net/hyperv: Add support for promiscuous mode setting
[linux-2.6.git] / drivers / net / hyperv / netvsc_drv.c
1 /*
2  * Copyright (c) 2009, Microsoft Corporation.
3  *
4  * This program is free software; you can redistribute it and/or modify it
5  * under the terms and conditions of the GNU General Public License,
6  * version 2, as published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope it will be useful, but WITHOUT
9  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
10  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
11  * more details.
12  *
13  * You should have received a copy of the GNU General Public License along with
14  * this program; if not, write to the Free Software Foundation, Inc., 59 Temple
15  * Place - Suite 330, Boston, MA 02111-1307 USA.
16  *
17  * Authors:
18  *   Haiyang Zhang <haiyangz@microsoft.com>
19  *   Hank Janssen  <hjanssen@microsoft.com>
20  */
21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22
23 #include <linux/init.h>
24 #include <linux/atomic.h>
25 #include <linux/module.h>
26 #include <linux/highmem.h>
27 #include <linux/device.h>
28 #include <linux/io.h>
29 #include <linux/delay.h>
30 #include <linux/netdevice.h>
31 #include <linux/inetdevice.h>
32 #include <linux/etherdevice.h>
33 #include <linux/skbuff.h>
34 #include <linux/in.h>
35 #include <linux/slab.h>
36 #include <net/arp.h>
37 #include <net/route.h>
38 #include <net/sock.h>
39 #include <net/pkt_sched.h>
40
41 #include "hyperv_net.h"
42
43 struct net_device_context {
44         /* point back to our device context */
45         struct hv_device *device_ctx;
46         atomic_t avail;
47         struct delayed_work dwork;
48 };
49
50
51 #define PACKET_PAGES_LOWATER  8
52 /* Need this many pages to handle worst case fragmented packet */
53 #define PACKET_PAGES_HIWATER  (MAX_SKB_FRAGS + 2)
54
55 static int ring_size = 128;
56 module_param(ring_size, int, S_IRUGO);
57 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
58
59 struct set_multicast_work {
60         struct work_struct work;
61         struct net_device *net;
62 };
63
64 static void do_set_multicast(struct work_struct *w)
65 {
66         struct set_multicast_work *swk =
67                 container_of(w, struct set_multicast_work, work);
68         struct net_device *net = swk->net;
69
70         struct net_device_context *ndevctx = netdev_priv(net);
71         struct netvsc_device *nvdev;
72         struct rndis_device *rdev;
73
74         nvdev = hv_get_drvdata(ndevctx->device_ctx);
75         if (nvdev == NULL)
76                 return;
77
78         rdev = nvdev->extension;
79         if (rdev == NULL)
80                 return;
81
82         if (net->flags & IFF_PROMISC)
83                 rndis_filter_set_packet_filter(rdev,
84                         NDIS_PACKET_TYPE_PROMISCUOUS);
85         else
86                 rndis_filter_set_packet_filter(rdev,
87                         NDIS_PACKET_TYPE_BROADCAST |
88                         NDIS_PACKET_TYPE_ALL_MULTICAST |
89                         NDIS_PACKET_TYPE_DIRECTED);
90
91         kfree(w);
92 }
93
94 static void netvsc_set_multicast_list(struct net_device *net)
95 {
96         struct set_multicast_work *swk =
97                 kmalloc(sizeof(struct set_multicast_work), GFP_ATOMIC);
98         if (swk == NULL)
99                 return;
100
101         swk->net = net;
102         INIT_WORK(&swk->work, do_set_multicast);
103         schedule_work(&swk->work);
104 }
105
106 static int netvsc_open(struct net_device *net)
107 {
108         struct net_device_context *net_device_ctx = netdev_priv(net);
109         struct hv_device *device_obj = net_device_ctx->device_ctx;
110         int ret = 0;
111
112         /* Open up the device */
113         ret = rndis_filter_open(device_obj);
114         if (ret != 0) {
115                 netdev_err(net, "unable to open device (ret %d).\n", ret);
116                 return ret;
117         }
118
119         netif_start_queue(net);
120
121         return ret;
122 }
123
124 static int netvsc_close(struct net_device *net)
125 {
126         struct net_device_context *net_device_ctx = netdev_priv(net);
127         struct hv_device *device_obj = net_device_ctx->device_ctx;
128         int ret;
129
130         netif_stop_queue(net);
131
132         ret = rndis_filter_close(device_obj);
133         if (ret != 0)
134                 netdev_err(net, "unable to close device (ret %d).\n", ret);
135
136         return ret;
137 }
138
139 static void netvsc_xmit_completion(void *context)
140 {
141         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
142         struct sk_buff *skb = (struct sk_buff *)
143                 (unsigned long)packet->completion.send.send_completion_tid;
144
145         kfree(packet);
146
147         if (skb) {
148                 struct net_device *net = skb->dev;
