xref: /openbmc/linux/drivers/net/hyperv/netvsc_drv.c (revision eb3fcf007fffe5830d815e713591f3e858f2a365)
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 int max_num_vrss_chns = 8;
50 
51 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE |
52 				NETIF_MSG_LINK | NETIF_MSG_IFUP |
53 				NETIF_MSG_IFDOWN | NETIF_MSG_RX_ERR |
54 				NETIF_MSG_TX_ERR;
55 
56 static int debug = -1;
57 module_param(debug, int, S_IRUGO);
58 MODULE_PARM_DESC(debug, "Debug level (0=none,...,16=all)");
59 
60 static void do_set_multicast(struct work_struct *w)
61 {
62 	struct net_device_context *ndevctx =
63 		container_of(w, struct net_device_context, work);
64 	struct netvsc_device *nvdev;
65 	struct rndis_device *rdev;
66 
67 	nvdev = hv_get_drvdata(ndevctx->device_ctx);
68 	if (nvdev == NULL || nvdev->ndev == NULL)
69 		return;
70 
71 	rdev = nvdev->extension;
72 	if (rdev == NULL)
73 		return;
74 
75 	if (nvdev->ndev->flags & IFF_PROMISC)
76 		rndis_filter_set_packet_filter(rdev,
77 			NDIS_PACKET_TYPE_PROMISCUOUS);
78 	else
79 		rndis_filter_set_packet_filter(rdev,
80 			NDIS_PACKET_TYPE_BROADCAST |
81 			NDIS_PACKET_TYPE_ALL_MULTICAST |
82 			NDIS_PACKET_TYPE_DIRECTED);
83 }
84 
85 static void netvsc_set_multicast_list(struct net_device *net)
86 {
87 	struct net_device_context *net_device_ctx = netdev_priv(net);
88 
89 	schedule_work(&net_device_ctx->work);
90 }
91 
92 static int netvsc_open(struct net_device *net)
93 {
94 	struct net_device_context *net_device_ctx = netdev_priv(net);
95 	struct hv_device *device_obj = net_device_ctx->device_ctx;
96 	struct netvsc_device *nvdev;
97 	struct rndis_device *rdev;
98 	int ret = 0;
99 
100 	netif_carrier_off(net);
101 
102 	/* Open up the device */
103 	ret = rndis_filter_open(device_obj);
104 	if (ret != 0) {
105 		netdev_err(net, "unable to open device (ret %d).\n", ret);
106 		return ret;
107 	}
108 
109 	netif_tx_wake_all_queues(net);
110 
111 	nvdev = hv_get_drvdata(device_obj);
112 	rdev = nvdev->extension;
113 	if (!rdev->link_state)
114 		netif_carrier_on(net);
115 
116 	return ret;
117 }
118 
119 static int netvsc_close(struct net_device *net)
120 {
121 	struct net_device_context *net_device_ctx = netdev_priv(net);
122 	struct hv_device *device_obj = net_device_ctx->device_ctx;
123 	struct netvsc_device *nvdev = hv_get_drvdata(device_obj);
124 	int ret;
125 	u32 aread, awrite, i, msec = 10, retry = 0, retry_max = 20;
126 	struct vmbus_channel *chn;
127 
128 	netif_tx_disable(net);
129 
130 	/* Make sure netvsc_set_multicast_list doesn't re-enable filter! */
131 	cancel_work_sync(&net_device_ctx->work);
132 	ret = rndis_filter_close(device_obj);
133 	if (ret != 0) {
134 		netdev_err(net, "unable to close device (ret %d).\n", ret);
135 		return ret;
136 	}
137 
138 	/* Ensure pending bytes in ring are read */
139 	while (true) {
140 		aread = 0;
141 		for (i = 0; i < nvdev->num_chn; i++) {
142 			chn = nvdev->chn_table[i];
143 			if (!chn)
144 				continue;
145 
146 			hv_get_ringbuffer_availbytes(&chn->inbound, &aread,
147 						     &awrite);
148 
149 			if (aread)
150 				break;
151 
152 			hv_get_ringbuffer_availbytes(&chn->outbound, &aread,
153 						     &awrite);
154 
155 			if (aread)
156 				break;
157 		}
158 
159 		retry++;
160 		if (retry > retry_max || aread == 0)
161 			break;
162 
163 		msleep(msec);
164 
165 		if (msec < 1000)
166 			msec *= 2;
167 	}
168 
169 	if (aread) {
170 		netdev_err(net, "Ring buffer not empty after closing rndis\n");
171 		ret = -ETIMEDOUT;
172 	}
173 
174 	return ret;
175 }
176 
177 static void *init_ppi_data(struct rndis_message *msg, u32 ppi_size,
178 				int pkt_type)
179 {
180 	struct rndis_packet *rndis_pkt;
181 	struct rndis_per_packet_info *ppi;
182 
183 	rndis_pkt = &msg->msg.pkt;
184 	rndis_pkt->data_offset += ppi_size;
185 
186 	ppi = (struct rndis_per_packet_info *)((void *)rndis_pkt +
187 		rndis_pkt->per_pkt_info_offset + rndis_pkt->per_pkt_info_len);
188 
189 	ppi->size = ppi_size;
190 	ppi->type = pkt_type;
191 	ppi->ppi_offset = sizeof(struct rndis_per_packet_info);
192 
193 	rndis_pkt->per_pkt_info_len += ppi_size;
194 
195 	return ppi;
196 }
197 
198 union sub_key {
199 	u64 k;
200 	struct {
201 		u8 pad[3];
202 		u8 kb;
203 		u32 ka;
204 	};
205 };
206 
207 /* Toeplitz hash function
208  * data: network byte order
209  * return: host byte order
210  */
211 static u32 comp_hash(u8 *key, int klen, void *data, int dlen)
212 {
213 	union sub_key subk;
214 	int k_next = 4;
215 	u8 dt;
216 	int i, j;
217 	u32 ret = 0;
218 
219 	subk.k = 0;
220 	subk.ka = ntohl(*(u32 *)key);
221 
222 	for (i = 0; i < dlen; i++) {
