1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (c) 2009, Microsoft Corporation.
4 *
5 * Authors:
6 *   Haiyang Zhang <haiyangz@microsoft.com>
7 *   Hank Janssen  <hjanssen@microsoft.com>
8 */
9#include <linux/ethtool.h>
10#include <linux/kernel.h>
11#include <linux/sched.h>
12#include <linux/wait.h>
13#include <linux/highmem.h>
14#include <linux/slab.h>
15#include <linux/io.h>
16#include <linux/if_ether.h>
17#include <linux/netdevice.h>
18#include <linux/if_vlan.h>
19#include <linux/nls.h>
20#include <linux/vmalloc.h>
21#include <linux/rtnetlink.h>
22#include <linux/ucs2_string.h>
23#include <linux/string.h>
24
25#include "hyperv_net.h"
26#include "netvsc_trace.h"
27
28static void rndis_set_multicast(struct work_struct *w);
29
30#define RNDIS_EXT_LEN HV_HYP_PAGE_SIZE
31struct rndis_request {
32	struct list_head list_ent;
33	struct completion  wait_event;
34
35	struct rndis_message response_msg;
36	/*
37	 * The buffer for extended info after the RNDIS response message. It's
38	 * referenced based on the data offset in the RNDIS message. Its size
39	 * is enough for current needs, and should be sufficient for the near
40	 * future.
41	 */
42	u8 response_ext[RNDIS_EXT_LEN];
43
44	/* Simplify allocation by having a netvsc packet inline */
45	struct hv_netvsc_packet	pkt;
46
47	struct rndis_message request_msg;
48	/*
49	 * The buffer for the extended info after the RNDIS request message.
50	 * It is referenced and sized in a similar way as response_ext.
51	 */
52	u8 request_ext[RNDIS_EXT_LEN];
53};
54
55static const u8 netvsc_hash_key[NETVSC_HASH_KEYLEN] = {
56	0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2,
57	0x41, 0x67, 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0,
58	0xd0, 0xca, 0x2b, 0xcb, 0xae, 0x7b, 0x30, 0xb4,
59	0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30, 0xf2, 0x0c,
60	0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa
61};
62
63static struct rndis_device *get_rndis_device(void)
64{
65	struct rndis_device *device;
66
67	device = kzalloc(sizeof(struct rndis_device), GFP_KERNEL);
68	if (!device)
69		return NULL;
70
71	spin_lock_init(&device->request_lock);
72
73	INIT_LIST_HEAD(&device->req_list);
74	INIT_WORK(&device->mcast_work, rndis_set_multicast);
75
76	device->state = RNDIS_DEV_UNINITIALIZED;
77
78	return device;
79}
80
81static struct rndis_request *get_rndis_request(struct rndis_device *dev,
82					     u32 msg_type,
83					     u32 msg_len)
84{
85	struct rndis_request *request;
86	struct rndis_message *rndis_msg;
87	struct rndis_set_request *set;
88	unsigned long flags;
89
90	request = kzalloc(sizeof(struct rndis_request), GFP_KERNEL);
91	if (!request)
92		return NULL;
93
94	init_completion(&request->wait_event);
95
96	rndis_msg = &request->request_msg;
97	rndis_msg->ndis_msg_type = msg_type;
98	rndis_msg->msg_len = msg_len;
99
100	request->pkt.q_idx = 0;
101
102	/*
103	 * Set the request id. This field is always after the rndis header for
104	 * request/response packet types so we just used the SetRequest as a
105	 * template
106	 */
107	set = &rndis_msg->msg.set_req;
108	set->req_id = atomic_inc_return(&dev->new_req_id);
109
110	/* Add to the request list */
111	spin_lock_irqsave(&dev->request_lock, flags);
112	list_add_tail(&request->list_ent, &dev->req_list);
113	spin_unlock_irqrestore(&dev->request_lock, flags);
114
115	return request;
116}
117
118static void put_rndis_request(struct rndis_device *dev,
119			    struct rndis_request *req)
120{
121	unsigned long flags;
122
123	spin_lock_irqsave(&dev->request_lock, flags);
124	list_del(&req->list_ent);
125	spin_unlock_irqrestore(&dev->request_lock, flags);
126
127	kfree(req);
128}
129
130static void dump_rndis_message(struct net_device *netdev,
131			       const struct rndis_message *rndis_msg,
132			       const void *data)
133{
134	switch (rndis_msg->ndis_msg_type) {
135	case RNDIS_MSG_PACKET:
136		if (rndis_msg->msg_len - RNDIS_HEADER_SIZE >= sizeof(struct rndis_packet)) {
137			const struct rndis_packet *pkt = data + RNDIS_HEADER_SIZE;
138			netdev_dbg(netdev, "RNDIS_MSG_PACKET (len %u, "
139				   "data offset %u data len %u, # oob %u, "
140				   "oob offset %u, oob len %u, pkt offset %u, "
141				   "pkt len %u\n",
142				   rndis_msg->msg_len,
143				   pkt->data_offset,
144				   pkt->data_len,
145				   pkt->num_oob_data_elements,
146				   pkt->oob_data_offset,
147				   pkt->oob_data_len,
148				   pkt->per_pkt_info_offset,
149				   pkt->per_pkt_info_len);
150		}
151		break;
152
153	case RNDIS_MSG_INIT_C:
154		if (rndis_msg->msg_len - RNDIS_HEADER_SIZE >=
155				sizeof(struct rndis_initialize_complete)) {
156			const struct rndis_initialize_complete *init_complete =
157				data + RNDIS_HEADER_SIZE;
158			netdev_dbg(netdev, "RNDIS_MSG_INIT_C "
159				"(len %u, id 0x%x, status 0x%x, major %d, minor %d, "
160				"device flags %d, max xfer size 0x%x, max pkts %u, "
161				"pkt aligned %u)\n",
162				rndis_msg->msg_len,
163				init_complete->req_id,
164				init_complete->status,
165				init_complete->major_ver,
166				init_complete->minor_ver,
167				init_complete->dev_flags,
168				init_complete->max_xfer_size,
169				init_complete->max_pkt_per_msg,
170				init_complete->pkt_alignment_factor);
171		}
172		break;
173
174	case RNDIS_MSG_QUERY_C:
175		if (rndis_msg->msg_len - RNDIS_HEADER_SIZE >=
176				sizeof(struct rndis_query_complete)) {
177			const struct rndis_query_complete *query_complete =
178				data + RNDIS_HEADER_SIZE;
179			netdev_dbg(netdev, "RNDIS_MSG_QUERY_C "
180				"(len %u, id 0x%x, status 0x%x, buf len %u, "
181				"buf offset %u)\n",
182				rndis_msg->msg_len,
183				query_complete->req_id,
184				query_complete->status,
185				query_complete->info_buflen,
186				query_complete->info_buf_offset);
187		}
188		break;
189
190	case RNDIS_MSG_SET_C:
191		if (rndis_msg->msg_len - RNDIS_HEADER_SIZE + sizeof(struct rndis_set_complete)) {
192			const struct rndis_set_complete *set_complete =
193				data + RNDIS_HEADER_SIZE;
194			netdev_dbg(netdev,
195				"RNDIS_MSG_SET_C (len %u, id 0x%x, status 0x%x)\n",
196				rndis_msg->msg_len,
197				set_complete->req_id,
198				set_complete->status);
199		}
200		break;
201
202	case RNDIS_MSG_INDICATE:
203		if (rndis_msg->msg_len - RNDIS_HEADER_SIZE >=
204				sizeof(struct rndis_indicate_status)) {
205			const struct rndis_indicate_status *indicate_status =
206				data + RNDIS_HEADER_SIZE;
207			netdev_dbg(netdev, "RNDIS_MSG_INDICATE "
208				"(len %u, status 0x%x, buf len %u, buf offset %u)\n",
209				rndis_msg->msg_len,
210				indicate_status->status,
211				indicate_status->status_buflen,
212				indicate_status->status_buf_offset);
213		}
214		break;
215
216	default:
217		netdev_dbg(netdev, "0x%x (len %u)\n",
218			rndis_msg->ndis_msg_type,
219			rndis_msg->msg_len);
220		break;
221	}
222}
223
224static int rndis_filter_send_request(struct rndis_device *dev,
225				  struct rndis_request *req)
226{
227	struct hv_netvsc_packet *packet;
228	struct hv_page_buffer page_buf[2];
229	struct hv_page_buffer *pb = page_buf;
230	int ret;
231
232	/* Setup the packet to send it */
233	packet = &req->pkt;
234
235	packet->total_data_buflen = req->request_msg.msg_len;
236	packet->page_buf_cnt = 1;
237
238	pb[0].pfn = virt_to_phys(&req->request_msg) >>
