netvsc_set_rss_hash_opts(ndc, info); 1633 1634 return -EOPNOTSUPP; 1635 } 1636 1637 static u32 netvsc_get_rxfh_key_size(struct net_device *dev) 1638 { 1639 return NETVSC_HASH_KEYLEN; 1640 } 1641 1642 static u32 netvsc_rss_indir_size(struct net_device *dev) 1643 { 1644 return ITAB_NUM; 1645 } 1646 1647 static int netvsc_get_rxfh(struct net_device *dev, u32 *indir, u8 *key, 1648 u8 *hfunc) 1649 { 1650 struct net_device_context *ndc = netdev_priv(dev); 1651 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev); 1652 struct rndis_device *rndis_dev; 1653 int i; 1654 1655 if (!ndev) 1656 return -ENODEV; 1657 1658 if (hfunc) 1659 *hfunc = ETH_RSS_HASH_TOP; /* Toeplitz */ 1660 1661 rndis_dev = ndev->extension; 1662 if (indir) { 1663 for (i = 0; i < ITAB_NUM; i++) 1664 indir[i] = rndis_dev->rx_table[i]; 1665 } 1666 1667 if (key) 1668 memcpy(key, rndis_dev->rss_key, NETVSC_HASH_KEYLEN); 1669 1670 return 0; 1671 } 1672 1673 static int netvsc_set_rxfh(struct net_device *dev, const u32 *indir, 1674 const u8 *key, const u8 hfunc) 1675 { 1676 struct net_device_context *ndc = netdev_priv(dev); 1677 struct netvsc_device *ndev = rtnl_dereference(ndc->nvdev); 1678 struct rndis_device *rndis_dev; 1679 int i; 1680 1681 if (!ndev) 1682 return -ENODEV; 1683 1684 if (hfunc != ETH_RSS_HASH_NO_CHANGE && hfunc != ETH_RSS_HASH_TOP) 1685 return -EOPNOTSUPP; 1686 1687 rndis_dev = ndev->extension; 1688 if (indir) { 1689 for (i = 0; i < ITAB_NUM; i++) 1690 if (indir[i] >= ndev->num_chn) 1691 return -EINVAL; 1692 1693 for (i = 0; i < ITAB_NUM; i++) 1694 rndis_dev->rx_table[i] = indir[i]; 1695 } 1696 1697 if (!key) { 1698 if (!indir) 1699 return 0; 1700 1701 key = rndis_dev->rss_key; 1702 } 1703 1704 return rndis_filter_set_rss_param(rndis_dev, key); 1705 } 1706 1707 /* Hyper-V RNDIS protocol does not have ring in the HW sense. 1708 * It does have pre-allocated receive area which is divided into sections. 1709 */ 1710 static void __netvsc_get_ringparam(struct netvsc_device *nvdev, 1711 struct ethtool_ringparam *ring) 1712 { 1713 u32 max_buf_size; 1714 1715 ring->rx_pending = nvdev->recv_section_cnt; 1716 ring->tx_pending = nvdev->send_section_cnt; 1717 1718 if (nvdev->nvsp_version <= NVSP_PROTOCOL_VERSION_2) 1719 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE_LEGACY; 1720 else 1721 max_buf_size = NETVSC_RECEIVE_BUFFER_SIZE; 1722 1723 ring->rx_max_pending = max_buf_size / nvdev->recv_section_size; 1724 ring->tx_max_pending = NETVSC_SEND_BUFFER_SIZE 1725 / nvdev->send_section_size; 1726 } 1727 1728 static void netvsc_get_ringparam(struct net_device *ndev, 1729 struct ethtool_ringparam *ring) 1730 { 1731 struct net_device_context *ndevctx = netdev_priv(ndev); 1732 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev); 1733 1734 if (!nvdev) 1735 return; 1736 1737 __netvsc_get_ringparam(nvdev, ring); 1738 } 1739 1740 static int netvsc_set_ringparam(struct net_device *ndev, 1741 struct ethtool_ringparam *ring) 1742 { 1743 struct net_device_context *ndevctx = netdev_priv(ndev); 1744 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev); 1745 struct netvsc_device_info *device_info; 1746 struct ethtool_ringparam orig; 1747 u32 new_tx, new_rx; 1748 int ret = 0; 1749 