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/>.
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>
33 #include "hyperv_net.h"
35 static void rndis_set_multicast(struct work_struct *w);
37 #define RNDIS_EXT_LEN PAGE_SIZE
38 struct rndis_request {
39 struct list_head list_ent;
40 struct completion wait_event;
42 struct rndis_message response_msg;
44 * The buffer for extended info after the RNDIS response message. It's
45 * referenced based on the data offset in the RNDIS message. Its size
46 * is enough for current needs, and should be sufficient for the near
49 u8 response_ext[RNDIS_EXT_LEN];
51 /* Simplify allocation by having a netvsc packet inline */
52 struct hv_netvsc_packet pkt;
54 struct rndis_message request_msg;
56 * The buffer for the extended info after the RNDIS request message.
57 * It is referenced and sized in a similar way as response_ext.
59 u8 request_ext[RNDIS_EXT_LEN];
62 static const u8 netvsc_hash_key[NETVSC_HASH_KEYLEN] = {
63 0x6d, 0x5a, 0x56, 0xda, 0x25, 0x5b, 0x0e, 0xc2,
64 0x41, 0x67, 0x25, 0x3d, 0x43, 0xa3, 0x8f, 0xb0,
65 0xd0, 0xca, 0x2b, 0xcb, 0xae, 0x7b, 0x30, 0xb4,
66 0x77, 0xcb, 0x2d, 0xa3, 0x80, 0x30, 0xf2, 0x0c,
67 0x6a, 0x42, 0xb7, 0x3b, 0xbe, 0xac, 0x01, 0xfa
70 static struct rndis_device *get_rndis_device(void)
72 struct rndis_device *device;
74 device = kzalloc(sizeof(struct rndis_device), GFP_KERNEL);
78 spin_lock_init(&device->request_lock);
80 INIT_LIST_HEAD(&device->req_list);
81 INIT_WORK(&device->mcast_work, rndis_set_multicast);
83 device->state = RNDIS_DEV_UNINITIALIZED;
88 static struct rndis_request *get_rndis_request(struct rndis_device *dev,
92 struct rndis_request *request;
93 struct rndis_message *rndis_msg;
94 struct rndis_set_request *set;
97 request = kzalloc(sizeof(struct rndis_request), GFP_KERNEL);
101 init_completion(&request->wait_event);
103 rndis_msg = &request->request_msg;
104 rndis_msg->ndis_msg_type = msg_type;
105 rndis_msg->msg_len = msg_len;
107 request->pkt.q_idx = 0;
110 * Set the request id. This field is always after the rndis header for
111 * request/response packet types so we just used the SetRequest as a
114 set = &rndis_msg->msg.set_req;
115 set->req_id = atomic_inc_return(&dev->new_req_id);
117 /* Add to the request list */
118 spin_lock_irqsave(&dev->request_lock, flags);
119 list_add_tail(&request->list_ent, &dev->req_list);
120 spin_unlock_irqrestore(&dev->request_lock, flags);
125 static void put_rndis_request(struct rndis_device *dev,
126 struct rndis_request *req)
130 spin_lock_irqsave(&dev->request_lock, flags);
131 list_del(&req->list_ent);
132 spin_unlock_irqrestore(&dev->request_lock, flags);
137 static void dump_rndis_message(struct hv_device *hv_dev,
138 const struct rndis_message *rndis_msg)
140 struct net_device *netdev = hv_get_drvdata(hv_dev);
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;
220 struct net_device_context *net_device_ctx = netdev_priv(dev->ndev);
223 /* Setup the packet to send it */
226 packet->total_data_buflen = req->request_msg.msg_len;
227 packet->page_buf_cnt = 1;
229 pb[0].pfn = virt_to_phys(&req->request_msg) >>
231 pb[0].len = req->request_msg.msg_len;
233 (unsigned long)&req->request_msg & (PAGE_SIZE - 1);
235 /* Add one page_buf when request_msg crossing page boundary */
236 if (pb[0].offset + pb[0].len > PAGE_SIZE) {
237 packet->page_buf_cnt++;
238 pb[0].len = PAGE_SIZE -
240 pb[1].pfn = virt_to_phys((void *)&req->request_msg
241 + pb[0].len) >> PAGE_SHIFT;
243 pb[1].len = req->request_msg.msg_len -
248 ret = netvsc_send(net_device_ctx, packet, NULL, pb, NULL);
249 rcu_read_unlock_bh();
254 static void rndis_set_link_state(struct rndis_device *rdev,
255 struct rndis_request *request)
258 struct rndis_query_complete *query_complete;
260 query_complete = &request->response_msg.msg.query_complete;
262 if (query_complete->status == RNDIS_STATUS_SUCCESS &&
263 query_complete->info_buflen == sizeof(u32)) {
264 memcpy(&link_status, (void *)((unsigned long)query_complete +
265 query_complete->info_buf_offset), sizeof(u32));
266 rdev->link_state = link_status != 0;
270 static void rndis_filter_receive_response(struct rndis_device *dev,
271 struct rndis_message *resp)
273 struct rndis_request *request = NULL;
276 struct net_device *ndev = dev->ndev;
278 spin_lock_irqsave(&dev->request_lock, flags);
279 list_for_each_entry(request, &dev->req_list, list_ent) {
281 * All request/response message contains RequestId as the 1st
284 if (request->request_msg.msg.init_req.req_id
285 == resp->msg.init_complete.req_id) {
290 spin_unlock_irqrestore(&dev->request_lock, flags);
