xref: /linux/drivers/net/hyperv/netvsc_drv.c (revision ec2212088c42ff7d1362629ec26dda4f3e8bdad3) 
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) +
163 			 NDIS_VLAN_PPI_SIZE, GFP_ATOMIC);
164 	if (!packet) {
165 		/* out of memory, drop packet */
166 		netdev_err(net, "unable to allocate hv_netvsc_packet\n");
167 
168 		dev_kfree_skb(skb);
169 		net->stats.tx_dropped++;
170 		return NETDEV_TX_OK;
171 	}
172 
173 	packet->vlan_tci = skb->vlan_tci;
174 
175 	packet->extension = (void *)(unsigned long)packet +
176 				sizeof(struct hv_netvsc_packet) +
177 				    (num_pages * sizeof(struct hv_page_buffer));
178 
179 	/* If the rndis msg goes beyond 1 page, we will add 1 later */
180 	packet->page_buf_cnt = num_pages - 1;
181 
182 	/* Initialize it from the skb */
183 	packet->total_data_buflen = skb->len;
184 
185 	/* Start filling in the page buffers starting after RNDIS buffer. */
186 	packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT;
187 	packet->page_buf[1].offset
188 		= (unsigned long)skb->data & (PAGE_SIZE - 1);
189 	if (npg_data == 1)
190 		packet->page_buf[1].len = skb_headlen(skb);
191 	else
192 		packet->page_buf[1].len = PAGE_SIZE
193 			- packet->page_buf[1].offset;
194 
195 	for (i = 2; i <= npg_data; i++) {
196 		packet->page_buf[i].pfn = virt_to_phys(skb->data
197 			+ PAGE_SIZE * (i-1)) >> PAGE_SHIFT;
198 		packet->page_buf[i].offset = 0;
199 		packet->page_buf[i].len = PAGE_SIZE;
200 	}
201 	if (npg_data > 1)
202 		packet->page_buf[npg_data].len = (((unsigned long)skb->data
203 			+ skb_headlen(skb) - 1) & (PAGE_SIZE - 1)) + 1;
204 
205 	/* Additional fragments are after SKB data */
206 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
207 		const skb_frag_t *f = &skb_shinfo(skb)->frags[i];
208 
209 		packet->page_buf[i+npg_data+1].pfn =
210 			page_to_pfn(skb_frag_page(f));
211 		packet->page_buf[i+npg_data+1].offset = f->page_offset;
212 		packet->page_buf[i+npg_data+1].len = skb_frag_size(f);
213 	}
214 
215 	/* Set the completion routine */
216 	packet->completion.send.send_completion = netvsc_xmit_completion;
217 	packet->completion.send.send_completion_ctx = packet;
218 	packet->completion.send.send_completion_tid = (unsigned long)skb;
219 
220 	ret = rndis_filter_send(net_device_ctx->device_ctx,
221 				  packet);
222 	if (ret == 0) {
223 		net->stats.tx_bytes += skb->len;
224 		net->stats.tx_packets++;
225 	} else {
226 		kfree(packet);
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 	skb->vlan_tci = packet->vlan_tci;
297 
298 	net->stats.rx_packets++;
299 	net->stats.rx_bytes += packet->total_data_buflen;
300 
301 	/*
302 	 * Pass the skb back up. Network stack will deallocate the skb when it
303 	 * is done.
304 	 * TODO - use NAPI?
305 	 */
306 	netif_rx(skb);
307 
308 	return 0;
309 }
310 
311 static void netvsc_get_drvinfo(struct net_device *net,
312 			       struct ethtool_drvinfo *info)
313 {
314 	strcpy(info->driver, KBUILD_MODNAME);
315 	strcpy(info->version, HV_DRV_VERSION);
316 	strcpy(info->fw_version, "N/A");
317 }
318 
319 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
320 {
321 	struct net_device_context *ndevctx = netdev_priv(ndev);
322 	struct hv_device *hdev =  ndevctx->device_ctx;
323 	struct netvsc_device *nvdev = hv_get_drvdata(hdev);
324 	struct netvsc_device_info device_info;
325 	int limit = ETH_DATA_LEN;
326 
327 	if (nvdev == NULL || nvdev->destroy)
328 		return -ENODEV;
329 
330 	if (nvdev->nvsp_version == NVSP_PROTOCOL_VERSION_2)
331 		limit = NETVSC_MTU;
332 
333 	if (mtu < 68 || mtu > limit)
334 		return -EINVAL;
335 
336 	nvdev->start_remove = true;
337 	cancel_delayed_work_sync(&ndevctx->dwork);
338 	netif_tx_disable(ndev);
339 	rndis_filter_device_remove(hdev);
340 
341 	ndev->mtu = mtu;
342 
343 	ndevctx->device_ctx = hdev;
344 	hv_set_drvdata(hdev, ndev);
345 	device_info.ring_size = ring_size;
346 	rndis_filter_device_add(hdev, &device_info);
347 	netif_wake_queue(ndev);
348 
349 	return 0;
350 }
351 
352 static const struct ethtool_ops ethtool_ops = {
353 	.get_drvinfo	= netvsc_get_drvinfo,
354 	.get_link	= ethtool_op_get_link,
355 };
356 
357 static const struct net_device_ops device_ops = {
358 	.ndo_open =			netvsc_open,
359 	.ndo_stop =			netvsc_close,
360 	.ndo_start_xmit =		netvsc_start_xmit,
361 	.ndo_set_rx_mode =		netvsc_set_multicast_list,
362 	.ndo_change_mtu =		netvsc_change_mtu,
363 	.ndo_validate_addr =		eth_validate_addr,
364 	.ndo_set_mac_address =		eth_mac_addr,
365 };
366 
367 /*
368  * Send GARP packet to network peers after migrations.