1750 if (!nvdev || nvdev->destroy) 1751 return -ENODEV; 1752 1753 memset(&orig, 0, sizeof(orig)); 1754 __netvsc_get_ringparam(nvdev, &orig); 1755 1756 new_tx = clamp_t(u32, ring->tx_pending, 1757 NETVSC_MIN_TX_SECTIONS, orig.tx_max_pending); 1758 new_rx = clamp_t(u32, ring->rx_pending, 1759 NETVSC_MIN_RX_SECTIONS, orig.rx_max_pending); 1760 1761 if (new_tx == orig.tx_pending && 1762 new_rx == orig.rx_pending) 1763 return 0; /* no change */ 1764 1765 device_info = netvsc_devinfo_get(nvdev); 1766 1767 if (!device_info) 1768 return -ENOMEM; 1769 1770 device_info->send_sections = new_tx; 1771 device_info->recv_sections = new_rx; 1772 1773 ret = netvsc_detach(ndev, nvdev); 1774 if (ret) 1775 goto out; 1776 1777 ret = netvsc_attach(ndev, device_info); 1778 if (ret) { 1779 device_info->send_sections = orig.tx_pending; 1780 device_info->recv_sections = orig.rx_pending; 1781 1782 if (netvsc_attach(ndev, device_info)) 1783 netdev_err(ndev, "restoring ringparam failed"); 1784 } 1785 1786 out: 1787 kfree(device_info); 1788 return ret; 1789 } 1790 1791 static int netvsc_set_features(struct net_device *ndev, 1792 netdev_features_t features) 1793 { 1794 netdev_features_t change = features ^ ndev->features; 1795 struct net_device_context *ndevctx = netdev_priv(ndev); 1796 struct netvsc_device *nvdev = rtnl_dereference(ndevctx->nvdev); 1797 struct ndis_offload_params offloads; 1798 1799 if (!nvdev || nvdev->destroy) 1800 return -ENODEV; 1801 1802 if (!(change & NETIF_F_LRO)) 1803 return 0; 1804 1805 memset(&offloads, 0, sizeof(struct ndis_offload_params)); 1806 1807 if (features & NETIF_F_LRO) { 1808 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED; 1809 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED; 1810 } else { 1811 offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED; 1812 offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED; 1813 } 1814 1815 return rndis_filter_set_offload_params(ndev, nvdev, &offloads); 1816 } 1817 1818 static u32 netvsc_get_msglevel(struct net_device *ndev) 1819 { 1820 struct net_device_context *ndev_ctx = netdev_priv(ndev); 1821 1822 return ndev_ctx->msg_enable; 1823 } 1824 1825 static void netvsc_set_msglevel(struct net_device *ndev, u32 val) 1826 { 1827 struct net_device_context *ndev_ctx = netdev_priv(ndev); 1828 1829 ndev_ctx->msg_enable = val; 1830 } 1831 1832 static const struct ethtool_ops ethtool_ops = { 1833 .get_drvinfo = netvsc_get_drvinfo, 1834 .get_msglevel = netvsc_get_msglevel, 1835 .set_msglevel = netvsc_set_msglevel, 1836 .get_link = ethtool_op_get_link, 1837 .get_ethtool_stats = netvsc_get_ethtool_stats, 1838 .get_sset_count = netvsc_get_sset_count, 1839 .get_strings = netvsc_get_strings, 1840 .get_channels = netvsc_get_channels, 1841 .set_channels = netvsc_set_channels, 1842 .get_ts_info = ethtool_op_get_ts_info, 1843 .get_rxnfc = netvsc_get_rxnfc, 1844 .set_rxnfc = netvsc_set_rxnfc, 1845 .get_rxfh_key_size = netvsc_get_rxfh_key_size, 1846 .get_rxfh_indir_size = netvsc_rss_indir_size, 1847 .get_rxfh = netvsc_get_rxfh, 1848 .set_rxfh = netvsc_set_rxfh, 1849 .get_link_ksettings = netvsc_get_link_ksettings, 1850 .set_link_ksettings = netvsc_set_link_ksettings, 1851 .get_ringparam = netvsc_get_ringparam, 1852 .set_ringparam = netvsc_set_ringparam, 1853 }; 1854 1855 static const struct net_device_ops device_ops = { 1856 .ndo_open = netvsc_open, 1857 .ndo_stop = netvsc_close, 1858 .ndo_start_xmit = netvsc_start_xmit, 1859 .ndo_change_rx_flags = netvsc_change_rx_flags, 1860 .ndo_set_rx_mode = netvsc_set_rx_mode, 1861 .ndo_set_features = netvsc_set_features, 1862 .ndo_change_mtu = netvsc_change_mtu, 1863 .ndo_validate_addr = eth_validate_addr, 1864 .ndo_set_mac_address = netvsc_set_mac_addr, 1865 .ndo_select_queue = netvsc_select_queue, 1866 .ndo_get_stats64 = netvsc_get_stats64, 1867 }; 1868 1869 /* 1870 * Handle link status changes. For RNDIS_STATUS_NETWORK_CHANGE emulate link 1871 * down/up sequence. In case of RNDIS_STATUS_MEDIA_CONNECT when carrier is 1872 * present send GARP packet to network peers with netif_notify_peers(). 1873 */ 1874 static void netvsc_link_change(struct work_struct *w) 1875 { 1876 struct net_device_context *ndev_ctx = 1877 container_of(w, struct net_device_context, dwork.work); 1878 struct hv_device *device_obj = ndev_ctx->device_ctx; 1879 struct net_device *net = hv_get_drvdata(device_obj); 1880 struct netvsc_device *net_device; 1881 struct rndis_device *rdev; 1882 struct netvsc_reconfig *event = NULL; 1883 bool notify = false, reschedule = false; 1884 unsigned long flags, next_reconfig, delay; 1885 1886 /* if changes are happening, comeback later */ 1887 if (!rtnl_trylock()) { 1888 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT); 1889 return; 1890 } 1891 1892 net_device = rtnl_dereference(ndev_ctx->nvdev); 1893 if (!net_device) 1894 goto out_unlock; 1895 1896 rdev = net_device->extension; 1897 1898 next_reconfig = ndev_ctx->last_reconfig + LINKCHANGE_INT; 1899 if (time_is_after_jiffies(next_reconfig)) { 1900 /* link_watch only sends one notification with current state 1901 * per second, avoid doing reconfig more frequently. Handle 1902 * wrap around. 1903 */ 1904 delay = next_reconfig - jiffies; 1905 delay = delay < LINKCHANGE_INT ? delay : LINKCHANGE_INT; 1906 schedule_delayed_work(&ndev_ctx->dwork, delay); 1907 goto out_unlock; 1908 } 1909 ndev_ctx->last_reconfig = jiffies; 1910 1911 spin_lock_irqsave(&ndev_ctx->lock, flags); 1912 if (!list_empty(&ndev_ctx->reconfig_events)) { 1913 event = list_first_entry(&ndev_ctx->reconfig_events, 1914 struct netvsc_reconfig, list); 1915 list_del(&event->list); 1916 reschedule = !list_empty(&ndev_ctx->reconfig_events); 1917 } 1918 spin_unlock_irqrestore(&ndev_ctx->lock, flags); 1919 1920 if (!event) 1921 goto out_unlock; 1922 1923 switch (event->event) { 1924 /* Only the following events are possible due to the check in 1925 * netvsc_linkstatus_callback() 1926 */ 1927 case RNDIS_STATUS_MEDIA_CONNECT: 1928 if (rdev->link_state) { 1929 rdev->link_state = false; 1930 netif_carrier_on(net); 1931 netvsc_tx_enable(net_device, net); 1932 } else { 1933 notify = true; 1934 } 1935 kfree(event); 1936 break; 1937 case RNDIS_STATUS_MEDIA_DISCONNECT: 1938 if (!rdev->link_state) { 1939 rdev->link_state = true; 1940 netif_carrier_off(net); 1941 netvsc_tx_disable(net_device, net); 1942 } 1943 kfree(event); 1944 break; 1945 case RNDIS_STATUS_NETWORK_CHANGE: 1946 /* Only