xref: /openbmc/linux/drivers/net/hyperv/netvsc_drv.c (revision 4bce6fce)
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, see <http://www.gnu.org/licenses/>.
15  *
16  * Authors:
17  *   Haiyang Zhang <haiyangz@microsoft.com>
18  *   Hank Janssen  <hjanssen@microsoft.com>
19  */
20 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
21 
22 #include <linux/init.h>
23 #include <linux/atomic.h>
24 #include <linux/module.h>
25 #include <linux/highmem.h>
26 #include <linux/device.h>
27 #include <linux/io.h>
28 #include <linux/delay.h>
29 #include <linux/netdevice.h>
30 #include <linux/inetdevice.h>
31 #include <linux/etherdevice.h>
32 #include <linux/skbuff.h>
33 #include <linux/if_vlan.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 	struct netvsc_device *nvdev;
92 	struct rndis_device *rdev;
93 	int ret = 0;
94 
95 	netif_carrier_off(net);
96 
97 	/* Open up the device */
98 	ret = rndis_filter_open(device_obj);
99 	if (ret != 0) {
100 		netdev_err(net, "unable to open device (ret %d).\n", ret);
101 		return ret;
102 	}
103 
104 	netif_tx_start_all_queues(net);
105 
106 	nvdev = hv_get_drvdata(device_obj);
107 	rdev = nvdev->extension;
108 	if (!rdev->link_state)
109 		netif_carrier_on(net);
110 
111 	return ret;
112 }
113 
114 static int netvsc_close(struct net_device *net)
115 {
116 	struct net_device_context *net_device_ctx = netdev_priv(net);
117 	struct hv_device *device_obj = net_device_ctx->device_ctx;
118 	int ret;
119 
120 	netif_tx_disable(net);
121 
122 	/* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
123 	cancel_work_sync(&net_device_ctx->work);
124 	ret = rndis_filter_close(device_obj);
125 	if (ret != 0)
126 		netdev_err(net, "unable to close device (ret %d).\n", ret);
127 
128 	return ret;
129 }
130 
131 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
132 				int pkt_type)
133 {
134 	struct rndis_packet *rndis_pkt;
135 	struct rndis_per_packet_info *ppi;
136 
137 	rndis_pkt = &msg->msg.pkt;
138 	rndis_pkt->data_offset += ppi_size;
139 
140 	ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
141 		rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
142 
143 	ppi->size = ppi_size;
144 	ppi->type = pkt_type;
145 	ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
146 
147 	rndis_pkt->per_pkt_info_len += ppi_size;
148 
149 	return ppi;
150 }
151 
152 union sub_key {
153 	u64 k;
154 	struct {
155 		u8 pad[3];
156 		u8 kb;
157 		u32 ka;
158 	};
159 };
160 
161 /* Toeplitz hash function
162  * data: network byte order
163  * return: host byte order
164  */
165 static u32 comp_hash(u8 *key, int klen, void *data, int dlen)
166 {
167 	union sub_key subk;
168 	int k_next = 4;
169 	u8 dt;
170 	int i, j;
171 	u32 ret = 0;
172 
173 	subk.k = 0;
174 	subk.ka = ntohl(*(u32 *)key);
175 
176 	for (i = 0; i < dlen; i++) {
177 		subk.kb = key[k_next];
178 		k_next = (k_next + 1) % klen;
179 		dt = ((u8 *)data)[i];
180 		for (j = 0; j < 8; j++) {
181 			if (dt & 0x80)
182 				ret ^= subk.ka;
183 			dt <<= 1;
184 			subk.k <<= 1;
185 		}
186 	}
187 
188 	return ret;
189 }
190 
191 static bool netvsc_set_hash(u32 *hash, struct sk_buff *skb)
192 {
193 	struct flow_keys flow;
194 	int data_len;
195 
196 	if (!skb_flow_dissect(skb, &flow) ||
197 	    !(flow.n_proto == htons(ETH_P_IP) ||
