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 struct work_struct work; 48 }; 49 50 #define RING_SIZE_MIN 64 51 static int ring_size = 128; 52 module_param(ring_size, int, S_IRUGO); 53 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)"); 54 55 static void do_set_multicast(struct work_struct *w) 56 { 57 struct net_device_context *ndevctx = 58 container_of(w, struct net_device_context, work); 59 struct netvsc_device *nvdev; 60 struct rndis_device *rdev; 61 62 nvdev = hv_get_drvdata(ndevctx->device_ctx); 63 if (nvdev == NULL || nvdev->ndev == NULL) 64 return; 65 66 rdev = nvdev->extension; 67 if (rdev == NULL) 68 return; 69 70 if (nvdev->ndev->flags & IFF_PROMISC) 71 rndis_filter_set_packet_filter(rdev, 72 NDIS_PACKET_TYPE_PROMISCUOUS); 73 else 74 rndis_filter_set_packet_filter(rdev, 75 NDIS_PACKET_TYPE_BROADCAST | 76 NDIS_PACKET_TYPE_ALL_MULTICAST | 77 NDIS_PACKET_TYPE_DIRECTED); 78 } 79 80 static void netvsc_set_multicast_list(struct net_device *net) 81 { 82 struct net_device_context *net_device_ctx = netdev_priv(net); 83 84 schedule_work(&net_device_ctx->work); 85 } 86 87 static int netvsc_open(struct net_device *net) 88 { 89 struct net_device_context *net_device_ctx = netdev_priv(net); 90 struct hv_device *device_obj = net_device_ctx->device_ctx; 91 int ret = 0; 92 93 /* Open up the device */ 94 ret = rndis_filter_open(device_obj); 95 if (ret != 0) { 96 netdev_err(net, "unable to open device (ret %d).\n", ret); 97 return ret; 98 } 99 100 netif_start_queue(net); 101 102 return ret; 103 } 104 105 static int netvsc_close(struct net_device *net) 106 { 107 struct net_device_context *net_device_ctx = netdev_priv(net); 108 struct hv_device *device_obj = net_device_ctx->device_ctx; 109 int ret; 110 111 netif_tx_disable(net); 112 113 /* Make sure netvsc_set_multicast_list doesn't re-enable filter! */ 114 cancel_work_sync(&net_device_ctx->work); 115 ret = rndis_filter_close(device_obj); 116 if (ret != 0) 117 netdev_err(net, "unable to close device (ret %d).\n", ret); 118 119 return ret; 120 } 121 122 static void netvsc_xmit_completion(void *context) 123 { 124 struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context; 125 struct sk_buff *skb = (struct sk_buff *) 126 (unsigned long)packet->completion.send.send_completion_tid; 127 128 kfree(packet); 129 130 if (skb) 131 dev_kfree_skb_any(skb); 132 } 133 134 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net) 135 { 136 struct net_device_context *net_device_ctx = netdev_priv(net); 137 struct hv_netvsc_packet *packet; 138 int ret; 139 unsigned int i, num_pages, npg_data; 140 141 /* Add multipages for skb->data and additional 2 for RNDIS */ 142 npg_data = (((unsigned long)skb->data + skb_headlen(skb) - 1) 143 >> PAGE_SHIFT) - ((unsigned long)skb->data >> PAGE_SHIFT) + 1; 144 num_pages = skb_shinfo(skb)->nr_frags + npg_data + 2; 145 146 /* Allocate a netvsc packet based on # of frags. */ 147 packet = kzalloc(sizeof(struct hv_netvsc_packet) + 148 (num_pages * sizeof(struct hv_page_buffer)) + 149 sizeof(struct rndis_filter_packet) + 150 NDIS_VLAN_PPI_SIZE, GFP_ATOMIC); 151 if (!packet) { 152 /* out of memory, drop packet */ 153 netdev_err(net, "unable to allocate hv_netvsc_packet\n"); 154 155 dev_kfree_skb(skb); 156 net->stats.tx_dropped++; 157 return NETDEV_TX_OK; 158 } 159 160 packet->vlan_tci = skb->vlan_tci; 161 162 packet->extension = (void *)(unsigned long)packet + 163 sizeof(struct hv_netvsc_packet) + 164 (num_pages * sizeof(struct