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
43struct net_device_context {
44	/* point back to our device context */
45	struct hv_device *device_ctx;
46	struct delayed_work dwork;
47};
48
49
50static int ring_size = 128;
51module_param(ring_size, int, S_IRUGO);
52MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
53
54struct set_multicast_work {
55	struct work_struct work;
56	struct net_device *net;
57};
58
59static 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
86out:
87	kfree(w);
88}
89
90static 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
102static 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
120static 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
135static 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
147static 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 */
235void 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 */
267int netvsc_recv_callback(struct hv_device *device_obj,
268				struct hv_netvsc_packet *packet)
269{
270	struct net_device *net = dev_get_drvdata(&device_obj->device);
271	struct sk_buff *skb;
272	struct netvsc_device *net_device;
273
274	net_device = hv_get_drvdata(device_obj);
275	net = net_device->ndev;
276
277	if (!net) {
278		netdev_err(net, "got receive callback but net device"
279			" not initialized yet\n");
280		return 0;
281	}
282
283	/* Allocate a skb - TODO direct I/O to pages? */
284	skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
285	if (unlikely(!skb)) {
286		++net->stats.rx_dropped;
287		return 0;
288	}
289
290	/*
291	 * Copy to skb. This copy is needed here since the memory pointed by
292	 * hv_netvsc_packet cannot be deallocated
293	 */
294	memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
295		packet->total_data_buflen);
296
297	skb->protocol = eth_type_trans(skb, net);
298	skb->ip_summed = CHECKSUM_NONE;
299
300	net->stats.rx_packets++;
301	net->stats.rx_bytes += packet->total_data_buflen;
302
303	/*
304	 * Pass the skb back up. Network stack will deallocate the skb when it
305	 * is done.
306	 * TODO - use NAPI?
307	 */
308	netif_rx(skb);
309
310	return 0;
311}
312
313static void netvsc_get_drvinfo(struct net_device *net,
314			       struct ethtool_drvinfo *info)
315{
316	strcpy(info->driver, "hv_netvsc");
317	strcpy(info->version, HV_DRV_VERSION);
318	strcpy(info->fw_version, "N/A");
319}
320
321static int netvsc_change_mtu(struct net_device *ndev, int mtu)
322{
323	struct net_device_context *ndevctx = netdev_priv(ndev);
324	struct hv_device *hdev =  ndevctx->device_ctx;
325	struct netvsc_device *nvdev = hv_get_drvdata(hdev);
326	struct netvsc_device_info device_info;
327	int limit = ETH_DATA_LEN;
328
329	if (nvdev == NULL || nvdev->destroy)
330		return -ENODEV;
331
332	if (nvdev->nvsp_version == NVSP_PROTOCOL_VERSION_2)
333		limit = NETVSC_MTU;
334
335	if (mtu < 68 || mtu > limit)
336		return -EINVAL;
337
338	nvdev->start_remove = true;
339	cancel_delayed_work_sync(&ndevctx->dwork);
340	netif_tx_disable(ndev);
341	rndis_filter_device_remove(hdev);
342
343	ndev->mtu = mtu;
344
345	ndevctx->device_ctx = hdev;
346	hv_set_drvdata(hdev, ndev);
347	device_info.ring_size = ring_size;
348	rndis_filter_device_add(hdev, &device_info);
349	netif_wake_queue(ndev);
350
351	return 0;
352}
353
354static const struct ethtool_ops ethtool_ops = {
355	.get_drvinfo	= netvsc_get_drvinfo,
356	.get_link	= ethtool_op_get_link,
357};
358
359static const struct net_device_ops device_ops = {
360	.ndo_open =			netvsc_open,
361	.ndo_stop =			netvsc_close,
362	.ndo_start_xmit =		netvsc_start_xmit,
363	.ndo_set_rx_mode =		netvsc_set_multicast_list,
364	.ndo_change_mtu =		netvsc_change_mtu,
365	.ndo_validate_addr =		eth_validate_addr,
366	.ndo_set_mac_address =		eth_mac_addr,
367};
368
369/*
370 * Send GARP packet to network peers after migrations.