239					HV_HYP_PAGE_SHIFT;
240	pb[0].len = req->request_msg.msg_len;
241	pb[0].offset = offset_in_hvpage(&req->request_msg);
242
243	/* Add one page_buf when request_msg crossing page boundary */
244	if (pb[0].offset + pb[0].len > HV_HYP_PAGE_SIZE) {
245		packet->page_buf_cnt++;
246		pb[0].len = HV_HYP_PAGE_SIZE -
247			pb[0].offset;
248		pb[1].pfn = virt_to_phys((void *)&req->request_msg
249			+ pb[0].len) >> HV_HYP_PAGE_SHIFT;
250		pb[1].offset = 0;
251		pb[1].len = req->request_msg.msg_len -
252			pb[0].len;
253	}
254
255	trace_rndis_send(dev->ndev, 0, &req->request_msg);
256
257	rcu_read_lock_bh();
258	ret = netvsc_send(dev->ndev, packet, NULL, pb, NULL, false);
259	rcu_read_unlock_bh();
260
261	return ret;
262}
263
264static void rndis_set_link_state(struct rndis_device *rdev,
265				 struct rndis_request *request)
266{
267	u32 link_status;
268	struct rndis_query_complete *query_complete;
269	u32 msg_len = request->response_msg.msg_len;
270
271	/* Ensure the packet is big enough to access its fields */
272	if (msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_query_complete))
273		return;
274
275	query_complete = &request->response_msg.msg.query_complete;
276
277	if (query_complete->status == RNDIS_STATUS_SUCCESS &&
278	    query_complete->info_buflen >= sizeof(u32) &&
279	    query_complete->info_buf_offset >= sizeof(*query_complete) &&
280	    msg_len - RNDIS_HEADER_SIZE >= query_complete->info_buf_offset &&
281	    msg_len - RNDIS_HEADER_SIZE - query_complete->info_buf_offset
282			>= query_complete->info_buflen) {
283		memcpy(&link_status, (void *)((unsigned long)query_complete +
284		       query_complete->info_buf_offset), sizeof(u32));
285		rdev->link_state = link_status != 0;
286	}
287}
288
289static void rndis_filter_receive_response(struct net_device *ndev,
290					  struct netvsc_device *nvdev,
291					  struct rndis_message *resp,
292					  void *data)
293{
294	u32 *req_id = &resp->msg.init_complete.req_id;
295	struct rndis_device *dev = nvdev->extension;
296	struct rndis_request *request = NULL;
297	bool found = false;
298	unsigned long flags;
299
300	/* This should never happen, it means control message
301	 * response received after device removed.
302	 */
303	if (dev->state == RNDIS_DEV_UNINITIALIZED) {
304		netdev_err(ndev,
305			   "got rndis message uninitialized\n");
306		return;
307	}
308
309	/* Ensure the packet is big enough to read req_id. Req_id is the 1st
310	 * field in any request/response message, so the payload should have at
311	 * least sizeof(u32) bytes
312	 */
313	if (resp->msg_len - RNDIS_HEADER_SIZE < sizeof(u32)) {
314		netdev_err(ndev, "rndis msg_len too small: %u\n",
315			   resp->msg_len);
316		return;
317	}
318
319	/* Copy the request ID into nvchan->recv_buf */
320	*req_id = *(u32 *)(data + RNDIS_HEADER_SIZE);
321
322	spin_lock_irqsave(&dev->request_lock, flags);
323	list_for_each_entry(request, &dev->req_list, list_ent) {
324		/*
325		 * All request/response message contains RequestId as the 1st
326		 * field
327		 */
328		if (request->request_msg.msg.init_req.req_id == *req_id) {
329			found = true;
330			break;
331		}
332	}
333	spin_unlock_irqrestore(&dev->request_lock, flags);
334
335	if (found) {
336		if (resp->msg_len <=
337		    sizeof(struct rndis_message) + RNDIS_EXT_LEN) {
338			memcpy(&request->response_msg, resp, RNDIS_HEADER_SIZE + sizeof(*req_id));
339			unsafe_memcpy((void *)&request->response_msg + RNDIS_HEADER_SIZE + sizeof(*req_id),
340			       data + RNDIS_HEADER_SIZE + sizeof(*req_id),
341			       resp->msg_len - RNDIS_HEADER_SIZE - sizeof(*req_id),
342			       "request->response_msg is followed by a padding of RNDIS_EXT_LEN inside rndis_request");
343			if (request->request_msg.ndis_msg_type ==
344			    RNDIS_MSG_QUERY && request->request_msg.msg.
345			    query_req.oid == RNDIS_OID_GEN_MEDIA_CONNECT_STATUS)
346				rndis_set_link_state(dev, request);
347		} else {
348			netdev_err(ndev,
349				"rndis response buffer overflow "
350				"detected (size %u max %zu)\n",
351				resp->msg_len,
352				sizeof(struct rndis_message));
353
354			if (resp->ndis_msg_type ==
355			    RNDIS_MSG_RESET_C) {
356				/* does not have a request id field */
357				request->response_msg.msg.reset_complete.
358					status = RNDIS_STATUS_BUFFER_OVERFLOW;
359			} else {
360				request->response_msg.msg.
361				init_complete.status =
362					RNDIS_STATUS_BUFFER_OVERFLOW;
363			}
364		}
365
366		netvsc_dma_unmap(((struct net_device_context *)
367			netdev_priv(ndev))->device_ctx, &request->pkt);
368		complete(&request->wait_event);
369	} else {
370		netdev_err(ndev,
371			"no rndis request found for this response "
372			"(id 0x%x res type 0x%x)\n",
373			*req_id,
374			resp->ndis_msg_type);
375	}
376}
377
378/*
379 * Get the Per-Packet-Info with the specified type
380 * return NULL if not found.
381 */
382static inline void *rndis_get_ppi(struct net_device *ndev,
383				  struct rndis_packet *rpkt,
384				  u32 rpkt_len, u32 type, u8 internal,
385				  u32 ppi_size, void *data)
386{
387	struct rndis_per_packet_info *ppi;
388	int len;
389
390	if (rpkt->per_pkt_info_offset == 0)
391		return NULL;
392
393	/* Validate info_offset and info_len */
394	if (rpkt->per_pkt_info_offset < sizeof(struct rndis_packet) ||
395	    rpkt->per_pkt_info_offset > rpkt_len) {
396		netdev_err(ndev, "Invalid per_pkt_info_offset: %u\n",
397			   rpkt->per_pkt_info_offset);
398		return NULL;
399	}
400
401	if (rpkt->per_pkt_info_len < sizeof(*ppi) ||
402	    rpkt->per_pkt_info_len > rpkt_len - rpkt->per_pkt_info_offset) {
403		netdev_err(ndev, "Invalid per_pkt_info_len: %u\n",
404			   rpkt->per_pkt_info_len);
405		return NULL;
406	}
407
408	ppi = (struct rndis_per_packet_info *)((ulong)rpkt +
409		rpkt->per_pkt_info_offset);
410	/* Copy the PPIs into nvchan->recv_buf */
411	memcpy(ppi, data + RNDIS_HEADER_SIZE + rpkt->per_pkt_info_offset, rpkt->per_pkt_info_len);
412	len = rpkt->per_pkt_info_len;
413
414	while (len > 0) {
415		/* Validate ppi_offset and ppi_size */
416		if (ppi->size > len) {
417			netdev_err(ndev, "Invalid ppi size: %u\n", ppi->size);
418			continue;
419		}
420
421		if (ppi->ppi_offset >= ppi->size) {
422			netdev_err(ndev, "Invalid ppi_offset: %u\n", ppi->ppi_offset);
423			continue;
424		}
425
426		if (ppi->type == type && ppi->internal == internal) {
427			/* ppi->size should be big enough to hold the returned object. */
428			if (ppi->size - ppi->ppi_offset < ppi_size) {
429				netdev_err(ndev, "Invalid ppi: size %u ppi_offset %u\n",
430					   ppi->size, ppi->ppi_offset);
431				continue;
432			}
433			return (void *)((ulong)ppi + ppi->ppi_offset);
434		}
435		len -= ppi->size;
436		ppi = (struct rndis_per_packet_info *)((ulong)ppi + ppi->size);
437	}
438
439	return NULL;
440}
441
442static inline
443void rsc_add_data(struct netvsc_channel *nvchan,
444		  const struct ndis_pkt_8021q_info *vlan,
445		  const struct ndis_tcp_ip_checksum_info *csum_info,
446		  const u32 *hash_info,
447		  void *data, u32 len)
448{
449	u32 cnt = nvchan->rsc.cnt;
450
451	if (cnt) {
452		nvchan->rsc.pktlen += len;
453	} else {
454		/* The data/values pointed by vlan, csum_info and hash_info are shared
455		 * across the different 'fragments' of the RSC packet; store them into
456		 * the packet itself.