makes sense if carrier is present */ 1947 if (!rdev->link_state) { 1948 rdev->link_state = true; 1949 netif_carrier_off(net); 1950 netvsc_tx_disable(net_device, net); 1951 event->event = RNDIS_STATUS_MEDIA_CONNECT; 1952 spin_lock_irqsave(&ndev_ctx->lock, flags); 1953 list_add(&event->list, &ndev_ctx->reconfig_events); 1954 spin_unlock_irqrestore(&ndev_ctx->lock, flags); 1955 reschedule = true; 1956 } 1957 break; 1958 } 1959 1960 rtnl_unlock(); 1961 1962 if (notify) 1963 netdev_notify_peers(net); 1964 1965 /* link_watch only sends one notification with current state per 1966 * second, handle next reconfig event in 2 seconds. 1967 */ 1968 if (reschedule) 1969 schedule_delayed_work(&ndev_ctx->dwork, LINKCHANGE_INT); 1970 1971 return; 1972 1973 out_unlock: 1974 rtnl_unlock(); 1975 } 1976 1977 static struct net_device *get_netvsc_byref(struct net_device *vf_netdev) 1978 { 1979 struct net_device_context *net_device_ctx; 1980 struct net_device *dev; 1981 1982 dev = netdev_master_upper_dev_get(vf_netdev); 1983 if (!dev || dev->netdev_ops != &device_ops) 1984 return NULL; /* not a netvsc device */ 1985 1986 net_device_ctx = netdev_priv(dev); 1987 if (!rtnl_dereference(net_device_ctx->nvdev)) 1988 return NULL; /* device is removed */ 1989 1990 return dev; 1991 } 1992 1993 /* Called when VF is injecting data into network stack. 1994 * Change the associated network device from VF to netvsc. 1995 * note: already called with rcu_read_lock 1996 */ 1997 static rx_handler_result_t netvsc_vf_handle_frame(struct sk_buff **pskb) 1998 { 1999 struct sk_buff *skb = *pskb; 2000 struct net_device *ndev = rcu_dereference(skb->dev->rx_handler_data); 2001 struct net_device_context *ndev_ctx = netdev_priv(ndev); 2002 struct netvsc_vf_pcpu_stats *pcpu_stats 2003 = this_cpu_ptr(ndev_ctx->vf_stats); 2004 2005 skb->dev = ndev; 2006 2007 u64_stats_update_begin(&pcpu_stats->syncp); 2008 pcpu_stats->rx_packets++; 2009 pcpu_stats->rx_bytes += skb->len; 2010 u64_stats_update_end(&pcpu_stats->syncp); 2011 2012 return RX_HANDLER_ANOTHER; 2013 } 2014 2015 static int netvsc_vf_join(struct net_device *vf_netdev, 2016 struct net_device *ndev) 2017 { 2018 struct net_device_context *ndev_ctx = netdev_priv(ndev); 2019 int ret; 2020 2021 ret = netdev_rx_handler_register(vf_netdev, 2022 netvsc_vf_handle_frame, ndev); 2023 if (ret != 0) { 2024 netdev_err(vf_netdev, 2025 "can not register netvsc VF receive handler (err = %d)\n", 2026 ret); 2027 goto rx_handler_failed; 2028 } 2029 2030 ret = netdev_master_upper_dev_link(vf_netdev, ndev, 2031 NULL, NULL, NULL); 2032 if (ret != 0) { 2033 netdev_err(vf_netdev, 2034 "can not set master device %s (err = %d)\n", 2035 ndev->name, ret); 2036 goto upper_link_failed; 2037 } 2038 2039 /* set slave flag before open to prevent IPv6 addrconf */ 2040 vf_netdev->flags |= IFF_SLAVE; 2041 2042 schedule_delayed_work(&ndev_ctx->vf_takeover, VF_TAKEOVER_INT); 2043 2044 call_netdevice_notifiers(NETDEV_JOIN, vf_netdev); 2045 2046 netdev_info(vf_netdev, "joined to %s\n", ndev->name); 2047 return 0; 2048 2049 upper_link_failed: 2050 netdev_rx_handler_unregister(vf_netdev); 2051 rx_handler_failed: 2052 return ret; 2053 } 2054 2055 static void __netvsc_vf_setup(struct