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/in.h>
  35 #include <linux/slab.h>
  36 #include <net/arp.h>
  37 #include <net/route.h>
  38 #include <net/sock.h>
  39 #include <net/pkt_sched.h>
  40
  41 #include "hyperv_net.h"
  42
  43 struct net_device_context {
  44         /* point back to our device context */
  45         struct hv_device *device_ctx;
  46         struct delayed_work dwork;
  47 };
  48
  49
  50 static int ring_size = 128;
  51 module_param(ring_size, int, S_IRUGO);
  52 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
  53
  54 struct set_multicast_work {
  55         struct work_struct work;
  56         struct net_device *net;
  57 };
  58
  59 static void do_set_multicast(struct work_struct *w)
  60 {
  61         struct set_multicast_work *swk =
  62                 container_of(w, struct set_multicast_work, work);
  63         struct net_device *net = swk->net;
  64
  65         struct net_device_context *ndevctx = netdev_priv(net);
  66         struct netvsc_device *nvdev;
  67         struct rndis_device *rdev;
  68
  69         nvdev = hv_get_drvdata(ndevctx->device_ctx);
  70         if (nvdev == NULL)
  71                 goto out;
  72
  73         rdev = nvdev->extension;
  74         if (rdev == NULL)
  75                 goto out;
  76
  77         if (net->flags & IFF_PROMISC)
  78                 rndis_filter_set_packet_filter(rdev,
  79                         NDIS_PACKET_TYPE_PROMISCUOUS);
  80         else
  81                 rndis_filter_set_packet_filter(rdev,
  82                         NDIS_PACKET_TYPE_BROADCAST |
  83                         NDIS_PACKET_TYPE_ALL_MULTICAST |
  84                         NDIS_PACKET_TYPE_DIRECTED);
  85
  86 out:
  87         kfree(w);
  88 }
  89
  90 static void netvsc_set_multicast_list(struct net_device *net)
  91 {
  92         struct set_multicast_work *swk =
  93                 kmalloc(sizeof(struct set_multicast_work), GFP_ATOMIC);
  94         if (swk == NULL)
  95                 return;
  96
  97         swk->net = net;
  98         INIT_WORK(&swk->work, do_set_multicast);
  99         schedule_work(&swk->work);
 100 }
 101
 102 static int netvsc_open(struct net_device *net)
 103 {
 104         struct net_device_context *net_device_ctx = netdev_priv(net);
 105         struct hv_device *device_obj = net_device_ctx->device_ctx;
 106         int ret = 0;
 107
 108         /* Open up the device */
 109         ret = rndis_filter_open(device_obj);
 110         if (ret != 0) {
 111                 netdev_err(net, "unable to open device (ret %d).\n", ret);
 112                 return ret;
 113         }
 114
 115         netif_start_queue(net);
 116
 117         return ret;
 118 }
 119
 120 static int netvsc_close(struct net_device *net)
 121 {
 122         struct net_device_context *net_device_ctx = netdev_priv(net);
 123         struct hv_device *device_obj = net_device_ctx->device_ctx;
 124         int ret;
 125
 126         netif_tx_disable(net);
 127
 128         ret = rndis_filter_close(device_obj);
 129         if (ret != 0)
 130                 netdev_err(net, "unable to close device (ret %d).\n", ret);
 131
 132         return ret;
 133 }
 134
 135 static void netvsc_xmit_completion(void *context)
 136 {
 137         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
 138         struct sk_buff *skb = (struct sk_buff *)
 139                 (unsigned long)packet->completion.send.send_completion_tid;
 140
 141         kfree(packet);
 142
 143         if (skb)
 144                 dev_kfree_skb_any(skb);
 145 }
 146
 147 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
 148 {
 149         struct net_device_context *net_device_ctx = netdev_priv(net);
 150         struct hv_netvsc_packet *packet;
 151         int ret;
 152         unsigned int i, num_pages, npg_data;
 153
 154         /* Add multipages for skb->data and additional 2 for RNDIS */
 155         npg_data = (((unsigned long)skb->data + skb_headlen(skb) - 1)
 156                 >> PAGE_SHIFT) - ((unsigned long)skb->data >> PAGE_SHIFT) + 1;
 157         num_pages = skb_shinfo(skb)->nr_frags + npg_data + 2;
 158
 159         /* Allocate a netvsc packet based on # of frags. */
 160         packet = kzalloc(sizeof(struct hv_netvsc_packet) +
 161                          (num_pages * sizeof(struct hv_page_buffer)) +