198 	      flow.n_proto == htons(ETH_P_IPV6)))
199 		return false;
200 
201 	if (flow.ip_proto == IPPROTO_TCP)
202 		data_len = 12;
203 	else
204 		data_len = 8;
205 
206 	*hash = comp_hash(netvsc_hash_key, HASH_KEYLEN, &flow, data_len);
207 
208 	return true;
209 }
210 
211 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
212 			void *accel_priv, select_queue_fallback_t fallback)
213 {
214 	struct net_device_context *net_device_ctx = netdev_priv(ndev);
215 	struct hv_device *hdev =  net_device_ctx->device_ctx;
216 	struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
217 	u32 hash;
218 	u16 q_idx = 0;
219 
220 	if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
221 		return 0;
222 
223 	if (netvsc_set_hash(&hash, skb)) {
224 		q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
225 			ndev->real_num_tx_queues;
226 		skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
227 	}
228 
229 	return q_idx;
230 }
231 
232 void netvsc_xmit_completion(void *context)
233 {
234 	struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
235 	struct sk_buff *skb = (struct sk_buff *)
236 		(unsigned long)packet->send_completion_tid;
237 
238 	if (!packet->part_of_skb)
239 		kfree(packet);
240 
241 	if (skb)
242 		dev_kfree_skb_any(skb);
243 }
244 
245 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
246 			struct hv_page_buffer *pb)
247 {
248 	int j = 0;
249 
250 	/* Deal with compund pages by ignoring unused part
251 	 * of the page.
252 	 */
253 	page += (offset >> PAGE_SHIFT);
254 	offset &= ~PAGE_MASK;
255 
256 	while (len > 0) {
257 		unsigned long bytes;
258 
259 		bytes = PAGE_SIZE - offset;
260 		if (bytes > len)
261 			bytes = len;
262 		pb[j].pfn = page_to_pfn(page);
263 		pb[j].offset = offset;
264 		pb[j].len = bytes;
265 
266 		offset += bytes;
267 		len -= bytes;
268 
269 		if (offset == PAGE_SIZE && len) {
270 			page++;
271 			offset = 0;
272 			j++;
273 		}
274 	}
275 
276 	return j + 1;
277 }
278 
279 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
280 			   struct hv_netvsc_packet *packet)
281 {
282 	struct hv_page_buffer *pb = packet->page_buf;
283 	u32 slots_used = 0;
284 	char *data = skb->data;
285 	int frags = skb_shinfo(skb)->nr_frags;
286 	int i;
287 
288 	/* The packet is laid out thus:
289 	 * 1. hdr: RNDIS header and PPI
290 	 * 2. skb linear data
291 	 * 3. skb fragment data
292 	 */
293 	if (hdr != NULL)
294 		slots_used += fill_pg_buf(virt_to_page(hdr),
295 					offset_in_page(hdr),
296 					len, &pb[slots_used]);
297 
298 	packet->rmsg_size = len;
299 	packet->rmsg_pgcnt = slots_used;
300 
301 	slots_used += fill_pg_buf(virt_to_page(data),
302 				offset_in_page(data),
303 				skb_headlen(skb), &pb[slots_used]);
304 
305 	for (i = 0; i < frags; i++) {
306 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
307 
308 		slots_used += fill_pg_buf(skb_frag_page(frag),
309 					frag->page_offset,
310 					skb_frag_size(frag), &pb[slots_used]);
311 	}
312 	return slots_used;
313 }
314 
315 static int count_skb_frag_slots(struct sk_buff *skb)
316 {
317 	int i, frags = skb_shinfo(skb)->nr_frags;
318 	int pages = 0;
319 
320 	for (i = 0; i < frags; i++) {
321 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
322 		unsigned long size = skb_frag_size(frag);
323 		unsigned long offset = frag->page_offset;
324 
