net/hyperv: Add support for jumbo frame up to 64KB
[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         struct delayed_work dwork;
47 };
48
49
50 static int ring_size = 128;
51 module_param(ring_size, int, S_IRUGO);
52 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
53
54 struct set_multicast_work {
55         struct work_struct work;
56         struct net_device *net;
57 };
58
59 static void do_set_multicast(struct work_struct *w)
60 {
61         struct set_multicast_work *swk =
62                 container_of(w, struct set_multicast_work, work);
63         struct net_device *net = swk->net;
64
65         struct net_device_context *ndevctx = netdev_priv(net);
66         struct netvsc_device *nvdev;
67         struct rndis_device *rdev;
68
69         nvdev = hv_get_drvdata(ndevctx->device_ctx);
70         if (nvdev == NULL)
71                 return;
72
73         rdev = nvdev->extension;
74         if (rdev == NULL)
75                 return;
76
77         if (net->flags & IFF_PROMISC)
78                 rndis_filter_set_packet_filter(rdev,
79                         NDIS_PACKET_TYPE_PROMISCUOUS);
80         else
81                 rndis_filter_set_packet_filter(rdev,
82                         NDIS_PACKET_TYPE_BROADCAST |
83                         NDIS_PACKET_TYPE_ALL_MULTICAST |
84                         NDIS_PACKET_TYPE_DIRECTED);
85
86         kfree(w);
87 }
88
89 static void netvsc_set_multicast_list(struct net_device *net)
90 {
91         struct set_multicast_work *swk =
92                 kmalloc(sizeof(struct set_multicast_work), GFP_ATOMIC);
93         if (swk == NULL)
94                 return;
95
96         swk->net = net;
97         INIT_WORK(&swk->work, do_set_multicast);
98         schedule_work(&swk->work);
99 }
100
101 static int netvsc_open(struct net_device *net)
102 {
103         struct net_device_context *net_device_ctx = netdev_priv(net);
104         struct hv_device *device_obj = net_device_ctx->device_ctx;
105         int ret = 0;
106
107         /* Open up the device */
108         ret = rndis_filter_open(device_obj);
109         if (ret != 0) {
110                 netdev_err(net, "unable to open device (ret %d).\n", ret);
111                 return ret;
112         }
113
114         netif_start_queue(net);
115
116         return ret;
117 }
118
119 static int netvsc_close(struct net_device *net)
120 {
121         struct net_device_context *net_device_ctx = netdev_priv(net);
122         struct hv_device *device_obj = net_device_ctx->device_ctx;
123         int ret;
124
125         netif_stop_queue(net);
126
127         ret = rndis_filter_close(device_obj);
128         if (ret != 0)
129                 netdev_err(net, "unable to close device (ret %d).\n", ret);
130
131         return ret;
132 }
133
134 static void netvsc_xmit_completion(void *context)
135 {
136         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
137         struct sk_buff *skb = (struct sk_buff *)
138                 (unsigned long)packet->completion.send.send_completion_tid;
139
140         kfree(packet);
141
142         if (skb)
143                 dev_kfree_skb_any(skb);
144 }
145
146 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
147 {
148         struct net_device_context *net_device_ctx = netdev_priv(net);
149         struct hv_netvsc_packet *packet;
150         int ret;
151         unsigned int i, num_pages, npg_data;
152
153         /* Add multipage for skb->data and additional one for RNDIS */
154         npg_data = (((unsigned long)skb->data + skb_headlen(skb) - 1)
155                 >> PAGE_SHIFT) - ((unsigned long)skb->data >> PAGE_SHIFT) + 1;
156         num_pages = skb_shinfo(skb)->nr_frags + npg_data + 1;
157
158         /* Allocate a netvsc packet based on # of frags. */
159         packet = kzalloc(sizeof(struct hv_netvsc_packet) +