457		 */
458		if (vlan != NULL) {
459			memcpy(&nvchan->rsc.vlan, vlan, sizeof(*vlan));
460			nvchan->rsc.ppi_flags |= NVSC_RSC_VLAN;
461		} else {
462			nvchan->rsc.ppi_flags &= ~NVSC_RSC_VLAN;
463		}
464		if (csum_info != NULL) {
465			memcpy(&nvchan->rsc.csum_info, csum_info, sizeof(*csum_info));
466			nvchan->rsc.ppi_flags |= NVSC_RSC_CSUM_INFO;
467		} else {
468			nvchan->rsc.ppi_flags &= ~NVSC_RSC_CSUM_INFO;
469		}
470		nvchan->rsc.pktlen = len;
471		if (hash_info != NULL) {
472			nvchan->rsc.hash_info = *hash_info;
473			nvchan->rsc.ppi_flags |= NVSC_RSC_HASH_INFO;
474		} else {
475			nvchan->rsc.ppi_flags &= ~NVSC_RSC_HASH_INFO;
476		}
477	}
478
479	nvchan->rsc.data[cnt] = data;
480	nvchan->rsc.len[cnt] = len;
481	nvchan->rsc.cnt++;
482}
483
484static int rndis_filter_receive_data(struct net_device *ndev,
485				     struct netvsc_device *nvdev,
486				     struct netvsc_channel *nvchan,
487				     struct rndis_message *msg,
488				     void *data, u32 data_buflen)
489{
490	struct rndis_packet *rndis_pkt = &msg->msg.pkt;
491	const struct ndis_tcp_ip_checksum_info *csum_info;
492	const struct ndis_pkt_8021q_info *vlan;
493	const struct rndis_pktinfo_id *pktinfo_id;
494	const u32 *hash_info;
495	u32 data_offset, rpkt_len;
496	bool rsc_more = false;
497	int ret;
498
499	/* Ensure data_buflen is big enough to read header fields */
500	if (data_buflen < RNDIS_HEADER_SIZE + sizeof(struct rndis_packet)) {
501		netdev_err(ndev, "invalid rndis pkt, data_buflen too small: %u\n",
502			   data_buflen);
503		return NVSP_STAT_FAIL;
504	}
505
506	/* Copy the RNDIS packet into nvchan->recv_buf */
507	memcpy(rndis_pkt, data + RNDIS_HEADER_SIZE, sizeof(*rndis_pkt));
508
509	/* Validate rndis_pkt offset */
510	if (rndis_pkt->data_offset >= data_buflen - RNDIS_HEADER_SIZE) {
511		netdev_err(ndev, "invalid rndis packet offset: %u\n",
512			   rndis_pkt->data_offset);
513		return NVSP_STAT_FAIL;
514	}
515
516	/* Remove the rndis header and pass it back up the stack */
517	data_offset = RNDIS_HEADER_SIZE + rndis_pkt->data_offset;
518
519	rpkt_len = data_buflen - RNDIS_HEADER_SIZE;
520	data_buflen -= data_offset;
521
522	/*
523	 * Make sure we got a valid RNDIS message, now total_data_buflen
524	 * should be the data packet size plus the trailer padding size
525	 */
526	if (unlikely(data_buflen < rndis_pkt->data_len)) {
527		netdev_err(ndev, "rndis message buffer "
528			   "overflow detected (got %u, min %u)"
529			   "...dropping this message!\n",
530			   data_buflen, rndis_pkt->data_len);
531		return NVSP_STAT_FAIL;
532	}
533
534	vlan = rndis_get_ppi(ndev, rndis_pkt, rpkt_len, IEEE_8021Q_INFO, 0, sizeof(*vlan),
535			     data);
536
537	csum_info = rndis_get_ppi(ndev, rndis_pkt, rpkt_len, TCPIP_CHKSUM_PKTINFO, 0,
538				  sizeof(*csum_info), data);
539
540	hash_info = rndis_get_ppi(ndev, rndis_pkt, rpkt_len, NBL_HASH_VALUE, 0,
541				  sizeof(*hash_info), data);
542
543	pktinfo_id = rndis_get_ppi(ndev, rndis_pkt, rpkt_len, RNDIS_PKTINFO_ID, 1,
544				   sizeof(*pktinfo_id), data);
545
546	/* Identify RSC frags, drop erroneous packets */
547	if (pktinfo_id && (pktinfo_id->flag & RNDIS_PKTINFO_SUBALLOC)) {
548		if (pktinfo_id->flag & RNDIS_PKTINFO_1ST_FRAG)
549			nvchan->rsc.cnt = 0;
550		else if (nvchan->rsc.cnt == 0)
551			goto drop;
552
553		rsc_more = true;
554
555		if (pktinfo_id->flag & RNDIS_PKTINFO_LAST_FRAG)
556			rsc_more = false;
557
558		if (rsc_more && nvchan->rsc.is_last)
559			goto drop;
560	} else {
561		nvchan->rsc.cnt = 0;
562	}
563
564	if (unlikely(nvchan->rsc.cnt >= NVSP_RSC_MAX))
565		goto drop;
566
567	/* Put data into per channel structure.