223 		subk.kb = key[k_next];
224 		k_next = (k_next + 1) % klen;
225 		dt = ((u8 *)data)[i];
226 		for (j = 0; j < 8; j++) {
227 			if (dt & 0x80)
228 				ret ^= subk.ka;
229 			dt <<= 1;
230 			subk.k <<= 1;
231 		}
232 	}
233 
234 	return ret;
235 }
236 
237 static bool netvsc_set_hash(u32 *hash, struct sk_buff *skb)
238 {
239 	struct flow_keys flow;
240 	int data_len;
241 
242 	if (!skb_flow_dissect_flow_keys(skb, &flow, 0) ||
243 	    !(flow.basic.n_proto == htons(ETH_P_IP) ||
244 	      flow.basic.n_proto == htons(ETH_P_IPV6)))
245 		return false;
246 
247 	if (flow.basic.ip_proto == IPPROTO_TCP)
248 		data_len = 12;
249 	else
250 		data_len = 8;
251 
252 	*hash = comp_hash(netvsc_hash_key, HASH_KEYLEN, &flow, data_len);
253 
254 	return true;
255 }
256 
257 static u16 netvsc_select_queue(struct net_device *ndev, struct sk_buff *skb,
258 			void *accel_priv, select_queue_fallback_t fallback)
259 {
260 	struct net_device_context *net_device_ctx = netdev_priv(ndev);
261 	struct hv_device *hdev =  net_device_ctx->device_ctx;
262 	struct netvsc_device *nvsc_dev = hv_get_drvdata(hdev);
263 	u32 hash;
264 	u16 q_idx = 0;
265 
266 	if (nvsc_dev == NULL || ndev->real_num_tx_queues <= 1)
267 		return 0;
268 
269 	if (netvsc_set_hash(&hash, skb)) {
270 		q_idx = nvsc_dev->send_table[hash % VRSS_SEND_TAB_SIZE] %
271 			ndev->real_num_tx_queues;
272 		skb_set_hash(skb, hash, PKT_HASH_TYPE_L3);
273 	}
274 
275 	return q_idx;
276 }
277 
278 void netvsc_xmit_completion(void *context)
279 {
280 	struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
281 	struct sk_buff *skb = (struct sk_buff *)
282 		(unsigned long)packet->send_completion_tid;
283 
284 	if (skb)
285 		dev_kfree_skb_any(skb);
286 }
287 
288 static u32 fill_pg_buf(struct page *page, u32 offset, u32 len,
289 			struct hv_page_buffer *pb)
290 {
291 	int j = 0;
292 
293 	/* Deal with compund pages by ignoring unused part
294 	 * of the page.
295 	 */
296 	page += (offset >> PAGE_SHIFT);
297 	offset &= ~PAGE_MASK;
298 
299 	while (len > 0) {
300 		unsigned long bytes;
301 
302 		bytes = PAGE_SIZE - offset;
303 		if (bytes > len)
304 			bytes = len;
305 		pb[j].pfn = page_to_pfn(page);
306 		pb[j].offset = offset;
307 		pb[j].len = bytes;
308 
309 		offset += bytes;
310 		len -= bytes;
311 
312 		if (offset == PAGE_SIZE && len) {
313 			page++;
314 			offset = 0;
315 			j++;
316 		}
317 	}
318 
319 	return j + 1;
320 }
321 
322 static u32 init_page_array(void *hdr, u32 len, struct sk_buff *skb,
323 			   struct hv_netvsc_packet *packet)
324 {
325 	struct hv_page_buffer *pb = packet->page_buf;
326 	u32 slots_used = 0;
327 	char *data = skb->data;
328 	int frags = skb_shinfo(skb)->nr_frags;
329 	int i;
330 
331 	/* The packet is laid out thus:
332 	 * 1. hdr: RNDIS header and PPI
333 	 * 2. skb linear data
334 	 * 3. skb fragment data
335 	 */
336 	if (hdr != NULL)
337 		slots_used += fill_pg_buf(virt_to_page(hdr),
338 					offset_in_page(hdr),
339 					len, &pb[slots_used]);
340 
341 	packet->rmsg_size = len;
342 	packet->rmsg_pgcnt = slots_used;
343 
344 	slots_used += fill_pg_buf(virt_to_page(data),
345 				offset_in_page(data),
346 				skb_headlen(skb), &pb[slots_used]);
347 
348 	for (i = 0; i < frags; i++) {
349 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
350 
351 		slots_used += fill_pg_buf(skb_frag_page(frag),
352 					frag->page_offset,
353 					skb_frag_size(frag), &pb[slots_used]);
354 	}
355 	return slots_used;
356 }
357 
358 static int count_skb_frag_slots(struct sk_buff *skb)
359 {
360 	int i, frags = skb_shinfo(skb)->nr_frags;
361 	int pages = 0;
362 
363 	for (i = 0; i < frags; i++) {
364 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
365 		unsigned long size = skb_frag_size(frag);
366 		unsigned long offset = frag->page_offset;
367 
368 		/* Skip unused frames from start of page */
369 		offset &= ~PAGE_MASK;
370 		pages += PFN_UP(offset + size);
371 	}
372 	return pages;
373 }
374 
375 static int netvsc_get_slots(struct sk_buff *skb)
376 {
377 	char *data = skb->data;
378 	unsigned int offset = offset_in_page(data);
379 	unsigned int len = skb_headlen(skb);
380 	int slots;
381 	int frag_slots;
382 
383 	slots = DIV_ROUND_UP(offset + len, PAGE_SIZE);
384 	frag_slots = count_skb_frag_slots(skb);
385 	return slots + frag_slots;
386 }
387 
388 static u32 get_net_transport_info(struct sk_buff *skb, u32 *trans_off)
389 {
390 	u32 ret_val = TRANSPORT_INFO_NOT_IP;
391 
392 	if ((eth_hdr(skb)->h_proto != htons(ETH_P_IP)) &&
393 		(eth_hdr(skb)->h_proto != htons(ETH_P_IPV6))) {
394 		goto not_ip;
395 	}
396 