149                 struct net_device_context *net_device_ctx = netdev_priv(net);
150                 unsigned int num_pages = skb_shinfo(skb)->nr_frags + 2;
151
152                 dev_kfree_skb_any(skb);
153
154                 atomic_add(num_pages, &net_device_ctx->avail);
155                 if (atomic_read(&net_device_ctx->avail) >=
156                                 PACKET_PAGES_HIWATER)
157                         netif_wake_queue(net);
158         }
159 }
160
161 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
162 {
163         struct net_device_context *net_device_ctx = netdev_priv(net);
164         struct hv_netvsc_packet *packet;
165         int ret;
166         unsigned int i, num_pages;
167
168         /* Add 1 for skb->data and additional one for RNDIS */
169         num_pages = skb_shinfo(skb)->nr_frags + 1 + 1;
170         if (num_pages > atomic_read(&net_device_ctx->avail))
171                 return NETDEV_TX_BUSY;
172
173         /* Allocate a netvsc packet based on # of frags. */
174         packet = kzalloc(sizeof(struct hv_netvsc_packet) +
175                          (num_pages * sizeof(struct hv_page_buffer)) +
176                          sizeof(struct rndis_filter_packet), GFP_ATOMIC);
177         if (!packet) {
178                 /* out of memory, drop packet */
179                 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
180
181                 dev_kfree_skb(skb);
182                 net->stats.tx_dropped++;
183                 return NETDEV_TX_BUSY;
184         }
185
186         packet->extension = (void *)(unsigned long)packet +
187                                 sizeof(struct hv_netvsc_packet) +
188                                     (num_pages * sizeof(struct hv_page_buffer));
189
190         /* Setup the rndis header */
191         packet->page_buf_cnt = num_pages;
192
193         /* Initialize it from the skb */
194         packet->total_data_buflen       = skb->len;
195
196         /* Start filling in the page buffers starting after RNDIS buffer. */
197         packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT;
198         packet->page_buf[1].offset
199                 = (unsigned long)skb->data & (PAGE_SIZE - 1);
200         packet->page_buf[1].len = skb_headlen(skb);
201
202         /* Additional fragments are after SKB data */
203         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
204                 const skb_frag_t *f = &skb_shinfo(skb)->frags[i];
205
206                 packet->page_buf[i+2].pfn = page_to_pfn(skb_frag_page(f));
207                 packet->page_buf[i+2].offset = f->page_offset;
208                 packet->page_buf[i+2].len = skb_frag_size(f);
209         }
210
211         /* Set the completion routine */
212         packet->completion.send.send_completion = netvsc_xmit_completion;
213         packet->completion.send.send_completion_ctx = packet;
214         packet->completion.send.send_completion_tid = (unsigned long)skb;
215
216         ret = rndis_filter_send(net_device_ctx->device_ctx,
217                                   packet);
218         if (ret == 0) {
219                 net->stats.tx_bytes += skb->len;
220                 net->stats.tx_packets++;
221
222                 atomic_sub(num_pages, &net_device_ctx->avail);
223                 if (atomic_read(&net_device_ctx->avail) < PACKET_PAGES_LOWATER)
224                         netif_stop_queue(net);
225         } else {
226                 /* we are shutting down or bus overloaded, just drop packet */
227                 net->stats.tx_dropped++;
228                 kfree(packet);
229                 dev_kfree_skb_any(skb);
230         }
231
232         return ret ? NETDEV_TX_BUSY : NETDEV_TX_OK;
233 }
234
235 /*
236  * netvsc_linkstatus_callback - Link up/down notification
237  */
238 void netvsc_linkstatus_callback(struct hv_device *device_obj,
239                                        unsigned int status)
240 {
241         struct net_device *net;
242         struct net_device_context *ndev_ctx;
243         struct netvsc_device *net_device;
244
245         net_device = hv_get_drvdata(device_obj);
246         net = net_device->ndev;
247
248         if (!net) {
249                 netdev_err(net, "got link status but net device "
250                                 "not initialized yet\n");
251                 return;
252         }
253
254         if (status == 1) {
255                 netif_carrier_on(net);
256                 netif_wake_queue(net);
257                 ndev_ctx = netdev_priv(net);
258                 schedule_delayed_work(&ndev_ctx->dwork, 0);
259                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
260         } else {
261                 netif_carrier_off(net);
262                 netif_stop_queue(net);
263         }
264 }
265
266 /*
267  * netvsc_recv_callback -  Callback when we receive a packet from the
268  * "wire" on the specified device.