568	 * Also, remove the rndis trailer padding from rndis packet message
569	 * rndis_pkt->data_len tell us the real data length, we only copy
570	 * the data packet to the stack, without the rndis trailer padding
571	 */
572	rsc_add_data(nvchan, vlan, csum_info, hash_info,
573		     data + data_offset, rndis_pkt->data_len);
574
575	if (rsc_more)
576		return NVSP_STAT_SUCCESS;
577
578	ret = netvsc_recv_callback(ndev, nvdev, nvchan);
579	nvchan->rsc.cnt = 0;
580
581	return ret;
582
583drop:
584	return NVSP_STAT_FAIL;
585}
586
587int rndis_filter_receive(struct net_device *ndev,
588			 struct netvsc_device *net_dev,
589			 struct netvsc_channel *nvchan,
590			 void *data, u32 buflen)
591{
592	struct net_device_context *net_device_ctx = netdev_priv(ndev);
593	struct rndis_message *rndis_msg = nvchan->recv_buf;
594
595	if (buflen < RNDIS_HEADER_SIZE) {
596		netdev_err(ndev, "Invalid rndis_msg (buflen: %u)\n", buflen);
597		return NVSP_STAT_FAIL;
598	}
599
600	/* Copy the RNDIS msg header into nvchan->recv_buf */
601	memcpy(rndis_msg, data, RNDIS_HEADER_SIZE);
602
603	/* Validate incoming rndis_message packet */
604	if (rndis_msg->msg_len < RNDIS_HEADER_SIZE ||
605	    buflen < rndis_msg->msg_len) {
606		netdev_err(ndev, "Invalid rndis_msg (buflen: %u, msg_len: %u)\n",
607			   buflen, rndis_msg->msg_len);
608		return NVSP_STAT_FAIL;
609	}
610
611	if (netif_msg_rx_status(net_device_ctx))
612		dump_rndis_message(ndev, rndis_msg, data);
613
614	switch (rndis_msg->ndis_msg_type) {
615	case RNDIS_MSG_PACKET:
616		return rndis_filter_receive_data(ndev, net_dev, nvchan,
617						 rndis_msg, data, buflen);
618	case RNDIS_MSG_INIT_C:
619	case RNDIS_MSG_QUERY_C:
620	case RNDIS_MSG_SET_C:
621		/* completion msgs */
622		rndis_filter_receive_response(ndev, net_dev, rndis_msg, data);
623		break;
624
625	case RNDIS_MSG_INDICATE:
626		/* notification msgs */
627		netvsc_linkstatus_callback(ndev, rndis_msg, data, buflen);
628		break;
629	default:
630		netdev_err(ndev,
631			"unhandled rndis message (type %u len %u)\n",
632			   rndis_msg->ndis_msg_type,
633			   rndis_msg->msg_len);
634		return NVSP_STAT_FAIL;
635	}
636
637	return NVSP_STAT_SUCCESS;
638}
639
640static int rndis_filter_query_device(struct rndis_device *dev,
641				     struct netvsc_device *nvdev,
642				     u32 oid, void *result, u32 *result_size)
643{
644	struct rndis_request *request;
645	u32 inresult_size = *result_size;
646	struct rndis_query_request *query;
647	struct rndis_query_complete *query_complete;
648	u32 msg_len;
649	int ret = 0;
650
651	if (!result)
652		return -EINVAL;
653
654	*result_size = 0;
655	request = get_rndis_request(dev, RNDIS_MSG_QUERY,
656			RNDIS_MESSAGE_SIZE(struct rndis_query_request));
657	if (!request) {
658		ret = -ENOMEM;
659		goto cleanup;
660	}
661
662	/* Setup the rndis query */
663	query = &request->request_msg.msg.query_req;
664	query->oid = oid;
665	query->info_buf_offset = sizeof(struct rndis_query_request);
666	query->info_buflen = 0;
667	query->dev_vc_handle = 0;
668
669	if (oid == OID_TCP_OFFLOAD_HARDWARE_CAPABILITIES) {
670		struct ndis_offload *hwcaps;
671		u32 nvsp_version = nvdev->nvsp_version;
672		u8 ndis_rev;
673		size_t size;
674
675		if (nvsp_version >= NVSP_PROTOCOL_VERSION_5) {
676			ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_3;
677			size = NDIS_OFFLOAD_SIZE;
678		} else if (nvsp_version >= NVSP_PROTOCOL_VERSION_4) {
679			ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_2;
680			size = NDIS_OFFLOAD_SIZE_6_1;
681		} else {
682			ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_1;
683			size = NDIS_OFFLOAD_SIZE_6_0;
684		}
685
686		request->request_msg.msg_len += size;
687		query->info_buflen = size;
688		hwcaps = (struct ndis_offload *)
689			((unsigned long)query + query->info_buf_offset);
690
691		hwcaps->header.type = NDIS_OBJECT_TYPE_OFFLOAD;
692		hwcaps->header.revision = ndis_rev;
693		hwcaps->header.size = size;
694
695	} else if (oid == OID_GEN_RECEIVE_SCALE_CAPABILITIES) {
696		struct ndis_recv_scale_cap *cap;
697
698		request->request_msg.msg_len +=
699			sizeof(struct ndis_recv_scale_cap);
700		query->info_buflen = sizeof(struct ndis_recv_scale_cap);
701		cap = (struct ndis_recv_scale_cap *)((unsigned long)query +
702						     query->info_buf_offset);
703		cap->hdr.type = NDIS_OBJECT_TYPE_RSS_CAPABILITIES;
704		cap->hdr.rev = NDIS_RECEIVE_SCALE_CAPABILITIES_REVISION_2;
705		cap->hdr.size = sizeof(struct ndis_recv_scale_cap);
706	}
707
708	ret = rndis_filter_send_request(dev, request);
709	if (ret != 0)
710		goto cleanup;
711
712	wait_for_completion(&request->wait_event);
713
714	/* Copy the response back */
715	query_complete = &request->response_msg.msg.query_complete;
716	msg_len = request->response_msg.msg_len;
717
718	/* Ensure the packet is big enough to access its fields */
719	if (msg_len - RNDIS_HEADER_SIZE < sizeof(struct rndis_query_complete)) {
720		ret = -1;
721		goto cleanup;
722	}
723
724	if (query_complete->info_buflen > inresult_size ||
725	    query_complete->info_buf_offset < sizeof(*query_complete) ||
726	    msg_len - RNDIS_HEADER_SIZE < query_complete->info_buf_offset ||
727	    msg_len - RNDIS_HEADER_SIZE - query_complete->info_buf_offset
728			< query_complete->info_buflen) {
729		ret = -1;
730		goto cleanup;
731	}
732
733	memcpy(result,
734	       (void *)((unsigned long)query_complete +
735			 query_complete->info_buf_offset),
736	       query_complete->info_buflen);
737
738	*result_size = query_complete->info_buflen;
739
740cleanup:
741	if (request)
742		put_rndis_request(dev, request);
743
744	return ret;
745}
746
747/* Get the hardware offload capabilities */
748static int
749rndis_query_hwcaps(struct rndis_device *dev, struct netvsc_device *net_device,
750		   struct ndis_offload *caps)
751{
752	u32 caps_len = sizeof(*caps);
753	int ret;
754
755	memset(caps, 0, sizeof(*caps));
756
757	ret = rndis_filter_query_device(dev, net_device,
758					OID_TCP_OFFLOAD_HARDWARE_CAPABILITIES,
759					caps, &caps_len);
760	if (ret)
761		return ret;
762
763	if (caps->header.type != NDIS_OBJECT_TYPE_OFFLOAD) {
764		netdev_warn(dev->ndev, "invalid NDIS objtype %#x\n",
765			    caps->header.type);
766		return -EINVAL;
767	}
768
769	if (caps->header.revision < NDIS_OFFLOAD_PARAMETERS_REVISION_1) {
770		netdev_warn(dev->ndev, "invalid NDIS objrev %x\n",
771			    caps->header.revision);
772		return -EINVAL;
773	}
774
775	if (caps->header.size > caps_len ||
776	    caps->header.size < NDIS_OFFLOAD_SIZE_6_0) {
777		netdev_warn(dev->ndev,
778			    "invalid NDIS objsize %u, data size %u\n",
779			    caps->header.size, caps_len);
780		return -EINVAL;
781	}
782
783	return 0;
784}
785
786static int rndis_filter_query_device_mac(struct rndis_device *dev,
787					 struct netvsc_device *net_device)
788{
789	u32 size = ETH_ALEN;
790
791	return rndis_filter_query_device(dev, net_device,
792				      RNDIS_OID_802_3_PERMANENT_ADDRESS,
793				      dev->hw_mac_adr, &size);
794}
795
796#define NWADR_STR "NetworkAddress"
797#define NWADR_STRLEN 14
798
799int rndis_filter_set_device_mac(struct netvsc_device *nvdev,
800				const char *mac)
801{
802	struct rndis_device *rdev = nvdev->extension;
803	struct rndis_request *request;
804	struct rndis_set_request *set;
805	struct rndis_config_parameter_info *cpi;
806	wchar_t *cfg_nwadr, *cfg_mac;
807	struct rndis_set_complete *set_complete;
808	char macstr[2*ETH_ALEN+1];
809	u32 extlen = sizeof(struct rndis_config_parameter_info) +
810		2*NWADR_STRLEN + 4*ETH_ALEN;
811	int ret;
812
813	request = get_rndis_request(rdev, RNDIS_MSG_SET,
814		RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
815	if (!request)
816		return -ENOMEM;
817
818	set = &request->request_msg.msg.set_req;
819	set->oid = RNDIS_OID_GEN_RNDIS_CONFIG_PARAMETER;
820	set->info_buflen = extlen;
821	set->info_buf_offset = sizeof(struct rndis_set_request);
822	set->dev_vc_handle = 0;
823
824	cpi = (struct rndis_config_parameter_info *)((ulong)set +
825		set->info_buf_offset);
826	cpi->parameter_name_offset =
827		sizeof(struct rndis_config_parameter_info);
828	/* Multiply by 2 because host needs 2 bytes (utf16) for each char */
829	cpi->parameter_name_length = 2*NWADR_STRLEN;
830	cpi->parameter_type = RNDIS_CONFIG_PARAM_TYPE_STRING;
831	cpi->parameter_value_offset =
832		cpi->parameter_name_offset + cpi->parameter_name_length;
833	/* Multiply by 4 because each MAC byte displayed as 2 utf16 chars */
834	cpi->parameter_value_length = 4*ETH_ALEN;
835
836	cfg_nwadr = (wchar_t *)((ulong)cpi + cpi->parameter_name_offset);
837	cfg_mac = (wchar_t *)((ulong)cpi + cpi->parameter_value_offset);
838	ret = utf8s_to_utf16s(NWADR_STR, NWADR_STRLEN, UTF16_HOST_ENDIAN,
839			      cfg_nwadr, NWADR_STRLEN);
840	if (ret < 0)
841		goto cleanup;
842	snprintf(macstr, 2*ETH_ALEN+1, "%pm", mac);
843	ret = utf8s_to_utf16s(macstr, 2*ETH_ALEN, UTF16_HOST_ENDIAN,
844			      cfg_mac, 2*ETH_ALEN);
845	if (ret < 0)
846		goto cleanup;
847
848	ret = rndis_filter_send_request(rdev, request);
849	if (ret != 0)
850		goto cleanup;
851
852	wait_for_completion(&request->wait_event);
853
854	set_complete = &request->response_msg.msg.set_complete;
855	if (set_complete->status != RNDIS_STATUS_SUCCESS)
856		ret = -EIO;
857
858cleanup:
859	put_rndis_request(rdev, request);
860	return ret;
861}
862
863int
864rndis_filter_set_offload_params(struct net_device *ndev,
865				struct netvsc_device *nvdev,
866				struct ndis_offload_params *req_offloads)
867{
868	struct rndis_device *rdev = nvdev->extension;
869	struct rndis_request *request;
870	struct rndis_set_request *set;
871	struct ndis_offload_params *offload_params;
872	struct rndis_set_complete *set_complete;
873	u32 extlen = sizeof(struct ndis_offload_params);
874	int ret;
875	u32 vsp_version = nvdev->nvsp_version;
876
877	if (vsp_version <= NVSP_PROTOCOL_VERSION_4) {
878		extlen = VERSION_4_OFFLOAD_SIZE;
879		/* On NVSP_PROTOCOL_VERSION_4 and below, we do not support
880		 * UDP checksum offload.
