staging: hv: Fix GARP not sent after Quick Migration
[linux-2.6.git] / drivers / staging / hv / 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 #include <linux/init.h>
22 #include <linux/module.h>
23 #include <linux/highmem.h>
24 #include <linux/device.h>
25 #include <linux/io.h>
26 #include <linux/delay.h>
27 #include <linux/netdevice.h>
28 #include <linux/inetdevice.h>
29 #include <linux/etherdevice.h>
30 #include <linux/skbuff.h>
31 #include <linux/in.h>
32 #include <linux/slab.h>
33 #include <linux/dmi.h>
34 #include <linux/pci.h>
35 #include <net/arp.h>
36 #include <net/route.h>
37 #include <net/sock.h>
38 #include <net/pkt_sched.h>
39 #include "hv_api.h"
40 #include "logging.h"
41 #include "version_info.h"
42 #include "vmbus.h"
43 #include "netvsc_api.h"
44
45 struct net_device_context {
46         /* point back to our device context */
47         struct hv_device *device_ctx;
48         unsigned long avail;
49         struct work_struct work;
50 };
51
52
53 #define PACKET_PAGES_LOWATER  8
54 /* Need this many pages to handle worst case fragmented packet */
55 #define PACKET_PAGES_HIWATER  (MAX_SKB_FRAGS + 2)
56
57 static int ring_size = 128;
58 module_param(ring_size, int, S_IRUGO);
59 MODULE_PARM_DESC(ring_size, "Ring buffer size (# of pages)");
60
61 /* The one and only one */
62 static struct  netvsc_driver g_netvsc_drv;
63
64 /* no-op so the netdev core doesn't return -EINVAL when modifying the the
65  * multicast address list in SIOCADDMULTI. hv is setup to get all multicast
66  * when it calls RndisFilterOnOpen() */
67 static void netvsc_set_multicast_list(struct net_device *net)
68 {
69 }
70
71 static int netvsc_open(struct net_device *net)
72 {
73         struct net_device_context *net_device_ctx = netdev_priv(net);
74         struct hv_device *device_obj = net_device_ctx->device_ctx;
75         int ret = 0;
76
77         if (netif_carrier_ok(net)) {
78                 /* Open up the device */
79                 ret = rndis_filter_open(device_obj);
80                 if (ret != 0) {
81                         DPRINT_ERR(NETVSC_DRV,
82                                    "unable to open device (ret %d).", ret);
83                         return ret;
84                 }
85
86                 netif_start_queue(net);
87         } else {
88                 DPRINT_ERR(NETVSC_DRV, "unable to open device...link is down.");
89         }
90
91         return ret;
92 }
93
94 static int netvsc_close(struct net_device *net)
95 {
96         struct net_device_context *net_device_ctx = netdev_priv(net);
97         struct hv_device *device_obj = net_device_ctx->device_ctx;
98         int ret;
99
100         netif_stop_queue(net);
101
102         ret = rndis_filter_close(device_obj);
103         if (ret != 0)
104                 DPRINT_ERR(NETVSC_DRV, "unable to close device (ret %d).", ret);
105
106         return ret;
107 }
108
109 static void netvsc_xmit_completion(void *context)
110 {
111         struct hv_netvsc_packet *packet = (struct hv_netvsc_packet *)context;
112         struct sk_buff *skb = (struct sk_buff *)
113                 (unsigned long)packet->completion.send.send_completion_tid;
114
115         kfree(packet);
116
117         if (skb) {
118                 struct net_device *net = skb->dev;
119                 struct net_device_context *net_device_ctx = netdev_priv(net);
120                 unsigned int num_pages = skb_shinfo(skb)->nr_frags + 2;
121
122                 dev_kfree_skb_any(skb);
123
124                 net_device_ctx->avail += num_pages;
125                 if (net_device_ctx->avail >= PACKET_PAGES_HIWATER)
126                         netif_wake_queue(net);
127         }
128 }
129
130 static int netvsc_start_xmit(struct sk_buff *skb, struct net_device *net)
131 {
132         struct net_device_context *net_device_ctx = netdev_priv(net);
133         struct hv_driver *drv =
134             drv_to_hv_drv(net_device_ctx->device_ctx->device.driver);
135         struct netvsc_driver *net_drv_obj = drv->priv;
136         struct hv_netvsc_packet *packet;
137         int ret;
138         unsigned int i, num_pages;
139
140         DPRINT_DBG(NETVSC_DRV, "xmit packet - len %d data_len %d",
141                    skb->len, skb->data_len);
142
