xref: /linux/drivers/net/hyperv/netvsc_drv.c (revision 827634added7f38b7d724cab1dccdb2b004c13c3)
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 
44 #define RING_SIZE_MIN 64
45 static int ring_size = 128;
46 module_param(ring_size, int, S_IRUGO);
47 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
48 
49 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
50 				NETIF_MSG_LINK | NETIF_MSG_IFUP |
51 				NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
52 				NETIF_MSG_TX_ERR;
53 
54 static int debug = -1;
55 module_param(debug, int, S_IRUGO);
56 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
57 
58 static void do_set_multicast(struct work_struct *w)
59 {
60 	struct net_device_context *ndevctx =
61 		container_of(w, struct net_device_context, work);
62 	struct netvsc_device *nvdev;
63 	struct rndis_device *rdev;
64 
65 	nvdev = hv_get_drvdata(ndevctx->device_ctx);
66 	if (nvdev == NULL || nvdev->ndev == NULL)
67 		return;
68 
69 	rdev = nvdev->extension;
70 	if (rdev == NULL)
71 		return;
72 
73 	if (nvdev->ndev->flags & IFF_PROMISC)
74 		rndis_filter_set_packet_filter(rdev,
75 			NDIS_PACKET_TYPE_PROMISCUOUS);
76 	else
77 		rndis_filter_set_packet_filter(rdev,
78 			NDIS_PACKET_TYPE_BROADCAST |
79 			NDIS_PACKET_TYPE_ALL_MULTICAST |
80 			NDIS_PACKET_TYPE_DIRECTED);
81 }
82 
83 static void netvsc_set_multicast_list(struct net_device *net)
84 {
85 	struct net_device_context *net_device_ctx = netdev_priv(net);
86 
87 	schedule_work(&net_device_ctx->work);
88 }
89 
90 static int netvsc_open(struct net_device *net)
91 {
92 	struct net_device_context *net_device_ctx = netdev_priv(net);
93 	struct hv_device *device_obj = net_device_ctx->device_ctx;
94 	struct netvsc_device *nvdev;
95 	struct rndis_device *rdev;
96 	int ret = 0;
97 
98 	netif_carrier_off(net);
99 
100 	/* Open up the device */
101 	ret = rndis_filter_open(device_obj);
102 	if (ret != 0) {
103 		netdev_err(net, "unable to open device (ret %d).\n", ret);
104 		return ret;
105 	}
106 
107 	netif_tx_start_all_queues(net);
108 
109 	nvdev = hv_get_drvdata(device_obj);
110 	rdev = nvdev->extension;
111 	if (!rdev->link_state)
112 		netif_carrier_on(net);
113 
114 	return ret;
115 }
116 
117 static int netvsc_close(struct net_device *net)
118 {
119 	struct net_device_context *net_device_ctx = netdev_priv(net);
120 	struct hv_device *device_obj = net_device_ctx->device_ctx;
121 	int ret;
122 
123 	netif_tx_disable(net);
124 
125 	/* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
126 	cancel_work_sync(&net_device_ctx->work);
127 	ret = rndis_filter_close(device_obj);
128 	if (ret != 0)
129 		netdev_err(net, "unable to close device (ret %d).\n", ret);
130 
131 	return ret;
132 }
133 
134 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
135 				int pkt_type)
136 {
137 	struct rndis_packet *rndis_pkt;
138 	struct rndis_per_packet_info *ppi;
139 
140 	rndis_pkt = &msg->msg.pkt;
141 	rndis_pkt->data_offset += ppi_size;
142 
143 	ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
144 		rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
145 
146 	ppi->size = ppi_size;
147 	ppi->type = pkt_type;
148 	ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
149 
150 	rndis_pkt->per_pkt_info_len += ppi_size;
151 
152 	return ppi;