881		 */
882		req_offloads->udp_ip_v4_csum = 0;
883		req_offloads->udp_ip_v6_csum = 0;
884	}
885
886	request = get_rndis_request(rdev, RNDIS_MSG_SET,
887		RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
888	if (!request)
889		return -ENOMEM;
890
891	set = &request->request_msg.msg.set_req;
892	set->oid = OID_TCP_OFFLOAD_PARAMETERS;
893	set->info_buflen = extlen;
894	set->info_buf_offset = sizeof(struct rndis_set_request);
895	set->dev_vc_handle = 0;
896
897	offload_params = (struct ndis_offload_params *)((ulong)set +
898				set->info_buf_offset);
899	*offload_params = *req_offloads;
900	offload_params->header.type = NDIS_OBJECT_TYPE_DEFAULT;
901	offload_params->header.revision = NDIS_OFFLOAD_PARAMETERS_REVISION_3;
902	offload_params->header.size = extlen;
903
904	ret = rndis_filter_send_request(rdev, request);
905	if (ret != 0)
906		goto cleanup;
907
908	wait_for_completion(&request->wait_event);
909	set_complete = &request->response_msg.msg.set_complete;
910	if (set_complete->status != RNDIS_STATUS_SUCCESS) {
911		netdev_err(ndev, "Fail to set offload on host side:0x%x\n",
912			   set_complete->status);
913		ret = -EINVAL;
914	}
915
916cleanup:
917	put_rndis_request(rdev, request);
918	return ret;
919}
920
921static int rndis_set_rss_param_msg(struct rndis_device *rdev,
922				   const u8 *rss_key, u16 flag)
923{
924	struct net_device *ndev = rdev->ndev;
925	struct net_device_context *ndc = netdev_priv(ndev);
926	struct rndis_request *request;
927	struct rndis_set_request *set;
928	struct rndis_set_complete *set_complete;
929	u32 extlen = sizeof(struct ndis_recv_scale_param) +
930		     4 * ndc->rx_table_sz + NETVSC_HASH_KEYLEN;
931	struct ndis_recv_scale_param *rssp;
932	u32 *itab;
933	u8 *keyp;
934	int i, ret;
935
936	request = get_rndis_request(
937			rdev, RNDIS_MSG_SET,
938			RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
939	if (!request)
940		return -ENOMEM;
941
942	set = &request->request_msg.msg.set_req;
943	set->oid = OID_GEN_RECEIVE_SCALE_PARAMETERS;
944	set->info_buflen = extlen;
945	set->info_buf_offset = sizeof(struct rndis_set_request);
946	set->dev_vc_handle = 0;
947
948	rssp = (struct ndis_recv_scale_param *)(set + 1);
949	rssp->hdr.type = NDIS_OBJECT_TYPE_RSS_PARAMETERS;
950	rssp->hdr.rev = NDIS_RECEIVE_SCALE_PARAMETERS_REVISION_2;
951	rssp->hdr.size = sizeof(struct ndis_recv_scale_param);
952	rssp->flag = flag;
953	rssp->hashinfo = NDIS_HASH_FUNC_TOEPLITZ | NDIS_HASH_IPV4 |
954			 NDIS_HASH_TCP_IPV4 | NDIS_HASH_IPV6 |
955			 NDIS_HASH_TCP_IPV6;
956	rssp->indirect_tabsize = 4 * ndc->rx_table_sz;
957	rssp->indirect_taboffset = sizeof(struct ndis_recv_scale_param);
958	rssp->hashkey_size = NETVSC_HASH_KEYLEN;
959	rssp->hashkey_offset = rssp->indirect_taboffset +
960			       rssp->indirect_tabsize;
961
962	/* Set indirection table entries */
963	itab = (u32 *)(rssp + 1);
964	for (i = 0; i < ndc->rx_table_sz; i++)
965		itab[i] = ndc->rx_table[i];
966
967	/* Set hask key values */
968	keyp = (u8 *)((unsigned long)rssp + rssp->hashkey_offset);
969	memcpy(keyp, rss_key, NETVSC_HASH_KEYLEN);
970
971	ret = rndis_filter_send_request(rdev, request);
972	if (ret != 0)
973		goto cleanup;
974
975	wait_for_completion(&request->wait_event);
976	set_complete = &request->response_msg.msg.set_complete;
977	if (set_complete->status == RNDIS_STATUS_SUCCESS) {
978		if (!(flag & NDIS_RSS_PARAM_FLAG_DISABLE_RSS) &&
979		    !(flag & NDIS_RSS_PARAM_FLAG_HASH_KEY_UNCHANGED))
980			memcpy(rdev->rss_key, rss_key, NETVSC_HASH_KEYLEN);
981
982	} else {
983		netdev_err(ndev, "Fail to set RSS parameters:0x%x\n",
984			   set_complete->status);
985		ret = -EINVAL;
986	}
987
988cleanup:
989	put_rndis_request(rdev, request);
990	return ret;
991}
992
993int rndis_filter_set_rss_param(struct rndis_device *rdev,
994			       const u8 *rss_key)
995{
996	/* Disable RSS before change */
997	rndis_set_rss_param_msg(rdev, rss_key,
998				NDIS_RSS_PARAM_FLAG_DISABLE_RSS);
999
1000	return rndis_set_rss_param_msg(rdev, rss_key, 0);
1001}
1002
1003static int rndis_filter_query_device_link_status(struct rndis_device *dev,
1004						 struct netvsc_device *net_device)
1005{
1006	u32 size = sizeof(u32);
1007	u32 link_status;
1008
1009	return rndis_filter_query_device(dev, net_device,
1010					 RNDIS_OID_GEN_MEDIA_CONNECT_STATUS,
1011					 &link_status, &size);
1012}
1013
1014static int rndis_filter_query_link_speed(struct rndis_device *dev,
1015					 struct netvsc_device *net_device)
1016{
1017	u32 size = sizeof(u32);
1018	u32 link_speed;
1019	struct net_device_context *ndc;
1020	int ret;
1021
1022	ret = rndis_filter_query_device(dev, net_device,
1023					RNDIS_OID_GEN_LINK_SPEED,
1024					&link_speed, &size);
1025
1026	if (!ret) {
1027		ndc = netdev_priv(dev->ndev);
1028
1029		/* The link speed reported from host is in 100bps unit, so
1030		 * we convert it to Mbps here.