397 	*trans_off = skb_transport_offset(skb);
398 
399 	if ((eth_hdr(skb)->h_proto == htons(ETH_P_IP))) {
400 		struct iphdr *iphdr = ip_hdr(skb);
401 
402 		if (iphdr->protocol == IPPROTO_TCP)
403 			ret_val = TRANSPORT_INFO_IPV4_TCP;
404 		else if (iphdr->protocol == IPPROTO_UDP)
405 			ret_val = TRANSPORT_INFO_IPV4_UDP;
406 	} else {
407 		if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
408 			ret_val = TRANSPORT_INFO_IPV6_TCP;
409 		else if (ipv6_hdr(skb)->nexthdr == IPPROTO_UDP)
410 			ret_val = TRANSPORT_INFO_IPV6_UDP;
411 	}
412 
413 not_ip:
414 	return ret_val;
415 }
416 
417 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
418 {
419 	struct net_device_context *net_device_ctx = netdev_priv(net);
420 	struct hv_netvsc_packet *packet = NULL;
421 	int ret;
422 	unsigned int num_data_pgs;
423 	struct rndis_message *rndis_msg;
424 	struct rndis_packet *rndis_pkt;
425 	u32 rndis_msg_size;
426 	bool isvlan;
427 	bool linear = false;
428 	struct rndis_per_packet_info *ppi;
429 	struct ndis_tcp_ip_checksum_info *csum_info;
430 	struct ndis_tcp_lso_info *lso_info;
431 	int  hdr_offset;
432 	u32 net_trans_info;
433 	u32 hash;
434 	u32 skb_length;
435 	u32 pkt_sz;
436 	struct hv_page_buffer page_buf[MAX_PAGE_BUFFER_COUNT];
437 	struct netvsc_stats *tx_stats = this_cpu_ptr(net_device_ctx->tx_stats);
438 
439 	/* We will atmost need two pages to describe the rndis
440 	 * header. We can only transmit MAX_PAGE_BUFFER_COUNT number
441 	 * of pages in a single packet. If skb is scattered around
442 	 * more pages we try linearizing it.
443 	 */
444 
445 check_size:
446 	skb_length = skb->len;
447 	num_data_pgs = netvsc_get_slots(skb) + 2;
448 	if (num_data_pgs > MAX_PAGE_BUFFER_COUNT && linear) {
449 		net_alert_ratelimited("packet too big: %u pages (%u bytes)\n",
450 				      num_data_pgs, skb->len);
451 		ret = -EFAULT;
452 		goto drop;
453 	} else if (num_data_pgs > MAX_PAGE_BUFFER_COUNT) {
454 		if (skb_linearize(skb)) {
455 			net_alert_ratelimited("failed to linearize skb\n");
456 			ret = -ENOMEM;
457 			goto drop;
458 		}
459 		linear = true;
460 		goto check_size;
461 	}
462 
463 	pkt_sz = sizeof(struct hv_netvsc_packet) + RNDIS_AND_PPI_SIZE;
464 
465 	ret = skb_cow_head(skb, pkt_sz);
466 	if (ret) {
467 		netdev_err(net, "unable to alloc hv_netvsc_packet\n");
468 		ret = -ENOMEM;
469 		goto drop;
470 	}
471 	/* Use the headroom for building up the packet */
472 	packet = (struct hv_netvsc_packet *)skb->head;
473 
474 	packet->status = 0;
475 	packet->xmit_more = skb->xmit_more;
476 
477 	packet->vlan_tci = skb->vlan_tci;
478 	packet->page_buf = page_buf;
479 
480 	packet->q_idx = skb_get_queue_mapping(skb);
481 
482 	packet->is_data_pkt = true;
483 	packet->total_data_buflen = skb->len;
484 
485 	packet->rndis_msg = (struct rndis_message *)((unsigned long)packet +
486 				sizeof(struct hv_netvsc_packet));
487 
488 	memset(packet->rndis_msg, 0, RNDIS_AND_PPI_SIZE);
489 
490 	/* Set the completion routine */
491 	packet->send_completion = netvsc_xmit_completion;
492 	packet->send_completion_ctx = packet;
493 	packet->send_completion_tid = (unsigned long)skb;
494 
495 	isvlan = packet->vlan_tci & VLAN_TAG_PRESENT;
496 
497 	/* Add the rndis header */
498 	rndis_msg = packet->rndis_msg;
499 	rndis_msg->ndis_msg_type = RNDIS_MSG_PACKET;
500 	rndis_msg->msg_len = packet->total_data_buflen;
501 	rndis_pkt = &rndis_msg->msg.pkt;
502 	rndis_pkt->data_offset = sizeof(struct rndis_packet);
503 	rndis_pkt->data_len = packet->total_data_buflen;
504 	rndis_pkt->per_pkt_info_offset = sizeof(struct rndis_packet);
505 
506 	rndis_msg_size = RNDIS_MESSAGE_SIZE(struct rndis_packet);
507 
508 	hash = skb_get_hash_raw(skb);
509 	if (hash != 0 && net->real_num_tx_queues > 1) {
510 		rndis_msg_size += NDIS_HASH_PPI_SIZE;
511 		ppi = init_ppi_data(rndis_msg, NDIS_HASH_PPI_SIZE,
512 				    NBL_HASH_VALUE);
513 		*(u32 *)((void *)ppi + ppi->ppi_offset) = hash;
514 	}
515 
516 	if (isvlan) {
517 		struct ndis_pkt_8021q_info *vlan;
518 
519 		rndis_msg_size += NDIS_VLAN_PPI_SIZE;
520 		ppi = init_ppi_data(rndis_msg, NDIS_VLAN_PPI_SIZE,
521 					IEEE_8021Q_INFO);
522 		vlan = (struct ndis_pkt_8021q_info *)((void *)ppi +
523 						ppi->ppi_offset);
524 		vlan->vlanid = packet->vlan_tci & VLAN_VID_MASK;
525 		vlan->pri = (packet->vlan_tci & VLAN_PRIO_MASK) >>
526 				VLAN_PRIO_SHIFT;
527 	}
528 
529 	net_trans_info = get_net_transport_info(skb, &hdr_offset);
530 	if (net_trans_info == TRANSPORT_INFO_NOT_IP)
531 		goto do_send;
532 
533 	/*
534 	 * Setup the sendside checksum offload only if this is not a
535 	 * GSO packet.
