2 * Copyright (c) 2009, Microsoft Corporation.
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.
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
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/>.
17 * Haiyang Zhang <haiyangz@microsoft.com>
18 * Hank Janssen <hjanssen@microsoft.com>
20 #include <linux/kernel.h>
21 #include <linux/sched.h>
22 #include <linux/wait.h>
23 #include <linux/highmem.h>
24 #include <linux/slab.h>
26 #include <linux/if_ether.h>
27 #include <linux/netdevice.h>
28 #include <linux/if_vlan.h>
29 #include <linux/nls.h>
30 #include <linux/vmalloc.h>
31 #include <linux/rtnetlink.h>
32 #include <linux/ucs2_string.h>
34 #include "hyperv_net.h"
35 #include "netvsc_trace.h"
37 static void rndis_set_multicast(struct work_struct *w);
39 #define RNDIS_EXT_LEN PAGE_SIZE
40 struct rndis_request {
41 struct list_head list_ent;
42 struct completion wait_event;
44 struct rndis_message response_msg;
46 * The buffer for extended info after the RNDIS response message. It's
47 * referenced based on the data offset in the RNDIS message. Its size
48 * is enough for current needs, and should be sufficient for the near
51 u8 response_ext[RNDIS_EXT_LEN];
53 /* Simplify allocation by having a netvsc packet inline */
54 struct hv_netvsc_packet pkt;
56 struct rndis_message request_msg;
58 * The buffer for the extended info after the RNDIS request message.
59 * It is referenced and sized in a similar way as response_ext.
61 u8 request_ext[RNDIS_EXT_LEN];
64 static const u8 netvsc_hash_key[NETVSC_HASH_KEYLEN] = {
65 0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2,
66 0x41, 0x67, 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0,
67 0xd0, 0xca, 0x2b, 0xcb, 0xae, 0x7b, 0x30, 0xb4,
68 0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30, 0xf2, 0x0c,
69 0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa
72 static struct rndis_device *get_rndis_device(void)
74 struct rndis_device *device;
76 device = kzalloc(sizeof(struct rndis_device), GFP_KERNEL);
80 spin_lock_init(&device->request_lock);
82 INIT_LIST_HEAD(&device->req_list);
83 INIT_WORK(&device->mcast_work, rndis_set_multicast);
85 device->state = RNDIS_DEV_UNINITIALIZED;
90 static struct rndis_request *get_rndis_request(struct rndis_device *dev,
94 struct rndis_request *request;
95 struct rndis_message *rndis_msg;
96 struct rndis_set_request *set;
99 request = kzalloc(sizeof(struct rndis_request), GFP_KERNEL);
103 init_completion(&request->wait_event);
105 rndis_msg = &request->request_msg;
106 rndis_msg->ndis_msg_type = msg_type;
107 rndis_msg->msg_len = msg_len;
109 request->pkt.q_idx = 0;
112 * Set the request id. This field is always after the rndis header for
113 * request/response packet types so we just used the SetRequest as a
116 set = &rndis_msg->msg.set_req;
117 set->req_id = atomic_inc_return(&dev->new_req_id);
119 /* Add to the request list */
120 spin_lock_irqsave(&dev->request_lock, flags);
121 list_add_tail(&request->list_ent, &dev->req_list);
122 spin_unlock_irqrestore(&dev->request_lock, flags);
127 static void put_rndis_request(struct rndis_device *dev,
128 struct rndis_request *req)
132 spin_lock_irqsave(&dev->request_lock, flags);
133 list_del(&req->list_ent);
134 spin_unlock_irqrestore(&dev->request_lock, flags);
139 static void dump_rndis_message(struct net_device *netdev,
140 const struct rndis_message *rndis_msg)
142 switch (rndis_msg->ndis_msg_type) {
143 case RNDIS_MSG_PACKET:
144 netdev_dbg(netdev, "RNDIS_MSG_PACKET (len %u, "
145 "data offset %u data len %u, # oob %u, "
146 "oob offset %u, oob len %u, pkt offset %u, "
149 rndis_msg->msg.pkt.data_offset,
150 rndis_msg->msg.pkt.data_len,
151 rndis_msg->msg.pkt.num_oob_data_elements,
152 rndis_msg->msg.pkt.oob_data_offset,
153 rndis_msg->msg.pkt.oob_data_len,
154 rndis_msg->msg.pkt.per_pkt_info_offset,
155 rndis_msg->msg.pkt.per_pkt_info_len);
158 case RNDIS_MSG_INIT_C:
159 netdev_dbg(netdev, "RNDIS_MSG_INIT_C "
160 "(len %u, id 0x%x, status 0x%x, major %d, minor %d, "
161 "device flags %d, max xfer size 0x%x, max pkts %u, "
164 rndis_msg->msg.init_complete.req_id,
165 rndis_msg->msg.init_complete.status,
166 rndis_msg->msg.init_complete.major_ver,
167 rndis_msg->msg.init_complete.minor_ver,
168 rndis_msg->msg.init_complete.dev_flags,
169 rndis_msg->msg.init_complete.max_xfer_size,
170 rndis_msg->msg.init_complete.
172 rndis_msg->msg.init_complete.
173 pkt_alignment_factor);
176 case RNDIS_MSG_QUERY_C:
177 netdev_dbg(netdev, "RNDIS_MSG_QUERY_C "
178 "(len %u, id 0x%x, status 0x%x, buf len %u, "
181 rndis_msg->msg.query_complete.req_id,
182 rndis_msg->msg.query_complete.status,
183 rndis_msg->msg.query_complete.