294 sizeof(struct rndis_message) + RNDIS_EXT_LEN) {
295 memcpy(&request->response_msg, resp,
297 if (request->request_msg.ndis_msg_type ==
298 RNDIS_MSG_QUERY && request->request_msg.msg.
299 query_req.oid == RNDIS_OID_GEN_MEDIA_CONNECT_STATUS)
300 rndis_set_link_state(dev, request);
303 "rndis response buffer overflow "
304 "detected (size %u max %zu)\n",
306 sizeof(struct rndis_message));
308 if (resp->ndis_msg_type ==
310 /* does not have a request id field */
311 request->response_msg.msg.reset_complete.
312 status = RNDIS_STATUS_BUFFER_OVERFLOW;
314 request->response_msg.msg.
315 init_complete.status =
316 RNDIS_STATUS_BUFFER_OVERFLOW;
320 complete(&request->wait_event);
323 "no rndis request found for this response "
324 "(id 0x%x res type 0x%x)\n",
325 resp->msg.init_complete.req_id,
326 resp->ndis_msg_type);
331 * Get the Per-Packet-Info with the specified type
332 * return NULL if not found.
334 static inline void *rndis_get_ppi(struct rndis_packet *rpkt, u32 type)
336 struct rndis_per_packet_info *ppi;
339 if (rpkt->per_pkt_info_offset == 0)
342 ppi = (struct rndis_per_packet_info *)((ulong)rpkt +
343 rpkt->per_pkt_info_offset);
344 len = rpkt->per_pkt_info_len;
347 if (ppi->type == type)
348 return (void *)((ulong)ppi + ppi->ppi_offset);
350 ppi = (struct rndis_per_packet_info *)((ulong)ppi + ppi->size);
356 static int rndis_filter_receive_data(struct net_device *ndev,
357 struct rndis_device *dev,
358 struct rndis_message *msg,
359 struct vmbus_channel *channel,
360 void *data, u32 data_buflen)
362 struct rndis_packet *rndis_pkt = &msg->msg.pkt;
363 const struct ndis_tcp_ip_checksum_info *csum_info;
364 const struct ndis_pkt_8021q_info *vlan;
367 /* Remove the rndis header and pass it back up the stack */
368 data_offset = RNDIS_HEADER_SIZE + rndis_pkt->data_offset;
370 data_buflen -= data_offset;
373 * Make sure we got a valid RNDIS message, now total_data_buflen
374 * should be the data packet size plus the trailer padding size
376 if (unlikely(data_buflen < rndis_pkt->data_len)) {
377 netdev_err(dev->ndev, "rndis message buffer "
378 "overflow detected (got %u, min %u)"
379 "...dropping this message!\n",
380 data_buflen, rndis_pkt->data_len);
381 return NVSP_STAT_FAIL;
384 vlan = rndis_get_ppi(rndis_pkt, IEEE_8021Q_INFO);
387 * Remove the rndis trailer padding from rndis packet message
388 * rndis_pkt->data_len tell us the real data length, we only copy
389 * the data packet to the stack, without the rndis trailer padding
391 data = (void *)((unsigned long)data + data_offset);
392 csum_info = rndis_get_ppi(rndis_pkt, TCPIP_CHKSUM_PKTINFO);
393 return netvsc_recv_callback(ndev, channel,
394 data, rndis_pkt->data_len,
398 int rndis_filter_receive(struct net_device *ndev,
399 struct netvsc_device *net_dev,
400 struct hv_device *dev,
401 struct vmbus_channel *channel,
402 void *data, u32 buflen)
404 struct net_device_context *net_device_ctx = netdev_priv(ndev);
405 struct rndis_device *rndis_dev = net_dev->extension;
406 struct rndis_message *rndis_msg = data;
408 /* Make sure the rndis device state is initialized */
409 if (unlikely(!rndis_dev)) {
410 netif_dbg(net_device_ctx, rx_err, ndev,
411 "got rndis message but no rndis device!\n");
412 return NVSP_STAT_FAIL;
415 if (unlikely(rndis_dev->state == RNDIS_DEV_UNINITIALIZED)) {
416 netif_dbg(net_device_ctx, rx_err, ndev,
417 "got rndis message uninitialized\n");
418 return NVSP_STAT_FAIL;
421 if (netif_msg_rx_status(net_device_ctx))
422 dump_rndis_message(dev, rndis_msg);
424 switch (rndis_msg->ndis_msg_type) {
425 case RNDIS_MSG_PACKET:
426 return rndis_filter_receive_data(ndev, rndis_dev, rndis_msg,
427 channel, data, buflen);
428 case RNDIS_MSG_INIT_C:
429 case RNDIS_MSG_QUERY_C:
430 case RNDIS_MSG_SET_C:
431 /* completion msgs */
432 rndis_filter_receive_response(rndis_dev, rndis_msg);
435 case RNDIS_MSG_INDICATE:
436 /* notification msgs */
437 netvsc_linkstatus_callback(dev, rndis_msg);
441 "unhandled rndis message (type %u len %u)\n",
442 rndis_msg->ndis_msg_type,
450 static int rndis_filter_query_device(struct rndis_device *dev,
451 struct netvsc_device *nvdev,
452 u32 oid, void *result, u32 *result_size)
454 struct rndis_request *request;
455 u32 inresult_size = *result_size;
456 struct rndis_query_request *query;
457 struct rndis_query_complete *query_complete;
464 request = get_rndis_request(dev, RNDIS_MSG_QUERY,
465 RNDIS_MESSAGE_SIZE(struct rndis_query_request));
471 /* Setup the rndis query */
472 query = &request->request_msg.msg.query_req;
474 query->info_buf_offset = sizeof(struct rndis_query_request);
475 query->info_buflen = 0;
476 query->dev_vc_handle = 0;
478 if (oid == OID_TCP_OFFLOAD_HARDWARE_CAPABILITIES) {
479 struct ndis_offload *hwcaps;
480 u32 nvsp_version = nvdev->nvsp_version;
484 if (nvsp_version >= NVSP_PROTOCOL_VERSION_5) {
485 ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_3;
486 size = NDIS_OFFLOAD_SIZE;
487 } else if (nvsp_version >= NVSP_PROTOCOL_VERSION_4) {
488 ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_2;
489 size = NDIS_OFFLOAD_SIZE_6_1;
491 ndis_rev = NDIS_OFFLOAD_PARAMETERS_REVISION_1;
492 size = NDIS_OFFLOAD_SIZE_6_0;
495 request->request_msg.msg_len += size;
496 query->info_buflen = size;
497 hwcaps = (struct ndis_offload *)
498 ((unsigned long)query + query->info_buf_offset);
500 hwcaps->header.type = NDIS_OBJECT_TYPE_OFFLOAD;
501 hwcaps->header.revision = ndis_rev;
502 hwcaps->header.size = size;
504 } else if (oid == OID_GEN_RECEIVE_SCALE_CAPABILITIES) {
505 struct ndis_recv_scale_cap *cap;
507 request->request_msg.msg_len +=
508 sizeof(struct ndis_recv_scale_cap);
509 query->info_buflen = sizeof(struct ndis_recv_scale_cap);
510 cap = (struct ndis_recv_scale_cap *)((unsigned long)query +
511 query->info_buf_offset);
512 cap->hdr.type = NDIS_OBJECT_TYPE_RSS_CAPABILITIES;
513 cap->hdr.rev = NDIS_RECEIVE_SCALE_CAPABILITIES_REVISION_2;
514 cap->hdr.size = sizeof(struct ndis_recv_scale_cap);
517 ret = rndis_filter_send_request(dev, request);
521 wait_for_completion(&request->wait_event);
523 /* Copy the response back */
524 query_complete = &request->response_msg.msg.query_complete;
526 if (query_complete->info_buflen > inresult_size) {
532 (void *)((unsigned long)query_complete +
533 query_complete->info_buf_offset),
534 query_complete->info_buflen);
536 *result_size = query_complete->info_buflen;
540 put_rndis_request(dev, request);
545 /* Get the hardware offload capabilities */
547 rndis_query_hwcaps(struct rndis_device *dev, struct netvsc_device *net_device,
548 struct ndis_offload *caps)
550 u32 caps_len = sizeof(*caps);
553 memset(caps, 0, sizeof(*caps));
555 ret = rndis_filter_query_device(dev, net_device,
556 OID_TCP_OFFLOAD_HARDWARE_CAPABILITIES,
561 if (caps->header.type != NDIS_OBJECT_TYPE_OFFLOAD) {
562 netdev_warn(dev->ndev, "invalid NDIS objtype %#x\n",
567 if (caps->header.revision < NDIS_OFFLOAD_PARAMETERS_REVISION_1) {
568 netdev_warn(dev->ndev, "invalid NDIS objrev %x\n",
569 caps->header.revision);
573 if (caps->header.size > caps_len ||
574 caps->header.size < NDIS_OFFLOAD_SIZE_6_0) {
575 netdev_warn(dev->ndev,
576 "invalid NDIS objsize %u, data size %u\n",
577 caps->header.size, caps_len);
584 static int rndis_filter_query_device_mac(struct rndis_device *dev,
585 struct netvsc_device *net_device)
589 return rndis_filter_query_device(dev, net_device,
590 RNDIS_OID_802_3_PERMANENT_ADDRESS,
591 dev->hw_mac_adr, &size);
594 #define NWADR_STR "NetworkAddress"
595 #define NWADR_STRLEN 14
597 int rndis_filter_set_device_mac(struct netvsc_device *nvdev,
600 struct rndis_device *rdev = nvdev->extension;
601 struct rndis_request *request;
602 struct rndis_set_request *set;
603 struct rndis_config_parameter_info *cpi;
604 wchar_t *cfg_nwadr, *cfg_mac;