hv_page_buffer)); 165 166 /* If the rndis msg goes beyond 1 page, we will add 1 later */ 167 packet->page_buf_cnt = num_pages - 1; 168 169 /* Initialize it from the skb */ 170 packet->total_data_buflen = skb->len; 171 172 /* Start filling in the page buffers starting after RNDIS buffer. */ 173 packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT; 174 packet->page_buf[1].offset 175 = (unsigned long)skb->data & (PAGE_SIZE - 1); 176 if (npg_data == 1) 177 packet->page_buf[1].len = skb_headlen(skb); 178 else 179 packet->page_buf[1].len = PAGE_SIZE 180 - packet->page_buf[1].offset; 181 182 for (i = 2; i <= npg_data; i++) { 183 packet->page_buf[i].pfn = virt_to_phys(skb->data 184 + PAGE_SIZE * (i-1)) >> PAGE_SHIFT; 185 packet->page_buf[i].offset = 0; 186 packet->page_buf[i].len = PAGE_SIZE; 187 } 188 if (npg_data > 1) 189 packet->page_buf[npg_data].len = (((unsigned long)skb->data 190 + skb_headlen(skb) - 1) & (PAGE_SIZE - 1)) + 1; 191 192 /* Additional fragments are after SKB data */ 193 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 194 const skb_frag_t *f = &skb_shinfo(skb)->frags[i]; 195 196 packet->page_buf[i+npg_data+1].pfn = 197 page_to_pfn(skb_frag_page(f)); 198 packet->page_buf[i+npg_data+1].offset = f->page_offset; 199 packet->page_buf[i+npg_data+1].len = skb_frag_size(f); 200 } 201 202 /* Set the completion routine */ 203 packet->completion.send.send_completion = netvsc_xmit_completion; 204 packet->completion.send.send_completion_ctx = packet; 205 packet->completion.send.send_completion_tid = (unsigned long)skb; 206 207 ret = rndis_filter_send(net_device_ctx->device_ctx, 208 packet); 209 if (ret == 0) { 210 net->stats.tx_bytes += skb->len; 211 net->stats.tx_packets++; 212 } else { 213 kfree(packet); 214 if (ret != -EAGAIN) { 215 dev_kfree_skb_any(skb); 216 net->stats.tx_dropped++; 217 } 218 } 219 220 return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK; 221 } 222 223 /* 224 * netvsc_linkstatus_callback - Link up/down notification 225 */ 226 void netvsc_linkstatus_callback(struct hv_device *device_obj, 227 unsigned int status) 228 { 229 struct net_device *net; 230 struct net_device_context *ndev_ctx; 231 struct netvsc_device *net_device; 232 233 net_device = hv_get_drvdata(device_obj); 234 net = net_device->ndev; 235 236 if (!net) { 237 netdev_err(net, "got link status but net device " 238 "not initialized yet\n"); 239 return; 240 } 241 242 if (status == 1) { 243 netif_carrier_on(net); 244 netif_wake_queue(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 netif_tx_disable(net); 251 } 252 } 253 254 /* 255 * netvsc_recv_callback - Callback when we receive a packet from the 256 * "wire" on the specified device. 257 */ 258 int netvsc_recv_callback(struct hv_device *device_obj, 259 struct hv_netvsc_packet *packet) 260 { 261 struct net_device *net; 262 struct sk_buff *skb; 263 264 net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev; 265 if (!net) { 266 netdev_err(net, "got receive callback but net device" 267 " not initialized yet\n"); 268 packet->status = NVSP_STAT_FAIL; 269 return 0; 270 } 271 272 /* Allocate a skb - TODO direct I/O to pages? */ 273 skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen); 274 if (unlikely(!skb)) { 275 ++net->stats.rx_dropped; 276 packet->status = NVSP_STAT_FAIL; 277 return 0; 278 } 279 280 /* 281 * Copy to skb. This copy is needed here since the memory pointed by 282 * hv_netvsc_packet cannot be deallocated 283 */ 284 