 162                          sizeof(struct rndis_filter_packet), GFP_ATOMIC);
 163         if (!packet) {
 164                 /* out of memory, drop packet */
 165                 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
 166
 167                 dev_kfree_skb(skb);
 168                 net->stats.tx_dropped++;
 169                 return NETDEV_TX_BUSY;
 170         }
 171
 172         packet->extension = (void *)(unsigned long)packet +
 173                                 sizeof(struct hv_netvsc_packet) +
 174                                     (num_pages * sizeof(struct hv_page_buffer));
 175
 176         /* If the rndis msg goes beyond 1 page, we will add 1 later */
 177         packet->page_buf_cnt = num_pages - 1;
 178
 179         /* Initialize it from the skb */
 180         packet->total_data_buflen = skb->len;
 181
 182         /* Start filling in the page buffers starting after RNDIS buffer. */
 183         packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT;
 184         packet->page_buf[1].offset
 185                 = (unsigned long)skb->data & (PAGE_SIZE - 1);
 186         if (npg_data == 1)
 187                 packet->page_buf[1].len = skb_headlen(skb);
 188         else
 189                 packet->page_buf[1].len = PAGE_SIZE
 190                         - packet->page_buf[1].offset;
 191
 192         for (i = 2; i <= npg_data; i++) {
 193                 packet->page_buf[i].pfn = virt_to_phys(skb->data
 194                         + PAGE_SIZE * (i-1)) >> PAGE_SHIFT;
 195                 packet->page_buf[i].offset = 0;
 196                 packet->page_buf[i].len = PAGE_SIZE;
 197         }
 198         if (npg_data > 1)
 199                 packet->page_buf[npg_data].len = (((unsigned long)skb->data
 200                         + skb_headlen(skb) - 1) & (PAGE_SIZE - 1)) + 1;
 201
 202         /* Additional fragments are after SKB data */
 203         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 204                 const skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 205
 206                 packet->page_buf[i+npg_data+1].pfn =
 207                         page_to_pfn(skb_frag_page(f));
 208                 packet->page_buf[i+npg_data+1].offset = f->page_offset;
 209                 packet->page_buf[i+npg_data+1].len = skb_frag_size(f);
 210         }
 211
 212         /* Set the completion routine */
 213         packet->completion.send.send_completion = netvsc_xmit_completion;
 214         packet->completion.send.send_completion_ctx = packet;
 215         packet->completion.send.send_completion_tid = (unsigned long)skb;
 216
 217         ret = rndis_filter_send(net_device_ctx->device_ctx,
 218                                   packet);
 219         if (ret == 0) {
 220                 net->stats.tx_bytes += skb->len;
 221                 net->stats.tx_packets++;
 222         } else {
 223                 /* we are shutting down or bus overloaded, just drop packet */
 224                 net->stats.tx_dropped++;
 225                 kfree(packet);
 226                 dev_kfree_skb_any(skb);
 227         }
 228
 229         return ret ? NETDEV_TX_BUSY : NETDEV_TX_OK;
 230 }
 231
 232 /*
 233  * netvsc_linkstatus_callback - Link up/down notification
 234  */
 235 void netvsc_linkstatus_callback(struct hv_device *device_obj,
 236                                        unsigned int status)
 237 {
 238         struct net_device *net;
 239         struct net_device_context *ndev_ctx;
 240         struct netvsc_device *net_device;
 241
 242         net_device = hv_get_drvdata(device_obj);
 243         net = net_device->ndev;
 244
 245         if (!net) {
 246                 netdev_err(net, "got link status but net device "
 247                                 "not initialized yet\n");
 248                 return;
 249         }
 250
 251         if (status == 1) {
 252                 netif_carrier_on(net);
 253                 netif_wake_queue(net);
 254                 ndev_ctx = netdev_priv(net);
 255                 schedule_delayed_work(&ndev_ctx->dwork, 0);
 256                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
 257         } else {
 258                 netif_carrier_off(net);
 259                 netif_tx_disable(net);
 260         }
 261 }
 262
 263 /*
 264  * netvsc_recv_callback -  Callback when we receive a packet from the
 265  * "wire" on the specified device.