536 	 */
537 	if (skb_is_gso(skb))
538 		goto do_lso;
539 
540 	if ((skb->ip_summed == CHECKSUM_NONE) ||
541 	    (skb->ip_summed == CHECKSUM_UNNECESSARY))
542 		goto do_send;
543 
544 	rndis_msg_size += NDIS_CSUM_PPI_SIZE;
545 	ppi = init_ppi_data(rndis_msg, NDIS_CSUM_PPI_SIZE,
546 			    TCPIP_CHKSUM_PKTINFO);
547 
548 	csum_info = (struct ndis_tcp_ip_checksum_info *)((void *)ppi +
549 			ppi->ppi_offset);
550 
551 	if (net_trans_info & (INFO_IPV4 << 16))
552 		csum_info->transmit.is_ipv4 = 1;
553 	else
554 		csum_info->transmit.is_ipv6 = 1;
555 
556 	if (net_trans_info & INFO_TCP) {
557 		csum_info->transmit.tcp_checksum = 1;
558 		csum_info->transmit.tcp_header_offset = hdr_offset;
559 	} else if (net_trans_info & INFO_UDP) {
560 		/* UDP checksum offload is not supported on ws2008r2.
561 		 * Furthermore, on ws2012 and ws2012r2, there are some
562 		 * issues with udp checksum offload from Linux guests.
563 		 * (these are host issues).
564 		 * For now compute the checksum here.
565 		 */
566 		struct udphdr *uh;
567 		u16 udp_len;
568 
569 		ret = skb_cow_head(skb, 0);
570 		if (ret)
571 			goto drop;
572 
573 		uh = udp_hdr(skb);
574 		udp_len = ntohs(uh->len);
575 		uh->check = 0;
576 		uh->check = csum_tcpudp_magic(ip_hdr(skb)->saddr,
577 					      ip_hdr(skb)->daddr,
578 					      udp_len, IPPROTO_UDP,
579 					      csum_partial(uh, udp_len, 0));
580 		if (uh->check == 0)
581 			uh->check = CSUM_MANGLED_0;
582 
583 		csum_info->transmit.udp_checksum = 0;
584 	}
585 	goto do_send;
586 
587 do_lso:
588 	rndis_msg_size += NDIS_LSO_PPI_SIZE;
589 	ppi = init_ppi_data(rndis_msg, NDIS_LSO_PPI_SIZE,
590 			    TCP_LARGESEND_PKTINFO);
591 
592 	lso_info = (struct ndis_tcp_lso_info *)((void *)ppi +
593 			ppi->ppi_offset);
594 
595 	lso_info->lso_v2_transmit.type = NDIS_TCP_LARGE_SEND_OFFLOAD_V2_TYPE;
596 	if (net_trans_info & (INFO_IPV4 << 16)) {
597 		lso_info->lso_v2_transmit.ip_version =
598 			NDIS_TCP_LARGE_SEND_OFFLOAD_IPV4;
599 		ip_hdr(skb)->tot_len = 0;
600 		ip_hdr(skb)->check = 0;
601 		tcp_hdr(skb)->check =
602 		~csum_tcpudp_magic(ip_hdr(skb)->saddr,
603 				   ip_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
604 	} else {
605 		lso_info->lso_v2_transmit.ip_version =
606 			NDIS_TCP_LARGE_SEND_OFFLOAD_IPV6;
607 		ipv6_hdr(skb)->payload_len = 0;
608 		tcp_hdr(skb)->check =
609 		~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
610 				&ipv6_hdr(skb)->daddr, 0, IPPROTO_TCP, 0);
611 	}
612 	lso_info->lso_v2_transmit.tcp_header_offset = hdr_offset;
613 	lso_info->lso_v2_transmit.mss = skb_shinfo(skb)->gso_size;
614 
615 do_send:
616 	/* Start filling in the page buffers with the rndis hdr */
617 	rndis_msg->msg_len += rndis_msg_size;
618 	packet->total_data_buflen = rndis_msg->msg_len;
619 	packet->page_buf_cnt = init_page_array(rndis_msg, rndis_msg_size,
620 					       skb, packet);
621 
622 	ret = netvsc_send(net_device_ctx->device_ctx, packet);
623 
624 drop:
625 	if (ret == 0) {
626 		u64_stats_update_begin(&tx_stats->syncp);
627 		tx_stats->packets++;
628 		tx_stats->bytes += skb_length;
629 		u64_stats_update_end(&tx_stats->syncp);
630 	} else {
631 		if (ret != -EAGAIN) {
632 			dev_kfree_skb_any(skb);
633 			net->stats.tx_dropped++;
634 		}
635 	}
636 
637 	return (ret == -EAGAIN) ? NETDEV_TX_BUSY : NETDEV_TX_OK;
638 }
639 
640 /*