160                          (num_pages * sizeof(struct hv_page_buffer)) +
161                          sizeof(struct rndis_filter_packet), GFP_ATOMIC);
162         if (!packet) {
163                 /* out of memory, drop packet */
164                 netdev_err(net, "unable to allocate hv_netvsc_packet\n");
165
166                 dev_kfree_skb(skb);
167                 net->stats.tx_dropped++;
168                 return NETDEV_TX_BUSY;
169         }
170
171         packet->extension = (void *)(unsigned long)packet +
172                                 sizeof(struct hv_netvsc_packet) +
173                                     (num_pages * sizeof(struct hv_page_buffer));
174
175         /* Setup the rndis header */
176         packet->page_buf_cnt = num_pages;
177
178         /* Initialize it from the skb */
179         packet->total_data_buflen = skb->len;
180
181         /* Start filling in the page buffers starting after RNDIS buffer. */
182         packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT;
183         packet->page_buf[1].offset
184                 = (unsigned long)skb->data & (PAGE_SIZE - 1);
185         if (npg_data == 1)
186                 packet->page_buf[1].len = skb_headlen(skb);
187         else
188                 packet->page_buf[1].len = PAGE_SIZE
189                         - packet->page_buf[1].offset;
190
191         for (i = 2; i <= npg_data; i++) {
192                 packet->page_buf[i].pfn = virt_to_phys(skb->data
193                         + PAGE_SIZE * (i-1)) >> PAGE_SHIFT;
194                 packet->page_buf[i].offset = 0;
195                 packet->page_buf[i].len = PAGE_SIZE;
196         }
197         if (npg_data > 1)
198                 packet->page_buf[npg_data].len = (((unsigned long)skb->data
199                         + skb_headlen(skb) - 1) & (PAGE_SIZE - 1)) + 1;
200
201         /* Additional fragments are after SKB data */
202         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
203                 const skb_frag_t *f = &skb_shinfo(skb)->frags[i];
204
205                 packet->page_buf[i+npg_data+1].pfn =
206                         page_to_pfn(skb_frag_page(f));
207                 packet->page_buf[i+npg_data+1].offset = f->page_offset;
208                 packet->page_buf[i+npg_data+1].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         } else {
222                 /* we are shutting down or bus overloaded, just drop packet */
223                 net->stats.tx_dropped++;
224                 kfree(packet);
225                 dev_kfree_skb_any(skb);
226         }
227
228         return ret ? NETDEV_TX_BUSY : NETDEV_TX_OK;
229 }
230
231 /*
232  * netvsc_linkstatus_callback - Link up/down notification
233  */
234 void netvsc_linkstatus_callback(struct hv_device *device_obj,
235                                        unsigned int status)
236 {
237         struct net_device *net;
238         struct net_device_context *ndev_ctx;
239         struct netvsc_device *net_device;
240
241         net_device = hv_get_drvdata(device_obj);
242         net = net_device->ndev;
243
244         if (!net) {
245                 netdev_err(net, "got link status but net device "
246                                 "not initialized yet\n");
247                 return;
248         }
249
250         if (status == 1) {
251                 netif_carrier_on(net);
252                 netif_wake_queue(net);
253                 ndev_ctx = netdev_priv(net);
254                 schedule_delayed_work(&ndev_ctx->dwork, 0);
255                 schedule_delayed_work(&ndev_ctx->dwork, msecs_to_jiffies(20));
256         } else {
257                 netif_carrier_off(net);
258                 netif_stop_queue(net);
259         }
260 }
261
262 /*
263  * netvsc_recv_callback -  Callback when we receive a packet from the
264  * "wire" on the specified device.
