drivers/net/hyperv/netvsc_drv.c

   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, write to the Free Software Foundation, Inc., 59 Temple
  15  * Place - Suite 330, Boston, MA 02111-1307 USA.
  16  *
  17  * Authors:
  18  *   Haiyang Zhang <haiyangz@microsoft.com>
  19  *   Hank Janssen  <hjanssen@microsoft.com>
  20  */
  21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  22
  23 #include <linux/init.h>
  24 #include <linux/atomic.h>
  25 #include <linux/module.h>
  26 #include <linux/highmem.h>
  27 #include <linux/device.h>
  28 #include <linux/io.h>
  29 #include <linux/delay.h>
  30 #include <linux/netdevice.h>
  31 #include <linux/inetdevice.h>
  32 #include <linux/etherdevice.h>
  33 #include <linux/skbuff.h>
  34 #include <linux/if_vlan.h>
  35 #include <linux/in.h>
  36 #include <linux/slab.h>
  37 #include <net/arp.h>
  38 #include <net/route.h>
  39 #include <net/sock.h>
  40 #include <net/pkt_sched.h>
  41
  42 #include "hyperv_net.h"
  43
  44 struct net_device_context {
  45         /* point back to our device context */
  46         struct hv_device *device_ctx;
  47         struct delayed_work dwork;
  48         struct work_struct work;
  49 };
  50
  51 #define RING_SIZE_MIN 64
  52 static int ring_size = 128;
  53 module_param(ring_size, int, S_IRUGO);
  54 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  55
  56 static void do_set_multicast(struct work_struct *w)
  57 {
  58         struct net_device_context *ndevctx =
  59                 container_of(w, struct net_device_context, work);
  60         struct netvsc_device *nvdev;
  61         struct rndis_device *rdev;
  62
  63         nvdev = hv_get_drvdata(ndevctx->device_ctx);
  64         if (nvdev == NULL || nvdev->ndev == NULL)
  65                 return;
  66
  67         rdev = nvdev->extension;
  68         if (rdev == NULL)
  69                 return;
  70
  71         if (nvdev->ndev->flags & IFF_PROMISC)
  72                 rndis_filter_set_packet_filter(rdev,
  73                         NDIS_PACKET_TYPE_PROMISCUOUS);
  74         else
  75                 rndis_filter_set_packet_filter(rdev,
  76                         NDIS_PACKET_TYPE_BROADCAST |
  77                         NDIS_PACKET_TYPE_ALL_MULTICAST |
  78                         NDIS_PACKET_TYPE_DIRECTED);
  79 }
  80
  81 static void netvsc_set_multicast_list(struct net_device *net)
  82 {
  83         struct net_device_context *net_device_ctx = netdev_priv(net);
  84
  85         schedule_work(&net_device_ctx->work);
  86 }
  87
  88 static int netvsc_open(struct net_device *net)
  89 {
  90         struct net_device_context *net_device_ctx = netdev_priv(net);
  91         struct hv_device *device_obj = net_device_ctx->device_ctx;
  92         int ret = 0;
  93
  94         /* Open up the device */
  95         ret = rndis_filter_open(device_obj);
  96         if (ret != 0) {
  97                 netdev_err(net, "unable to open device (ret %d).\n", ret);
  98                 return ret;
  99         }
 100
 101         netif_start_queue(net);
 102
 103         return ret;
 104 }
 105
 106 static int netvsc_close(struct net_device *net)
 107 {
 108         struct net_device_context *net_device_ctx = netdev_priv(net);
 109         struct hv_device *device_obj = net_device_ctx->device_ctx;
 110         int ret;
 111
 112         netif_tx_disable(net);
 113
 114         /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
 115         cancel_work_sync(&net_device_ctx->work);
 116         ret = rndis_filter_close(device_obj);
 117         if (ret != 0)
 118                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 119
 120         return ret;
 121 }
 122
 123 static void netvsc_xmit_completion(void *context)
 124 {
 125         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 126         struct sk_buff *skb = (struct sk_buff *)
 127                 (unsigned long)packet->completion.send.send_completion_tid;
 128
 129         kfree(packet);
 130
 131         if (skb)
 132                 dev_kfree_skb_any(skb);
 133 }
 134
 135 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 136 {
 137         struct net_device_context *net_device_ctx = netdev_priv(net);
 138         struct hv_netvsc_packet *packet;
 139         int ret;
 140         unsigned int i, num_pages, npg_data;
 141
 142         /* Add multipages for skb->data and additional 2 for RNDIS */
 143         npg_data = (((unsigned long)skb->data + skb_headlen(skb) - 1)
 144                 >> PAGE_SHIFT) - ((unsigned long)skb->data >> PAGE_SHIFT) + 1;
 145         num_pages = skb_shinfo(skb)->nr_frags + npg_data + 2;
 146