641  * netvsc_linkstatus_callback - Link up/down notification
642  */
643 void netvsc_linkstatus_callback(struct hv_device *device_obj,
644 				struct rndis_message *resp)
645 {
646 	struct rndis_indicate_status *indicate = &resp->msg.indicate_status;
647 	struct net_device *net;
648 	struct net_device_context *ndev_ctx;
649 	struct netvsc_device *net_device;
650 	struct rndis_device *rdev;
651 
652 	net_device = hv_get_drvdata(device_obj);
653 	rdev = net_device->extension;
654 
655 	switch (indicate->status) {
656 	case RNDIS_STATUS_MEDIA_CONNECT:
657 		rdev->link_state = false;
658 		break;
659 	case RNDIS_STATUS_MEDIA_DISCONNECT:
660 		rdev->link_state = true;
661 		break;
662 	case RNDIS_STATUS_NETWORK_CHANGE:
663 		rdev->link_change = true;
664 		break;
665 	default:
666 		return;
667 	}
668 
669 	net = net_device->ndev;
670 
671 	if (!net || net->reg_state != NETREG_REGISTERED)
672 		return;
673 
674 	ndev_ctx = netdev_priv(net);
675 	if (!rdev->link_state) {
676 		schedule_delayed_work(&ndev_ctx->dwork, 0);
677 		schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
678 	} else {
679 		schedule_delayed_work(&ndev_ctx->dwork, 0);
680 	}
681 }
682 
683 /*
684  * netvsc_recv_callback -  Callback when we receive a packet from the
685  * "wire" on the specified device.
686  */
687 int netvsc_recv_callback(struct hv_device *device_obj,
688 				struct hv_netvsc_packet *packet,
689 				struct ndis_tcp_ip_checksum_info *csum_info)
690 {
691 	struct net_device *net;
692 	struct net_device_context *net_device_ctx;
693 	struct sk_buff *skb;
694 	struct netvsc_stats *rx_stats;
695 
696 	net = ((struct netvsc_device *)hv_get_drvdata(device_obj))->ndev;
697 	if (!net || net->reg_state != NETREG_REGISTERED) {
698 		packet->status = NVSP_STAT_FAIL;
699 		return 0;
700 	}
701 	net_device_ctx = netdev_priv(net);
702 	rx_stats = this_cpu_ptr(net_device_ctx->rx_stats);
703 
704 	/* Allocate a skb - TODO direct I/O to pages? */
705 	skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
706 	if (unlikely(!skb)) {
707 		++net->stats.rx_dropped;
708 		packet->status = NVSP_STAT_FAIL;
709 		return 0;
710 	}
711 
712 	/*
713 	 * Copy to skb. This copy is needed here since the memory pointed by
714 	 * hv_netvsc_packet cannot be deallocated
715 	 */
716 	memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
717 		packet->total_data_buflen);
718 
719 	skb->protocol = eth_type_trans(skb, net);
720 	if (csum_info) {
721 		/* We only look at the IP checksum here.
722 		 * Should we be dropping the packet if checksum
723 		 * failed? How do we deal with other checksums - TCP/UDP?
724 		 */
725 		if (csum_info->receive.ip_checksum_succeeded)
726 			skb->ip_summed = CHECKSUM_UNNECESSARY;
727 		else
728 			skb->ip_summed = CHECKSUM_NONE;
729 	}
730 
731 	if (packet->vlan_tci & VLAN_TAG_PRESENT)
732 		__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
733 				       packet->vlan_tci);
734 
735 	skb_record_rx_queue(skb, packet->channel->
736 			    offermsg.offer.sub_channel_index);
737 
738 	u64_stats_update_begin(&rx_stats->syncp);
739 	rx_stats->packets++;
740 	rx_stats->bytes += packet->total_data_buflen;
741 	u64_stats_update_end(&rx_stats->syncp);
742 
743 	/*
744 	 * Pass the skb back up. Network stack will deallocate the skb when it
745 	 * is done.
746 	 * TODO - use NAPI?