185 rndis_msg->msg.query_complete.
189 case RNDIS_MSG_SET_C:
191 "RNDIS_MSG_SET_C (len %u, id 0x%x, status 0x%x)\n",
193 rndis_msg->msg.set_complete.req_id,
194 rndis_msg->msg.set_complete.status);
197 case RNDIS_MSG_INDICATE:
198 netdev_dbg(netdev, "RNDIS_MSG_INDICATE "
199 "(len %u, status 0x%x, buf len %u, buf offset %u)\n",
201 rndis_msg->msg.indicate_status.status,
202 rndis_msg->msg.indicate_status.status_buflen,
203 rndis_msg->msg.indicate_status.status_buf_offset);
207 netdev_dbg(netdev, "0x%x (len %u)\n",
208 rndis_msg->ndis_msg_type,
214 static int rndis_filter_send_request(struct rndis_device *dev,
215 struct rndis_request *req)
217 struct hv_netvsc_packet *packet;
218 struct hv_page_buffer page_buf[2];
219 struct hv_page_buffer *pb = page_buf;
222 /* Setup the packet to send it */
225 packet->total_data_buflen = req->request_msg.msg_len;
226 packet->page_buf_cnt = 1;
228 pb[0].pfn = virt_to_phys(&req->request_msg) >>
230 pb[0].len = req->request_msg.msg_len;
232 (unsigned long)&req->request_msg & (PAGE_SIZE - 1);
234 /* Add one page_buf when request_msg crossing page boundary */
235 if (pb[0].offset + pb[0].len > PAGE_SIZE) {
236 packet->page_buf_cnt++;
237 pb[0].len = PAGE_SIZE -
239 pb[1].pfn = virt_to_phys((void *)&req->request_msg
240 + pb[0].len) >> PAGE_SHIFT;
242 pb[1].len = req->request_msg.msg_len -
246 trace_rndis_send(dev->ndev, 0, &req->request_msg);
249 ret = netvsc_send(dev->ndev, packet, NULL, pb, NULL);
250 rcu_read_unlock_bh();
255 static void rndis_set_link_state(struct rndis_device *rdev,
256 struct rndis_request *request)
259 struct rndis_query_complete *query_complete;
261 query_complete = &request->response_msg.msg.query_complete;
263 if (query_complete->status == RNDIS_STATUS_SUCCESS &&
264 query_complete->info_buflen == sizeof(u32)) {
265 memcpy(&link_status, (void *)((unsigned long)query_complete +
266 query_complete->info_buf_offset), sizeof(u32));
267 rdev->link_state = link_status != 0;
271 static void rndis_filter_receive_response(struct net_device *ndev,
272 struct netvsc_device *nvdev,
273 const struct rndis_message *resp)
275 struct rndis_device *dev = nvdev->extension;
276 struct rndis_request *request = NULL;
280 /* This should never happen, it means control message
281 * response received after device removed.
283 if (dev->state == RNDIS_DEV_UNINITIALIZED) {
285 "got rndis message uninitialized\n");
289 spin_lock_irqsave(&dev->request_lock, flags);
290 list_for_each_entry(request, &dev->req_list, list_ent) {
292 * All request/response message contains RequestId as the 1st
295 if (request->request_msg.msg.init_req.req_id
296 == resp->msg.init_complete.req_id) {
301 spin_unlock_irqrestore(&dev->request_lock, flags);
305 sizeof(struct rndis_message) + RNDIS_EXT_LEN) {
306 memcpy(&request->response_msg, resp,
308 if (request->request_msg.ndis_msg_type ==
309 RNDIS_MSG_QUERY && request->request_msg.msg.
310 query_req.oid == RNDIS_OID_GEN_MEDIA_CONNECT_STATUS)
311 rndis_set_link_state(dev, request);
314 "rndis response buffer overflow "
315 "detected (size %u max %zu)\n",
317 sizeof(struct rndis_message));
319 if (resp->ndis_msg_type ==
321 /* does not have a request id field */
322 request->response_msg.msg.reset_complete.
323 status = RNDIS_STATUS_BUFFER_OVERFLOW;
325 request->response_msg.msg.
326 init_complete.status =
327 RNDIS_STATUS_BUFFER_OVERFLOW;
331 complete(&request->wait_event);
334 "no rndis request found for this response "
335 "(id 0x%x res type 0x%x)\n",
336 resp->msg.init_complete.req_id,
337 resp->ndis_msg_type);
342 * Get the Per-Packet-Info with the specified type
343 * return NULL if not found.