605 struct rndis_set_complete *set_complete;
606 char macstr[2*ETH_ALEN+1];
607 u32 extlen = sizeof(struct rndis_config_parameter_info) +
608 2*NWADR_STRLEN + 4*ETH_ALEN;
611 request = get_rndis_request(rdev, RNDIS_MSG_SET,
612 RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
616 set = &request->request_msg.msg.set_req;
617 set->oid = RNDIS_OID_GEN_RNDIS_CONFIG_PARAMETER;
618 set->info_buflen = extlen;
619 set->info_buf_offset = sizeof(struct rndis_set_request);
620 set->dev_vc_handle = 0;
622 cpi = (struct rndis_config_parameter_info *)((ulong)set +
623 set->info_buf_offset);
624 cpi->parameter_name_offset =
625 sizeof(struct rndis_config_parameter_info);
626 /* Multiply by 2 because host needs 2 bytes (utf16) for each char */
627 cpi->parameter_name_length = 2*NWADR_STRLEN;
628 cpi->parameter_type = RNDIS_CONFIG_PARAM_TYPE_STRING;
629 cpi->parameter_value_offset =
630 cpi->parameter_name_offset + cpi->parameter_name_length;
631 /* Multiply by 4 because each MAC byte displayed as 2 utf16 chars */
632 cpi->parameter_value_length = 4*ETH_ALEN;
634 cfg_nwadr = (wchar_t *)((ulong)cpi + cpi->parameter_name_offset);
635 cfg_mac = (wchar_t *)((ulong)cpi + cpi->parameter_value_offset);
636 ret = utf8s_to_utf16s(NWADR_STR, NWADR_STRLEN, UTF16_HOST_ENDIAN,
637 cfg_nwadr, NWADR_STRLEN);
640 snprintf(macstr, 2*ETH_ALEN+1, "%pm", mac);
641 ret = utf8s_to_utf16s(macstr, 2*ETH_ALEN, UTF16_HOST_ENDIAN,
642 cfg_mac, 2*ETH_ALEN);
646 ret = rndis_filter_send_request(rdev, request);
650 wait_for_completion(&request->wait_event);
652 set_complete = &request->response_msg.msg.set_complete;
653 if (set_complete->status != RNDIS_STATUS_SUCCESS)
657 put_rndis_request(rdev, request);
662 rndis_filter_set_offload_params(struct net_device *ndev,
663 struct netvsc_device *nvdev,
664 struct ndis_offload_params *req_offloads)
666 struct rndis_device *rdev = nvdev->extension;
667 struct rndis_request *request;
668 struct rndis_set_request *set;
669 struct ndis_offload_params *offload_params;
670 struct rndis_set_complete *set_complete;
671 u32 extlen = sizeof(struct ndis_offload_params);
673 u32 vsp_version = nvdev->nvsp_version;
675 if (vsp_version <= NVSP_PROTOCOL_VERSION_4) {
676 extlen = VERSION_4_OFFLOAD_SIZE;
677 /* On NVSP_PROTOCOL_VERSION_4 and below, we do not support
678 * UDP checksum offload.
680 req_offloads->udp_ip_v4_csum = 0;
681 req_offloads->udp_ip_v6_csum = 0;
684 request = get_rndis_request(rdev, RNDIS_MSG_SET,
685 RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
689 set = &request->request_msg.msg.set_req;
690 set->oid = OID_TCP_OFFLOAD_PARAMETERS;
691 set->info_buflen = extlen;
692 set->info_buf_offset = sizeof(struct rndis_set_request);
693 set->dev_vc_handle = 0;
695 offload_params = (struct ndis_offload_params *)((ulong)set +
696 set->info_buf_offset);
697 *offload_params = *req_offloads;
698 offload_params->header.type = NDIS_OBJECT_TYPE_DEFAULT;
699 offload_params->header.revision = NDIS_OFFLOAD_PARAMETERS_REVISION_3;
700 offload_params->header.size = extlen;
702 ret = rndis_filter_send_request(rdev, request);
706 wait_for_completion(&request->wait_event);
707 set_complete = &request->response_msg.msg.set_complete;
708 if (set_complete->status != RNDIS_STATUS_SUCCESS) {
709 netdev_err(ndev, "Fail to set offload on host side:0x%x\n",
710 set_complete->status);
715 put_rndis_request(rdev, request);
719 int rndis_filter_set_rss_param(struct rndis_device *rdev,
722 struct net_device *ndev = rdev->ndev;
723 struct rndis_request *request;
724 struct rndis_set_request *set;
725 struct rndis_set_complete *set_complete;
726 u32 extlen = sizeof(struct ndis_recv_scale_param) +
727 4 * ITAB_NUM + NETVSC_HASH_KEYLEN;
728 struct ndis_recv_scale_param *rssp;
733 request = get_rndis_request(
735 RNDIS_MESSAGE_SIZE(struct rndis_set_request) + extlen);
739 set = &request->request_msg.msg.set_req;