1031		 */
1032		ndc->speed = link_speed / 10000;
1033	}
1034
1035	return ret;
1036}
1037
1038static int rndis_filter_set_packet_filter(struct rndis_device *dev,
1039					  u32 new_filter)
1040{
1041	struct rndis_request *request;
1042	struct rndis_set_request *set;
1043	int ret;
1044
1045	if (dev->filter == new_filter)
1046		return 0;
1047
1048	request = get_rndis_request(dev, RNDIS_MSG_SET,
1049			RNDIS_MESSAGE_SIZE(struct rndis_set_request) +
1050			sizeof(u32));
1051	if (!request)
1052		return -ENOMEM;
1053
1054	/* Setup the rndis set */
1055	set = &request->request_msg.msg.set_req;
1056	set->oid = RNDIS_OID_GEN_CURRENT_PACKET_FILTER;
1057	set->info_buflen = sizeof(u32);
1058	set->info_buf_offset = offsetof(typeof(*set), info_buf);
1059	memcpy(set->info_buf, &new_filter, sizeof(u32));
1060
1061	ret = rndis_filter_send_request(dev, request);
1062	if (ret == 0) {
1063		wait_for_completion(&request->wait_event);
1064		dev->filter = new_filter;
1065	}
1066
1067	put_rndis_request(dev, request);
1068
1069	return ret;
1070}
1071
1072static void rndis_set_multicast(struct work_struct *w)
1073{
1074	struct rndis_device *rdev
1075		= container_of(w, struct rndis_device, mcast_work);
1076	u32 filter = NDIS_PACKET_TYPE_DIRECTED;
1077	unsigned int flags = rdev->ndev->flags;
1078
1079	if (flags & IFF_PROMISC) {
1080		filter = NDIS_PACKET_TYPE_PROMISCUOUS;
1081	} else {
1082		if (!netdev_mc_empty(rdev->ndev) || (flags & IFF_ALLMULTI))
1083			filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
1084		if (flags & IFF_BROADCAST)
1085			filter |= NDIS_PACKET_TYPE_BROADCAST;
1086	}
1087
1088	rndis_filter_set_packet_filter(rdev, filter);
1089}
1090
1091void rndis_filter_update(struct netvsc_device *nvdev)
1092{
1093	struct rndis_device *rdev = nvdev->extension;
1094
1095	schedule_work(&rdev->mcast_work);
1096}
1097
1098static int rndis_filter_init_device(struct rndis_device *dev,
1099				    struct netvsc_device *nvdev)
1100{
1101	struct rndis_request *request;
1102	struct rndis_initialize_request *init;
1103	struct rndis_initialize_complete *init_complete;
1104	u32 status;
1105	int ret;
1106
1107	request = get_rndis_request(dev, RNDIS_MSG_INIT,
1108			RNDIS_MESSAGE_SIZE(struct rndis_initialize_request));
1109	if (!request) {
1110		ret = -ENOMEM;
1111		goto cleanup;
1112	}
1113
1114	/* Setup the rndis set */
1115	init = &request->request_msg.msg.init_req;
1116	init->major_ver = RNDIS_MAJOR_VERSION;
1117	init->minor_ver = RNDIS_MINOR_VERSION;
1118	init->max_xfer_size = 0x4000;
1119
1120	dev->state = RNDIS_DEV_INITIALIZING;
1121
1122	ret = rndis_filter_send_request(dev, request);
1123	if (ret != 0) {
1124		dev->state = RNDIS_DEV_UNINITIALIZED;
1125		goto cleanup;
1126	}
1127
1128	wait_for_completion(&request->wait_event);
1129
1130	init_complete = &request->response_msg.msg.init_complete;
1131	status = init_complete->status;
1132	if (status == RNDIS_STATUS_SUCCESS) {
1133		dev->state = RNDIS_DEV_INITIALIZED;
1134		nvdev->max_pkt = init_complete->max_pkt_per_msg;
1135		nvdev->pkt_align = 1 << init_complete->pkt_alignment_factor;
1136		ret = 0;
1137	} else {
1138		dev->state = RNDIS_DEV_UNINITIALIZED;
1139		ret = -EINVAL;
1140	}
1141
1142cleanup:
1143	if (request)
1144		put_rndis_request(dev, request);
1145
1146	return ret;
1147}
1148
1149static bool netvsc_device_idle(const struct netvsc_device *nvdev)
1150{
1151	int i;
1152
1153	for (i = 0; i < nvdev->num_chn; i++) {
1154		const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
1155
1156		if (nvchan->mrc.first != nvchan->mrc.next)
1157			return false;
1158
1159		if (atomic_read(&nvchan->queue_sends) > 0)
1160			return false;
1161	}
1162
1163	return true;
1164}
1165
1166static void rndis_filter_halt_device(struct netvsc_device *nvdev,
1167				     struct rndis_device *dev)
1168{
1169	struct rndis_request *request;
1170	struct rndis_halt_request *halt;
1171
1172	/* Attempt to do a rndis device halt */
1173	request = get_rndis_request(dev, RNDIS_MSG_HALT,
1174				RNDIS_MESSAGE_SIZE(struct rndis_halt_request));
1175	if (!request)
1176		goto cleanup;
1177
1178	/* Setup the rndis set */
1179	halt = &request->request_msg.msg.halt_req;
1180	halt->req_id = atomic_inc_return(&dev->new_req_id);
1181
1182	/* Ignore return since this msg is optional. */
1183	rndis_filter_send_request(dev, request);
1184
1185	dev->state = RNDIS_DEV_UNINITIALIZED;
1186
1187cleanup:
1188	nvdev->destroy = true;
1189
1190	/* Force flag to be ordered before waiting */
1191	wmb();
1192
1193	/* Wait for all send completions */
1194	wait_event(nvdev->wait_drain, netvsc_device_idle(nvdev));
1195
1196	if (request)
1197		put_rndis_request(dev, request);
1198}
1199
1200static int rndis_filter_open_device(struct rndis_device *dev)
1201{
1202	int ret;
1203
1204	if (dev->state != RNDIS_DEV_INITIALIZED)
1205		return 0;
1206
1207	ret = rndis_filter_set_packet_filter(dev,
1208					 NDIS_PACKET_TYPE_BROADCAST |
1209					 NDIS_PACKET_TYPE_ALL_MULTICAST |
1210					 NDIS_PACKET_TYPE_DIRECTED);
1211	if (ret == 0)
1212		dev->state = RNDIS_DEV_DATAINITIALIZED;
1213
1214	return ret;
1215}
1216
1217static int rndis_filter_close_device(struct rndis_device *dev)
1218{
1219	int ret;
1220
1221	if (dev->state != RNDIS_DEV_DATAINITIALIZED)
1222		return 0;
1223
1224	/* Make sure rndis_set_multicast doesn't re-enable filter! */
1225	cancel_work_sync(&dev->mcast_work);
1226
1227	ret = rndis_filter_set_packet_filter(dev, 0);
1228	if (ret == -ENODEV)
1229		ret = 0;
1230
1231	if (ret == 0)
1232		dev->state = RNDIS_DEV_INITIALIZED;
1233
1234	return ret;
1235}
1236
1237static void netvsc_sc_open(struct vmbus_channel *new_sc)
1238{
1239	struct net_device *ndev =
1240		hv_get_drvdata(new_sc->primary_channel->device_obj);
1241	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1242	struct netvsc_device *nvscdev;
1243	u16 chn_index = new_sc->offermsg.offer.sub_channel_index;
1244	struct netvsc_channel *nvchan;
1245	int ret;
1246
1247	/* This is safe because this callback only happens when
1248	 * new device is being setup and waiting on the channel_init_wait.