325 		/* Skip unused frames from start of page */
326 		offset &= ~PAGE_MASK;
327 		pages += PFN_UP(offset + size);
328 	}
329 	return pages;
330 }
331 
332 static int netvsc_get_slots(struct sk_buff *skb)
333 {
334 	char *data = skb->data;
335 	unsigned int offset = offset_in_page(data);
336 	unsigned int len = skb_headlen(skb);
337 	int slots;
338 	int frag_slots;
339 
340 	slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
341 	frag_slots = count_skb_frag_slots(skb);
342 	return slots + frag_slots;
343 }
344 
345 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
346 {
347 	u32 ret_val = TRANSPORT_INFO_NOT_IP;
348 
349 	if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
350 		(eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
351 		goto not_ip;
352 	}
353 
354 	*trans_off = skb_transport_offset(skb);
355 
356 	if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
357 		struct iphdr *iphdr = ip_hdr(skb);
358 
359 		if (iphdr->protocol == IPPROTO_TCP)
360 			ret_val = TRANSPORT_INFO_IPV4_TCP;
361 		else if (iphdr->protocol == IPPROTO_UDP)
362 			ret_val = TRANSPORT_INFO_IPV4_UDP;
363 	} else {
364 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
365 			ret_val = TRANSPORT_INFO_IPV6_TCP;
366 		else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
367 			ret_val = TRANSPORT_INFO_IPV6_UDP;
368 	}
369 
370 not_ip:
371 	return ret_val;
372 }
373 
374 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
375 {
376 	struct net_device_context *net_device_ctx = netdev_priv(net);
377 	struct hv_netvsc_packet *packet = NULL;
378 	int ret;
379 	unsigned int num_data_pgs;
380 	struct rndis_message *rndis_msg;
381 	struct rndis_packet *rndis_pkt;
382 	u32 rndis_msg_size;
383 	bool isvlan;
384 	bool linear = false;
385 	struct rndis_per_packet_info *ppi;
386 	struct ndis_tcp_ip_checksum_info *csum_info;
387 	struct ndis_tcp_lso_info *lso_info;
388 	int  hdr_offset;
389 	u32 net_trans_info;
390 	u32 hash;
391 	u32 skb_length;
392 	u32 head_room;
393 	u32 pkt_sz;
394 	struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
395 
396 
397 	/* We will atmost need two pages to describe the rndis
398 	 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
399 	 * of pages in a single packet. If skb is scattered around
400 	 * more pages we try linearizing it.
401 	 */
402 
403 check_size:
404 	skb_length = skb->len;
405 	head_room = skb_headroom(skb);
406 	num_data_pgs = netvsc_get_slots(skb) + 2;
407 	if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
408 		net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
409 				      num_data_pgs, skb->len);
410 		ret = -EFAULT;
411 		goto drop;
412 	} else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
413 		if (skb_linearize(skb)) {
414 			net_alert_ratelimited("failed to linearize skb\n");
415 			ret = -ENOMEM;
416 			goto drop;
417 		}
418 		linear = true;
419 		goto check_size;
420 	}
421 
422 	pkt_sz = sizeof(struct hv_netvsc_packet) + RNDIS_AND_PPI_SIZE;
423 
424 	if (head_room < pkt_sz) {
425 		packet = kmalloc(pkt_sz, GFP_ATOMIC);
426 		if (!packet) {
427 			/* out of memory, drop packet */
428 			netdev_err(net, "unable to alloc hv_netvsc_packet\n");
429 			ret = -ENOMEM;
430 			goto drop;
431 		}
432 		packet->part_of_skb = false;
433 	} else {