memcpy(skb_put(skb, packet->total_data_buflen), packet->data, 285 packet->total_data_buflen); 286 287 skb->protocol = eth_type_trans(skb, net); 288 skb->ip_summed = CHECKSUM_NONE; 289 skb->vlan_tci = packet->vlan_tci; 290 291 net->stats.rx_packets++; 292 net->stats.rx_bytes += packet->total_data_buflen; 293 294 /* 295 * Pass the skb back up. Network stack will deallocate the skb when it 296 * is done. 297 * TODO - use NAPI? 298 */ 299 netif_rx(skb); 300 301 return 0; 302 } 303 304 static void netvsc_get_drvinfo(struct net_device *net, 305 struct ethtool_drvinfo *info) 306 { 307 strcpy(info->driver, KBUILD_MODNAME); 308 strcpy(info->version, HV_DRV_VERSION); 309 strcpy(info->fw_version, "N/A"); 310 } 311 312 static int netvsc_change_mtu(struct net_device *ndev, int mtu) 313 { 314 struct net_device_context *ndevctx = netdev_priv(ndev); 315 struct hv_device *hdev = ndevctx->device_ctx; 316 struct netvsc_device *nvdev = hv_get_drvdata(hdev); 317 struct netvsc_device_info device_info; 318 int limit = ETH_DATA_LEN; 319 320 if (nvdev == NULL || nvdev->destroy) 321 return -ENODEV; 322 323 if (nvdev->nvsp_version == NVSP_PROTOCOL_VERSION_2) 324 limit = NETVSC_MTU; 325 326 if (mtu < 68 || mtu > limit) 327 return -EINVAL; 328 329 nvdev->start_remove = true; 330 cancel_delayed_work_sync(&ndevctx->dwork); 331 cancel_work_sync(&ndevctx->work); 332 netif_tx_disable(ndev); 333 rndis_filter_device_remove(hdev); 334 335 ndev->mtu = mtu; 336 337 ndevctx->device_ctx = hdev; 338 hv_set_drvdata(hdev, ndev); 339 device_info.ring_size = ring_size; 340 rndis_filter_device_add(hdev, &device_info); 341 netif_wake_queue(ndev); 342 343 return 0; 344 } 345 346 347 static int netvsc_set_mac_addr(struct net_device *ndev, void *p) 348 { 349 struct net_device_context *ndevctx = netdev_priv(ndev); 350 struct hv_device *hdev = ndevctx->device_ctx; 351 struct sockaddr *addr = p; 352 char save_adr[14]; 353 unsigned char save_aatype; 354 int err; 355 356 memcpy(save_adr, ndev->dev_addr, ETH_ALEN); 357 save_aatype = ndev->addr_assign_type; 358 359 err = eth_mac_addr(ndev, p); 360 if (err != 0) 361 return err; 362 363 err = rndis_filter_set_device_mac(hdev, addr->sa_data); 364 if (err != 0) { 365 /* roll back to saved MAC */ 366 memcpy(ndev->dev_addr, save_adr, ETH_ALEN); 367 ndev->addr_assign_type = save_aatype; 368 } 369 370 return err; 371 } 372 373 374 static const struct ethtool_ops ethtool_ops = { 375 .get_drvinfo = netvsc_get_drvinfo, 376 .get_link = ethtool_op_get_link, 377 }; 378 379 static const struct net_device_ops device_ops = { 380 .ndo_open = netvsc_open, 381 .ndo_stop = netvsc_close, 382 .ndo_start_xmit = netvsc_start_xmit, 383 .ndo_set_rx_mode = netvsc_set_multicast_list, 384 .ndo_change_mtu = netvsc_change_mtu, 385 .ndo_validate_addr = eth_validate_addr, 386 .ndo_set_mac_address = netvsc_set_mac_addr, 387 }; 388 389 /* 390 * Send GARP packet to network peers after migrations. 391 * After Quick Migration, the network is not immediately operational in the 392 * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add 393 * another netif_notify_peers() into a delayed work, otherwise GARP packet 394 * will not be sent after quick migration, and cause network disconnection. 