1249	 */
1250	nvscdev = rcu_dereference_raw(ndev_ctx->nvdev);
1251	if (!nvscdev || chn_index >= nvscdev->num_chn)
1252		return;
1253
1254	nvchan = nvscdev->chan_table + chn_index;
1255
1256	/* Because the device uses NAPI, all the interrupt batching and
1257	 * control is done via Net softirq, not the channel handling
1258	 */
1259	set_channel_read_mode(new_sc, HV_CALL_ISR);
1260
1261	/* Set the channel before opening.*/
1262	nvchan->channel = new_sc;
1263
1264	new_sc->next_request_id_callback = vmbus_next_request_id;
1265	new_sc->request_addr_callback = vmbus_request_addr;
1266	new_sc->rqstor_size = netvsc_rqstor_size(netvsc_ring_bytes);
1267	new_sc->max_pkt_size = NETVSC_MAX_PKT_SIZE;
1268
1269	ret = vmbus_open(new_sc, netvsc_ring_bytes,
1270			 netvsc_ring_bytes, NULL, 0,
1271			 netvsc_channel_cb, nvchan);
1272	if (ret == 0)
1273		napi_enable(&nvchan->napi);
1274	else
1275		netdev_notice(ndev, "sub channel open failed: %d\n", ret);
1276
1277	if (atomic_inc_return(&nvscdev->open_chn) == nvscdev->num_chn)
1278		wake_up(&nvscdev->subchan_open);
1279}
1280
1281/* Open sub-channels after completing the handling of the device probe.
1282 * This breaks overlap of processing the host message for the
1283 * new primary channel with the initialization of sub-channels.
1284 */
1285int rndis_set_subchannel(struct net_device *ndev,
1286			 struct netvsc_device *nvdev,
1287			 struct netvsc_device_info *dev_info)
1288{
1289	struct nvsp_message *init_packet = &nvdev->channel_init_pkt;
1290	struct net_device_context *ndev_ctx = netdev_priv(ndev);
1291	struct hv_device *hv_dev = ndev_ctx->device_ctx;
1292	struct rndis_device *rdev = nvdev->extension;
1293	int i, ret;
1294
1295	ASSERT_RTNL();
1296
1297	memset(init_packet, 0, sizeof(struct nvsp_message));
1298	init_packet->hdr.msg_type = NVSP_MSG5_TYPE_SUBCHANNEL;
1299	init_packet->msg.v5_msg.subchn_req.op = NVSP_SUBCHANNEL_ALLOCATE;
1300	init_packet->msg.v5_msg.subchn_req.num_subchannels =
1301						nvdev->num_chn - 1;
1302	trace_nvsp_send(ndev, init_packet);
1303
1304	ret = vmbus_sendpacket(hv_dev->channel, init_packet,
1305			       sizeof(struct nvsp_message),
1306			       (unsigned long)init_packet,
1307			       VM_PKT_DATA_INBAND,
1308			       VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1309	if (ret) {
1310		netdev_err(ndev, "sub channel allocate send failed: %d\n", ret);
1311		return ret;
1312	}
1313
1314	wait_for_completion(&nvdev->channel_init_wait);
1315	if (init_packet->msg.v5_msg.subchn_comp.status != NVSP_STAT_SUCCESS) {
1316		netdev_err(ndev, "sub channel request failed\n");
1317		return -EIO;
1318	}
1319
1320	/* Check that number of allocated sub channel is within the expected range */
1321	if (init_packet->msg.v5_msg.subchn_comp.num_subchannels > nvdev->num_chn - 1) {
1322		netdev_err(ndev, "invalid number of allocated sub channel\n");
1323		return -EINVAL;
1324	}
1325	nvdev->num_chn = 1 +
1326		init_packet->msg.v5_msg.subchn_comp.num_subchannels;
1327
1328	/* wait for all sub channels to open */
1329	wait_event(nvdev->subchan_open,
1330		   atomic_read(&nvdev->open_chn) == nvdev->num_chn);
1331
1332	for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1333		ndev_ctx->tx_table[i] = i % nvdev->num_chn;
1334
1335	/* ignore failures from setting rss parameters, still have channels */
1336	if (dev_info)
1337		rndis_filter_set_rss_param(rdev, dev_info->rss_key);
1338	else
1339		rndis_filter_set_rss_param(rdev, netvsc_hash_key);
1340
1341	netif_set_real_num_tx_queues(ndev, nvdev->num_chn);
1342	netif_set_real_num_rx_queues(ndev, nvdev->num_chn);
1343
1344	return 0;
1345}
1346
1347static int rndis_netdev_set_hwcaps(struct rndis_device *rndis_device,
1348				   struct netvsc_device *nvdev)
1349{
1350	struct net_device *net = rndis_device->ndev;
1351	struct net_device_context *net_device_ctx = netdev_priv(net);
1352	struct ndis_offload hwcaps;
1353	struct ndis_offload_params offloads;
1354	unsigned int gso_max_size = GSO_LEGACY_MAX_SIZE;
1355	int ret;
1356
1357	/* Find HW offload capabilities */
1358	ret = rndis_query_hwcaps(rndis_device, nvdev, &hwcaps);
1359	if (ret != 0)
1360		return ret;
1361
1362	/* A value of zero means "no change"; now turn on what we want. */
1363	memset(&offloads, 0, sizeof(struct ndis_offload_params));
1364
1365	/* Linux does not care about IP checksum, always does in kernel */
1366	offloads.ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_DISABLED;
1367
1368	/* Reset previously set hw_features flags */
1369	net->hw_features &= ~NETVSC_SUPPORTED_HW_FEATURES;
1370	net_device_ctx->tx_checksum_mask = 0;
1371
1372	/* Compute tx offload settings based on hw capabilities */
1373	net->hw_features |= NETIF_F_RXCSUM;
1374	net->hw_features |= NETIF_F_SG;
1375	net->hw_features |= NETIF_F_RXHASH;
1376
1377	if ((hwcaps.csum.ip4_txcsum & NDIS_TXCSUM_ALL_TCP4) == NDIS_TXCSUM_ALL_TCP4) {
1378		/* Can checksum TCP */
1379		net->hw_features |= NETIF_F_IP_CSUM;
1380		net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV4_TCP;
1381
1382		offloads.tcp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1383
1384		if (hwcaps.lsov2.ip4_encap & NDIS_OFFLOAD_ENCAP_8023) {
1385			offloads.lso_v2_ipv4 = NDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED;
1386			net->hw_features |= NETIF_F_TSO;
1387
1388			if (hwcaps.lsov2.ip4_maxsz < gso_max_size)
1389				gso_max_size = hwcaps.lsov2.ip4_maxsz;
1390		}
1391
1392		if (hwcaps.csum.ip4_txcsum & NDIS_TXCSUM_CAP_UDP4) {
1393			offloads.udp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1394			net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV4_UDP;
1395		}
1396	}
1397
1398	if ((hwcaps.csum.ip6_txcsum & NDIS_TXCSUM_ALL_TCP6) == NDIS_TXCSUM_ALL_TCP6) {
1399		net->hw_features |= NETIF_F_IPV6_CSUM;
1400
1401		offloads.tcp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1402		net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV6_TCP;
1403
1404		if ((hwcaps.lsov2.ip6_encap & NDIS_OFFLOAD_ENCAP_8023) &&
1405		    (hwcaps.lsov2.ip6_opts & NDIS_LSOV2_CAP_IP6) == NDIS_LSOV2_CAP_IP6) {
1406			offloads.lso_v2_ipv6 = NDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED;
1407			net->hw_features |= NETIF_F_TSO6;
1408
1409			if (hwcaps.lsov2.ip6_maxsz < gso_max_size)
1410				gso_max_size = hwcaps.lsov2.ip6_maxsz;
1411		}
1412
1413		if (hwcaps.csum.ip6_txcsum & NDIS_TXCSUM_CAP_UDP6) {
1414			offloads.udp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1415			net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV6_UDP;
1416		}
1417	}
1418
1419	if (hwcaps.rsc.ip4 && hwcaps.rsc.ip6) {
1420		net->hw_features |= NETIF_F_LRO;
1421
1422		if (net->features & NETIF_F_LRO) {
1423			offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1424			offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_ENABLED;
1425		} else {
1426			offloads.rsc_ip_v4 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1427			offloads.rsc_ip_v6 = NDIS_OFFLOAD_PARAMETERS_RSC_DISABLED;
1428		}
1429	}
1430
1431	/* In case some hw_features disappeared we need to remove them from
1432	 * net->features list as they're no longer supported.