434 		/* Use the headroom for building up the packet */
435 		packet = (struct hv_netvsc_packet *)skb->head;
436 		packet->part_of_skb = true;
437 	}
438 
439 	packet->status = 0;
440 	packet->xmit_more = skb->xmit_more;
441 
442 	packet->vlan_tci = skb->vlan_tci;
443 	packet->page_buf = page_buf;
444 
445 	packet->q_idx = skb_get_queue_mapping(skb);
446 
447 	packet->is_data_pkt = true;
448 	packet->total_data_buflen = skb->len;
449 
450 	packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
451 				sizeof(struct hv_netvsc_packet));
452 
453 	memset(packet->rndis_msg, 0, RNDIS_AND_PPI_SIZE);
454 
455 	/* Set the completion routine */
456 	packet->send_completion = netvsc_xmit_completion;
457 	packet->send_completion_ctx = packet;
458 	packet->send_completion_tid = (unsigned long)skb;
459 
460 	isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
461 
462 	/* Add the rndis header */
463 	rndis_msg = packet->rndis_msg;
464 	rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
465 	rndis_msg->msg_len = packet->total_data_buflen;
466 	rndis_pkt = &rndis_msg->msg.pkt;
467 	rndis_pkt->data_offset = sizeof(struct rndis_packet);
468 	rndis_pkt->data_len = packet->total_data_buflen;
469 	rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
470 
471 	rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
472 
473 	hash = skb_get_hash_raw(skb);
474 	if (hash != 0 && net->real_num_tx_queues > 1) {
475 		rndis_msg_size += NDIS_HASH_PPI_SIZE;
476 		ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
477 				    NBL_HASH_VALUE);
478 		*(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
479 	}
480 
481 	if (isvlan) {
482 		struct ndis_pkt_8021q_info *vlan;
483 
484 		rndis_msg_size += NDIS_VLAN_PPI_SIZE;
485 		ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
486 					IEEE_8021Q_INFO);
487 		vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
488 						ppi->ppi_offset);
489 		vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
490 		vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
491 				VLAN_PRIO_SHIFT;
492 	}
493 
494 	net_trans_info = get_net_transport_info(skb, &hdr_offset);
495 	if (net_trans_info == TRANSPORT_INFO_NOT_IP)
496 		goto do_send;
497 
498 	/*
499 	 * Setup the sendside checksum offload only if this is not a
500 	 * GSO packet.
501 	 */
502 	if (skb_is_gso(skb))
503 		goto do_lso;
504 
505 	if ((skb->ip_summed == CHECKSUM_NONE) ||
506 	    (skb->ip_summed == CHECKSUM_UNNECESSARY))
507 		goto do_send;
508 
509 	rndis_msg_size += NDIS_CSUM_PPI_SIZE;
510 	ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
511 			    TCPIP_CHKSUM_PKTINFO);
512 
513 	csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
514 			ppi->ppi_offset);
515 
516 	if (net_trans_info & (INFO_IPV4 << 16))
517 		csum_info->transmit.is_ipv4 = 1;
518 	else
519 		csum_info->transmit.is_ipv6 = 1;
520 
521 	if (net_trans_info & INFO_TCP) {
522 		csum_info->transmit.tcp_checksum = 1;
523 		csum_info->transmit.tcp_header_offset = hdr_offset;
524 	} else if (net_trans_info & INFO_UDP) {
525 		/* UDP checksum offload is not supported on ws2008r2.
526 		 * Furthermore, on ws2012 and ws2012r2, there are some
527 		 * issues with udp checksum offload from Linux guests.
528 		 * (these are host issues).
529 		 * For now compute the checksum here.