395 */ 396 static void netvsc_send_garp(struct work_struct *w) 397 { 398 struct net_device_context *ndev_ctx; 399 struct net_device *net; 400 struct netvsc_device *net_device; 401 402 ndev_ctx = container_of(w, struct net_device_context, dwork.work); 403 net_device = hv_get_drvdata(ndev_ctx->device_ctx); 404 net = net_device->ndev; 405 netdev_notify_peers(net); 406 } 407 408 409 static int netvsc_probe(struct hv_device *dev, 410 const struct hv_vmbus_device_id *dev_id) 411 { 412 struct net_device *net = NULL; 413 struct net_device_context *net_device_ctx; 414 struct netvsc_device_info device_info; 415 int ret; 416 417 net = alloc_etherdev(sizeof(struct net_device_context)); 418 if (!net) 419 return -ENOMEM; 420 421 /* Set initial state */ 422 netif_carrier_off(net); 423 424 net_device_ctx = netdev_priv(net); 425 net_device_ctx->device_ctx = dev; 426 hv_set_drvdata(dev, net); 427 INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_send_garp); 428 INIT_WORK(&net_device_ctx->work, do_set_multicast); 429 430 net->netdev_ops = &device_ops; 431 432 /* TODO: Add GSO and Checksum offload */ 433 net->hw_features = NETIF_F_SG; 434 net->features = NETIF_F_SG | NETIF_F_HW_VLAN_TX; 435 436 SET_ETHTOOL_OPS(net, ðtool_ops); 437 SET_NETDEV_DEV(net, &dev->device); 438 439 ret = register_netdev(net); 440 if (ret != 0) { 441 pr_err("Unable to register netdev.\n"); 442 free_netdev(net); 443 goto out; 444 } 445 446 /* Notify the netvsc driver of the new device */ 447 device_info.ring_size = ring_size; 448 ret = rndis_filter_device_add(dev, &device_info); 449 if (ret != 0) { 450 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret); 451 unregister_netdev(net); 452 free_netdev(net); 453 hv_set_drvdata(dev, NULL); 454 return ret; 455 } 456 memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN); 457 458 netif_carrier_on(net); 459 460 out: 461 return ret; 462 } 463 464 static int netvsc_remove(struct hv_device *dev) 465 { 466 struct net_device *net; 467 struct net_device_context *ndev_ctx; 468 struct netvsc_device *net_device; 469 470 net_device = hv_get_drvdata(dev); 471 net = net_device->ndev; 472 473 if (net == NULL) { 474 dev_err(&dev->device, "No net device to remove\n"); 475 return 0; 476 } 477 478 net_device->start_remove = true; 479 480 ndev_ctx = netdev_priv(net); 481 cancel_delayed_work_sync(&ndev_ctx->dwork); 482 cancel_work_sync(&ndev_ctx->work); 483 484 /* Stop outbound asap */ 485 netif_tx_disable(net); 486 487 unregister_netdev(net); 488 489 /* 490 * Call to the vsc driver to let it know that the device is being 491 * removed 492 */ 493 rndis_filter_device_remove(dev); 494 495 free_netdev(net); 496 return 0; 497 } 498 499 static const struct hv_vmbus_device_id id_table[] = { 500 /* Network guid */ 501 { VMBUS_DEVICE(0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46, 502 0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E) }, 503 { }, 504 }; 505 506 MODULE_DEVICE_TABLE(vmbus, id_table); 507 508 /* The one and only one */ 509 static struct hv_driver netvsc_drv = { 510 .name = KBUILD_MODNAME, 511 .id_table = id_table, 512 .probe = netvsc_probe, 513 .remove = netvsc_remove, 514 }; 515 516 static void __exit netvsc_drv_exit(void) 517 { 518 vmbus_driver_unregister(&netvsc_drv); 519 } 520 521 static int __init netvsc_drv_init(void) 522 { 523 if (ring_size < RING_SIZE_MIN) { 524 ring_size = RING_SIZE_MIN; 525 pr_info("Increased ring_size to %d (min allowed)\n", 526 ring_size); 527 } 528 return vmbus_driver_register(&netvsc_drv); 529 } 530 531 MODULE_LICENSE("GPL"); 532 MODULE_VERSION(HV_DRV_VERSION); 533 MODULE_DESCRIPTION("Microsoft Hyper-V network driver"); 534 535 module_init(netvsc_drv_init); 536 module_exit(netvsc_drv_exit); 537