371 * After Quick Migration, the network is not immediately operational in the
372 * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
373 * another netif_notify_peers() into a delayed work, otherwise GARP packet
374 * will not be sent after quick migration, and cause network disconnection.
375 */
376static void netvsc_send_garp(struct work_struct *w)
377{
378	struct net_device_context *ndev_ctx;
379	struct net_device *net;
380	struct netvsc_device *net_device;
381
382	ndev_ctx = container_of(w, struct net_device_context, dwork.work);
383	net_device = hv_get_drvdata(ndev_ctx->device_ctx);
384	net = net_device->ndev;
385	netif_notify_peers(net);
386}
387
388
389static int netvsc_probe(struct hv_device *dev,
390			const struct hv_vmbus_device_id *dev_id)
391{
392	struct net_device *net = NULL;
393	struct net_device_context *net_device_ctx;
394	struct netvsc_device_info device_info;
395	int ret;
396
397	net = alloc_etherdev(sizeof(struct net_device_context));
398	if (!net)
399		return -ENOMEM;
400
401	/* Set initial state */
402	netif_carrier_off(net);
403
404	net_device_ctx = netdev_priv(net);
405	net_device_ctx->device_ctx = dev;
406	hv_set_drvdata(dev, net);
407	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_send_garp);
408
409	net->netdev_ops = &device_ops;
410
411	/* TODO: Add GSO and Checksum offload */
412	net->hw_features = NETIF_F_SG;
413	net->features = NETIF_F_SG;
414
415	SET_ETHTOOL_OPS(net, &ethtool_ops);
416	SET_NETDEV_DEV(net, &dev->device);
417
418	ret = register_netdev(net);
419	if (ret != 0) {
420		pr_err("Unable to register netdev.\n");
421		free_netdev(net);
422		goto out;
423	}
424
425	/* Notify the netvsc driver of the new device */
426	device_info.ring_size = ring_size;
427	ret = rndis_filter_device_add(dev, &device_info);
428	if (ret != 0) {
429		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
430		unregister_netdev(net);
431		free_netdev(net);
432		hv_set_drvdata(dev, NULL);
433		return ret;
434	}
435	memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
436
437	netif_carrier_on(net);
438
439out:
440	return ret;
441}
442
443static int netvsc_remove(struct hv_device *dev)
444{
445	struct net_device *net;
446	struct net_device_context *ndev_ctx;
447	struct netvsc_device *net_device;
448
449	net_device = hv_get_drvdata(dev);
450	net = net_device->ndev;
451
452	if (net == NULL) {
453		dev_err(&dev->device, "No net device to remove\n");
454		return 0;
455	}
456
457	net_device->start_remove = true;
458
459	ndev_ctx = netdev_priv(net);
460	cancel_delayed_work_sync(&ndev_ctx->dwork);
461
462	/* Stop outbound asap */
463	netif_tx_disable(net);
464
465	unregister_netdev(net);
466
467	/*
468	 * Call to the vsc driver to let it know that the device is being
469	 * removed
470	 */
471	rndis_filter_device_remove(dev);
472
473	free_netdev(net);
474	return 0;
475}
476
477static const struct hv_vmbus_device_id id_table[] = {
478	/* Network guid */
479	{ VMBUS_DEVICE(0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
480		       0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E) },
481	{ },
482};
483
484MODULE_DEVICE_TABLE(vmbus, id_table);
485
486/* The one and only one */
487static struct  hv_driver netvsc_drv = {
488	.name = "netvsc",
489	.id_table = id_table,
490	.probe = netvsc_probe,
491	.remove = netvsc_remove,
492};
493
494static void __exit netvsc_drv_exit(void)
495{
496	vmbus_driver_unregister(&netvsc_drv);
497}
498
499static int __init netvsc_drv_init(void)
500{
501	return vmbus_driver_register(&netvsc_drv);
502}
503
504MODULE_LICENSE("GPL");
505MODULE_VERSION(HV_DRV_VERSION);
506MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
507
508module_init(netvsc_drv_init);
509module_exit(netvsc_drv_exit);
510