 266  */
 267 int netvsc_recv_callback(struct hv_device *device_obj,
 268                                 struct hv_netvsc_packet *packet)
 269 {
 270         struct net_device *net;
 271         struct sk_buff *skb;
 272
 273         net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
 274         if (!net) {
 275                 netdev_err(net, "got receive callback but net device"
 276                         " not initialized yet\n");
 277                 return 0;
 278         }
 279
 280         /* Allocate a skb - TODO direct I/O to pages? */
 281         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
 282         if (unlikely(!skb)) {
 283                 ++net->stats.rx_dropped;
 284                 return 0;
 285         }
 286
 287         /*
 288          * Copy to skb. This copy is needed here since the memory pointed by
 289          * hv_netvsc_packet cannot be deallocated
 290          */
 291         memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
 292                 packet->total_data_buflen);
 293
 294         skb->protocol = eth_type_trans(skb, net);
 295         skb->ip_summed = CHECKSUM_NONE;
 296
 297         net->stats.rx_packets++;
 298         net->stats.rx_bytes += packet->total_data_buflen;
 299
 300         /*
 301          * Pass the skb back up. Network stack will deallocate the skb when it
 302          * is done.
 303          * TODO - use NAPI?
 304          */
 305         netif_rx(skb);
 306
 307         return 0;
 308 }
 309
 310 static void netvsc_get_drvinfo(struct net_device *net,
 311                                struct ethtool_drvinfo *info)
 312 {
 313         strcpy(info->driver, KBUILD_MODNAME);
 314         strcpy(info->version, HV_DRV_VERSION);
 315         strcpy(info->fw_version, "N/A");
 316 }
 317
 318 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
 319 {
 320         struct net_device_context *ndevctx = netdev_priv(ndev);
 321         struct hv_device *hdev =  ndevctx->device_ctx;
 322         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
 323         struct netvsc_device_info device_info;
 324         int limit = ETH_DATA_LEN;
 325
 326         if (nvdev == NULL || nvdev->destroy)
 327                 return -ENODEV;
 328
 329         if (nvdev->nvsp_version == NVSP_PROTOCOL_VERSION_2)
 330                 limit = NETVSC_MTU;
 331
 332         if (mtu < 68 || mtu > limit)
 333                 return -EINVAL;
 334
 335         nvdev->start_remove = true;
 336         cancel_delayed_work_sync(&ndevctx->dwork);
 337         netif_tx_disable(ndev);
 338         rndis_filter_device_remove(hdev);
 339
 340         ndev->mtu = mtu;
 341
 342         ndevctx->device_ctx = hdev;
 343         hv_set_drvdata(hdev, ndev);
 344         device_info.ring_size = ring_size;
 345         rndis_filter_device_add(hdev, &device_info);
 346         netif_wake_queue(ndev);
 347
 348         return 0;
 349 }
 350
 351 static const struct ethtool_ops ethtool_ops = {
 352         .get_drvinfo    = netvsc_get_drvinfo,
 353         .get_link       = ethtool_op_get_link,
 354 };
 355
 356 static const struct net_device_ops device_ops = {
 357         .ndo_open =                     netvsc_open,
 358         .ndo_stop =                     netvsc_close,
 359         .ndo_start_xmit =               netvsc_start_xmit,
 360         .ndo_set_rx_mode =              netvsc_set_multicast_list,
 361         .ndo_change_mtu =               netvsc_change_mtu,
 362         .ndo_validate_addr =            eth_validate_addr,
 363         .ndo_set_mac_address =          eth_mac_addr,
 364 };
 365
 366 /*
 367  * Send GARP packet to network peers after migrations.
 368  * After Quick Migration, the network is not immediately operational in the
 369  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
 370  * another netif_notify_peers() into a delayed work, otherwise GARP packet
 371  * will not be sent after quick migration, and cause network disconnection.