740 set->oid = OID_GEN_RECEIVE_SCALE_PARAMETERS;
741 set->info_buflen = extlen;
742 set->info_buf_offset = sizeof(struct rndis_set_request);
743 set->dev_vc_handle = 0;
745 rssp = (struct ndis_recv_scale_param *)(set + 1);
746 rssp->hdr.type = NDIS_OBJECT_TYPE_RSS_PARAMETERS;
747 rssp->hdr.rev = NDIS_RECEIVE_SCALE_PARAMETERS_REVISION_2;
748 rssp->hdr.size = sizeof(struct ndis_recv_scale_param);
750 rssp->hashinfo = NDIS_HASH_FUNC_TOEPLITZ | NDIS_HASH_IPV4 |
751 NDIS_HASH_TCP_IPV4 | NDIS_HASH_IPV6 |
753 rssp->indirect_tabsize = 4*ITAB_NUM;
754 rssp->indirect_taboffset = sizeof(struct ndis_recv_scale_param);
755 rssp->hashkey_size = NETVSC_HASH_KEYLEN;
756 rssp->kashkey_offset = rssp->indirect_taboffset +
757 rssp->indirect_tabsize;
759 /* Set indirection table entries */
760 itab = (u32 *)(rssp + 1);
761 for (i = 0; i < ITAB_NUM; i++)
762 itab[i] = rdev->rx_table[i];
764 /* Set hask key values */
765 keyp = (u8 *)((unsigned long)rssp + rssp->kashkey_offset);
766 memcpy(keyp, rss_key, NETVSC_HASH_KEYLEN);
768 ret = rndis_filter_send_request(rdev, request);
772 wait_for_completion(&request->wait_event);
773 set_complete = &request->response_msg.msg.set_complete;
774 if (set_complete->status == RNDIS_STATUS_SUCCESS)
775 memcpy(rdev->rss_key, rss_key, NETVSC_HASH_KEYLEN);
777 netdev_err(ndev, "Fail to set RSS parameters:0x%x\n",
778 set_complete->status);
783 put_rndis_request(rdev, request);
787 static int rndis_filter_query_device_link_status(struct rndis_device *dev,
788 struct netvsc_device *net_device)
790 u32 size = sizeof(u32);
793 return rndis_filter_query_device(dev, net_device,
794 RNDIS_OID_GEN_MEDIA_CONNECT_STATUS,
795 &link_status, &size);
798 static int rndis_filter_query_link_speed(struct rndis_device *dev,
799 struct netvsc_device *net_device)
801 u32 size = sizeof(u32);
803 struct net_device_context *ndc;
806 ret = rndis_filter_query_device(dev, net_device,
807 RNDIS_OID_GEN_LINK_SPEED,
811 ndc = netdev_priv(dev->ndev);
813 /* The link speed reported from host is in 100bps unit, so
814 * we convert it to Mbps here.
816 ndc->speed = link_speed / 10000;
822 static int rndis_filter_set_packet_filter(struct rndis_device *dev,
825 struct rndis_request *request;
826 struct rndis_set_request *set;
829 request = get_rndis_request(dev, RNDIS_MSG_SET,
830 RNDIS_MESSAGE_SIZE(struct rndis_set_request) +
836 /* Setup the rndis set */
837 set = &request->request_msg.msg.set_req;
838 set->oid = RNDIS_OID_GEN_CURRENT_PACKET_FILTER;
839 set->info_buflen = sizeof(u32);
840 set->info_buf_offset = sizeof(struct rndis_set_request);
842 memcpy((void *)(unsigned long)set + sizeof(struct rndis_set_request),
843 &new_filter, sizeof(u32));
845 ret = rndis_filter_send_request(dev, request);
847 wait_for_completion(&request->wait_event);
849 put_rndis_request(dev, request);
854 static void rndis_set_multicast(struct work_struct *w)
856 struct rndis_device *rdev
857 = container_of(w, struct rndis_device, mcast_work);
859 if (rdev->ndev->flags & IFF_PROMISC)
860 rndis_filter_set_packet_filter(rdev,
861 NDIS_PACKET_TYPE_PROMISCUOUS);
863 rndis_filter_set_packet_filter(rdev,
864 NDIS_PACKET_TYPE_BROADCAST |
865 NDIS_PACKET_TYPE_ALL_MULTICAST |
866 NDIS_PACKET_TYPE_DIRECTED);
869 void rndis_filter_update(struct netvsc_device *nvdev)
871 struct rndis_device *rdev = nvdev->extension;
873 schedule_work(&rdev->mcast_work);
876 static int rndis_filter_init_device(struct rndis_device *dev,
877 struct netvsc_device *nvdev)
879 struct rndis_request *request;
880 struct rndis_initialize_request *init;
881 struct rndis_initialize_complete *init_complete;
885 request = get_rndis_request(dev, RNDIS_MSG_INIT,
886 RNDIS_MESSAGE_SIZE(struct rndis_initialize_request));
892 /* Setup the rndis set */
893 init = &request->request_msg.msg.init_req;
894 init->major_ver = RNDIS_MAJOR_VERSION;
895 init->minor_ver = RNDIS_MINOR_VERSION;