153 }
154 
155 union sub_key {
156 	u64 k;
157 	struct {
158 		u8 pad[3];
159 		u8 kb;
160 		u32 ka;
161 	};
162 };
163 
164 /* Toeplitz hash function
165  * data: network byte order
166  * return: host byte order
167  */
168 static u32 comp_hash(u8 *key, int klen, void *data, int dlen)
169 {
170 	union sub_key subk;
171 	int k_next = 4;
172 	u8 dt;
173 	int i, j;
174 	u32 ret = 0;
175 
176 	subk.k = 0;
177 	subk.ka = ntohl(*(u32 *)key);
178 
179 	for (i = 0; i < dlen; i++) {
180 		subk.kb = key[k_next];
181 		k_next = (k_next + 1) % klen;
182 		dt = ((u8 *)data)[i];
183 		for (j = 0; j < 8; j++) {
184 			if (dt & 0x80)
185 				ret ^= subk.ka;
186 			dt <<= 1;
187 			subk.k <<= 1;
188 		}
189 	}
190 
191 	return ret;
192 }
193 
194 static bool netvsc_set_hash(u32 *hash, struct sk_buff *skb)
195 {
196 	struct flow_keys flow;
197 	int data_len;
198 
199 	if (!skb_flow_dissect(skb, &flow) ||
200 	    !(flow.n_proto == htons(ETH_P_IP) ||
201 	      flow.n_proto == htons(ETH_P_IPV6)))
202 		return false;
203 
204 	if (flow.ip_proto == IPPROTO_TCP)
205 		data_len = 12;
206 	else
207 		data_len = 8;
208 
209 	*hash = comp_hash(netvsc_hash_key, HASH_KEYLEN, &flow, data_len);
210 
211 	return true;
212 }
213 
214 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
215 			void *accel_priv, select_queue_fallback_t fallback)
216 {
217 	struct net_device_context *net_device_ctx = netdev_priv(ndev);
218 	struct hv_device *hdev =  net_device_ctx->device_ctx;
219 	struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
220 	u32 hash;
221 	u16 q_idx = 0;
222 
223 	if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
224 		return 0;
225 
226 	if (netvsc_set_hash(&hash, skb)) {
227 		q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
228 			ndev->real_num_tx_queues;
229 		skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
230 	}
231 
232 	return q_idx;
233 }
234 
235 void netvsc_xmit_completion(void *context)
236 {
237 	struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
238 	struct sk_buff *skb = (struct sk_buff *)
239 		(unsigned long)packet->send_completion_tid;
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 pkt_sz;
393 	struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
394 
395 
396 	/* We will atmost need two pages to describe the rndis
397 	 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
398 	 * of pages in a single packet. If skb is scattered around
399 	 * more pages we try linearizing it.
400 	 */
401 
402 check_size:
403 	skb_length = skb->len;
404 	num_data_pgs = netvsc_get_slots(skb) + 2;
405 	if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
406 		net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
407 				      num_data_pgs, skb->len);
408 		ret = -EFAULT;
409 		goto drop;
410 	} else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
411 		if (skb_linearize(skb)) {
412 			net_alert_ratelimited("failed to linearize skb\n");
413 			ret = -ENOMEM;
414 			goto drop;
415 		}
416 		linear = true;
417 		goto check_size;
418 	}
419 
420 	pkt_sz = sizeof(struct hv_netvsc_packet) + RNDIS_AND_PPI_SIZE;
421 
422 	ret = skb_cow_head(skb, pkt_sz);
423 	if (ret) {
424 		netdev_err(net, "unable to alloc hv_netvsc_packet\n");
425 		ret = -ENOMEM;
426 		goto drop;