 147         /* Allocate a netvsc packet based on # of frags. */
 148         packet = kzalloc(sizeof(struct hv_netvsc_packet) +
 149                          (num_pages * sizeof(struct hv_page_buffer)) +
 150                          sizeof(struct rndis_filter_packet) +
 151                          NDIS_VLAN_PPI_SIZE, GFP_ATOMIC);
 152         if (!packet) {
 153                 /* out of memory, drop packet */
 154                 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
 155
 156                 dev_kfree_skb(skb);
 157                 net->stats.tx_dropped++;
 158                 return NETDEV_TX_OK;
 159         }
 160
 161         packet->vlan_tci = skb->vlan_tci;
 162
 163         packet->extension = (void *)(unsigned long)packet +
 164                                 sizeof(struct hv_netvsc_packet) +
 165                                     (num_pages * sizeof(struct hv_page_buffer));
 166
 167         /* If the rndis msg goes beyond 1 page, we will add 1 later */
 168         packet->page_buf_cnt = num_pages - 1;
 169
 170         /* Initialize it from the skb */
 171         packet->total_data_buflen = skb->len;
 172
 173         /* Start filling in the page buffers starting after RNDIS buffer. */
 174         packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT;
 175         packet->page_buf[1].offset
 176                 = (unsigned long)skb->data & (PAGE_SIZE - 1);
 177         if (npg_data == 1)
 178                 packet->page_buf[1].len = skb_headlen(skb);
 179         else
 180                 packet->page_buf[1].len = PAGE_SIZE
 181                         - packet->page_buf[1].offset;
 182
 183         for (i = 2; i <= npg_data; i++) {
 184                 packet->page_buf[i].pfn = virt_to_phys(skb->data
 185                         + PAGE_SIZE * (i-1)) >> PAGE_SHIFT;
 186                 packet->page_buf[i].offset = 0;
 187                 packet->page_buf[i].len = PAGE_SIZE;
 188         }
 189         if (npg_data > 1)
 190                 packet->page_buf[npg_data].len = (((unsigned long)skb->data
 191                         + skb_headlen(skb) - 1) & (PAGE_SIZE - 1)) + 1;
 192
 193         /* Additional fragments are after SKB data */
 194         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 195                 const skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 196
 197                 packet->page_buf[i+npg_data+1].pfn =
 198                         page_to_pfn(skb_frag_page(f));
 199                 packet->page_buf[i+npg_data+1].offset = f->page_offset;
 200                 packet->page_buf[i+npg_data+1].len = skb_frag_size(f);
 201         }
 202
 203         /* Set the completion routine */
 204         packet->completion.send.send_completion = netvsc_xmit_completion;
 205         packet->completion.send.send_completion_ctx = packet;
 206         packet->completion.send.send_completion_tid = (unsigned long)skb;
 207
 208         ret = rndis_filter_send(net_device_ctx->device_ctx,
 209                                   packet);
 210         if (ret == 0) {
 211                 net->stats.tx_bytes += skb->len;
 212                 net->stats.tx_packets++;
 213         } else {
 214                 kfree(packet);
 215                 if (ret != -EAGAIN) {
 216                         dev_kfree_skb_any(skb);
 217                         net->stats.tx_dropped++;
 218                 }
 219         }
 220
 221         return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 222 }
 223
 224 /*
 225  * netvsc_linkstatus_callback - Link up/down notification
 226  */
 227 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 228                                        unsigned int status)
 229 {
 230         struct net_device *net;
 231         struct net_device_context *ndev_ctx;
 232         struct netvsc_device *net_device;
 233
 234         net_device = hv_get_drvdata(device_obj);
 235         net = net_device->ndev;
 236
 237         if (!net) {
 238                 netdev_err(net, "got link status but net device "
 239                                 "not initialized yet\n");
 240                 return;
 241         }
 242
 243         if (status == 1) {
 244                 netif_carrier_on(net);
 245                 ndev_ctx = netdev_priv(net);
 246                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 247                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
 248         } else {
 249                 netif_carrier_off(net);
 250         }
 251 }
 252
 253 /*
 254  * netvsc_recv_callback -  Callback when we receive a packet from the
 255  * "wire" on the specified device.