265  */
266 int netvsc_recv_callback(struct hv_device *device_obj,
267                                 struct hv_netvsc_packet *packet)
268 {
269         struct net_device *net = dev_get_drvdata(&device_obj->device);
270         struct sk_buff *skb;
271         struct netvsc_device *net_device;
272
273         net_device = hv_get_drvdata(device_obj);
274         net = net_device->ndev;
275
276         if (!net) {
277                 netdev_err(net, "got receive callback but net device"
278                         " not initialized yet\n");
279                 return 0;
280         }
281
282         /* Allocate a skb - TODO direct I/O to pages? */
283         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
284         if (unlikely(!skb)) {
285                 ++net->stats.rx_dropped;
286                 return 0;
287         }
288
289         /*
290          * Copy to skb. This copy is needed here since the memory pointed by
291          * hv_netvsc_packet cannot be deallocated
292          */
293         memcpy(skb_put(skb, packet->total_data_buflen), packet->data,
294                 packet->total_data_buflen);
295
296         skb->protocol = eth_type_trans(skb, net);
297         skb->ip_summed = CHECKSUM_NONE;
298
299         net->stats.rx_packets++;
300         net->stats.rx_bytes += skb->len;
301
302         /*
303          * Pass the skb back up. Network stack will deallocate the skb when it
304          * is done.
305          * TODO - use NAPI?
306          */
307         netif_rx(skb);
308
309         return 0;
310 }
311
312 static void netvsc_get_drvinfo(struct net_device *net,
313                                struct ethtool_drvinfo *info)
314 {
315         strcpy(info->driver, "hv_netvsc");
316         strcpy(info->version, HV_DRV_VERSION);
317         strcpy(info->fw_version, "N/A");
318 }
319
320 static int netvsc_change_mtu(struct net_device *ndev, int mtu)
321 {
322         struct net_device_context *ndevctx = netdev_priv(ndev);
323         struct hv_device *hdev =  ndevctx->device_ctx;
324         struct netvsc_device *nvdev = hv_get_drvdata(hdev);
325         struct netvsc_device_info device_info;
326         int limit = ETH_DATA_LEN;
327
328         if (nvdev == NULL || nvdev->destroy)
329                 return -ENODEV;
330
331         if (nvdev->nvsp_version == NVSP_PROTOCOL_VERSION_2)
332                 limit = NETVSC_MTU;
333
334         if (mtu < 68 || mtu > limit)
335                 return -EINVAL;
336
337         nvdev->start_remove = true;
338         cancel_delayed_work_sync(&ndevctx->dwork);
339         netif_stop_queue(ndev);
340         rndis_filter_device_remove(hdev);
341
342         ndev->mtu = mtu;
343
344         ndevctx->device_ctx = hdev;
345         hv_set_drvdata(hdev, ndev);
346         device_info.ring_size = ring_size;
347         rndis_filter_device_add(hdev, &device_info);
348         netif_wake_queue(ndev);
349
350         return 0;
351 }
352
353 static const struct ethtool_ops ethtool_ops = {
354         .get_drvinfo    = netvsc_get_drvinfo,
355         .get_link       = ethtool_op_get_link,
356 };
357
358 static const struct net_device_ops device_ops = {
359         .ndo_open =                     netvsc_open,
360         .ndo_stop =                     netvsc_close,
361         .ndo_start_xmit =               netvsc_start_xmit,
362         .ndo_set_rx_mode =              netvsc_set_multicast_list,
363         .ndo_change_mtu =               netvsc_change_mtu,
364         .ndo_validate_addr =            eth_validate_addr,
365         .ndo_set_mac_address =          eth_mac_addr,
366 };
367
368 /*
369  * Send GARP packet to network peers after migrations.
370  * After Quick Migration, the network is not immediately operational in the
371  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
372  * another netif_notify_peers() into a delayed work, otherwise GARP packet
373  * will not be sent after quick migration, and cause network disconnection.