 372  */
 373 static void netvsc_send_garp(struct work_struct *w)
 374 {
 375         struct net_device_context *ndev_ctx;
 376         struct net_device *net;
 377         struct netvsc_device *net_device;
 378
 379         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
 380         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
 381         net = net_device->ndev;
 382         netif_notify_peers(net);
 383 }
 384
 385
 386 static int netvsc_probe(struct hv_device *dev,
 387                         const struct hv_vmbus_device_id *dev_id)
 388 {
 389         struct net_device *net = NULL;
 390         struct net_device_context *net_device_ctx;
 391         struct netvsc_device_info device_info;
 392         int ret;
 393
 394         net = alloc_etherdev(sizeof(struct net_device_context));
 395         if (!net)
 396                 return -ENOMEM;
 397
 398         /* Set initial state */
 399         netif_carrier_off(net);
 400
 401         net_device_ctx = netdev_priv(net);
 402         net_device_ctx->device_ctx = dev;
 403         hv_set_drvdata(dev, net);
 404         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_send_garp);
 405
 406         net->netdev_ops = &device_ops;
 407
 408         /* TODO: Add GSO and Checksum offload */
 409         net->hw_features = NETIF_F_SG;
 410         net->features = NETIF_F_SG;
 411
 412         SET_ETHTOOL_OPS(net, &ethtool_ops);
 413         SET_NETDEV_DEV(net, &dev->device);
 414
 415         ret = register_netdev(net);
 416         if (ret != 0) {
 417                 pr_err("Unable to register netdev.\n");
 418                 free_netdev(net);
 419                 goto out;
 420         }
 421
 422         /* Notify the netvsc driver of the new device */
 423         device_info.ring_size = ring_size;
 424         ret = rndis_filter_device_add(dev, &device_info);
 425         if (ret != 0) {
 426                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
 427                 unregister_netdev(net);
 428                 free_netdev(net);
 429                 hv_set_drvdata(dev, NULL);
 430                 return ret;
 431         }
 432         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
 433
 434         netif_carrier_on(net);
 435
 436 out:
 437         return ret;
 438 }
 439
 440 static int netvsc_remove(struct hv_device *dev)
 441 {
 442         struct net_device *net;
 443         struct net_device_context *ndev_ctx;
 444         struct netvsc_device *net_device;
 445
 446         net_device = hv_get_drvdata(dev);
 447         net = net_device->ndev;
 448
 449         if (net == NULL) {
 450                 dev_err(&dev->device, "No net device to remove\n");
 451                 return 0;
 452         }
 453
 454         net_device->start_remove = true;
 455
 456         ndev_ctx = netdev_priv(net);
 457         cancel_delayed_work_sync(&ndev_ctx->dwork);
 458
 459         /* Stop outbound asap */
 460         netif_tx_disable(net);
 461
 462         unregister_netdev(net);
 463
 464         /*
 465          * Call to the vsc driver to let it know that the device is being
 466          * removed
 467          */
 468         rndis_filter_device_remove(dev);
 469
 470         free_netdev(net);
 471         return 0;
 472 }
 473
 474 static const struct hv_vmbus_device_id id_table[] = {
 475         /* Network guid */
 476         { VMBUS_DEVICE(0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
 477                        0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E) },
 478         { },
 479 };
 480
 481 MODULE_DEVICE_TABLE(vmbus, id_table);
 482
 483 /* The one and only one */
 484 static struct  hv_driver netvsc_drv = {
 485         .name = KBUILD_MODNAME,
 486         .id_table = id_table,
 487         .probe = netvsc_probe,
 488         .remove = netvsc_remove,
 489 };
 490
 491 static void __exit netvsc_drv_exit(void)
 492 {
 493         vmbus_driver_unregister(&netvsc_drv);
 494 }
 495
 496 static int __init netvsc_drv_init(void)
 497 {
 498         return vmbus_driver_register(&netvsc_drv);
 499 }
 500
 501 MODULE_LICENSE("GPL");
 502 MODULE_VERSION(HV_DRV_VERSION);
 503 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
 504
 505 module_init(netvsc_drv_init);
 506 module_exit(netvsc_drv_exit);