 256  */
 257 int netvsc_recv_callback(struct hv_device *device_obj,
 258                                 struct hv_netvsc_packet *packet)
 259 {
 260         struct net_device *net;
 261         struct sk_buff *skb;
 262
 263         net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
 264         if (!net || net->reg_state != NETREG_REGISTERED) {
 265                 packet->status = NVSP_STAT_FAIL;
 266                 return 0;
 267         }
 268
 269         /* Allocate a skb - TODO direct I/O to pages? */
 270         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 271         if (unlikely(!skb)) {
 272                 ++net->stats.rx_dropped;
 273                 packet->status = NVSP_STAT_FAIL;
 274                 return 0;
 275         }
 276
 277         /*
 278          * Copy to skb. This copy is needed here since the memory pointed by
 279          * hv_netvsc_packet cannot be deallocated
 280          */
 281         memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
 282                 packet->total_data_buflen);
 283
 284         skb->protocol = eth_type_trans(skb, net);
 285         skb->ip_summed = CHECKSUM_NONE;
 286         if (packet->vlan_tci & VLAN_TAG_PRESENT)
 287                 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
 288                                        packet->vlan_tci);
 289
 290         net->stats.rx_packets++;
 291         net->stats.rx_bytes += packet->total_data_buflen;
 292
 293         /*
 294          * Pass the skb back up. Network stack will deallocate the skb when it
 295          * is done.
 296          * TODO - use NAPI?
 297          */
 298         netif_rx(skb);
 299
 300         return 0;
 301 }
 302
 303 static void netvsc_get_drvinfo(struct net_device *net,
 304                                struct ethtool_drvinfo *info)
 305 {
 306         strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
 307         strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
 308 }
 309
 310 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 311 {
 312         struct net_device_context *ndevctx = netdev_priv(ndev);
 313         struct hv_device *hdev =  ndevctx->device_ctx;
 314         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
 315         struct netvsc_device_info device_info;
 316         int limit = ETH_DATA_LEN;
 317
 318         if (nvdev == NULL || nvdev->destroy)
 319                 return -ENODEV;
 320
 321         if (nvdev->nvsp_version == NVSP_PROTOCOL_VERSION_2)
 322                 limit = NETVSC_MTU;
 323
 324         if (mtu < 68 || mtu > limit)
 325                 return -EINVAL;
 326
 327         nvdev->start_remove = true;
 328         cancel_work_sync(&ndevctx->work);
 329         netif_tx_disable(ndev);
 330         rndis_filter_device_remove(hdev);
 331
 332         ndev->mtu = mtu;
 333
 334         ndevctx->device_ctx = hdev;
 335         hv_set_drvdata(hdev, ndev);
 336         device_info.ring_size = ring_size;
 337         rndis_filter_device_add(hdev, &device_info);
 338         netif_wake_queue(ndev);
 339
 340         return 0;
 341 }
 342
 343
 344 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
 345 {
 346         struct net_device_context *ndevctx = netdev_priv(ndev);
 347         struct hv_device *hdev =  ndevctx->device_ctx;
 348         struct sockaddr *addr = p;
 349         char save_adr[ETH_ALEN];
 350         unsigned char save_aatype;
 351         int err;
 352
 353         memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
 354         save_aatype = ndev->addr_assign_type;
 355
 356         err = eth_mac_addr(ndev, p);
 357         if (err != 0)
 358                 return err;
 359
 360         err = rndis_filter_set_device_mac(hdev, addr->sa_data);
 361         if (err != 0) {
 362                 /* roll back to saved MAC */
 363                 memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
 364                 ndev->addr_assign_type = save_aatype;
 365         }
 366
 367         return err;
 368 }
 369
 370
 371 static const struct ethtool_ops ethtool_ops = {
 372         .get_drvinfo    = netvsc_get_drvinfo,
 373         .get_link       = ethtool_op_get_link,
 374 };
 375
 376 static const struct net_device_ops device_ops = {
 377         .ndo_open =                     netvsc_open,
 378         .ndo_stop =                     netvsc_close,
 379         .ndo_start_xmit =               netvsc_start_xmit,
 380         .ndo_set_rx_mode =              netvsc_set_multicast_list,
 381         .ndo_change_mtu =               netvsc_change_mtu,
 382         .ndo_validate_addr =            eth_validate_addr,
 383         .ndo_set_mac_address =          netvsc_set_mac_addr,
 384 };
 385
 386 /*
 387  * Send GARP packet to network peers after migrations.