345 static inline void *rndis_get_ppi(struct rndis_packet *rpkt, u32 type)
347 struct rndis_per_packet_info *ppi;
350 if (rpkt->per_pkt_info_offset == 0)
353 ppi = (struct rndis_per_packet_info *)((ulong)rpkt +
354 rpkt->per_pkt_info_offset);
355 len = rpkt->per_pkt_info_len;
358 if (ppi->type == type)
359 return (void *)((ulong)ppi + ppi->ppi_offset);
361 ppi = (struct rndis_per_packet_info *)((ulong)ppi + ppi->size);
367 static int rndis_filter_receive_data(struct net_device *ndev,
368 struct netvsc_device *nvdev,
369 struct vmbus_channel *channel,
370 struct rndis_message *msg,
373 struct rndis_packet *rndis_pkt = &msg->msg.pkt;
374 const struct ndis_tcp_ip_checksum_info *csum_info;
375 const struct ndis_pkt_8021q_info *vlan;
379 /* Remove the rndis header and pass it back up the stack */
380 data_offset = RNDIS_HEADER_SIZE + rndis_pkt->data_offset;
382 data_buflen -= data_offset;
385 * Make sure we got a valid RNDIS message, now total_data_buflen
386 * should be the data packet size plus the trailer padding size
388 if (unlikely(data_buflen < rndis_pkt->data_len)) {
389 netdev_err(ndev, "rndis message buffer "
390 "overflow detected (got %u, min %u)"
391 "...dropping this message!\n",
392 data_buflen, rndis_pkt->data_len);
393 return NVSP_STAT_FAIL;
396 vlan = rndis_get_ppi(rndis_pkt, IEEE_8021Q_INFO);
398 csum_info = rndis_get_ppi(rndis_pkt, TCPIP_CHKSUM_PKTINFO);
400 data = (void *)msg + data_offset;
403 * Remove the rndis trailer padding from rndis packet message
404 * rndis_pkt->data_len tell us the real data length, we only copy
405 * the data packet to the stack, without the rndis trailer padding
407 return netvsc_recv_callback(ndev, nvdev, channel,
408 data, rndis_pkt->data_len,
412 int rndis_filter_receive(struct net_device *ndev,
413 struct netvsc_device *net_dev,
414 struct vmbus_channel *channel,
415 void *data, u32 buflen)
417 struct net_device_context *net_device_ctx = netdev_priv(ndev);
418 struct rndis_message *rndis_msg = data;
420 if (netif_msg_rx_status(net_device_ctx))
421 dump_rndis_message(ndev, rndis_msg);
423 switch (rndis_msg->ndis_msg_type) {
424 case RNDIS_MSG_PACKET:
425 return rndis_filter_receive_data(ndev, net_dev, channel,
427 case RNDIS_MSG_INIT_C:
428 case RNDIS_MSG_QUERY_C:
429 case RNDIS_MSG_SET_C:
430 /* completion msgs */
431 rndis_filter_receive_response(ndev, net_dev, rndis_msg);
434 case RNDIS_MSG_INDICATE:
435 /* notification msgs */
436 netvsc_linkstatus_callback(ndev, rndis_msg);
440 "unhandled rndis message (type %u len %u)\n",
441 rndis_msg->ndis_msg_type,
443 return NVSP_STAT_FAIL;
446 return NVSP_STAT_SUCCESS;
449 static int rndis_filter_query_device(struct rndis_device *dev,
450 struct netvsc_device *nvdev,
451 u32 oid, void *result, u32 *result_size)
453 struct rndis_request *request;
454 u32 inresult_size = *result_size;
455 struct rndis_query_request *query;
456 struct rndis_query_complete *query_complete;
463 request = get_rndis_request(dev, RNDIS_MSG_QUERY,
464 RNDIS_MESSAGE_SIZE(struct rndis_query_request));
470 /* Setup the rndis query */
471 query = &request->request_msg.msg.query_req;
473 query->info_buf_offset = sizeof(struct rndis_query_request);
474 query->info_buflen = 0;
475 query->dev_vc_handle = 0;
477 if (oid == OID_TCP_OFFLOAD_HARDWARE_CAPABILITIES) {
478 struct ndis_offload *hwcaps;
479 u32 nvsp_version = nvdev->nvsp_version;
483 if (nvsp_version >= NVSP_PROTOCOL_VERSION_5) {
484 ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_3;
485 size = NDIS_OFFLOAD_SIZE;
486 } else if (nvsp_version >= NVSP_PROTOCOL_VERSION_4) {
487 ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_2;
488 size = NDIS_OFFLOAD_SIZE_6_1;
490 ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_1;
491 size = NDIS_OFFLOAD_SIZE_6_0;
494 request->request_msg.msg_len += size;
495 query->info_buflen = size;
496 hwcaps = (struct ndis_offload *)
497 ((unsigned long)query + query->info_buf_offset);
499 hwcaps->header.type = NDIS_OBJECT_TYPE_OFFLOAD;
500 hwcaps->header.revision = ndis_rev;
501 hwcaps->header.size = size;
503 } else if (oid == OID_GEN_RECEIVE_SCALE_CAPABILITIES) {
504 struct ndis_recv_scale_cap *cap;
506 request->request_msg.msg_len +=
507 sizeof(struct ndis_recv_scale_cap);
508 query->info_buflen = sizeof(struct ndis_recv_scale_cap);
509 cap = (struct ndis_recv_scale_cap *)((unsigned long)query +