143         /* Add 1 for skb->data and additional one for RNDIS */
144         num_pages = skb_shinfo(skb)->nr_frags + 1 + 1;
145         if (num_pages > net_device_ctx->avail)
146                 return NETDEV_TX_BUSY;
147
148         /* Allocate a netvsc packet based on # of frags. */
149         packet = kzalloc(sizeof(struct hv_netvsc_packet) +
150                          (num_pages * sizeof(struct hv_page_buffer)) +
151                          net_drv_obj->req_ext_size, GFP_ATOMIC);
152         if (!packet) {
153                 /* out of memory, silently drop packet */
154                 DPRINT_ERR(NETVSC_DRV, "unable to allocate hv_netvsc_packet");
155
156                 dev_kfree_skb(skb);
157                 net->stats.tx_dropped++;
158                 return NETDEV_TX_OK;
159         }
160
161         packet->extension = (void *)(unsigned long)packet +
162                                 sizeof(struct hv_netvsc_packet) +
163                                     (num_pages * sizeof(struct hv_page_buffer));
164
165         /* Setup the rndis header */
166         packet->page_buf_cnt = num_pages;
167
168         /* TODO: Flush all write buffers/ memory fence ??? */
169         /* wmb(); */
170
171         /* Initialize it from the skb */
172         packet->total_data_buflen       = skb->len;
173
174         /* Start filling in the page buffers starting after RNDIS buffer. */
175         packet->page_buf[1].pfn = virt_to_phys(skb->data) >> PAGE_SHIFT;
176         packet->page_buf[1].offset
177                 = (unsigned long)skb->data & (PAGE_SIZE - 1);
178         packet->page_buf[1].len = skb_headlen(skb);
179
180         /* Additional fragments are after SKB data */
181         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
182                 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
183
184                 packet->page_buf[i+2].pfn = page_to_pfn(f->page);
185                 packet->page_buf[i+2].offset = f->page_offset;
186                 packet->page_buf[i+2].len = f->size;
187         }
188
189         /* Set the completion routine */
190         packet->completion.send.send_completion = netvsc_xmit_completion;
191         packet->completion.send.send_completion_ctx = packet;
192         packet->completion.send.send_completion_tid = (unsigned long)skb;
193
194         ret = net_drv_obj->send(net_device_ctx->device_ctx,
195                                   packet);
196         if (ret == 0) {
197                 net->stats.tx_bytes += skb->len;
198                 net->stats.tx_packets++;
199
200                 DPRINT_DBG(NETVSC_DRV, "# of xmits %lu total size %lu",
201                            net->stats.tx_packets,
202                            net->stats.tx_bytes);
203
204                 net_device_ctx->avail -= num_pages;
205                 if (net_device_ctx->avail < PACKET_PAGES_LOWATER)
206                         netif_stop_queue(net);
207         } else {
208                 /* we are shutting down or bus overloaded, just drop packet */
209                 net->stats.tx_dropped++;
210                 netvsc_xmit_completion(packet);
211         }
212
213         return NETDEV_TX_OK;
214 }
215
216 /*
217  * netvsc_linkstatus_callback - Link up/down notification
218  */
219 static void netvsc_linkstatus_callback(struct hv_device *device_obj,
220                                        unsigned int status)
221 {
222         struct net_device *net = dev_get_drvdata(&device_obj->device);
223         struct net_device_context *ndev_ctx;
224
225         if (!net) {
226                 DPRINT_ERR(NETVSC_DRV, "got link status but net device "
227                                 "not initialized yet");
228                 return;
229         }
230
231         if (status == 1) {
232                 netif_carrier_on(net);
233                 netif_wake_queue(net);
234                 netif_notify_peers(net);
235                 ndev_ctx = netdev_priv(net);
236                 schedule_work(&ndev_ctx->work);
237         } else {
238                 netif_carrier_off(net);
239                 netif_stop_queue(net);
240         }
241 }
242
243 /*
244  * netvsc_recv_callback -  Callback when we receive a packet from the
245  * "wire" on the specified device.