530 		 */
531 		struct udphdr *uh;
532 		u16 udp_len;
533 
534 		ret = skb_cow_head(skb, 0);
535 		if (ret)
536 			goto drop;
537 
538 		uh = udp_hdr(skb);
539 		udp_len = ntohs(uh->len);
540 		uh->check = 0;
541 		uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
542 					      ip_hdr(skb)->daddr,
543 					      udp_len, IPPROTO_UDP,
544 					      csum_partial(uh, udp_len, 0));
545 		if (uh->check == 0)
546 			uh->check = CSUM_MANGLED_0;
547 
548 		csum_info->transmit.udp_checksum = 0;
549 	}
550 	goto do_send;
551 
552 do_lso:
553 	rndis_msg_size += NDIS_LSO_PPI_SIZE;
554 	ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
555 			    TCP_LARGESEND_PKTINFO);
556 
557 	lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
558 			ppi->ppi_offset);
559 
560 	lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
561 	if (net_trans_info & (INFO_IPV4 << 16)) {
562 		lso_info->lso_v2_transmit.ip_version =
563 			NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
564 		ip_hdr(skb)->tot_len = 0;
565 		ip_hdr(skb)->check = 0;
566 		tcp_hdr(skb)->check =
567 		~csum_tcpudp_magic(ip_hdr(skb)->saddr,
568 				   ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
569 	} else {
570 		lso_info->lso_v2_transmit.ip_version =
571 			NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
572 		ipv6_hdr(skb)->payload_len = 0;
573 		tcp_hdr(skb)->check =
574 		~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
575 				&ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
576 	}
577 	lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
578 	lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
579 
580 do_send:
581 	/* Start filling in the page buffers with the rndis hdr */
582 	rndis_msg->msg_len += rndis_msg_size;
583 	packet->total_data_buflen = rndis_msg->msg_len;
584 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
585 					       skb, packet);
586 
587 	ret = netvsc_send(net_device_ctx->device_ctx, packet);
588 
589 drop:
590 	if (ret == 0) {
591 		net->stats.tx_bytes += skb_length;
592 		net->stats.tx_packets++;
593 	} else {
594 		if (packet && !packet->part_of_skb)
595 			kfree(packet);
596 		if (ret != -EAGAIN) {
597 			dev_kfree_skb_any(skb);
598 			net->stats.tx_dropped++;
599 		}
600 	}
601 
602 	return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
603 }
604 
605 /*
606  * netvsc_linkstatus_callback - Link up/down notification
607  */
608 void netvsc_linkstatus_callback(struct hv_device *device_obj,
609 				struct rndis_message *resp)
610 {
611 	struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
612 	struct net_device *net;
613 	struct net_device_context *ndev_ctx;
614 	struct netvsc_device *net_device;
615 	struct rndis_device *rdev;
616 
617 	net_device = hv_get_drvdata(device_obj);
618 	rdev = net_device->extension;
619 
620 	switch (indicate->status) {
621 	case RNDIS_STATUS_MEDIA_CONNECT:
622 		rdev->link_state = false;
623 		break;
624 	case RNDIS_STATUS_MEDIA_DISCONNECT:
625 		rdev->link_state = true;
626 		break;
627 	case RNDIS_STATUS_NETWORK_CHANGE:
628 		rdev->link_change = true;
629 		break;
630 	default:
631 		return;
632 	}
633 
634 	net = net_device->ndev;
635 
636 	if (!net || net->reg_state != NETREG_REGISTERED)
637 		return;
638 
639 	ndev_ctx = netdev_priv(net);
640 	if (!rdev->link_state) {
641 		schedule_delayed_work(&ndev_ctx->dwork, 0);
642 		schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
643 	} else {
644 		schedule_delayed_work(&ndev_ctx->dwork, 0);
645 	}
646 }
647 
648 /*
649  * netvsc_recv_callback -  Callback when we receive a packet from the
650  * "wire" on the specified device.
651  */
652 int netvsc_recv_callback(struct hv_device *device_obj,
653 				struct hv_netvsc_packet *packet,
654 				struct ndis_tcp_ip_checksum_info *csum_info)
655 {
656 	struct net_device *net;
657 	struct sk_buff *skb;
658 
659 	net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
660 	if (!net || net->reg_state != NETREG_REGISTERED) {
661 		packet->status = NVSP_STAT_FAIL;
662 		return 0;
663 	}
664 
665 	/* Allocate a skb - TODO direct I/O to pages? */
666 	skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
667 	if (unlikely(!skb)) {
668 		++net->stats.rx_dropped;
669 		packet->status = NVSP_STAT_FAIL;
670 		return 0;
671 	}
672 
673 	/*
674 	 * Copy to skb. This copy is needed here since the memory pointed by
675 	 * hv_netvsc_packet cannot be deallocated
676 	 */
677 	memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
678 		packet->total_data_buflen);
679 
680 	skb->protocol = eth_type_trans(skb, net);
681 	if (csum_info) {
682 		/* We only look at the IP checksum here.