427 	}
428 	/* Use the headroom for building up the packet */
429 	packet = (struct hv_netvsc_packet *)skb->head;
430 
431 	packet->status = 0;
432 	packet->xmit_more = skb->xmit_more;
433 
434 	packet->vlan_tci = skb->vlan_tci;
435 	packet->page_buf = page_buf;
436 
437 	packet->q_idx = skb_get_queue_mapping(skb);
438 
439 	packet->is_data_pkt = true;
440 	packet->total_data_buflen = skb->len;
441 
442 	packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
443 				sizeof(struct hv_netvsc_packet));
444 
445 	memset(packet->rndis_msg, 0, RNDIS_AND_PPI_SIZE);
446 
447 	/* Set the completion routine */
448 	packet->send_completion = netvsc_xmit_completion;
449 	packet->send_completion_ctx = packet;
450 	packet->send_completion_tid = (unsigned long)skb;
451 
452 	isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
453 
454 	/* Add the rndis header */
455 	rndis_msg = packet->rndis_msg;
456 	rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
457 	rndis_msg->msg_len = packet->total_data_buflen;
458 	rndis_pkt = &rndis_msg->msg.pkt;
459 	rndis_pkt->data_offset = sizeof(struct rndis_packet);
460 	rndis_pkt->data_len = packet->total_data_buflen;
461 	rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
462 
463 	rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
464 
465 	hash = skb_get_hash_raw(skb);
466 	if (hash != 0 && net->real_num_tx_queues > 1) {
467 		rndis_msg_size += NDIS_HASH_PPI_SIZE;
468 		ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
469 				    NBL_HASH_VALUE);
470 		*(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
471 	}
472 
473 	if (isvlan) {
474 		struct ndis_pkt_8021q_info *vlan;
475 
476 		rndis_msg_size += NDIS_VLAN_PPI_SIZE;
477 		ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
478 					IEEE_8021Q_INFO);
479 		vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
480 						ppi->ppi_offset);
481 		vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
482 		vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
483 				VLAN_PRIO_SHIFT;
484 	}
485 
486 	net_trans_info = get_net_transport_info(skb, &hdr_offset);
487 	if (net_trans_info == TRANSPORT_INFO_NOT_IP)
488 		goto do_send;
489 
490 	/*
491 	 * Setup the sendside checksum offload only if this is not a
492 	 * GSO packet.
493 	 */
494 	if (skb_is_gso(skb))
495 		goto do_lso;
496 
497 	if ((skb->ip_summed == CHECKSUM_NONE) ||
498 	    (skb->ip_summed == CHECKSUM_UNNECESSARY))
499 		goto do_send;
500 
501 	rndis_msg_size += NDIS_CSUM_PPI_SIZE;
502 	ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
503 			    TCPIP_CHKSUM_PKTINFO);
504 
505 	csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
506 			ppi->ppi_offset);
507 
508 	if (net_trans_info & (INFO_IPV4 << 16))
509 		csum_info->transmit.is_ipv4 = 1;
510 	else
511 		csum_info->transmit.is_ipv6 = 1;
512 
513 	if (net_trans_info & INFO_TCP) {
514 		csum_info->transmit.tcp_checksum = 1;
515 		csum_info->transmit.tcp_header_offset = hdr_offset;
516 	} else if (net_trans_info & INFO_UDP) {
517 		/* UDP checksum offload is not supported on ws2008r2.
518 		 * Furthermore, on ws2012 and ws2012r2, there are some
519 		 * issues with udp checksum offload from Linux guests.
520 		 * (these are host issues).
521 		 * For now compute the checksum here.