246  */
247 static int netvsc_recv_callback(struct hv_device *device_obj,
248                                 struct hv_netvsc_packet *packet)
249 {
250         struct net_device *net = dev_get_drvdata(&device_obj->device);
251         struct sk_buff *skb;
252         void *data;
253         int i;
254         unsigned long flags;
255
256         if (!net) {
257                 DPRINT_ERR(NETVSC_DRV, "got receive callback but net device "
258                                 "not initialized yet");
259                 return 0;
260         }
261
262         /* Allocate a skb - TODO direct I/O to pages? */
263         skb = netdev_alloc_skb_ip_align(net, packet->total_data_buflen);
264         if (unlikely(!skb)) {
265                 ++net->stats.rx_dropped;
266                 return 0;
267         }
268
269         /* for kmap_atomic */
270         local_irq_save(flags);
271
272         /*
273          * Copy to skb. This copy is needed here since the memory pointed by
274          * hv_netvsc_packet cannot be deallocated
275          */
276         for (i = 0; i < packet->page_buf_cnt; i++) {
277                 data = kmap_atomic(pfn_to_page(packet->page_buf[i].pfn),
278                                                KM_IRQ1);
279                 data = (void *)(unsigned long)data +
280                                 packet->page_buf[i].offset;
281
282                 memcpy(skb_put(skb, packet->page_buf[i].len), data,
283                        packet->page_buf[i].len);
284
285                 kunmap_atomic((void *)((unsigned long)data -
286                                        packet->page_buf[i].offset), KM_IRQ1);
287         }
288
289         local_irq_restore(flags);
290
291         skb->protocol = eth_type_trans(skb, net);
292         skb->ip_summed = CHECKSUM_NONE;
293
294         net->stats.rx_packets++;
295         net->stats.rx_bytes += skb->len;
296
297         /*
298          * Pass the skb back up. Network stack will deallocate the skb when it
299          * is done.
300          * TODO - use NAPI?
301          */
302         netif_rx(skb);
303
304         DPRINT_DBG(NETVSC_DRV, "# of recvs %lu total size %lu",
305                    net->stats.rx_packets, net->stats.rx_bytes);
306
307         return 0;
308 }
309
310 static void netvsc_get_drvinfo(struct net_device *net,
311                                struct ethtool_drvinfo *info)
312 {
313         strcpy(info->driver, "hv_netvsc");
314         strcpy(info->version, HV_DRV_VERSION);
315         strcpy(info->fw_version, "N/A");
316 }
317
318 static const struct ethtool_ops ethtool_ops = {
319         .get_drvinfo    = netvsc_get_drvinfo,
320         .get_sg         = ethtool_op_get_sg,
321         .set_sg         = ethtool_op_set_sg,
322         .get_link       = ethtool_op_get_link,
323 };
324
325 static const struct net_device_ops device_ops = {
326         .ndo_open =                     netvsc_open,
327         .ndo_stop =                     netvsc_close,
328         .ndo_start_xmit =               netvsc_start_xmit,
329         .ndo_set_multicast_list =       netvsc_set_multicast_list,
330         .ndo_change_mtu =               eth_change_mtu,
331         .ndo_validate_addr =            eth_validate_addr,
332         .ndo_set_mac_address =          eth_mac_addr,
333 };
334
335 /*
336  * Send GARP packet to network peers after migrations.