net_device *ndev, 2056 struct net_device *vf_netdev) 2057 { 2058 int ret; 2059 2060 /* Align MTU of VF with master */ 2061 ret = dev_set_mtu(vf_netdev, ndev->mtu); 2062 if (ret) 2063 netdev_warn(vf_netdev, 2064 "unable to change mtu to %u\n", ndev->mtu); 2065 2066 /* set multicast etc flags on VF */ 2067 dev_change_flags(vf_netdev, ndev->flags | IFF_SLAVE, NULL); 2068 2069 /* sync address list from ndev to VF */ 2070 netif_addr_lock_bh(ndev); 2071 dev_uc_sync(vf_netdev, ndev); 2072 dev_mc_sync(vf_netdev, ndev); 2073 netif_addr_unlock_bh(ndev); 2074 2075 if (netif_running(ndev)) { 2076 ret = dev_open(vf_netdev, NULL); 2077 if (ret) 2078 netdev_warn(vf_netdev, 2079 "unable to open: %d\n", ret); 2080 } 2081 } 2082 2083 /* Setup VF as slave of the synthetic device. 2084 * Runs in workqueue to avoid recursion in netlink callbacks. 2085 */ 2086 static void netvsc_vf_setup(struct work_struct *w) 2087 { 2088 struct net_device_context *ndev_ctx 2089 = container_of(w, struct net_device_context, vf_takeover.work); 2090 struct net_device *ndev = hv_get_drvdata(ndev_ctx->device_ctx); 2091 struct net_device *vf_netdev; 2092 2093 if (!rtnl_trylock()) { 2094 schedule_delayed_work(&ndev_ctx->vf_takeover, 0); 2095 return; 2096 } 2097 2098 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev); 2099 if (vf_netdev) 2100 __netvsc_vf_setup(ndev, vf_netdev); 2101 2102 rtnl_unlock(); 2103 } 2104 2105 /* Find netvsc by VF serial number. 2106 * The PCI hyperv controller records the serial number as the slot kobj name. 2107 */ 2108 static struct net_device *get_netvsc_byslot(const struct net_device *vf_netdev) 2109 { 2110 struct device *parent = vf_netdev->dev.parent; 2111 struct net_device_context *ndev_ctx; 2112 struct pci_dev *pdev; 2113 u32 serial; 2114 2115 if (!parent || !dev_is_pci(parent)) 2116 return NULL; /* not a PCI device */ 2117 2118 pdev = to_pci_dev(parent); 2119 if (!pdev->slot) { 2120 netdev_notice(vf_netdev, "no PCI slot information\n"); 2121 return NULL; 2122 } 2123 2124 if (kstrtou32(pci_slot_name(pdev->slot), 10, &serial)) { 2125 netdev_notice(vf_netdev, "Invalid vf serial:%s\n", 2126 pci_slot_name(pdev->slot)); 2127 return NULL; 2128 } 2129 2130 list_for_each_entry(ndev_ctx, &netvsc_dev_list, list) { 2131 if (!ndev_ctx->vf_alloc) 2132 continue; 2133 2134 if (ndev_ctx->vf_serial == serial) 2135 return hv_get_drvdata(ndev_ctx->device_ctx); 2136 } 2137 2138 netdev_notice(vf_netdev, 2139 "no netdev found for vf serial:%u\n", serial); 2140 return NULL; 2141 } 2142 2143 static int netvsc_register_vf(struct net_device *vf_netdev) 2144 { 2145 struct net_device_context *net_device_ctx; 2146 struct netvsc_device *netvsc_dev; 2147 struct net_device *ndev; 2148 int ret; 2149 2150 if (vf_netdev->addr_len != ETH_ALEN) 2151 return NOTIFY_DONE; 2152 2153 ndev = get_netvsc_byslot(vf_netdev); 2154 if (!ndev) 2155 return NOTIFY_DONE; 2156 2157 net_device_ctx = netdev_priv(ndev); 2158 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev); 2159 if (!netvsc_dev || rtnl_dereference(net_device_ctx->vf_netdev)) 2160 return NOTIFY_DONE; 2161 2162 /* if synthetic interface is a different namespace, 2163 * then move the VF to that namespace; join will be 2164 * done again in that context. 