510 query->info_buf_offset);
511 cap->hdr.type = NDIS_OBJECT_TYPE_RSS_CAPABILITIES;
512 cap->hdr.rev = NDIS_RECEIVE_SCALE_CAPABILITIES_REVISION_2;
513 cap->hdr.size = sizeof(struct ndis_recv_scale_cap);
516 ret = rndis_filter_send_request(dev, request);
520 wait_for_completion(&request->wait_event);
522 /* Copy the response back */
523 query_complete = &request->response_msg.msg.query_complete;
525 if (query_complete->info_buflen > inresult_size) {
531 (void *)((unsigned long)query_complete +
532 query_complete->info_buf_offset),
533 query_complete->info_buflen);
535 *result_size = query_complete->info_buflen;
539 put_rndis_request(dev, request);
544 /* Get the hardware offload capabilities */
546 rndis_query_hwcaps(struct rndis_device *dev, struct netvsc_device *net_device,
547 struct ndis_offload *caps)
549 u32 caps_len = sizeof(*caps);
552 memset(caps, 0, sizeof(*caps));
554 ret = rndis_filter_query_device(dev, net_device,
555 OID_TCP_OFFLOAD_HARDWARE_CAPABILITIES,
560 if (caps->header.type != NDIS_OBJECT_TYPE_OFFLOAD) {
561 netdev_warn(dev->ndev, "invalid NDIS objtype %#x\n",
566 if (caps->header.revision < NDIS_OFFLOAD_PARAMETERS_REVISION_1) {
567 netdev_warn(dev->ndev, "invalid NDIS objrev %x\n",
568 caps->header.revision);
572 if (caps->header.size > caps_len ||
573 caps->header.size < NDIS_OFFLOAD_SIZE_6_0) {
574 netdev_warn(dev->ndev,
575 "invalid NDIS objsize %u, data size %u\n",
576 caps->header.size, caps_len);
583 static int rndis_filter_query_device_mac(struct rndis_device *dev,
584 struct netvsc_device *net_device)
588 return rndis_filter_query_device(dev, net_device,
589 RNDIS_OID_802_3_PERMANENT_ADDRESS,
590 dev->hw_mac_adr, &size);
593 #define NWADR_STR "NetworkAddress"
594 #define NWADR_STRLEN 14
596 int rndis_filter_set_device_mac(struct netvsc_device *nvdev,
599 struct rndis_device *rdev = nvdev->extension;
600 struct rndis_request *request;
601 struct rndis_set_request *set;
602 struct rndis_config_parameter_info *cpi;
603 wchar_t *cfg_nwadr, *cfg_mac;
604 struct rndis_set_complete *set_complete;
605 char macstr[2*ETH_ALEN+1];
606 u32 extlen = sizeof(struct rndis_config_parameter_info) +
607 2*NWADR_STRLEN + 4*ETH_ALEN;
610 request = get_rndis_request(rdev, RNDIS_MSG_SET,
611 RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
615 set = &request->request_msg.msg.set_req;
616 set->oid = RNDIS_OID_GEN_RNDIS_CONFIG_PARAMETER;
617 set->info_buflen = extlen;
618 set->info_buf_offset = sizeof(struct rndis_set_request);
619 set->dev_vc_handle = 0;
621 cpi = (struct rndis_config_parameter_info *)((ulong)set +
622 set->info_buf_offset);
623 cpi->parameter_name_offset =
624 sizeof(struct rndis_config_parameter_info);
625 /* Multiply by 2 because host needs 2 bytes (utf16) for each char */
626 cpi->parameter_name_length = 2*NWADR_STRLEN;
627 cpi->parameter_type = RNDIS_CONFIG_PARAM_TYPE_STRING;
628 cpi->parameter_value_offset =
629 cpi->parameter_name_offset + cpi->parameter_name_length;
630 /* Multiply by 4 because each MAC byte displayed as 2 utf16 chars */
631 cpi->parameter_value_length = 4*ETH_ALEN;
633 cfg_nwadr = (wchar_t *)((ulong)cpi + cpi->parameter_name_offset);
634 cfg_mac = (wchar_t *)((ulong)cpi + cpi->parameter_value_offset);
635 ret = utf8s_to_utf16s(NWADR_STR, NWADR_STRLEN, UTF16_HOST_ENDIAN,
636 cfg_nwadr, NWADR_STRLEN);
639 snprintf(macstr, 2*ETH_ALEN+1, "%pm", mac);
640 ret = utf8s_to_utf16s(macstr, 2*ETH_ALEN, UTF16_HOST_ENDIAN,
641 cfg_mac, 2*ETH_ALEN);
645 ret = rndis_filter_send_request(rdev, request);
649 wait_for_completion(&request->wait_event);
651 set_complete = &request->response_msg.msg.set_complete;
652 if (set_complete->status != RNDIS_STATUS_SUCCESS)
656 put_rndis_request(rdev, request);
661 rndis_filter_set_offload_params(struct net_device *ndev,
662 struct netvsc_device *nvdev,
663 struct ndis_offload_params *req_offloads)
665 struct rndis_device *rdev = nvdev->extension;
666 struct rndis_request *request;
667 struct rndis_set_request *set;
668 struct ndis_offload_params *offload_params;
669 struct rndis_set_complete *set_complete;
670 u32 extlen = sizeof(struct ndis_offload_params);
672 u32 vsp_version = nvdev->nvsp_version;
674 if (vsp_version <= NVSP_PROTOCOL_VERSION_4) {
675 extlen = VERSION_4_OFFLOAD_SIZE;
676 /* On NVSP_PROTOCOL_VERSION_4 and below, we do not support
677 * UDP checksum offload.