337  * After Quick Migration, the network is not immediately operational in the
338  * current context when receiving RNDIS_STATUS_MEDIA_CONNECT event. So, add
339  * another netif_notify_peers() into a scheduled work, otherwise GARP packet
340  * will not be sent after quick migration, and cause network disconnection.
341  */
342 static void netvsc_send_garp(struct work_struct *w)
343 {
344         struct net_device_context *ndev_ctx;
345         struct net_device *net;
346
347         msleep(20);
348         ndev_ctx = container_of(w, struct net_device_context, work);
349         net = dev_get_drvdata(&ndev_ctx->device_ctx->device);
350         netif_notify_peers(net);
351 }
352
353
354 static int netvsc_probe(struct device *device)
355 {
356         struct hv_driver *drv =
357                 drv_to_hv_drv(device->driver);
358         struct netvsc_driver *net_drv_obj = drv->priv;
359         struct hv_device *device_obj = device_to_hv_device(device);
360         struct net_device *net = NULL;
361         struct net_device_context *net_device_ctx;
362         struct netvsc_device_info device_info;
363         int ret;
364
365         if (!net_drv_obj->base.dev_add)
366                 return -1;
367
368         net = alloc_etherdev(sizeof(struct net_device_context));
369         if (!net)
370                 return -1;
371
372         /* Set initial state */
373         netif_carrier_off(net);
374
375         net_device_ctx = netdev_priv(net);
376         net_device_ctx->device_ctx = device_obj;
377         net_device_ctx->avail = ring_size;
378         dev_set_drvdata(device, net);
379         INIT_WORK(&net_device_ctx->work, netvsc_send_garp);
380
381         /* Notify the netvsc driver of the new device */
382         ret = net_drv_obj->base.dev_add(device_obj, &device_info);
383         if (ret != 0) {
384                 free_netdev(net);
385                 dev_set_drvdata(device, NULL);
386
387                 DPRINT_ERR(NETVSC_DRV, "unable to add netvsc device (ret %d)",
388                            ret);
389                 return ret;
390         }
391
392         /*
393          * If carrier is still off ie we did not get a link status callback,
394          * update it if necessary
395          */
396         /*
397          * FIXME: We should use a atomic or test/set instead to avoid getting
398          * out of sync with the device's link status
399          */
400         if (!netif_carrier_ok(net))
401                 if (!device_info.link_state)
402                         netif_carrier_on(net);
403
404         memcpy(net->dev_addr, device_info.mac_adr, ETH_ALEN);
405
406         net->netdev_ops = &device_ops;
407
408         /* TODO: Add GSO and Checksum offload */
409         net->features = NETIF_F_SG;
410
411         SET_ETHTOOL_OPS(net, &ethtool_ops);
412         SET_NETDEV_DEV(net, device);
413
414         ret = register_netdev(net);
415         if (ret != 0) {
416                 /* Remove the device and release the resource */
417                 net_drv_obj->base.dev_rm(device_obj);
418                 free_netdev(net);
419         }
420
421         return ret;
422 }
423
424 static int netvsc_remove(struct device *device)
425 {
426         struct hv_driver *drv =
427                 drv_to_hv_drv(device->driver);
428         struct netvsc_driver *net_drv_obj = drv->priv;
429         struct hv_device *device_obj = device_to_hv_device(device);
430         struct net_device *net = dev_get_drvdata(&device_obj->device);
431         int ret;
432
433         if (net == NULL) {
434                 DPRINT_INFO(NETVSC, "no net device to remove");
435                 return 0;
436         }
437
438         if (!net_drv_obj->base.dev_rm)
439                 return -1;
440
441         /* Stop outbound asap */
442         netif_stop_queue(net);
443         /* netif_carrier_off(net); */
444
445         unregister_netdev(net);
446
447         /*
448          * Call to the vsc driver to let it know that the device is being
