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