2165 */ 2166 if (!net_eq(dev_net(ndev), dev_net(vf_netdev))) { 2167 ret = dev_change_net_namespace(vf_netdev, 2168 dev_net(ndev), "eth%d"); 2169 if (ret) 2170 netdev_err(vf_netdev, 2171 "could not move to same namespace as %s: %d\n", 2172 ndev->name, ret); 2173 else 2174 netdev_info(vf_netdev, 2175 "VF moved to namespace with: %s\n", 2176 ndev->name); 2177 return NOTIFY_DONE; 2178 } 2179 2180 netdev_info(ndev, "VF registering: %s\n", vf_netdev->name); 2181 2182 if (netvsc_vf_join(vf_netdev, ndev) != 0) 2183 return NOTIFY_DONE; 2184 2185 dev_hold(vf_netdev); 2186 rcu_assign_pointer(net_device_ctx->vf_netdev, vf_netdev); 2187 return NOTIFY_OK; 2188 } 2189 2190 /* VF up/down change detected, schedule to change data path */ 2191 static int netvsc_vf_changed(struct net_device *vf_netdev) 2192 { 2193 struct net_device_context *net_device_ctx; 2194 struct netvsc_device *netvsc_dev; 2195 struct net_device *ndev; 2196 bool vf_is_up = netif_running(vf_netdev); 2197 2198 ndev = get_netvsc_byref(vf_netdev); 2199 if (!ndev) 2200 return NOTIFY_DONE; 2201 2202 net_device_ctx = netdev_priv(ndev); 2203 netvsc_dev = rtnl_dereference(net_device_ctx->nvdev); 2204 if (!netvsc_dev) 2205 return NOTIFY_DONE; 2206 2207 netvsc_switch_datapath(ndev, vf_is_up); 2208 netdev_info(ndev, "Data path switched %s VF: %s\n", 2209 vf_is_up ? "to" : "from", vf_netdev->name); 2210 2211 return NOTIFY_OK; 2212 } 2213 2214 static int netvsc_unregister_vf(struct net_device *vf_netdev) 2215 { 2216 struct net_device *ndev; 2217 struct net_device_context *net_device_ctx; 2218 2219 ndev = get_netvsc_byref(vf_netdev); 2220 if (!ndev) 2221 return NOTIFY_DONE; 2222 2223 net_device_ctx = netdev_priv(ndev); 2224 cancel_delayed_work_sync(&net_device_ctx->vf_takeover); 2225 2226 netdev_info(ndev, "VF unregistering: %s\n", vf_netdev->name); 2227 2228 netdev_rx_handler_unregister(vf_netdev); 2229 netdev_upper_dev_unlink(vf_netdev, ndev); 2230 RCU_INIT_POINTER(net_device_ctx->vf_netdev, NULL); 2231 dev_put(vf_netdev); 2232 2233 return NOTIFY_OK; 2234 } 2235 2236 static int netvsc_probe(struct hv_device *dev, 2237 const struct hv_vmbus_device_id *dev_id) 2238 { 2239 struct net_device *net = NULL; 2240 struct net_device_context *net_device_ctx; 2241 struct netvsc_device_info *device_info = NULL; 2242 struct netvsc_device *nvdev; 2243 int ret = -ENOMEM; 2244 2245 net = alloc_etherdev_mq(sizeof(struct net_device_context), 2246 VRSS_CHANNEL_MAX); 2247 if (!net) 2248 goto no_net; 2249 2250 netif_carrier_off(net); 2251 2252 netvsc_init_settings(net); 2253 2254 net_device_ctx = netdev_priv(net); 2255 net_device_ctx->device_ctx = dev; 2256 net_device_ctx->msg_enable = netif_msg_init(debug, default_msg); 2257 if (netif_msg_probe(net_device_ctx)) 2258 netdev_dbg(net, "netvsc msg_enable: %d\n", 2259 net_device_ctx->msg_enable); 2260 2261 hv_set_drvdata(dev, net); 2262 2263 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change); 2264 2265 spin_lock_init(&net_device_ctx->lock); 2266 INIT_LIST_HEAD(&net_device_ctx->reconfig_events); 2267 INIT_DELAYED_WORK(&net_device_ctx->vf_takeover, netvsc_vf_setup); 2268 2269 net_device_ctx->vf_stats 2270 = netdev_alloc_pcpu_stats(struct netvsc_vf_pcpu_stats); 2271 if (!net_device_ctx->vf_stats) 2272 goto no_stats; 2273 2274 net->netdev_ops = &device_ops; 2275 net->ethtool_ops = ðtool_ops; 2276 SET_NETDEV_DEV(net, &dev->device); 2277 2278 /* We always need headroom for rndis header */ 2279 net->needed_headroom = RNDIS_AND_PPI_SIZE; 2280 2281 /* Initialize the number of queues to be 1, we may change it if more 2282 * channels are offered later. 