522 		 */
523 		struct udphdr *uh;
524 		u16 udp_len;
525 
526 		ret = skb_cow_head(skb, 0);
527 		if (ret)
528 			goto drop;
529 
530 		uh = udp_hdr(skb);
531 		udp_len = ntohs(uh->len);
532 		uh->check = 0;
533 		uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
534 					      ip_hdr(skb)->daddr,
535 					      udp_len, IPPROTO_UDP,
536 					      csum_partial(uh, udp_len, 0));
537 		if (uh->check == 0)
538 			uh->check = CSUM_MANGLED_0;
539 
540 		csum_info->transmit.udp_checksum = 0;
541 	}
542 	goto do_send;
543 
544 do_lso:
545 	rndis_msg_size += NDIS_LSO_PPI_SIZE;
546 	ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
547 			    TCP_LARGESEND_PKTINFO);
548 
549 	lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
550 			ppi->ppi_offset);
551 
552 	lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
553 	if (net_trans_info & (INFO_IPV4 << 16)) {
554 		lso_info->lso_v2_transmit.ip_version =
555 			NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
556 		ip_hdr(skb)->tot_len = 0;
557 		ip_hdr(skb)->check = 0;
558 		tcp_hdr(skb)->check =
559 		~csum_tcpudp_magic(ip_hdr(skb)->saddr,
560 				   ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
561 	} else {
562 		lso_info->lso_v2_transmit.ip_version =
563 			NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
564 		ipv6_hdr(skb)->payload_len = 0;
565 		tcp_hdr(skb)->check =
566 		~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
567 				&ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
568 	}
569 	lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
570 	lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
571 
572 do_send:
573 	/* Start filling in the page buffers with the rndis hdr */
574 	rndis_msg->msg_len += rndis_msg_size;
575 	packet->total_data_buflen = rndis_msg->msg_len;
576 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
577 					       skb, packet);
578 
579 	ret = netvsc_send(net_device_ctx->device_ctx, packet);
580 
581 drop:
582 	if (ret == 0) {
583 		net->stats.tx_bytes += skb_length;
584 		net->stats.tx_packets++;
585 	} else {
586 		if (ret != -EAGAIN) {
587 			dev_kfree_skb_any(skb);
588 			net->stats.tx_dropped++;
589 		}
590 	}
591 
592 	return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
593 }
594 
595 /*
596  * netvsc_linkstatus_callback - Link up/down notification
597  */
598 void netvsc_linkstatus_callback(struct hv_device *device_obj,
599 				struct rndis_message *resp)
600 {
601 	struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
602 	struct net_device *net;
603 	struct net_device_context *ndev_ctx;
604 	struct netvsc_device *net_device;
605 	struct rndis_device *rdev;
606 
607 	net_device = hv_get_drvdata(device_obj);
608 	rdev = net_device->extension;
609 
610 	switch (indicate->status) {
611 	case RNDIS_STATUS_MEDIA_CONNECT:
612 		rdev->link_state = false;
613 		break;
614 	case RNDIS_STATUS_MEDIA_DISCONNECT:
615 		rdev->link_state = true;
616 		break;
617 	case RNDIS_STATUS_NETWORK_CHANGE:
618 		rdev->link_change = true;
619 		break;
620 	default:
621 		return;
622 	}
623 
624 	net = net_device->ndev;
625 
626 	if (!net || net->reg_state != NETREG_REGISTERED)
627 		return;
628 
629 	ndev_ctx = netdev_priv(net);
630 	if (!rdev->link_state) {
631 		schedule_delayed_work(&ndev_ctx->dwork, 0);
632 		schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
633 	} else {
634 		schedule_delayed_work(&ndev_ctx->dwork, 0);
635 	}
636 }
637 
638 /*
639  * netvsc_recv_callback -  Callback when we receive a packet from the
640  * "wire" on the specified device.
641  */
642 int netvsc_recv_callback(struct hv_device *device_obj,
643 				struct hv_netvsc_packet *packet,
644 				struct ndis_tcp_ip_checksum_info *csum_info)
645 {
646 	struct net_device *net;
647 	struct sk_buff *skb;
648 
649 	net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
650 	if (!net || net->reg_state != NETREG_REGISTERED) {
651 		packet->status = NVSP_STAT_FAIL;
652 		return 0;
653 	}
654 
655 	/* Allocate a skb - TODO direct I/O to pages? */
656 	skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
657 	if (unlikely(!skb)) {
658 		++net->stats.rx_dropped;
659 		packet->status = NVSP_STAT_FAIL;
660 		return 0;
661 	}
662 
663 	/*
664 	 * Copy to skb. This copy is needed here since the memory pointed by
665 	 * hv_netvsc_packet cannot be deallocated
666 	 */
667 	memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
668 		packet->total_data_buflen);
669 
670 	skb->protocol = eth_type_trans(skb, net);
671 	if (csum_info) {
672 		/* We only look at the IP checksum here.