747 	 */
748 	netif_rx(skb);
749 
750 	return 0;
751 }
752 
753 static void netvsc_get_drvinfo(struct net_device *net,
754 			       struct ethtool_drvinfo *info)
755 {
756 	strlcpy(info->driver, KBUILD_MODNAME, sizeof(info->driver));
757 	strlcpy(info->fw_version, "N/A", sizeof(info->fw_version));
758 }
759 
760 static void netvsc_get_channels(struct net_device *net,
761 				struct ethtool_channels *channel)
762 {
763 	struct net_device_context *net_device_ctx = netdev_priv(net);
764 	struct hv_device *dev = net_device_ctx->device_ctx;
765 	struct netvsc_device *nvdev = hv_get_drvdata(dev);
766 
767 	if (nvdev) {
768 		channel->max_combined	= nvdev->max_chn;
769 		channel->combined_count = nvdev->num_chn;
770 	}
771 }
772 
773 static int netvsc_set_channels(struct net_device *net,
774 			       struct ethtool_channels *channels)
775 {
776 	struct net_device_context *net_device_ctx = netdev_priv(net);
777 	struct hv_device *dev = net_device_ctx->device_ctx;
778 	struct netvsc_device *nvdev = hv_get_drvdata(dev);
779 	struct netvsc_device_info device_info;
780 	u32 num_chn;
781 	u32 max_chn;
782 	int ret = 0;
783 	bool recovering = false;
784 
785 	if (!nvdev || nvdev->destroy)
786 		return -ENODEV;
787 
788 	num_chn = nvdev->num_chn;
789 	max_chn = min_t(u32, nvdev->max_chn, num_online_cpus());
790 
791 	if (nvdev->nvsp_version < NVSP_PROTOCOL_VERSION_5) {
792 		pr_info("vRSS unsupported before NVSP Version 5\n");
793 		return -EINVAL;
794 	}
795 
796 	/* We do not support rx, tx, or other */
797 	if (!channels ||
798 	    channels->rx_count ||
799 	    channels->tx_count ||
800 	    channels->other_count ||
801 	    (channels->combined_count < 1))
802 		return -EINVAL;
803 
804 	if (channels->combined_count > max_chn) {
805 		pr_info("combined channels too high, using %d\n", max_chn);
806 		channels->combined_count = max_chn;
807 	}
808 
809 	ret = netvsc_close(net);
810 	if (ret)
811 		goto out;
812 
813  do_set:
814 	nvdev->start_remove = true;
815 	rndis_filter_device_remove(dev);
816 
817 	nvdev->num_chn = channels->combined_count;
818 
819 	net_device_ctx->device_ctx = dev;
820 	hv_set_drvdata(dev, net);
821 
822 	memset(&device_info, 0, sizeof(device_info));
823 	device_info.num_chn = nvdev->num_chn; /* passed to RNDIS */
824 	device_info.ring_size = ring_size;
825 	device_info.max_num_vrss_chns = max_num_vrss_chns;
826 
827 	ret = rndis_filter_device_add(dev, &device_info);
828 	if (ret) {
829 		if (recovering) {
830 			netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
831 			return ret;
832 		}
833 		goto recover;
834 	}
835 
836 	nvdev = hv_get_drvdata(dev);
837 
838 	ret = netif_set_real_num_tx_queues(net, nvdev->num_chn);
839 	if (ret) {
840 		if (recovering) {
841 			netdev_err(net, "could not set tx queue count (ret %d)\n", ret);
842 			return ret;
843 		}
844 		goto recover;
845 	}
846 
847 	ret = netif_set_real_num_rx_queues(net, nvdev->num_chn);
848 	if (ret) {
849 		if (recovering) {
850 			netdev_err(net, "could not set rx queue count (ret %d)\n", ret);
851 			return ret;
852 		}
853 		goto recover;
854 	}
855 
856  out:
857 	netvsc_open(net);
858 
859 	return ret;
860 
861  recover:
862 	/* If the above failed, we attempt to recover through the same
863 	 * process but with the original number of channels.
864 	 */
865 	netdev_err(net, "could not set channels, recovering\n");
866 	recovering = true;
867 	channels->combined_count = num_chn;
868 	goto do_set;
869 }
870 
871 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
872 {
873 	struct net_device_context *ndevctx = netdev_priv(ndev);
874 	struct hv_device *hdev =  ndevctx->device_ctx;
875 	struct netvsc_device *nvdev = hv_get_drvdata(hdev);
876 	struct netvsc_device_info device_info;
877 	int limit = ETH_DATA_LEN;
878 	int ret = 0;
879 
880 	if (nvdev == NULL || nvdev->destroy)
881 		return -ENODEV;
882 
883 	if (nvdev->nvsp_version >= NVSP_PROTOCOL_VERSION_2)
884 		limit = NETVSC_MTU - ETH_HLEN;
885 
886 	if (mtu < NETVSC_MTU_MIN || mtu > limit)
887 		return -EINVAL;
888 
889 	ret = netvsc_close(ndev);
890 	if (ret)
891 		goto out;
892 
893 	nvdev->start_remove = true;
894 	rndis_filter_device_remove(hdev);
895 
896 	ndev->mtu = mtu;
897 
898 	ndevctx->device_ctx = hdev;
899 	hv_set_drvdata(hdev, ndev);
900 
901 	memset(&device_info, 0, sizeof(device_info));
902 	device_info.ring_size = ring_size;
903 	device_info.num_chn = nvdev->num_chn;
904 	device_info.max_num_vrss_chns = max_num_vrss_chns;
905 	rndis_filter_device_add(hdev, &device_info);
906 
907 out:
908 	netvsc_open(ndev);
909 
910 	return ret;
911 }
912 
913 static struct rtnl_link_stats64 *netvsc_get_stats64(struct net_device *net,
914 						    struct rtnl_link_stats64 *t)
915 {