896 init->max_xfer_size = 0x4000;
898 dev->state = RNDIS_DEV_INITIALIZING;
900 ret = rndis_filter_send_request(dev, request);
902 dev->state = RNDIS_DEV_UNINITIALIZED;
906 wait_for_completion(&request->wait_event);
908 init_complete = &request->response_msg.msg.init_complete;
909 status = init_complete->status;
910 if (status == RNDIS_STATUS_SUCCESS) {
911 dev->state = RNDIS_DEV_INITIALIZED;
912 nvdev->max_pkt = init_complete->max_pkt_per_msg;
913 nvdev->pkt_align = 1 << init_complete->pkt_alignment_factor;
916 dev->state = RNDIS_DEV_UNINITIALIZED;
922 put_rndis_request(dev, request);
927 static bool netvsc_device_idle(const struct netvsc_device *nvdev)
931 for (i = 0; i < nvdev->num_chn; i++) {
932 const struct netvsc_channel *nvchan = &nvdev->chan_table[i];
934 if (nvchan->mrc.first != nvchan->mrc.next)
937 if (atomic_read(&nvchan->queue_sends) > 0)
944 static void rndis_filter_halt_device(struct rndis_device *dev)
946 struct rndis_request *request;
947 struct rndis_halt_request *halt;
948 struct net_device_context *net_device_ctx = netdev_priv(dev->ndev);
949 struct netvsc_device *nvdev = rtnl_dereference(net_device_ctx->nvdev);
951 /* Attempt to do a rndis device halt */
952 request = get_rndis_request(dev, RNDIS_MSG_HALT,
953 RNDIS_MESSAGE_SIZE(struct rndis_halt_request));
957 /* Setup the rndis set */
958 halt = &request->request_msg.msg.halt_req;
959 halt->req_id = atomic_inc_return(&dev->new_req_id);
961 /* Ignore return since this msg is optional. */
962 rndis_filter_send_request(dev, request);
964 dev->state = RNDIS_DEV_UNINITIALIZED;
967 nvdev->destroy = true;
969 /* Force flag to be ordered before waiting */
972 /* Wait for all send completions */
973 wait_event(nvdev->wait_drain, netvsc_device_idle(nvdev));
976 put_rndis_request(dev, request);
979 static int rndis_filter_open_device(struct rndis_device *dev)
983 if (dev->state != RNDIS_DEV_INITIALIZED)
986 ret = rndis_filter_set_packet_filter(dev,
987 NDIS_PACKET_TYPE_BROADCAST |
988 NDIS_PACKET_TYPE_ALL_MULTICAST |
989 NDIS_PACKET_TYPE_DIRECTED);
991 dev->state = RNDIS_DEV_DATAINITIALIZED;
996 static int rndis_filter_close_device(struct rndis_device *dev)
1000 if (dev->state != RNDIS_DEV_DATAINITIALIZED)
1003 /* Make sure rndis_set_multicast doesn't re-enable filter! */
1004 cancel_work_sync(&dev->mcast_work);
1006 ret = rndis_filter_set_packet_filter(dev, 0);
1011 dev->state = RNDIS_DEV_INITIALIZED;
1016 static void netvsc_sc_open(struct vmbus_channel *new_sc)
1018 struct net_device *ndev =
1019 hv_get_drvdata(new_sc->primary_channel->device_obj);
1020 struct net_device_context *ndev_ctx = netdev_priv(ndev);
1021 struct netvsc_device *nvscdev;
1022 u16 chn_index = new_sc->offermsg.offer.sub_channel_index;
1023 struct netvsc_channel *nvchan;
1026 /* This is safe because this callback only happens when
1027 * new device is being setup and waiting on the channel_init_wait.
1029 nvscdev = rcu_dereference_raw(ndev_ctx->nvdev);
1030 if (!nvscdev || chn_index >= nvscdev->num_chn)
1033 nvchan = nvscdev->chan_table + chn_index;
1035 /* Because the device uses NAPI, all the interrupt batching and
1036 * control is done via Net softirq, not the channel handling
1038 set_channel_read_mode(new_sc, HV_CALL_ISR);
1040 /* Set the channel before opening.*/
1041 nvchan->channel = new_sc;
1043 ret = vmbus_open(new_sc, nvscdev->ring_size * PAGE_SIZE,
1044 nvscdev->ring_size * PAGE_SIZE, NULL, 0,
1045 netvsc_channel_cb, nvchan);
1047 napi_enable(&nvchan->napi);
1049 netdev_notice(ndev, "sub channel open failed: %d\n", ret);
1051 if (atomic_inc_return(&nvscdev->open_chn) == nvscdev->num_chn)
1052 wake_up(&nvscdev->subchan_open);
1055 /* Open sub-channels after completing the handling of the device probe.
1056 * This breaks overlap of processing the host message for the
1057 * new primary channel with the initialization of sub-channels.