673 		 * Should we be dropping the packet if checksum
674 		 * failed? How do we deal with other checksums - TCP/UDP?
675 		 */
676 		if (csum_info->receive.ip_checksum_succeeded)
677 			skb->ip_summed = CHECKSUM_UNNECESSARY;
678 		else
679 			skb->ip_summed = CHECKSUM_NONE;
680 	}
681 
682 	if (packet->vlan_tci & VLAN_TAG_PRESENT)
683 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
684 				       packet->vlan_tci);
685 
686 	skb_record_rx_queue(skb, packet->channel->
687 			    offermsg.offer.sub_channel_index);
688 
689 	net->stats.rx_packets++;
690 	net->stats.rx_bytes += packet->total_data_buflen;
691 
692 	/*
693 	 * Pass the skb back up. Network stack will deallocate the skb when it
694 	 * is done.
695 	 * TODO - use NAPI?
696 	 */
697 	netif_rx(skb);
698 
699 	return 0;
700 }
701 
702 static void netvsc_get_drvinfo(struct net_device *net,
703 			       struct ethtool_drvinfo *info)
704 {
705 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
706 	strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
707 }
708 
709 static void netvsc_get_channels(struct net_device *net,
710 				struct ethtool_channels *channel)
711 {
712 	struct net_device_context *net_device_ctx = netdev_priv(net);
713 	struct hv_device *dev = net_device_ctx->device_ctx;
714 	struct netvsc_device *nvdev = hv_get_drvdata(dev);
715 
716 	if (nvdev) {
717 		channel->max_combined	= nvdev->max_chn;
718 		channel->combined_count = nvdev->num_chn;
719 	}
720 }
721 
722 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
723 {
724 	struct net_device_context *ndevctx = netdev_priv(ndev);
725 	struct hv_device *hdev =  ndevctx->device_ctx;
726 	struct netvsc_device *nvdev = hv_get_drvdata(hdev);
727 	struct netvsc_device_info device_info;
728 	int limit = ETH_DATA_LEN;
729 
730 	if (nvdev == NULL || nvdev->destroy)
731 		return -ENODEV;
732 
733 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
734 		limit = NETVSC_MTU - ETH_HLEN;
735 
736 	/* Hyper-V hosts don't support MTU < ETH_DATA_LEN (1500) */
737 	if (mtu < ETH_DATA_LEN || mtu > limit)
738 		return -EINVAL;
739 
740 	nvdev->start_remove = true;
741 	cancel_work_sync(&ndevctx->work);
742 	netif_tx_disable(ndev);
743 	rndis_filter_device_remove(hdev);
744 
745 	ndev->mtu = mtu;
746 
747 	ndevctx->device_ctx = hdev;
748 	hv_set_drvdata(hdev, ndev);
749 	device_info.ring_size = ring_size;
750 	rndis_filter_device_add(hdev, &device_info);
751 	netif_tx_wake_all_queues(ndev);
752 
753 	return 0;
754 }
755 
756 
757 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
758 {
759 	struct net_device_context *ndevctx = netdev_priv(ndev);
760 	struct hv_device *hdev =  ndevctx->device_ctx;
761 	struct sockaddr *addr = p;
762 	char save_adr[ETH_ALEN];
763 	unsigned char save_aatype;
764 	int err;
765 
766 	memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
767 	save_aatype = ndev->addr_assign_type;
768 
769 	err = eth_mac_addr(ndev, p);
770 	if (err != 0)
771 		return err;
772 
773 	err = rndis_filter_set_device_mac(hdev, addr->sa_data);
774 	if (err != 0) {
775 		/* roll back to saved MAC */
776 		memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
777 		ndev->addr_assign_type = save_aatype;
778 	}
779 
780 	return err;
781 }
782 
783 #ifdef CONFIG_NET_POLL_CONTROLLER
784 static void netvsc_poll_controller(struct net_device *net)
785 {
786 	/* As netvsc_start_xmit() works synchronous we don't have to
787 	 * trigger anything here.