916 	struct net_device_context *ndev_ctx = netdev_priv(net);
917 	int cpu;
918 
919 	for_each_possible_cpu(cpu) {
920 		struct netvsc_stats *tx_stats = per_cpu_ptr(ndev_ctx->tx_stats,
921 							    cpu);
922 		struct netvsc_stats *rx_stats = per_cpu_ptr(ndev_ctx->rx_stats,
923 							    cpu);
924 		u64 tx_packets, tx_bytes, rx_packets, rx_bytes;
925 		unsigned int start;
926 
927 		do {
928 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
929 			tx_packets = tx_stats->packets;
930 			tx_bytes = tx_stats->bytes;
931 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
932 
933 		do {
934 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
935 			rx_packets = rx_stats->packets;
936 			rx_bytes = rx_stats->bytes;
937 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
938 
939 		t->tx_bytes	+= tx_bytes;
940 		t->tx_packets	+= tx_packets;
941 		t->rx_bytes	+= rx_bytes;
942 		t->rx_packets	+= rx_packets;
943 	}
944 
945 	t->tx_dropped	= net->stats.tx_dropped;
946 	t->tx_errors	= net->stats.tx_dropped;
947 
948 	t->rx_dropped	= net->stats.rx_dropped;
949 	t->rx_errors	= net->stats.rx_errors;
950 
951 	return t;
952 }
953 
954 static int netvsc_set_mac_addr(struct net_device *ndev, void *p)
955 {
956 	struct net_device_context *ndevctx = netdev_priv(ndev);
957 	struct hv_device *hdev =  ndevctx->device_ctx;
958 	struct sockaddr *addr = p;
959 	char save_adr[ETH_ALEN];
960 	unsigned char save_aatype;
961 	int err;
962 
963 	memcpy(save_adr, ndev->dev_addr, ETH_ALEN);
964 	save_aatype = ndev->addr_assign_type;
965 
966 	err = eth_mac_addr(ndev, p);
967 	if (err != 0)
968 		return err;
969 
970 	err = rndis_filter_set_device_mac(hdev, addr->sa_data);
971 	if (err != 0) {
972 		/* roll back to saved MAC */
973 		memcpy(ndev->dev_addr, save_adr, ETH_ALEN);
974 		ndev->addr_assign_type = save_aatype;
975 	}
976 
977 	return err;
978 }
979 
980 #ifdef CONFIG_NET_POLL_CONTROLLER
981 static void netvsc_poll_controller(struct net_device *net)
982 {
983 	/* As netvsc_start_xmit() works synchronous we don't have to
984 	 * trigger anything here.
985 	 */
986 }
987 #endif
988 
989 static const struct ethtool_ops ethtool_ops = {
990 	.get_drvinfo	= netvsc_get_drvinfo,
991 	.get_link	= ethtool_op_get_link,
992 	.get_channels   = netvsc_get_channels,
993 	.set_channels   = netvsc_set_channels,
994 };
995 
996 static const struct net_device_ops device_ops = {
997 	.ndo_open =			netvsc_open,
998 	.ndo_stop =			netvsc_close,
999 	.ndo_start_xmit =		netvsc_start_xmit,
1000 	.ndo_set_rx_mode =		netvsc_set_multicast_list,
1001 	.ndo_change_mtu =		netvsc_change_mtu,
1002 	.ndo_validate_addr =		eth_validate_addr,
1003 	.ndo_set_mac_address =		netvsc_set_mac_addr,
1004 	.ndo_select_queue =		netvsc_select_queue,
1005 	.ndo_get_stats64 =		netvsc_get_stats64,
1006 #ifdef CONFIG_NET_POLL_CONTROLLER
1007 	.ndo_poll_controller =		netvsc_poll_controller,
1008 #endif
1009 };
1010 
1011 /*
1012  * Send GARP packet to network peers after migrations.
1013  * After Quick Migration, the network is not immediately operational in the
1014  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
1015  * another netif_notify_peers() into a delayed work, otherwise GARP packet
1016  * will not be sent after quick migration, and cause network disconnection.
1017  * Also, we update the carrier status here.
1018  */
1019 static void netvsc_link_change(struct work_struct *w)
1020 {
1021 	struct net_device_context *ndev_ctx;
1022 	struct net_device *net;
1023 	struct netvsc_device *net_device;
1024 	struct rndis_device *rdev;
1025 	bool notify, refresh = false;
1026 	char *argv[] = { "/etc/init.d/network", "restart", NULL };
1027 	char *envp[] = { "HOME=/", "PATH=/sbin:/usr/sbin:/bin:/usr/bin", NULL };
1028 
1029 	rtnl_lock();
1030 
1031 	ndev_ctx = container_of(w, struct net_device_context, dwork.work);
1032 	net_device = hv_get_drvdata(ndev_ctx->device_ctx);
1033 	rdev = net_device->extension;
1034 	net = net_device->ndev;
1035 
1036 	if (rdev->link_state) {
1037 		netif_carrier_off(net);
1038 		notify = false;
1039 	} else {
1040 		netif_carrier_on(net);
1041 		notify = true;
1042 		if (rdev->link_change) {
1043 			rdev->link_change = false;
1044 			refresh = true;
1045 		}
1046 	}
1047 
1048 	rtnl_unlock();
1049 
1050 	if (refresh)
1051 		call_usermodehelper(argv[0], argv, envp, UMH_WAIT_EXEC);
1052 
1053 	if (notify)
1054 		netdev_notify_peers(net);
1055 }
1056 
1057 static void netvsc_free_netdev(struct net_device *netdev)
1058 {
1059 	struct net_device_context *net_device_ctx = netdev_priv(netdev);
1060 
1061 	free_percpu(net_device_ctx->tx_stats);
1062 	free_percpu(net_device_ctx->rx_stats);
1063 	free_netdev(netdev);
1064 }
1065 
1066 static int netvsc_probe(struct hv_device *dev,