1059 void rndis_set_subchannel(struct work_struct *w)
1061 struct netvsc_device *nvdev
1062 = container_of(w, struct netvsc_device, subchan_work);
1063 struct nvsp_message *init_packet = &nvdev->channel_init_pkt;
1064 struct net_device_context *ndev_ctx;
1065 struct rndis_device *rdev;
1066 struct net_device *ndev;
1067 struct hv_device *hv_dev;
1070 if (!rtnl_trylock()) {
1075 rdev = nvdev->extension;
1077 goto unlock; /* device was removed */
1080 ndev_ctx = netdev_priv(ndev);
1081 hv_dev = ndev_ctx->device_ctx;
1083 memset(init_packet, 0, sizeof(struct nvsp_message));
1084 init_packet->hdr.msg_type = NVSP_MSG5_TYPE_SUBCHANNEL;
1085 init_packet->msg.v5_msg.subchn_req.op = NVSP_SUBCHANNEL_ALLOCATE;
1086 init_packet->msg.v5_msg.subchn_req.num_subchannels =
1088 ret = vmbus_sendpacket(hv_dev->channel, init_packet,
1089 sizeof(struct nvsp_message),
1090 (unsigned long)init_packet,
1092 VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED);
1094 netdev_err(ndev, "sub channel allocate send failed: %d\n", ret);
1098 wait_for_completion(&nvdev->channel_init_wait);
1099 if (init_packet->msg.v5_msg.subchn_comp.status != NVSP_STAT_SUCCESS) {
1100 netdev_err(ndev, "sub channel request failed\n");
1104 nvdev->num_chn = 1 +
1105 init_packet->msg.v5_msg.subchn_comp.num_subchannels;
1107 /* wait for all sub channels to open */
1108 wait_event(nvdev->subchan_open,
1109 atomic_read(&nvdev->open_chn) == nvdev->num_chn);
1111 /* ignore failues from setting rss parameters, still have channels */
1112 rndis_filter_set_rss_param(rdev, netvsc_hash_key);
1114 netif_set_real_num_tx_queues(ndev, nvdev->num_chn);
1115 netif_set_real_num_rx_queues(ndev, nvdev->num_chn);
1117 for (i = 0; i < VRSS_SEND_TAB_SIZE; i++)
1118 ndev_ctx->tx_table[i] = i % nvdev->num_chn;
1124 /* fallback to only primary channel */
1125 for (i = 1; i < nvdev->num_chn; i++)
1126 netif_napi_del(&nvdev->chan_table[i].napi);
1134 struct netvsc_device *rndis_filter_device_add(struct hv_device *dev,
1135 struct netvsc_device_info *device_info)
1137 struct net_device *net = hv_get_drvdata(dev);
1138 struct net_device_context *net_device_ctx = netdev_priv(net);
1139 struct netvsc_device *net_device;
1140 struct rndis_device *rndis_device;
1141 struct ndis_offload hwcaps;
1142 struct ndis_offload_params offloads;
1143 struct ndis_recv_scale_cap rsscap;
1144 u32 rsscap_size = sizeof(struct ndis_recv_scale_cap);
1145 unsigned int gso_max_size = GSO_MAX_SIZE;
1147 const struct cpumask *node_cpu_mask;
1148 u32 num_possible_rss_qs;
1151 rndis_device = get_rndis_device();
1153 return ERR_PTR(-ENODEV);
1156 * Let the inner driver handle this first to create the netvsc channel
1157 * NOTE! Once the channel is created, we may get a receive callback
1158 * (RndisFilterOnReceive()) before this call is completed
1160 net_device = netvsc_device_add(dev, device_info);
1161 if (IS_ERR(net_device)) {
1162 kfree(rndis_device);
1166 /* Initialize the rndis device */
1167 net_device->max_chn = 1;
1168 net_device->num_chn = 1;
1170 net_device->extension = rndis_device;
1171 rndis_device->ndev = net;
1173 /* Send the rndis initialization message */
1174 ret = rndis_filter_init_device(rndis_device, net_device);
1178 /* Get the MTU from the host */
1180 ret = rndis_filter_query_device(rndis_device, net_device,
1181 RNDIS_OID_GEN_MAXIMUM_FRAME_SIZE,
1183 if (ret == 0 && size == sizeof(u32) && mtu < net->mtu)
1186 /* Get the mac address */
1187 ret = rndis_filter_query_device_mac(rndis_device, net_device);
1191 memcpy(device_info->mac_adr, rndis_device->hw_mac_adr, ETH_ALEN);
1193 /* Find HW offload capabilities */
1194 ret = rndis_query_hwcaps(rndis_device, net_device, &hwcaps);
1198 /* A value of zero means "no change"; now turn on what we want. */
1199 memset(&offloads, 0, sizeof(struct ndis_offload_params));
1201 /* Linux does not care about IP checksum, always does in kernel */
1202 offloads.ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_DISABLED;
1204 /* Compute tx offload settings based on hw capabilities */
1205 net->hw_features = NETIF_F_RXCSUM;
1207 if ((hwcaps.csum.ip4_txcsum & NDIS_TXCSUM_ALL_TCP4) == NDIS_TXCSUM_ALL_TCP4) {
1208 /* Can checksum TCP */
1209 net->hw_features |= NETIF_F_IP_CSUM;