679 req_offloads->udp_ip_v4_csum = 0;
680 req_offloads->udp_ip_v6_csum = 0;
683 request = get_rndis_request(rdev, RNDIS_MSG_SET,
684 RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
688 set = &request->request_msg.msg.set_req;
689 set->oid = OID_TCP_OFFLOAD_PARAMETERS;
690 set->info_buflen = extlen;
691 set->info_buf_offset = sizeof(struct rndis_set_request);
692 set->dev_vc_handle = 0;
694 offload_params = (struct ndis_offload_params *)((ulong)set +
695 set->info_buf_offset);
696 *offload_params = *req_offloads;
697 offload_params->header.type = NDIS_OBJECT_TYPE_DEFAULT;
698 offload_params->header.revision = NDIS_OFFLOAD_PARAMETERS_REVISION_3;
699 offload_params->header.size = extlen;
701 ret = rndis_filter_send_request(rdev, request);
705 wait_for_completion(&request->wait_event);
706 set_complete = &request->response_msg.msg.set_complete;
707 if (set_complete->status != RNDIS_STATUS_SUCCESS) {
708 netdev_err(ndev, "Fail to set offload on host side:0x%x\n",
709 set_complete->status);
714 put_rndis_request(rdev, request);
718 int rndis_filter_set_rss_param(struct rndis_device *rdev,
721 struct net_device *ndev = rdev->ndev;
722 struct rndis_request *request;
723 struct rndis_set_request *set;
724 struct rndis_set_complete *set_complete;
725 u32 extlen = sizeof(struct ndis_recv_scale_param) +
726 4 * ITAB_NUM + NETVSC_HASH_KEYLEN;
727 struct ndis_recv_scale_param *rssp;
732 request = get_rndis_request(
734 RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
738 set = &request->request_msg.msg.set_req;
739 set->oid = OID_GEN_RECEIVE_SCALE_PARAMETERS;
740 set->info_buflen = extlen;
741 set->info_buf_offset = sizeof(struct rndis_set_request);
742 set->dev_vc_handle = 0;
744 rssp = (struct ndis_recv_scale_param *)(set + 1);
745 rssp->hdr.type = NDIS_OBJECT_TYPE_RSS_PARAMETERS;
746 rssp->hdr.rev = NDIS_RECEIVE_SCALE_PARAMETERS_REVISION_2;
747 rssp->hdr.size = sizeof(struct ndis_recv_scale_param);
749 rssp->hashinfo = NDIS_HASH_FUNC_TOEPLITZ | NDIS_HASH_IPV4 |
750 NDIS_HASH_TCP_IPV4 | NDIS_HASH_IPV6 |
752 rssp->indirect_tabsize = 4*ITAB_NUM;
753 rssp->indirect_taboffset = sizeof(struct ndis_recv_scale_param);
754 rssp->hashkey_size = NETVSC_HASH_KEYLEN;
755 rssp->hashkey_offset = rssp->indirect_taboffset +
756 rssp->indirect_tabsize;
758 /* Set indirection table entries */
759 itab = (u32 *)(rssp + 1);
760 for (i = 0; i < ITAB_NUM; i++)
761 itab[i] = rdev->rx_table[i];
763 /* Set hask key values */
764 keyp = (u8 *)((unsigned long)rssp + rssp->hashkey_offset);
765 memcpy(keyp, rss_key, NETVSC_HASH_KEYLEN);
767 ret = rndis_filter_send_request(rdev, request);
771 wait_for_completion(&request->wait_event);
772 set_complete = &request->response_msg.msg.set_complete;
773 if (set_complete->status == RNDIS_STATUS_SUCCESS)
774 memcpy(rdev->rss_key, rss_key, NETVSC_HASH_KEYLEN);
776 netdev_err(ndev, "Fail to set RSS parameters:0x%x\n",
777 set_complete->status);
782 put_rndis_request(rdev, request);
786 static int rndis_filter_query_device_link_status(struct rndis_device *dev,
787 struct netvsc_device *net_device)
789 u32 size = sizeof(u32);
792 return rndis_filter_query_device(dev, net_device,
793 RNDIS_OID_GEN_MEDIA_CONNECT_STATUS,
794 &link_status, &size);
797 static int rndis_filter_query_link_speed(struct rndis_device *dev,
798 struct netvsc_device *net_device)
800 u32 size = sizeof(u32);
802 struct net_device_context *ndc;
805 ret = rndis_filter_query_device(dev, net_device,
806 RNDIS_OID_GEN_LINK_SPEED,
810 ndc = netdev_priv(dev->ndev);
812 /* The link speed reported from host is in 100bps unit, so
813 * we convert it to Mbps here.
815 ndc->speed = link_speed / 10000;
821 static int rndis_filter_set_packet_filter(struct rndis_device *dev,
824 struct rndis_request *request;
825 struct rndis_set_request *set;
828 if (dev->filter == new_filter)
831 request = get_rndis_request(dev, RNDIS_MSG_SET,
832 RNDIS_MESSAGE_SIZE(struct rndis_set_request) +
837 /* Setup the rndis set */
838 set = &request->request_msg.msg.set_req;
839 set->oid = RNDIS_OID_GEN_CURRENT_PACKET_FILTER;
840 set->info_buflen = sizeof(u32);
841 set->info_buf_offset = sizeof(struct rndis_set_request);
843 memcpy((void *)(unsigned long)set + sizeof(struct rndis_set_request),
844 &new_filter, sizeof(u32));
846 ret = rndis_filter_send_request(dev, request);
848 wait_for_completion(&request->wait_event);
849 dev->filter = new_filter;
852 put_rndis_request(dev, request);
857 static void rndis_set_multicast(struct work_struct *w)
859 struct rndis_device *rdev
860 = container_of(w, struct rndis_device, mcast_work);
861 u32 filter = NDIS_PACKET_TYPE_DIRECTED;
862 unsigned int flags = rdev->ndev->flags;