683 		 * Should we be dropping the packet if checksum
684 		 * failed? How do we deal with other checksums - TCP/UDP?
685 		 */
686 		if (csum_info->receive.ip_checksum_succeeded)
687 			skb->ip_summed = CHECKSUM_UNNECESSARY;
688 		else
689 			skb->ip_summed = CHECKSUM_NONE;
690 	}
691 
692 	if (packet->vlan_tci & VLAN_TAG_PRESENT)
693 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
694 				       packet->vlan_tci);
695 
696 	skb_record_rx_queue(skb, packet->channel->
697 			    offermsg.offer.sub_channel_index);
698 
699 	net->stats.rx_packets++;
700 	net->stats.rx_bytes += packet->total_data_buflen;
701 
702 	/*
703 	 * Pass the skb back up. Network stack will deallocate the skb when it
704 	 * is done.
705 	 * TODO - use NAPI?
706 	 */
707 	netif_rx(skb);
708 
709 	return 0;
710 }
711 
712 static void netvsc_get_drvinfo(struct net_device *net,
713 			       struct ethtool_drvinfo *info)
714 {
715 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
716 	strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
717 }
718 
719 static void netvsc_get_channels(struct net_device *net,
720 				struct ethtool_channels *channel)
721 {
722 	struct net_device_context *net_device_ctx = netdev_priv(net);
723 	struct hv_device *dev = net_device_ctx->device_ctx;
724 	struct netvsc_device *nvdev = hv_get_drvdata(dev);
725 
726 	if (nvdev) {
727 		channel->max_combined	= nvdev->max_chn;
728 		channel->combined_count = nvdev->num_chn;
729 	}
730 }
731 
732 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
733 {
734 	struct net_device_context *ndevctx = netdev_priv(ndev);
735 	struct hv_device *hdev =  ndevctx->device_ctx;
736 	struct netvsc_device *nvdev = hv_get_drvdata(hdev);
737 	struct netvsc_device_info device_info;
738 	int limit = ETH_DATA_LEN;
739 
740 	if (nvdev == NULL || nvdev->destroy)
741 		return -ENODEV;
742 
743 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
744 		limit = NETVSC_MTU - ETH_HLEN;
745 
746 	/* Hyper-V hosts don't support MTU < ETH_DATA_LEN (1500) */
747 	if (mtu < ETH_DATA_LEN || mtu > limit)
748 		return -EINVAL;
749 
750 	nvdev->start_remove = true;
751 	cancel_work_sync(&ndevctx->work);
752 	netif_tx_disable(ndev);
753 	rndis_filter_device_remove(hdev);
754 
755 	ndev->mtu = mtu;
756 
757 	ndevctx->device_ctx = hdev;
758 	hv_set_drvdata(hdev, ndev);
759 	device_info.ring_size = ring_size;
760 	rndis_filter_device_add(hdev, &device_info);
761 	netif_tx_wake_all_queues(ndev);
762 
763 	return 0;
764 }
765 
766 
767 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
768 {
769 	struct net_device_context *ndevctx = netdev_priv(ndev);
770 	struct hv_device *hdev =  ndevctx->device_ctx;
771 	struct sockaddr *addr = p;
772 	char save_adr[ETH_ALEN];
773 	unsigned char save_aatype;
774 	int err;
775 
776 	memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
777 	save_aatype = ndev->addr_assign_type;
778 
779 	err = eth_mac_addr(ndev, p);
780 	if (err != 0)
781 		return err;
782 
783 	err = rndis_filter_set_device_mac(hdev, addr->sa_data);
784 	if (err != 0) {
785 		/* roll back to saved MAC */
786 		memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
787 		ndev->addr_assign_type = save_aatype;
788 	}
789 
790 	return err;
791 }
792 
793 #ifdef CONFIG_NET_POLL_CONTROLLER
794 static void netvsc_poll_controller(struct net_device *net)
795 {
796 	/* As netvsc_start_xmit() works synchronous we don't have to
797 	 * trigger anything here.