788 	 */
789 }
790 #endif
791 
792 static const struct ethtool_ops ethtool_ops = {
793 	.get_drvinfo	= netvsc_get_drvinfo,
794 	.get_link	= ethtool_op_get_link,
795 	.get_channels   = netvsc_get_channels,
796 };
797 
798 static const struct net_device_ops device_ops = {
799 	.ndo_open =			netvsc_open,
800 	.ndo_stop =			netvsc_close,
801 	.ndo_start_xmit =		netvsc_start_xmit,
802 	.ndo_set_rx_mode =		netvsc_set_multicast_list,
803 	.ndo_change_mtu =		netvsc_change_mtu,
804 	.ndo_validate_addr =		eth_validate_addr,
805 	.ndo_set_mac_address =		netvsc_set_mac_addr,
806 	.ndo_select_queue =		netvsc_select_queue,
807 #ifdef CONFIG_NET_POLL_CONTROLLER
808 	.ndo_poll_controller =		netvsc_poll_controller,
809 #endif
810 };
811 
812 /*
813  * Send GARP packet to network peers after migrations.
814  * After Quick Migration, the network is not immediately operational in the
815  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
816  * another netif_notify_peers() into a delayed work, otherwise GARP packet
817  * will not be sent after quick migration, and cause network disconnection.
818  * Also, we update the carrier status here.
819  */
820 static void netvsc_link_change(struct work_struct *w)
821 {
822 	struct net_device_context *ndev_ctx;
823 	struct net_device *net;
824 	struct netvsc_device *net_device;
825 	struct rndis_device *rdev;
826 	bool notify, refresh = false;
827 	char *argv[] = { "/etc/init.d/network", "restart", NULL };
828 	char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
829 
830 	rtnl_lock();
831 
832 	ndev_ctx = container_of(w, struct net_device_context, dwork.work);
833 	net_device = hv_get_drvdata(ndev_ctx->device_ctx);
834 	rdev = net_device->extension;
835 	net = net_device->ndev;
836 
837 	if (rdev->link_state) {
838 		netif_carrier_off(net);
839 		notify = false;
840 	} else {
841 		netif_carrier_on(net);
842 		notify = true;
843 		if (rdev->link_change) {
844 			rdev->link_change = false;
845 			refresh = true;
846 		}
847 	}
848 
849 	rtnl_unlock();
850 
851 	if (refresh)
852 		call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
853 
854 	if (notify)
855 		netdev_notify_peers(net);
856 }
857 
858 
859 static int netvsc_probe(struct hv_device *dev,
860 			const struct hv_vmbus_device_id *dev_id)
861 {
862 	struct net_device *net = NULL;
863 	struct net_device_context *net_device_ctx;
864 	struct netvsc_device_info device_info;
865 	struct netvsc_device *nvdev;
866 	int ret;
867 	u32 max_needed_headroom;
868 
869 	net = alloc_etherdev_mq(sizeof(struct net_device_context),
870 				num_online_cpus());
871 	if (!net)
872 		return -ENOMEM;
873 
874 	max_needed_headroom = sizeof(struct hv_netvsc_packet) +
875 			      RNDIS_AND_PPI_SIZE;
876 
877 	netif_carrier_off(net);
878 
879 	net_device_ctx = netdev_priv(net);
880 	net_device_ctx->device_ctx = dev;
881 	net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
882 	if (netif_msg_probe(net_device_ctx))
883 		netdev_dbg(net, "netvsc msg_enable: %d\n",
884 			   net_device_ctx->msg_enable);
885 
886 	hv_set_drvdata(dev, net);
887 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
888 	INIT_WORK(&net_device_ctx->work, do_set_multicast);
889 
890 	net->netdev_ops = &device_ops;
891 
892 	net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
893 				NETIF_F_TSO;
894 	net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
895 			NETIF_F_IP_CSUM | NETIF_F_TSO;
896 
897 	net->ethtool_ops = &ethtool_ops;
898 	SET_NETDEV_DEV(net, &dev->device);
899 
900 	/*
901 	 * Request additional head room in the skb.