1210 net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV4_TCP;
1212 offloads.tcp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1214 if (hwcaps.lsov2.ip4_encap & NDIS_OFFLOAD_ENCAP_8023) {
1215 offloads.lso_v2_ipv4 = NDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED;
1216 net->hw_features |= NETIF_F_TSO;
1218 if (hwcaps.lsov2.ip4_maxsz < gso_max_size)
1219 gso_max_size = hwcaps.lsov2.ip4_maxsz;
1222 if (hwcaps.csum.ip4_txcsum & NDIS_TXCSUM_CAP_UDP4) {
1223 offloads.udp_ip_v4_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1224 net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV4_UDP;
1228 if ((hwcaps.csum.ip6_txcsum & NDIS_TXCSUM_ALL_TCP6) == NDIS_TXCSUM_ALL_TCP6) {
1229 net->hw_features |= NETIF_F_IPV6_CSUM;
1231 offloads.tcp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1232 net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV6_TCP;
1234 if ((hwcaps.lsov2.ip6_encap & NDIS_OFFLOAD_ENCAP_8023) &&
1235 (hwcaps.lsov2.ip6_opts & NDIS_LSOV2_CAP_IP6) == NDIS_LSOV2_CAP_IP6) {
1236 offloads.lso_v2_ipv6 = NDIS_OFFLOAD_PARAMETERS_LSOV2_ENABLED;
1237 net->hw_features |= NETIF_F_TSO6;
1239 if (hwcaps.lsov2.ip6_maxsz < gso_max_size)
1240 gso_max_size = hwcaps.lsov2.ip6_maxsz;
1243 if (hwcaps.csum.ip6_txcsum & NDIS_TXCSUM_CAP_UDP6) {
1244 offloads.udp_ip_v6_csum = NDIS_OFFLOAD_PARAMETERS_TX_RX_ENABLED;
1245 net_device_ctx->tx_checksum_mask |= TRANSPORT_INFO_IPV6_UDP;
1249 netif_set_gso_max_size(net, gso_max_size);
1251 ret = rndis_filter_set_offload_params(net, net_device, &offloads);
1255 rndis_filter_query_device_link_status(rndis_device, net_device);
1257 netdev_dbg(net, "Device MAC %pM link state %s\n",
1258 rndis_device->hw_mac_adr,
1259 rndis_device->link_state ? "down" : "up");
1261 if (net_device->nvsp_version < NVSP_PROTOCOL_VERSION_5)
1264 rndis_filter_query_link_speed(rndis_device, net_device);
1267 memset(&rsscap, 0, rsscap_size);
1268 ret = rndis_filter_query_device(rndis_device, net_device,
1269 OID_GEN_RECEIVE_SCALE_CAPABILITIES,
1270 &rsscap, &rsscap_size);
1271 if (ret || rsscap.num_recv_que < 2)
1275 * We will limit the VRSS channels to the number CPUs in the NUMA node
1276 * the primary channel is currently bound to.
1278 * This also guarantees that num_possible_rss_qs <= num_online_cpus
1280 node_cpu_mask = cpumask_of_node(cpu_to_node(dev->channel->target_cpu));
1281 num_possible_rss_qs = min_t(u32, cpumask_weight(node_cpu_mask),
1282 rsscap.num_recv_que);
1284 net_device->max_chn = min_t(u32, VRSS_CHANNEL_MAX, num_possible_rss_qs);
1286 /* We will use the given number of channels if available. */
1287 net_device->num_chn = min(net_device->max_chn, device_info->num_chn);
1289 for (i = 0; i < ITAB_NUM; i++)
1290 rndis_device->rx_table[i] = ethtool_rxfh_indir_default(
1291 i, net_device->num_chn);
1293 atomic_set(&net_device->open_chn, 1);
1294 vmbus_set_sc_create_callback(dev->channel, netvsc_sc_open);
1296 for (i = 1; i < net_device->num_chn; i++) {
1297 ret = netvsc_alloc_recv_comp_ring(net_device, i);
1300 vfree(net_device->chan_table[i].mrc.slots);
1305 for (i = 1; i < net_device->num_chn; i++)
1306 netif_napi_add(net, &net_device->chan_table[i].napi,
1307 netvsc_poll, NAPI_POLL_WEIGHT);
1309 if (net_device->num_chn > 1)
1310 schedule_work(&net_device->subchan_work);
1313 /* if unavailable, just proceed with one queue */
1315 net_device->max_chn = 1;
1316 net_device->num_chn = 1;
1322 rndis_filter_device_remove(dev, net_device);
1323 return ERR_PTR(ret);
1326 void rndis_filter_device_remove(struct hv_device *dev,
1327 struct netvsc_device *net_dev)
1329 struct rndis_device *rndis_dev = net_dev->extension;
1331 /* Halt and release the rndis device */
1332 rndis_filter_halt_device(rndis_dev);
1334 net_dev->extension = NULL;
1336 netvsc_device_remove(dev);
1340 int rndis_filter_open(struct netvsc_device *nvdev)
1345 if (atomic_inc_return(&nvdev->open_cnt) != 1)
1348 return rndis_filter_open_device(nvdev->extension);
1351 int rndis_filter_close(struct netvsc_device *nvdev)
1356 if (atomic_dec_return(&nvdev->open_cnt) != 0)
1359 return rndis_filter_close_device(nvdev->extension);
1362 bool rndis_filter_opened(const struct netvsc_device *nvdev)
1364 return atomic_read(&nvdev->open_cnt) > 0;