864 if (flags & IFF_PROMISC) {
865 filter = NDIS_PACKET_TYPE_PROMISCUOUS;
867 if (!netdev_mc_empty(rdev->ndev) || (flags & IFF_ALLMULTI))
868 filter |= NDIS_PACKET_TYPE_ALL_MULTICAST;
869 if (flags & IFF_BROADCAST)
870 filter |= NDIS_PACKET_TYPE_BROADCAST;
873 rndis_filter_set_packet_filter(rdev, filter);
876 void rndis_filter_update(struct netvsc_device *nvdev)
878 struct rndis_device *rdev = nvdev->extension;
880 schedule_work(&rdev->mcast_work);
883 static int rndis_filter_init_device(struct rndis_device *dev,
884 struct netvsc_device *nvdev)
886 struct rndis_request *request;
887 struct rndis_initialize_request *init;
888 struct rndis_initialize_complete *init_complete;
892 request = get_rndis_request(dev, RNDIS_MSG_INIT,
893 RNDIS_MESSAGE_SIZE(struct rndis_initialize_request));
899 /* Setup the rndis set */
900 init = &request->request_msg.msg.init_req;
901 init->major_ver = RNDIS_MAJOR_VERSION;
902 init->minor_ver = RNDIS_MINOR_VERSION;
903 init->max_xfer_size = 0x4000;
905 dev->state = RNDIS_DEV_INITIALIZING;
907 ret = rndis_filter_send_request(dev, request);
909 dev->state = RNDIS_DEV_UNINITIALIZED;
913 wait_for_completion(&request->wait_event);
915 init_complete = &request->response_msg.msg.init_complete;
916 status = init_complete->status;
917 if (status == RNDIS_STATUS_SUCCESS) {
918 dev->state = RNDIS_DEV_INITIALIZED;
919 nvdev->max_pkt = init_complete->max_pkt_per_msg;
920 nvdev->pkt_align = 1 << init_complete->pkt_alignment_factor;
923 dev->state = RNDIS_DEV_UNINITIALIZED;
929 put_rndis_request(dev, request);
934 static bool netvsc_device_idle(const struct netvsc_device *nvdev)
938 for (i = 0; i < nvdev->num_chn; i++) {
939 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
941 if (nvchan->mrc.first != nvchan->mrc.next)
944 if (atomic_read(&nvchan->queue_sends) > 0)
951 static void rndis_filter_halt_device(struct netvsc_device *nvdev,
952 struct rndis_device *dev)
954 struct rndis_request *request;
955 struct rndis_halt_request *halt;
957 /* Attempt to do a rndis device halt */
958 request = get_rndis_request(dev, RNDIS_MSG_HALT,
959 RNDIS_MESSAGE_SIZE(struct rndis_halt_request));
963 /* Setup the rndis set */
964 halt = &request->request_msg.msg.halt_req;
965 halt->req_id = atomic_inc_return(&dev->new_req_id);
967 /* Ignore return since this msg is optional. */
968 rndis_filter_send_request(dev, request);
970 dev->state = RNDIS_DEV_UNINITIALIZED;
973 nvdev->destroy = true;
975 /* Force flag to be ordered before waiting */
978 /* Wait for all send completions */
979 wait_event(nvdev->wait_drain, netvsc_device_idle(nvdev));
982 put_rndis_request(dev, request);
985 static int rndis_filter_open_device(struct rndis_device *dev)
989 if (dev->state != RNDIS_DEV_INITIALIZED)
992 ret = rndis_filter_set_packet_filter(dev,
993 NDIS_PACKET_TYPE_BROADCAST |
994 NDIS_PACKET_TYPE_ALL_MULTICAST |
995 NDIS_PACKET_TYPE_DIRECTED);
997 dev->state = RNDIS_DEV_DATAINITIALIZED;
1002 static int rndis_filter_close_device(struct rndis_device *dev)
1006 if (dev->state != RNDIS_DEV_DATAINITIALIZED)
1009 /* Make sure rndis_set_multicast doesn't re-enable filter! */
1010 cancel_work_sync(&dev->mcast_work);
1012 ret = rndis_filter_set_packet_filter(dev, 0);
1017 dev->state = RNDIS_DEV_INITIALIZED;
1022 static void netvsc_sc_open(struct vmbus_channel *new_sc)
1024 struct net_device *ndev =
1025 hv_get_drvdata(new_sc->primary_channel->device_obj);
1026 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1027 struct netvsc_device *nvscdev;
1028 u16 chn_index = new_sc->offermsg.offer.sub_channel_index;
1029 struct netvsc_channel *nvchan;
1032 /* This is safe because this callback only happens when
1033 * new device is being setup and waiting on the channel_init_wait.
1035 nvscdev = rcu_dereference_raw(ndev_ctx->nvdev);
1036 if (!nvscdev || chn_index >= nvscdev->num_chn)
1039 nvchan = nvscdev->chan_table + chn_index;
1041 /* Because the device uses NAPI, all the interrupt batching and
1042 * control is done via Net softirq, not the channel handling
1044 set_channel_read_mode(new_sc, HV_CALL_ISR);
1046 /* Set the channel before opening.*/
1047 nvchan->channel = new_sc;
1049 ret = vmbus_open(new_sc, netvsc_ring_bytes,
1050 netvsc_ring_bytes, NULL, 0,
1051 netvsc_channel_cb, nvchan);
1053 napi_enable(&nvchan->napi);
1055 netdev_notice(ndev, "sub channel open failed: %d\n", ret);
1057 if (atomic_inc_return(&nvscdev->open_chn) == nvscdev->num_chn)
1058 wake_up(&nvscdev->subchan_open);
1061 /* Open sub-channels after completing the handling of the device probe.
1062 * This breaks overlap of processing the host message for the
1063 * new primary channel with the initialization of sub-channels.