1433	 */
1434	net->features &= ~NETVSC_SUPPORTED_HW_FEATURES | net->hw_features;
1435
1436	netif_set_tso_max_size(net, gso_max_size);
1437
1438	ret = rndis_filter_set_offload_params(net, nvdev, &offloads);
1439
1440	return ret;
1441}
1442
1443static void rndis_get_friendly_name(struct net_device *net,
1444				    struct rndis_device *rndis_device,
1445				    struct netvsc_device *net_device)
1446{
1447	ucs2_char_t wname[256];
1448	unsigned long len;
1449	u8 ifalias[256];
1450	u32 size;
1451
1452	size = sizeof(wname);
1453	if (rndis_filter_query_device(rndis_device, net_device,
1454				      RNDIS_OID_GEN_FRIENDLY_NAME,
1455				      wname, &size) != 0)
1456		return;	/* ignore if host does not support */
1457
1458	if (size == 0)
1459		return;	/* name not set */
1460
1461	/* Convert Windows Unicode string to UTF-8 */
1462	len = ucs2_as_utf8(ifalias, wname, sizeof(ifalias));
1463
1464	/* ignore the default value from host */
1465	if (strcmp(ifalias, "Network Adapter") != 0)
1466		dev_set_alias(net, ifalias, len);
1467}
1468
1469struct netvsc_device *rndis_filter_device_add(struct hv_device *dev,
1470				      struct netvsc_device_info *device_info)
1471{
1472	struct net_device *net = hv_get_drvdata(dev);
1473	struct net_device_context *ndc = netdev_priv(net);
1474	struct netvsc_device *net_device;
1475	struct rndis_device *rndis_device;
1476	struct ndis_recv_scale_cap rsscap;
1477	u32 rsscap_size = sizeof(struct ndis_recv_scale_cap);
1478	u32 mtu, size;
1479	u32 num_possible_rss_qs;
1480	int i, ret;
1481
1482	rndis_device = get_rndis_device();
1483	if (!rndis_device)
1484		return ERR_PTR(-ENODEV);
1485
1486	/* Let the inner driver handle this first to create the netvsc channel
1487	 * NOTE! Once the channel is created, we may get a receive callback
1488	 * (RndisFilterOnReceive()) before this call is completed
1489	 */
1490	net_device = netvsc_device_add(dev, device_info);
1491	if (IS_ERR(net_device)) {
1492		kfree(rndis_device);
1493		return net_device;
1494	}
1495
1496	/* Initialize the rndis device */
1497	net_device->max_chn = 1;
1498	net_device->num_chn = 1;
1499
1500	net_device->extension = rndis_device;
1501	rndis_device->ndev = net;
1502
1503	/* Send the rndis initialization message */
1504	ret = rndis_filter_init_device(rndis_device, net_device);
1505	if (ret != 0)
1506		goto err_dev_remv;
1507
1508	/* Get the MTU from the host */
1509	size = sizeof(u32);
1510	ret = rndis_filter_query_device(rndis_device, net_device,
1511					RNDIS_OID_GEN_MAXIMUM_FRAME_SIZE,
1512					&mtu, &size);
1513	if (ret == 0 && size == sizeof(u32) && mtu < net->mtu)
1514		net->mtu = mtu;
1515
1516	/* Get the mac address */
1517	ret = rndis_filter_query_device_mac(rndis_device, net_device);
1518	if (ret != 0)
1519		goto err_dev_remv;
1520
1521	memcpy(device_info->mac_adr, rndis_device->hw_mac_adr, ETH_ALEN);
1522
1523	/* Get friendly name as ifalias*/
1524	if (!net->ifalias)
1525		rndis_get_friendly_name(net, rndis_device, net_device);
1526
1527	/* Query and set hardware capabilities */
1528	ret = rndis_netdev_set_hwcaps(rndis_device, net_device);
1529	if (ret != 0)
1530		goto err_dev_remv;
1531
1532	rndis_filter_query_device_link_status(rndis_device, net_device);
1533
1534	netdev_dbg(net, "Device MAC %pM link state %s\n",
1535		   rndis_device->hw_mac_adr,
1536		   rndis_device->link_state ? "down" : "up");
1537
1538	if (net_device->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1539		goto out;
1540
1541	rndis_filter_query_link_speed(rndis_device, net_device);
1542
1543	/* vRSS setup */
1544	memset(&rsscap, 0, rsscap_size);
1545	ret = rndis_filter_query_device(rndis_device, net_device,
1546					OID_GEN_RECEIVE_SCALE_CAPABILITIES,
1547					&rsscap, &rsscap_size);
1548	if (ret || rsscap.num_recv_que < 2)
1549		goto out;
1550
1551	if (rsscap.num_indirect_tabent &&
1552	    rsscap.num_indirect_tabent <= ITAB_NUM_MAX)
1553		ndc->rx_table_sz = rsscap.num_indirect_tabent;
1554	else
1555		ndc->rx_table_sz = ITAB_NUM;
1556
1557	ndc->rx_table = kcalloc(ndc->rx_table_sz, sizeof(u16), GFP_KERNEL);
1558	if (!ndc->rx_table) {
1559		ret = -ENOMEM;
1560		goto err_dev_remv;
1561	}
1562
1563	/* This guarantees that num_possible_rss_qs <= num_online_cpus */
1564	num_possible_rss_qs = min_t(u32, num_online_cpus(),
1565				    rsscap.num_recv_que);
1566
1567	net_device->max_chn = min_t(u32, VRSS_CHANNEL_MAX, num_possible_rss_qs);
1568
1569	/* We will use the given number of channels if available. */
1570	net_device->num_chn = min(net_device->max_chn, device_info->num_chn);
1571
1572	if (!netif_is_rxfh_configured(net)) {
1573		for (i = 0; i < ndc->rx_table_sz; i++)
1574			ndc->rx_table[i] = ethtool_rxfh_indir_default(
1575						i, net_device->num_chn);
1576	}
1577
1578	atomic_set(&net_device->open_chn, 1);
1579	vmbus_set_sc_create_callback(dev->channel, netvsc_sc_open);
1580
1581	for (i = 1; i < net_device->num_chn; i++) {
1582		ret = netvsc_alloc_recv_comp_ring(net_device, i);
1583		if (ret) {
1584			while (--i != 0)
1585				vfree(net_device->chan_table[i].mrc.slots);
1586			goto out;
1587		}
1588	}
1589
1590	for (i = 1; i < net_device->num_chn; i++)
1591		netif_napi_add(net, &net_device->chan_table[i].napi,
1592			       netvsc_poll);
1593
1594	return net_device;
1595
1596out:
1597	/* setting up multiple channels failed */
1598	net_device->max_chn = 1;
1599	net_device->num_chn = 1;
1600	return net_device;
1601
1602err_dev_remv:
1603	rndis_filter_device_remove(dev, net_device);
1604	return ERR_PTR(ret);
1605}
1606
1607void rndis_filter_device_remove(struct hv_device *dev,
1608				struct netvsc_device *net_dev)
1609{
1610	struct rndis_device *rndis_dev = net_dev->extension;
1611	struct net_device *net = hv_get_drvdata(dev);
1612	struct net_device_context *ndc;
1613
1614	ndc = netdev_priv(net);
1615
1616	/* Halt and release the rndis device */
1617	rndis_filter_halt_device(net_dev, rndis_dev);
1618
1619	netvsc_device_remove(dev);
1620
1621	ndc->rx_table_sz = 0;
1622	kfree(ndc->rx_table);
1623	ndc->rx_table = NULL;
1624}
1625
1626int rndis_filter_open(struct netvsc_device *nvdev)
1627{
1628	if (!nvdev)
1629		return -EINVAL;
1630
1631	return rndis_filter_open_device(nvdev->extension);
1632}
1633
1634int rndis_filter_close(struct netvsc_device *nvdev)
1635{
1636	if (!nvdev)
1637		return -EINVAL;
1638
1639	return rndis_filter_close_device(nvdev->extension);
1640}
1641