2283 */ 2284 netif_set_real_num_tx_queues(net, 1); 2285 netif_set_real_num_rx_queues(net, 1); 2286 2287 /* Notify the netvsc driver of the new device */ 2288 device_info = netvsc_devinfo_get(NULL); 2289 2290 if (!device_info) { 2291 ret = -ENOMEM; 2292 goto devinfo_failed; 2293 } 2294 2295 nvdev = rndis_filter_device_add(dev, device_info); 2296 if (IS_ERR(nvdev)) { 2297 ret = PTR_ERR(nvdev); 2298 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret); 2299 goto rndis_failed; 2300 } 2301 2302 memcpy(net->dev_addr, device_info->mac_adr, ETH_ALEN); 2303 2304 /* We must get rtnl lock before scheduling nvdev->subchan_work, 2305 * otherwise netvsc_subchan_work() can get rtnl lock first and wait 2306 * all subchannels to show up, but that may not happen because 2307 * netvsc_probe() can't get rtnl lock and as a result vmbus_onoffer() 2308 * -> ... -> device_add() -> ... -> __device_attach() can't get 2309 * the device lock, so all the subchannels can't be processed -- 2310 * finally netvsc_subchan_work() hangs forever. 2311 */ 2312 rtnl_lock(); 2313 2314 if (nvdev->num_chn > 1) 2315 schedule_work(&nvdev->subchan_work); 2316 2317 /* hw_features computed in rndis_netdev_set_hwcaps() */ 2318 net->features = net->hw_features | 2319 NETIF_F_HIGHDMA | NETIF_F_SG | 2320 NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_HW_VLAN_CTAG_RX; 2321 net->vlan_features = net->features; 2322 2323 netdev_lockdep_set_classes(net); 2324 2325 /* MTU range: 68 - 1500 or 65521 */ 2326 net->min_mtu = NETVSC_MTU_MIN; 2327 if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2) 2328 net->max_mtu = NETVSC_MTU - ETH_HLEN; 2329 else 2330 net->max_mtu = ETH_DATA_LEN; 2331 2332 ret = register_netdevice(net); 2333 if (ret != 0) { 2334 pr_err("Unable to register netdev.\n"); 2335 goto register_failed; 2336 } 2337 2338 list_add(&net_device_ctx->list, &netvsc_dev_list); 2339 rtnl_unlock(); 2340 2341 kfree(device_info); 2342 return 0; 2343 2344 register_failed: 2345 rtnl_unlock(); 2346 rndis_filter_device_remove(dev, nvdev); 2347 rndis_failed: 2348 kfree(device_info); 2349 devinfo_failed: 2350 free_percpu(net_device_ctx->vf_stats); 2351 no_stats: 2352 hv_set_drvdata(dev, NULL); 2353 free_netdev(net); 2354 no_net: 2355 return ret; 2356 } 2357 2358 static int netvsc_remove(struct hv_device *dev) 2359 { 2360 struct net_device_context *ndev_ctx; 2361 struct net_device *vf_netdev, *net; 2362 struct netvsc_device *nvdev; 2363 2364 net = hv_get_drvdata(dev); 2365 if (net == NULL) { 2366 dev_err(&dev->device, "No net device to remove\n"); 2367 return 0; 2368 } 2369 2370 ndev_ctx = netdev_priv(net); 2371 2372 cancel_delayed_work_sync(&ndev_ctx->dwork); 2373 2374 rtnl_lock(); 2375 nvdev = rtnl_dereference(ndev_ctx->nvdev); 2376 if (nvdev) 2377 cancel_work_sync(&nvdev->subchan_work); 2378 2379 /* 2380 * Call to the vsc driver to let it know that the device is being 2381 * removed. Also blocks mtu and channel changes. 