374  */
375 static void netvsc_send_garp(struct work_struct *w)
376 {
377         struct net_device_context *ndev_ctx;
378         struct net_device *net;
379         struct netvsc_device *net_device;
380
381         ndev_ctx = container_of(w, struct net_device_context, dwork.work);
382         net_device = hv_get_drvdata(ndev_ctx->device_ctx);
383         net = net_device->ndev;
384         netif_notify_peers(net);
385 }
386
387
388 static int netvsc_probe(struct hv_device *dev,
389                         const struct hv_vmbus_device_id *dev_id)
390 {
391         struct net_device *net = NULL;
392         struct net_device_context *net_device_ctx;
393         struct netvsc_device_info device_info;
394         int ret;
395
396         net = alloc_etherdev(sizeof(struct net_device_context));
397         if (!net)
398                 return -ENOMEM;
399
400         /* Set initial state */
401         netif_carrier_off(net);
402
403         net_device_ctx = netdev_priv(net);
404         net_device_ctx->device_ctx = dev;
405         hv_set_drvdata(dev, net);
406         INIT_DELAYED_WORK(&net_device_ctx->dwork, netvsc_send_garp);
407
408         net->netdev_ops = &device_ops;
409
410         /* TODO: Add GSO and Checksum offload */
411         net->hw_features = NETIF_F_SG;
412         net->features = NETIF_F_SG;
413
414         SET_ETHTOOL_OPS(net, &ethtool_ops);
415         SET_NETDEV_DEV(net, &dev->device);
416
417         ret = register_netdev(net);
418         if (ret != 0) {
419                 pr_err("Unable to register netdev.\n");
420                 free_netdev(net);
421                 goto out;
422         }
423
424         /* Notify the netvsc driver of the new device */
425         device_info.ring_size = ring_size;
426         ret = rndis_filter_device_add(dev, &device_info);
427         if (ret != 0) {
428                 netdev_err(net, "unable to add netvsc device (ret %d)\n", ret);
429                 unregister_netdev(net);
430                 free_netdev(net);
431                 hv_set_drvdata(dev, NULL);
432                 return ret;
433         }
434         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
435
436         netif_carrier_on(net);
437
438 out:
439         return ret;
440 }
441
442 static int netvsc_remove(struct hv_device *dev)
443 {
444         struct net_device *net;
445         struct net_device_context *ndev_ctx;
446         struct netvsc_device *net_device;
447
448         net_device = hv_get_drvdata(dev);
449         net = net_device->ndev;
450
451         if (net == NULL) {
452                 dev_err(&dev->device, "No net device to remove\n");
453                 return 0;
454         }
455
456         net_device->start_remove = true;
457
458         ndev_ctx = netdev_priv(net);
459         cancel_delayed_work_sync(&ndev_ctx->dwork);
460
461         /* Stop outbound asap */
462         netif_stop_queue(net);
463
464         unregister_netdev(net);
465
466         /*
467          * Call to the vsc driver to let it know that the device is being
468          * removed
469          */
470         rndis_filter_device_remove(dev);
471
472         free_netdev(net);
473         return 0;
474 }
475
476 static const struct hv_vmbus_device_id id_table[] = {
477         /* Network guid */
478         { VMBUS_DEVICE(0x63, 0x51, 0x61, 0xF8, 0x3E, 0xDF, 0xc5, 0x46,
479                        0x91, 0x3F, 0xF2, 0xD2, 0xF9, 0x65, 0xED, 0x0E) },
480         { },
481 };
482
483 MODULE_DEVICE_TABLE(vmbus, id_table);
484
485 /* The one and only one */
486 static struct  hv_driver netvsc_drv = {
487         .name = "netvsc",
488         .id_table = id_table,
489         .probe = netvsc_probe,
490         .remove = netvsc_remove,
491 };
492
493 static void __exit netvsc_drv_exit(void)
494 {
495         vmbus_driver_unregister(&netvsc_drv);
496 }
497
498 static int __init netvsc_drv_init(void)
499 {
500         return vmbus_driver_register(&netvsc_drv);
501 }
502
503 MODULE_LICENSE("GPL");
504 MODULE_VERSION(HV_DRV_VERSION);
505 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
506
507 module_init(netvsc_drv_init);
508 module_exit(netvsc_drv_exit);