 388  * After Quick Migration, the network is not immediately operational in the
 389  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 390  * another netif_notify_peers() into a delayed work, otherwise GARP packet
 391  * will not be sent after quick migration, and cause network disconnection.
 392  */
 393 static void netvsc_send_garp(struct work_struct *w)
 394 {
 395         struct net_device_context *ndev_ctx;
 396         struct net_device *net;
 397         struct netvsc_device *net_device;
 398
 399         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
 400         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
 401         net = net_device->ndev;
 402         netdev_notify_peers(net);
 403 }
 404
 405
 406 static int netvsc_probe(struct hv_device *dev,
 407                         const struct hv_vmbus_device_id *dev_id)
 408 {
 409         struct net_device *net = NULL;
 410         struct net_device_context *net_device_ctx;
 411         struct netvsc_device_info device_info;
 412         int ret;
 413
 414         net = alloc_etherdev(sizeof(struct net_device_context));
 415         if (!net)
 416                 return -ENOMEM;
 417
 418         /* Set initial state */
 419         netif_carrier_off(net);
 420
 421         net_device_ctx = netdev_priv(net);
 422         net_device_ctx->device_ctx = dev;
 423         hv_set_drvdata(dev, net);
 424         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_send_garp);
 425         INIT_WORK(&net_device_ctx->work, do_set_multicast);
 426
 427         net->netdev_ops = &device_ops;
 428
 429         /* TODO: Add GSO and Checksum offload */
 430         net->hw_features = 0;
 431         net->features = NETIF_F_HW_VLAN_CTAG_TX;
 432
 433         SET_ETHTOOL_OPS(net, &ethtool_ops);
 434         SET_NETDEV_DEV(net, &dev->device);
 435
 436         /* Notify the netvsc driver of the new device */
 437         device_info.ring_size = ring_size;
 438         ret = rndis_filter_device_add(dev, &device_info);
 439         if (ret != 0) {
 440                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 441                 free_netdev(net);
 442                 hv_set_drvdata(dev, NULL);
 443                 return ret;
 444         }
 445         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 446
 447         netif_carrier_on(net);
 448
 449         ret = register_netdev(net);
 450         if (ret != 0) {
 451                 pr_err("Unable to register netdev.\n");
 452                 rndis_filter_device_remove(dev);
 453                 free_netdev(net);
 454         }
 455
 456         return ret;
 457 }
 458
 459 static int netvsc_remove(struct hv_device *dev)
 460 {
 461         struct net_device *net;
 462         struct net_device_context *ndev_ctx;
 463         struct netvsc_device *net_device;
 464
 465         net_device = hv_get_drvdata(dev);
 466         net = net_device->ndev;
 467
 468         if (net == NULL) {
 469                 dev_err(&dev->device, "No net device to remove\n");
 470                 return 0;
 471         }
 472
 473         net_device->start_remove = true;
 474
 475         ndev_ctx = netdev_priv(net);
 476         cancel_delayed_work_sync(&ndev_ctx->dwork);
 477         cancel_work_sync(&ndev_ctx->work);
 478
 479         /* Stop outbound asap */
 480         netif_tx_disable(net);
 481
 482         unregister_netdev(net);
 483
 484         /*
 485          * Call to the vsc driver to let it know that the device is being
 486          * removed
 487          */
 488         rndis_filter_device_remove(dev);
 489
 490         free_netdev(net);
 491         return 0;
 492 }
 493
 494 static const struct hv_vmbus_device_id id_table[] = {
 495         /* Network guid */
 496         { HV_NIC_GUID, },
 497         { },
 498 };
 499
 500 MODULE_DEVICE_TABLE(vmbus, id_table);
 501
 502 /* The one and only one */
 503 static struct  hv_driver netvsc_drv = {
 504         .name = KBUILD_MODNAME,
 505         .id_table = id_table,
 506         .probe = netvsc_probe,
 507         .remove = netvsc_remove,
 508 };
 509
 510 static void __exit netvsc_drv_exit(void)
 511 {
 512         vmbus_driver_unregister(&netvsc_drv);
 513 }
 514
 515 static int __init netvsc_drv_init(void)
 516 {
 517         if (ring_size < RING_SIZE_MIN) {
 518                 ring_size = RING_SIZE_MIN;
 519                 pr_info("Increased ring_size to %d (min allowed)\n",
 520                         ring_size);
 521         }
 522         return vmbus_driver_register(&netvsc_drv);
 523 }
 524
 525 MODULE_LICENSE("GPL");
 526 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
 527
 528 module_init(netvsc_drv_init);
 529 module_exit(netvsc_drv_exit);