2382 */ 2383 vf_netdev = rtnl_dereference(ndev_ctx->vf_netdev); 2384 if (vf_netdev) 2385 netvsc_unregister_vf(vf_netdev); 2386 2387 if (nvdev) 2388 rndis_filter_device_remove(dev, nvdev); 2389 2390 unregister_netdevice(net); 2391 list_del(&ndev_ctx->list); 2392 2393 rtnl_unlock(); 2394 2395 hv_set_drvdata(dev, NULL); 2396 2397 free_percpu(ndev_ctx->vf_stats); 2398 free_netdev(net); 2399 return 0; 2400 } 2401 2402 static const struct hv_vmbus_device_id id_table[] = { 2403 /* Network guid */ 2404 { HV_NIC_GUID, }, 2405 { }, 2406 }; 2407 2408 MODULE_DEVICE_TABLE(vmbus, id_table); 2409 2410 /* The one and only one */ 2411 static struct hv_driver netvsc_drv = { 2412 .name = KBUILD_MODNAME, 2413 .id_table = id_table, 2414 .probe = netvsc_probe, 2415 .remove = netvsc_remove, 2416 .driver = { 2417 .probe_type = PROBE_PREFER_ASYNCHRONOUS, 2418 }, 2419 }; 2420 2421 /* 2422 * On Hyper-V, every VF interface is matched with a corresponding 2423 * synthetic interface. The synthetic interface is presented first 2424 * to the guest. When the corresponding VF instance is registered, 2425 * we will take care of switching the data path. 2426 */ 2427 static int netvsc_netdev_event(struct notifier_block *this, 2428 unsigned long event, void *ptr) 2429 { 2430 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr); 2431 2432 /* Skip our own events */ 2433 if (event_dev->netdev_ops == &device_ops) 2434 return NOTIFY_DONE; 2435 2436 /* Avoid non-Ethernet type devices */ 2437 if (event_dev->type != ARPHRD_ETHER) 2438 return NOTIFY_DONE; 2439 2440 /* Avoid Vlan dev with same MAC registering as VF */ 2441 if (is_vlan_dev(event_dev)) 2442 return NOTIFY_DONE; 2443 2444 /* Avoid Bonding master dev with same MAC registering as VF */ 2445 if ((event_dev->priv_flags & IFF_BONDING) && 2446 (event_dev->flags & IFF_MASTER)) 2447 return NOTIFY_DONE; 2448 2449 switch (event) { 2450 case NETDEV_REGISTER: 2451 return netvsc_register_vf(event_dev); 2452 case NETDEV_UNREGISTER: 2453 return netvsc_unregister_vf(event_dev); 2454 case NETDEV_UP: 2455 case NETDEV_DOWN: 2456 return netvsc_vf_changed(event_dev); 2457 default: 2458 return NOTIFY_DONE; 2459 } 2460 } 2461 2462 static struct notifier_block netvsc_netdev_notifier = { 2463 .notifier_call = netvsc_netdev_event, 2464 }; 2465 2466 static void __exit netvsc_drv_exit(void) 2467 { 2468 unregister_netdevice_notifier(&netvsc_netdev_notifier); 2469 vmbus_driver_unregister(&netvsc_drv); 2470 } 2471 2472 static int __init netvsc_drv_init(void) 2473 { 2474 int ret; 2475 2476 if (ring_size < RING_SIZE_MIN) { 2477 ring_size = RING_SIZE_MIN; 2478 pr_info("Increased ring_size to %u (min allowed)\n", 2479 ring_size); 2480 } 2481 netvsc_ring_bytes = ring_size * PAGE_SIZE; 2482 2483 ret = vmbus_driver_register(&netvsc_drv); 2484 if (ret) 2485 return ret; 2486 2487 register_netdevice_notifier(&netvsc_netdev_notifier); 2488 return 0; 2489 } 2490 2491 MODULE_LICENSE("GPL"); 2492 MODULE_DESCRIPTION("Microsoft Hyper-V network driver"); 2493 2494 module_init(netvsc_drv_init); 2495 module_exit(netvsc_drv_exit); 2496