369  * After Quick Migration, the network is not immediately operational in the
370  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
371  * another netif_notify_peers() into a delayed work, otherwise GARP packet
372  * will not be sent after quick migration, and cause network disconnection.
373  */
374 static void netvsc_send_garp(struct work_struct *w)
375 {
376 	struct net_device_context *ndev_ctx;
377 	struct net_device *net;
378 	struct netvsc_device *net_device;
379 
380 	ndev_ctx = container_of(w, struct net_device_context, dwork.work);
381 	net_device = hv_get_drvdata(ndev_ctx->device_ctx);
382 	net = net_device->ndev;
383 	netif_notify_peers(net);
384 }
385 
386 
387 static int netvsc_probe(struct hv_device *dev,
388 			const struct hv_vmbus_device_id *dev_id)
389 {
390 	struct net_device *net = NULL;
391 	struct net_device_context *net_device_ctx;
392 	struct netvsc_device_info device_info;
393 	int ret;
394 
395 	net = alloc_etherdev(sizeof(struct net_device_context));
396 	if (!net)
397 		return -ENOMEM;
398 
399 	/* Set initial state */
400 	netif_carrier_off(net);
401 
402 	net_device_ctx = netdev_priv(net);
403 	net_device_ctx->device_ctx = dev;
404 	hv_set_drvdata(dev, net);
405 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_send_garp);
406 
407 	net->netdev_ops = &device_ops;
408 
409 	/* TODO: Add GSO and Checksum offload */
410 	net->hw_features = NETIF_F_SG;
411 	net->features = NETIF_F_SG | NETIF_F_HW_VLAN_TX;
412 
413 	SET_ETHTOOL_OPS(net, &ethtool_ops);
414 	SET_NETDEV_DEV(net, &dev->device);
415 
416 	ret = register_netdev(net);
417 	if (ret != 0) {
418 		pr_err("Unable to register netdev.\n");
419 		free_netdev(net);
420 		goto out;
421 	}
422 
423 	/* Notify the netvsc driver of the new device */
424 	device_info.ring_size = ring_size;
425 	ret = rndis_filter_device_add(dev, &device_info);
426 	if (ret != 0) {
427 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
428 		unregister_netdev(net);
429 		free_netdev(net);
430 		hv_set_drvdata(dev, NULL);
431 		return ret;
432 	}
433 	memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
434 
435 	netif_carrier_on(net);
436 
437 out:
438 	return ret;
439 }
440 
441 static int netvsc_remove(struct hv_device *dev)
442 {
443 	struct net_device *net;
444 	struct net_device_context *ndev_ctx;
445 	struct netvsc_device *net_device;
446 
447 	net_device = hv_get_drvdata(dev);
448 	net = net_device->ndev;
449 
450 	if (net == NULL) {
451 		dev_err(&dev->device, "No net device to remove\n");
452 		return 0;
453 	}
454 
455 	net_device->start_remove = true;
456 
457 	ndev_ctx = netdev_priv(net);
458 	cancel_delayed_work_sync(&ndev_ctx->dwork);
459 
460 	/* Stop outbound asap */
461 	netif_tx_disable(net);
462 
463 	unregister_netdev(net);
464 
465 	/*
466 	 * Call to the vsc driver to let it know that the device is being
467 	 * removed
468 	 */
469 	rndis_filter_device_remove(dev);
470 
471 	free_netdev(net);
472 	return 0;
473 }
474 
475 static const struct hv_vmbus_device_id id_table[] = {
476 	/* Network guid */
477 	{ VMBUS_DEVICE(0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
478 		       0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E) },
479 	{ },
480 };
481 
482 MODULE_DEVICE_TABLE(vmbus, id_table);
483 
484 /* The one and only one */
485 static struct  hv_driver netvsc_drv = {
486 	.name = KBUILD_MODNAME,
487 	.id_table = id_table,
488 	.probe = netvsc_probe,
489 	.remove = netvsc_remove,
490 };
491 
492 static void __exit netvsc_drv_exit(void)
493 {
494 	vmbus_driver_unregister(&netvsc_drv);
495 }
496 
497 static int __init netvsc_drv_init(void)
498 {
499 	return vmbus_driver_register(&netvsc_drv);
500 }
501 
502 MODULE_LICENSE("GPL");
503 MODULE_VERSION(HV_DRV_VERSION);
504 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
505 
506 module_init(netvsc_drv_init);
507 module_exit(netvsc_drv_exit);
508