449          * removed
450          */
451         ret = net_drv_obj->base.dev_rm(device_obj);
452         if (ret != 0) {
453                 /* TODO: */
454                 DPRINT_ERR(NETVSC, "unable to remove vsc device (ret %d)", ret);
455         }
456
457         free_netdev(net);
458         return ret;
459 }
460
461 static int netvsc_drv_exit_cb(struct device *dev, void *data)
462 {
463         struct device **curr = (struct device **)data;
464
465         *curr = dev;
466         /* stop iterating */
467         return 1;
468 }
469
470 static void netvsc_drv_exit(void)
471 {
472         struct netvsc_driver *netvsc_drv_obj = &g_netvsc_drv;
473         struct hv_driver *drv = &g_netvsc_drv.base;
474         struct device *current_dev;
475         int ret;
476
477         while (1) {
478                 current_dev = NULL;
479
480                 /* Get the device */
481                 ret = driver_for_each_device(&drv->driver, NULL,
482                                              &current_dev, netvsc_drv_exit_cb);
483                 if (ret)
484                         DPRINT_WARN(NETVSC_DRV,
485                                     "driver_for_each_device returned %d", ret);
486
487                 if (current_dev == NULL)
488                         break;
489
490                 /* Initiate removal from the top-down */
491                 DPRINT_INFO(NETVSC_DRV, "unregistering device (%p)...",
492                             current_dev);
493
494                 device_unregister(current_dev);
495         }
496
497         if (netvsc_drv_obj->base.cleanup)
498                 netvsc_drv_obj->base.cleanup(&netvsc_drv_obj->base);
499
500         vmbus_child_driver_unregister(&drv->driver);
501
502         return;
503 }
504
505 static int netvsc_drv_init(int (*drv_init)(struct hv_driver *drv))
506 {
507         struct netvsc_driver *net_drv_obj = &g_netvsc_drv;
508         struct hv_driver *drv = &g_netvsc_drv.base;
509         int ret;
510
511         net_drv_obj->ring_buf_size = ring_size * PAGE_SIZE;
512         net_drv_obj->recv_cb = netvsc_recv_callback;
513         net_drv_obj->link_status_change = netvsc_linkstatus_callback;
514         drv->priv = net_drv_obj;
515
516         /* Callback to client driver to complete the initialization */
517         drv_init(&net_drv_obj->base);
518
519         drv->driver.name = net_drv_obj->base.name;
520
521         drv->driver.probe = netvsc_probe;
522         drv->driver.remove = netvsc_remove;
523
524         /* The driver belongs to vmbus */
525         ret = vmbus_child_driver_register(&drv->driver);
526
527         return ret;
528 }
529
530 static const struct dmi_system_id __initconst
531 hv_netvsc_dmi_table[] __maybe_unused  = {
532         {
533                 .ident = "Hyper-V",
534                 .matches = {
535                         DMI_MATCH(DMI_SYS_VENDOR, "Microsoft Corporation"),
536                         DMI_MATCH(DMI_PRODUCT_NAME, "Virtual Machine"),
537                         DMI_MATCH(DMI_BOARD_NAME, "Virtual Machine"),
538                 },
539         },
540         { },
541 };
542 MODULE_DEVICE_TABLE(dmi, hv_netvsc_dmi_table);
543
544 static int __init netvsc_init(void)
545 {
546         DPRINT_INFO(NETVSC_DRV, "Netvsc initializing....");
547
548         if (!dmi_check_system(hv_netvsc_dmi_table))
549                 return -ENODEV;
550
551         return netvsc_drv_init(netvsc_initialize);
552 }
553
554 static void __exit netvsc_exit(void)
555 {
556         netvsc_drv_exit();
557 }
558
559 static const struct pci_device_id __initconst
560 hv_netvsc_pci_table[] __maybe_unused = {
561         { PCI_DEVICE(0x1414, 0x5353) }, /* VGA compatible controller */
562         { 0 }
563 };
564 MODULE_DEVICE_TABLE(pci, hv_netvsc_pci_table);
565
566 MODULE_LICENSE("GPL");
567 MODULE_VERSION(HV_DRV_VERSION);
568 MODULE_DESCRIPTION("Microsoft Hyper-V network driver");
569
570 module_init(netvsc_init);
571 module_exit(netvsc_exit);