1067 			const struct hv_vmbus_device_id *dev_id)
1068 {
1069 	struct net_device *net = NULL;
1070 	struct net_device_context *net_device_ctx;
1071 	struct netvsc_device_info device_info;
1072 	struct netvsc_device *nvdev;
1073 	int ret;
1074 	u32 max_needed_headroom;
1075 
1076 	net = alloc_etherdev_mq(sizeof(struct net_device_context),
1077 				num_online_cpus());
1078 	if (!net)
1079 		return -ENOMEM;
1080 
1081 	max_needed_headroom = sizeof(struct hv_netvsc_packet) +
1082 			      RNDIS_AND_PPI_SIZE;
1083 
1084 	netif_carrier_off(net);
1085 
1086 	net_device_ctx = netdev_priv(net);
1087 	net_device_ctx->device_ctx = dev;
1088 	net_device_ctx->msg_enable = netif_msg_init(debug, default_msg);
1089 	if (netif_msg_probe(net_device_ctx))
1090 		netdev_dbg(net, "netvsc msg_enable: %d\n",
1091 			   net_device_ctx->msg_enable);
1092 
1093 	net_device_ctx->tx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1094 	if (!net_device_ctx->tx_stats) {
1095 		free_netdev(net);
1096 		return -ENOMEM;
1097 	}
1098 	net_device_ctx->rx_stats = netdev_alloc_pcpu_stats(struct netvsc_stats);
1099 	if (!net_device_ctx->rx_stats) {
1100 		free_percpu(net_device_ctx->tx_stats);
1101 		free_netdev(net);
1102 		return -ENOMEM;
1103 	}
1104 
1105 	hv_set_drvdata(dev, net);
1106 	INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_link_change);
1107 	INIT_WORK(&net_device_ctx->work, do_set_multicast);
1108 
1109 	net->netdev_ops = &device_ops;
1110 
1111 	net->hw_features = NETIF_F_RXCSUM | NETIF_F_SG | NETIF_F_IP_CSUM |
1112 				NETIF_F_TSO;
1113 	net->features = NETIF_F_HW_VLAN_CTAG_TX | NETIF_F_SG | NETIF_F_RXCSUM |
1114 			NETIF_F_IP_CSUM | NETIF_F_TSO;
1115 
1116 	net->ethtool_ops = &ethtool_ops;
1117 	SET_NETDEV_DEV(net, &dev->device);
1118 
1119 	/*
1120 	 * Request additional head room in the skb.
1121 	 * We will use this space to build the rndis
1122 	 * heaser and other state we need to maintain.
1123 	 */
1124 	net->needed_headroom = max_needed_headroom;
1125 
1126 	/* Notify the netvsc driver of the new device */
1127 	memset(&device_info, 0, sizeof(device_info));
1128 	device_info.ring_size = ring_size;
1129 	device_info.max_num_vrss_chns = max_num_vrss_chns;
1130 	ret = rndis_filter_device_add(dev, &device_info);
1131 	if (ret != 0) {
1132 		netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
1133 		netvsc_free_netdev(net);
1134 		hv_set_drvdata(dev, NULL);
1135 		return ret;
1136 	}
1137 	memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
1138 
1139 	nvdev = hv_get_drvdata(dev);
1140 	netif_set_real_num_tx_queues(net, nvdev->num_chn);
1141 	netif_set_real_num_rx_queues(net, nvdev->num_chn);
1142 
1143 	ret = register_netdev(net);
1144 	if (ret != 0) {
1145 		pr_err("Unable to register netdev.\n");
1146 		rndis_filter_device_remove(dev);
1147 		netvsc_free_netdev(net);
1148 	} else {
1149 		schedule_delayed_work(&net_device_ctx->dwork, 0);
1150 	}
1151 
1152 	return ret;
1153 }
1154 
1155 static int netvsc_remove(struct hv_device *dev)
1156 {
1157 	struct net_device *net;
1158 	struct net_device_context *ndev_ctx;
1159 	struct netvsc_device *net_device;
1160 
1161 	net_device = hv_get_drvdata(dev);
1162 	net = net_device->ndev;
1163 
1164 	if (net == NULL) {
1165 		dev_err(&dev->device, "No net device to remove\n");
1166 		return 0;
1167 	}
1168 
1169 	net_device->start_remove = true;
1170 
1171 	ndev_ctx = netdev_priv(net);
1172 	cancel_delayed_work_sync(&ndev_ctx->dwork);
1173 	cancel_work_sync(&ndev_ctx->work);
1174 
1175 	/* Stop outbound asap */
1176 	netif_tx_disable(net);
1177 
1178 	unregister_netdev(net);
1179 
1180 	/*
1181 	 * Call to the vsc driver to let it know that the device is being
1182 	 * removed
1183 	 */
1184 	rndis_filter_device_remove(dev);
1185 
1186 	netvsc_free_netdev(net);
1187 	return 0;
1188 }
1189 
1190 static const struct hv_vmbus_device_id id_table[] = {
1191 	/* Network guid */
1192 	{ HV_NIC_GUID, },
1193 	{ },
1194 };
1195 
1196 MODULE_DEVICE_TABLE(vmbus, id_table);
1197 
1198 /* The one and only one */
1199 static struct  hv_driver netvsc_drv = {
1200 	.name = KBUILD_MODNAME,
1201 	.id_table = id_table,
1202 	.probe = netvsc_probe,
1203 	.remove = netvsc_remove,
1204 };
1205 
1206 static void __exit netvsc_drv_exit(void)
1207 {
1208 	vmbus_driver_unregister(&netvsc_drv);
1209 }
1210 
1211 static int __init netvsc_drv_init(void)
1212 {
1213 	if (ring_size < RING_SIZE_MIN) {
1214 		ring_size = RING_SIZE_MIN;
1215 		pr_info("Increased ring_size to %d (min allowed)\n",
1216 			ring_size);
1217 	}
1218 	return vmbus_driver_register(&netvsc_drv);
1219 }
1220 
1221 MODULE_LICENSE("GPL");
1222 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
1223 
1224 module_init(netvsc_drv_init);
1225 module_exit(netvsc_drv_exit);
1226