1065 void rndis_set_subchannel(struct work_struct *w)
1067 struct netvsc_device *nvdev
1068 = container_of(w, struct netvsc_device, subchan_work);
1069 struct nvsp_message *init_packet = &nvdev->channel_init_pkt;
1070 struct net_device_context *ndev_ctx;
1071 struct rndis_device *rdev;
1072 struct net_device *ndev;
1073 struct hv_device *hv_dev;
1076 if (!rtnl_trylock()) {
1081 rdev = nvdev->extension;
1083 goto unlock; /* device was removed */
1086 ndev_ctx = netdev_priv(ndev);
1087 hv_dev = ndev_ctx->device_ctx;
1089 memset(init_packet, 0, sizeof(struct nvsp_message));
1090 init_packet->hdr.msg_type = NVSP_MSG5_TYPE_SUBCHANNEL;
1091 init_packet->msg.v5_msg.subchn_req.op = NVSP_SUBCHANNEL_ALLOCATE;
1092 init_packet->msg.v5_msg.subchn_req.num_subchannels =
1094 trace_nvsp_send(ndev, init_packet);
1096 ret = vmbus_sendpacket(hv_dev->channel, init_packet,
1097 sizeof(struct nvsp_message),
1098 (unsigned long)init_packet,
1100 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1102 netdev_err(ndev, "sub channel allocate send failed: %d\n", ret);
1106 wait_for_completion(&nvdev->channel_init_wait);
1107 if (init_packet->msg.v5_msg.subchn_comp.status != NVSP_STAT_SUCCESS) {
1108 netdev_err(ndev, "sub channel request failed\n");
1112 nvdev->num_chn = 1 +
1113 init_packet->msg.v5_msg.subchn_comp.num_subchannels;
1115 /* wait for all sub channels to open */
1116 wait_event(nvdev->subchan_open,
1117 atomic_read(&nvdev->open_chn) == nvdev->num_chn);
1119 /* ignore failues from setting rss parameters, still have channels */
1120 rndis_filter_set_rss_param(rdev, netvsc_hash_key);
1122 netif_set_real_num_tx_queues(ndev, nvdev->num_chn);
1123 netif_set_real_num_rx_queues(ndev, nvdev->num_chn);
1125 for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1126 ndev_ctx->tx_table[i] = i % nvdev->num_chn;
1128 netif_device_attach(ndev);
1133 /* fallback to only primary channel */
1134 for (i = 1; i < nvdev->num_chn; i++)
1135 netif_napi_del(&nvdev->chan_table[i].napi);
1140 netif_device_attach(ndev);
1145 static int rndis_netdev_set_hwcaps(struct rndis_device *rndis_device,
1146 struct netvsc_device *nvdev)
1148 struct net_device *net = rndis_device->ndev;
1149 struct net_device_context *net_device_ctx = netdev_priv(net);
1150 struct ndis_offload hwcaps;
1151 struct ndis_offload_params offloads;
1152 unsigned int gso_max_size = GSO_MAX_SIZE;
1155 /* Find HW offload capabilities */
1156 ret = rndis_query_hwcaps(rndis_device, nvdev, &hwcaps);
1160 /* A value of zero means "no change"; now turn on what we want. */
1161 memset(&offloads, 0, sizeof(struct ndis_offload_params));
1163 /* Linux does not care about IP checksum, always does in kernel */
1164 offloads.ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_DISABLED;
1166 /* Reset previously set hw_features flags */
1167 net->hw_features &= ~NETVSC_SUPPORTED_HW_FEATURES;
1168 net_device_ctx->tx_checksum_mask = 0;
1170 /* Compute tx offload settings based on hw capabilities */
1171 net->hw_features |= NETIF_F_RXCSUM;
1173 if ((hwcaps.csum.ip4_txcsum & NDIS_TXCSUM_ALL_TCP4) == NDIS_TXCSUM_ALL_TCP4) {
1174 /* Can checksum TCP */
1175 net->hw_features |= NETIF_F_IP_CSUM;
1176 net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV4_TCP;
1178 offloads.tcp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1180 if (hwcaps.lsov2.ip4_encap & NDIS_OFFLOAD_ENCAP_8023) {
1181 offloads.lso_v2_ipv4 = NDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED;
1182 net->hw_features |= NETIF_F_TSO;
1184 if (hwcaps.lsov2.ip4_maxsz < gso_max_size)
1185 gso_max_size = hwcaps.lsov2.ip4_maxsz;
1188 if (hwcaps.csum.ip4_txcsum & NDIS_TXCSUM_CAP_UDP4) {
1189 offloads.udp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1190 net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV4_UDP;
1194 if ((hwcaps.csum.ip6_txcsum & NDIS_TXCSUM_ALL_TCP6) == NDIS_TXCSUM_ALL_TCP6) {
1195 net->hw_features |= NETIF_F_IPV6_CSUM;
1197 offloads.tcp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1198 net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV6_TCP;
1200 if ((hwcaps.lsov2.ip6_encap & NDIS_OFFLOAD_ENCAP_8023) &&
1201 (hwcaps.lsov2.ip6_opts & NDIS_LSOV2_CAP_IP6) == NDIS_LSOV2_CAP_IP6) {
1202 offloads.lso_v2_ipv6 = NDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED;
1203 net->hw_features |= NETIF_F_TSO6;
1205 if (hwcaps.lsov2.ip6_maxsz < gso_max_size)
1206 gso_max_size = hwcaps.lsov2.ip6_maxsz;
1209 if (hwcaps.csum.ip6_txcsum & NDIS_TXCSUM_CAP_UDP6) {
1210 offloads.udp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1211 net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV6_UDP;
1215 /* In case some hw_features disappeared we need to remove them from
1216 * net->features list as they're no longer supported.