798 	 */
799 }
800 #endif
801 
802 static const struct ethtool_ops ethtool_ops = {
803 	.get_drvinfo	= netvsc_get_drvinfo,
804 	.get_link	= ethtool_op_get_link,
805 	.get_channels   = netvsc_get_channels,
806 };
807 
808 static const struct net_device_ops device_ops = {
809 	.ndo_open =			netvsc_open,
810 	.ndo_stop =			netvsc_close,
811 	.ndo_start_xmit =		netvsc_start_xmit,
812 	.ndo_set_rx_mode =		netvsc_set_multicast_list,
813 	.ndo_change_mtu =		netvsc_change_mtu,
814 	.ndo_validate_addr =		eth_validate_addr,
815 	.ndo_set_mac_address =		netvsc_set_mac_addr,
816 	.ndo_select_queue =		netvsc_select_queue,
817 #ifdef CONFIG_NET_POLL_CONTROLLER
818 	.ndo_poll_controller =		netvsc_poll_controller,
819 #endif
820 };
821 
822 /*
823  * Send GARP packet to network peers after migrations.
824  * After Quick Migration, the network is not immediately operational in the
825  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
826  * another netif_notify_peers() into a delayed work, otherwise GARP packet
827  * will not be sent after quick migration, and cause network disconnection.
828  * Also, we update the carrier status here.
829  */
830 static void netvsc_link_change(struct work_struct *w)
831 {
832 	struct net_device_context *ndev_ctx;
833 	struct net_device *net;
834 	struct netvsc_device *net_device;
835 	struct rndis_device *rdev;
836 	bool notify, refresh = false;
837 	char *argv[] = { "/etc/init.d/network", "restart", NULL };
838 	char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
839 
840 	rtnl_lock();
841 
842 	ndev_ctx = container_of(w, struct net_device_context, dwork.work);
843 	net_device = hv_get_drvdata(ndev_ctx->device_ctx);
844 	rdev = net_device->extension;
845 	net = net_device->ndev;
846 
847 	if (rdev->link_state) {
848 		netif_carrier_off(net);
849 		notify = false;
850 	} else {
851 		netif_carrier_on(net);
852 		notify = true;
853 		if (rdev->link_change) {
854 			rdev->link_change = false;
855 			refresh = true;
856 		}
857 	}
858 
859 	rtnl_unlock();
860 
861 	if (refresh)
862 		call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
863 
864 	if (notify)
865 		netdev_notify_peers(net);
866 }
867 
868 
869 static int netvsc_probe(struct hv_device *dev,
870 			const struct hv_vmbus_device_id *dev_id)
871 {
872 	struct net_device *net = NULL;
873 	struct net_device_context *net_device_ctx;
874 	struct netvsc_device_info device_info;
875 	struct netvsc_device *nvdev;
876 	int ret;
877 	u32 max_needed_headroom;
878 
879 	net = alloc_etherdev_mq(sizeof(struct net_device_context),
880 				num_online_cpus());
881 	if (!net)
882 		return -ENOMEM;
883 
884 	max_needed_headroom = sizeof(struct hv_netvsc_packet) +
885 			      RNDIS_AND_PPI_SIZE;
886 
887 	netif_carrier_off(net);
888 
889 	net_device_ctx = netdev_priv(net);
890 	net_device_ctx->device_ctx = dev;
891 	hv_set_drvdata(dev, net);
892 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
893 	INIT_WORK(&net_device_ctx->work, do_set_multicast);
894 
895 	net->netdev_ops = &device_ops;
896 
897 	net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
898 				NETIF_F_TSO;
899 	net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
900 			NETIF_F_IP_CSUM | NETIF_F_TSO;
901 
902 	net->ethtool_ops = &ethtool_ops;
903 	SET_NETDEV_DEV(net, &dev->device);
904 
905 	/*
906 	 * Request additional head room in the skb.