269  */
270 int netvsc_recv_callback(struct hv_device *device_obj,
271                                 struct hv_netvsc_packet *packet)
272 {
273         struct net_device *net = dev_get_drvdata(&device_obj->device);
274         struct sk_buff *skb;
275         void *data;
276         int i;
277         unsigned long flags;
278         struct netvsc_device *net_device;
279
280         net_device = hv_get_drvdata(device_obj);
281         net = net_device->ndev;
282
283         if (!net) {
284                 netdev_err(net, "got receive callback but net device"
285                         " not initialized yet\n");
286                 return 0;
287         }
288
289         /* Allocate a skb - TODO direct I/O to pages? */
290         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
291         if (unlikely(!skb)) {
292                 ++net->stats.rx_dropped;
293                 return 0;
294         }
295
296         /* for kmap_atomic */
297         local_irq_save(flags);
298
299         /*
300          * Copy to skb. This copy is needed here since the memory pointed by
301          * hv_netvsc_packet cannot be deallocated
302          */
303         for (i = 0; i < packet->page_buf_cnt; i++) {
304                 data = kmap_atomic(pfn_to_page(packet->page_buf[i].pfn),
305                                                KM_IRQ1);
306                 data = (void *)(unsigned long)data +
307                                 packet->page_buf[i].offset;
308
309                 memcpy(skb_put(skb, packet->page_buf[i].len), data,
310                        packet->page_buf[i].len);
311
312                 kunmap_atomic((void *)((unsigned long)data -
313                                        packet->page_buf[i].offset), KM_IRQ1);
314         }
315
316         local_irq_restore(flags);
317
318         skb->protocol = eth_type_trans(skb, net);
319         skb->ip_summed = CHECKSUM_NONE;
320
321         net->stats.rx_packets++;
322         net->stats.rx_bytes += skb->len;
323
324         /*
325          * Pass the skb back up. Network stack will deallocate the skb when it
326          * is done.
327          * TODO - use NAPI?
328          */
329         netif_rx(skb);
330
331         return 0;
332 }
333
334 static void netvsc_get_drvinfo(struct net_device *net,
335                                struct ethtool_drvinfo *info)
336 {
337         strcpy(info->driver, "hv_netvsc");
338         strcpy(info->version, HV_DRV_VERSION);
339         strcpy(info->fw_version, "N/A");
340 }
341
342 static const struct ethtool_ops ethtool_ops = {
343         .get_drvinfo    = netvsc_get_drvinfo,
344         .get_link       = ethtool_op_get_link,
345 };
346
347 static const struct net_device_ops device_ops = {
348         .ndo_open =                     netvsc_open,
349         .ndo_stop =                     netvsc_close,
350         .ndo_start_xmit =               netvsc_start_xmit,
351         .ndo_set_rx_mode =              netvsc_set_multicast_list,
352         .ndo_change_mtu =               eth_change_mtu,
353         .ndo_validate_addr =            eth_validate_addr,
354         .ndo_set_mac_address =          eth_mac_addr,
355 };
356
357 /*
358  * Send GARP packet to network peers after migrations.