1218 net->features &= ~NETVSC_SUPPORTED_HW_FEATURES | net->hw_features;
1220 netif_set_gso_max_size(net, gso_max_size);
1222 ret = rndis_filter_set_offload_params(net, nvdev, &offloads);
1227 static void rndis_get_friendly_name(struct net_device *net,
1228 struct rndis_device *rndis_device,
1229 struct netvsc_device *net_device)
1231 ucs2_char_t wname[256];
1236 size = sizeof(wname);
1237 if (rndis_filter_query_device(rndis_device, net_device,
1238 RNDIS_OID_GEN_FRIENDLY_NAME,
1240 return; /* ignore if host does not support */
1243 return; /* name not set */
1245 /* Convert Windows Unicode string to UTF-8 */
1246 len = ucs2_as_utf8(ifalias, wname, sizeof(ifalias));
1248 /* ignore the default value from host */
1249 if (strcmp(ifalias, "Network Adapter") != 0)
1250 dev_set_alias(net, ifalias, len);
1253 struct netvsc_device *rndis_filter_device_add(struct hv_device *dev,
1254 struct netvsc_device_info *device_info)
1256 struct net_device *net = hv_get_drvdata(dev);
1257 struct netvsc_device *net_device;
1258 struct rndis_device *rndis_device;
1259 struct ndis_recv_scale_cap rsscap;
1260 u32 rsscap_size = sizeof(struct ndis_recv_scale_cap);
1262 u32 num_possible_rss_qs;
1265 rndis_device = get_rndis_device();
1267 return ERR_PTR(-ENODEV);
1269 /* Let the inner driver handle this first to create the netvsc channel
1270 * NOTE! Once the channel is created, we may get a receive callback
1271 * (RndisFilterOnReceive()) before this call is completed
1273 net_device = netvsc_device_add(dev, device_info);
1274 if (IS_ERR(net_device)) {
1275 kfree(rndis_device);
1279 /* Initialize the rndis device */
1280 net_device->max_chn = 1;
1281 net_device->num_chn = 1;
1283 net_device->extension = rndis_device;
1284 rndis_device->ndev = net;
1286 /* Send the rndis initialization message */
1287 ret = rndis_filter_init_device(rndis_device, net_device);
1291 /* Get the MTU from the host */
1293 ret = rndis_filter_query_device(rndis_device, net_device,
1294 RNDIS_OID_GEN_MAXIMUM_FRAME_SIZE,
1296 if (ret == 0 && size == sizeof(u32) && mtu < net->mtu)
1299 /* Get the mac address */
1300 ret = rndis_filter_query_device_mac(rndis_device, net_device);
1304 memcpy(device_info->mac_adr, rndis_device->hw_mac_adr, ETH_ALEN);
1306 /* Get friendly name as ifalias*/
1308 rndis_get_friendly_name(net, rndis_device, net_device);
1310 /* Query and set hardware capabilities */
1311 ret = rndis_netdev_set_hwcaps(rndis_device, net_device);
1315 rndis_filter_query_device_link_status(rndis_device, net_device);
1317 netdev_dbg(net, "Device MAC %pM link state %s\n",
1318 rndis_device->hw_mac_adr,
1319 rndis_device->link_state ? "down" : "up");
1321 if (net_device->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1324 rndis_filter_query_link_speed(rndis_device, net_device);
1327 memset(&rsscap, 0, rsscap_size);
1328 ret = rndis_filter_query_device(rndis_device, net_device,
1329 OID_GEN_RECEIVE_SCALE_CAPABILITIES,
1330 &rsscap, &rsscap_size);
1331 if (ret || rsscap.num_recv_que < 2)
1334 /* This guarantees that num_possible_rss_qs <= num_online_cpus */
1335 num_possible_rss_qs = min_t(u32, num_online_cpus(),
1336 rsscap.num_recv_que);
1338 net_device->max_chn = min_t(u32, VRSS_CHANNEL_MAX, num_possible_rss_qs);
1340 /* We will use the given number of channels if available. */
1341 net_device->num_chn = min(net_device->max_chn, device_info->num_chn);
1343 for (i = 0; i < ITAB_NUM; i++)
1344 rndis_device->rx_table[i] = ethtool_rxfh_indir_default(
1345 i, net_device->num_chn);
1347 atomic_set(&net_device->open_chn, 1);
1348 vmbus_set_sc_create_callback(dev->channel, netvsc_sc_open);
1350 for (i = 1; i < net_device->num_chn; i++) {
1351 ret = netvsc_alloc_recv_comp_ring(net_device, i);
1354 vfree(net_device->chan_table[i].mrc.slots);
1359 for (i = 1; i < net_device->num_chn; i++)
1360 netif_napi_add(net, &net_device->chan_table[i].napi,
1361 netvsc_poll, NAPI_POLL_WEIGHT);
1363 if (net_device->num_chn > 1)
1364 schedule_work(&net_device->subchan_work);
1367 /* if unavailable, just proceed with one queue */
1369 net_device->max_chn = 1;
1370 net_device->num_chn = 1;
1373 /* No sub channels, device is ready */
1374 if (net_device->num_chn == 1)
1375 netif_device_attach(net);
1380 rndis_filter_device_remove(dev, net_device);
1381 return ERR_PTR(ret);
1384 void rndis_filter_device_remove(struct hv_device *dev,
1385 struct netvsc_device *net_dev)
1387 struct rndis_device *rndis_dev = net_dev->extension;
1389 /* Halt and release the rndis device */
1390 rndis_filter_halt_device(net_dev, rndis_dev);
1392 net_dev->extension = NULL;
1394 netvsc_device_remove(dev);
1397 int rndis_filter_open(struct netvsc_device *nvdev)
1402 return rndis_filter_open_device(nvdev->extension);
1405 int rndis_filter_close(struct netvsc_device *nvdev)
1410 return rndis_filter_close_device(nvdev->extension);