907 	 * We will use this space to build the rndis
908 	 * heaser and other state we need to maintain.
909 	 */
910 	net->needed_headroom = max_needed_headroom;
911 
912 	/* Notify the netvsc driver of the new device */
913 	device_info.ring_size = ring_size;
914 	ret = rndis_filter_device_add(dev, &device_info);
915 	if (ret != 0) {
916 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
917 		free_netdev(net);
918 		hv_set_drvdata(dev, NULL);
919 		return ret;
920 	}
921 	memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
922 
923 	nvdev = hv_get_drvdata(dev);
924 	netif_set_real_num_tx_queues(net, nvdev->num_chn);
925 	netif_set_real_num_rx_queues(net, nvdev->num_chn);
926 
927 	ret = register_netdev(net);
928 	if (ret != 0) {
929 		pr_err("Unable to register netdev.\n");
930 		rndis_filter_device_remove(dev);
931 		free_netdev(net);
932 	} else {
933 		schedule_delayed_work(&net_device_ctx->dwork, 0);
934 	}
935 
936 	return ret;
937 }
938 
939 static int netvsc_remove(struct hv_device *dev)
940 {
941 	struct net_device *net;
942 	struct net_device_context *ndev_ctx;
943 	struct netvsc_device *net_device;
944 
945 	net_device = hv_get_drvdata(dev);
946 	net = net_device->ndev;
947 
948 	if (net == NULL) {
949 		dev_err(&dev->device, "No net device to remove\n");
950 		return 0;
951 	}
952 
953 	net_device->start_remove = true;
954 
955 	ndev_ctx = netdev_priv(net);
956 	cancel_delayed_work_sync(&ndev_ctx->dwork);
957 	cancel_work_sync(&ndev_ctx->work);
958 
959 	/* Stop outbound asap */
960 	netif_tx_disable(net);
961 
962 	unregister_netdev(net);
963 
964 	/*
965 	 * Call to the vsc driver to let it know that the device is being
966 	 * removed
967 	 */
968 	rndis_filter_device_remove(dev);
969 
970 	free_netdev(net);
971 	return 0;
972 }
973 
974 static const struct hv_vmbus_device_id id_table[] = {
975 	/* Network guid */
976 	{ HV_NIC_GUID, },
977 	{ },
978 };
979 
980 MODULE_DEVICE_TABLE(vmbus, id_table);
981 
982 /* The one and only one */
983 static struct  hv_driver netvsc_drv = {
984 	.name = KBUILD_MODNAME,
985 	.id_table = id_table,
986 	.probe = netvsc_probe,
987 	.remove = netvsc_remove,
988 };
989 
990 static void __exit netvsc_drv_exit(void)
991 {
992 	vmbus_driver_unregister(&netvsc_drv);
993 }
994 
995 static int __init netvsc_drv_init(void)
996 {
997 	if (ring_size < RING_SIZE_MIN) {
998 		ring_size = RING_SIZE_MIN;
999 		pr_info("Increased ring_size to %d (min allowed)\n",
1000 			ring_size);
1001 	}
1002 	return vmbus_driver_register(&netvsc_drv);
1003 }
1004 
1005 MODULE_LICENSE("GPL");
1006 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1007 
1008 module_init(netvsc_drv_init);
1009 module_exit(netvsc_drv_exit);
1010