359  * After Quick Migration, the network is not immediately operational in the
360  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
361  * another netif_notify_peers() into a delayed work, otherwise GARP packet
362  * will not be sent after quick migration, and cause network disconnection.
363  */
364 static void netvsc_send_garp(struct work_struct *w)
365 {
366         struct net_device_context *ndev_ctx;
367         struct net_device *net;
368         struct netvsc_device *net_device;
369
370         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
371         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
372         net = net_device->ndev;
373         netif_notify_peers(net);
374 }
375
376
377 static int netvsc_probe(struct hv_device *dev,
378                         const struct hv_vmbus_device_id *dev_id)
379 {
380         struct net_device *net = NULL;
381         struct net_device_context *net_device_ctx;
382         struct netvsc_device_info device_info;
383         int ret;
384
385         net = alloc_etherdev(sizeof(struct net_device_context));
386         if (!net)
387                 return -ENOMEM;
388
389         /* Set initial state */
390         netif_carrier_off(net);
391
392         net_device_ctx = netdev_priv(net);
393         net_device_ctx->device_ctx = dev;
394         atomic_set(&net_device_ctx->avail, ring_size);
395         hv_set_drvdata(dev, net);
396         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_send_garp);
397
398         net->netdev_ops = &device_ops;
399
400         /* TODO: Add GSO and Checksum offload */
401         net->hw_features = NETIF_F_SG;
402         net->features = NETIF_F_SG;
403
404         SET_ETHTOOL_OPS(net, &ethtool_ops);
405         SET_NETDEV_DEV(net, &dev->device);
406
407         ret = register_netdev(net);
408         if (ret != 0) {
409                 pr_err("Unable to register netdev.\n");
410                 free_netdev(net);
411                 goto out;
412         }
413
414         /* Notify the netvsc driver of the new device */
415         device_info.ring_size = ring_size;
416         ret = rndis_filter_device_add(dev, &device_info);
417         if (ret != 0) {
418                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
419                 unregister_netdev(net);
420                 free_netdev(net);
421                 hv_set_drvdata(dev, NULL);
422                 return ret;
423         }
424         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
425
426         netif_carrier_on(net);
427
428 out:
429         return ret;
430 }
431
432 static int netvsc_remove(struct hv_device *dev)
433 {
434         struct net_device *net;
435         struct net_device_context *ndev_ctx;
436         struct netvsc_device *net_device;
437
438         net_device = hv_get_drvdata(dev);
439         net = net_device->ndev;
440
441         if (net == NULL) {
442                 dev_err(&dev->device, "No net device to remove\n");
443                 return 0;
444         }
445
446         ndev_ctx = netdev_priv(net);
447         cancel_delayed_work_sync(&ndev_ctx->dwork);
448
449         /* Stop outbound asap */
450         netif_stop_queue(net);
451
452         unregister_netdev(net);
453
454         /*
455          * Call to the vsc driver to let it know that the device is being
456          * removed
457          */
458         rndis_filter_device_remove(dev);
459
460         free_netdev(net);
461         return 0;
462 }
463
464 static const struct hv_vmbus_device_id id_table[] = {
465         /* Network guid */
466         { VMBUS_DEVICE(0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
467                        0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E) },
468         { },
469 };
470
471 MODULE_DEVICE_TABLE(vmbus, id_table);
472
473 /* The one and only one */
474 static struct  hv_driver netvsc_drv = {
475         .name = "netvsc",
476         .id_table = id_table,
477         .probe = netvsc_probe,
478         .remove = netvsc_remove,
479 };
480
481 static void __exit netvsc_drv_exit(void)
482 {
483         vmbus_driver_unregister(&netvsc_drv);
484 }
485
486 static int __init netvsc_drv_init(void)
487 {
488         return vmbus_driver_register(&netvsc_drv);
489 }
490
491 MODULE_LICENSE("GPL");
492 MODULE_VERSION(HV_DRV_VERSION);
493 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
494
495 module_init(netvsc_drv_init);
496 module_exit(netvsc_drv_exit);