902 	 * We will use this space to build the rndis
903 	 * heaser and other state we need to maintain.
904 	 */
905 	net->needed_headroom = max_needed_headroom;
906 
907 	/* Notify the netvsc driver of the new device */
908 	device_info.ring_size = ring_size;
909 	ret = rndis_filter_device_add(dev, &device_info);
910 	if (ret != 0) {
911 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
912 		free_netdev(net);
913 		hv_set_drvdata(dev, NULL);
914 		return ret;
915 	}
916 	memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
917 
918 	nvdev = hv_get_drvdata(dev);
919 	netif_set_real_num_tx_queues(net, nvdev->num_chn);
920 	netif_set_real_num_rx_queues(net, nvdev->num_chn);
921 
922 	ret = register_netdev(net);
923 	if (ret != 0) {
924 		pr_err("Unable to register netdev.\n");
925 		rndis_filter_device_remove(dev);
926 		free_netdev(net);
927 	} else {
928 		schedule_delayed_work(&net_device_ctx->dwork, 0);
929 	}
930 
931 	return ret;
932 }
933 
934 static int netvsc_remove(struct hv_device *dev)
935 {
936 	struct net_device *net;
937 	struct net_device_context *ndev_ctx;
938 	struct netvsc_device *net_device;
939 
940 	net_device = hv_get_drvdata(dev);
941 	net = net_device->ndev;
942 
943 	if (net == NULL) {
944 		dev_err(&dev->device, "No net device to remove\n");
945 		return 0;
946 	}
947 
948 	net_device->start_remove = true;
949 
950 	ndev_ctx = netdev_priv(net);
951 	cancel_delayed_work_sync(&ndev_ctx->dwork);
952 	cancel_work_sync(&ndev_ctx->work);
953 
954 	/* Stop outbound asap */
955 	netif_tx_disable(net);
956 
957 	unregister_netdev(net);
958 
959 	/*
960 	 * Call to the vsc driver to let it know that the device is being
961 	 * removed
962 	 */
963 	rndis_filter_device_remove(dev);
964 
965 	free_netdev(net);
966 	return 0;
967 }
968 
969 static const struct hv_vmbus_device_id id_table[] = {
970 	/* Network guid */
971 	{ HV_NIC_GUID, },
972 	{ },
973 };
974 
975 MODULE_DEVICE_TABLE(vmbus, id_table);
976 
977 /* The one and only one */
978 static struct  hv_driver netvsc_drv = {
979 	.name = KBUILD_MODNAME,
980 	.id_table = id_table,
981 	.probe = netvsc_probe,
982 	.remove = netvsc_remove,
983 };
984 
985 static void __exit netvsc_drv_exit(void)
986 {
987 	vmbus_driver_unregister(&netvsc_drv);
988 }
989 
990 static int __init netvsc_drv_init(void)
991 {
992 	if (ring_size < RING_SIZE_MIN) {
993 		ring_size = RING_SIZE_MIN;
994 		pr_info("Increased ring_size to %d (min allowed)\n",
995 			ring_size);
996 	}
997 	return vmbus_driver_register(&netvsc_drv);
998 }
999 
1000 MODULE_LICENSE("GPL");
1001 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1002 
1003 module_init(netvsc_drv_init);
1004 module_exit(netvsc_drv_exit);
1005