1 /*
2  * Network-device interface management.
3  *
4  * Copyright (c) 2004-2005, Keir Fraser
5  *
6  * This program is free software; you can redistribute it and/or
7  * modify it under the terms of the GNU General Public License version 2
8  * as published by the Free Software Foundation; or, when distributed
9  * separately from the Linux kernel or incorporated into other
10  * software packages, subject to the following license:
11  *
12  * Permission is hereby granted, free of charge, to any person obtaining a copy
13  * of this source file (the "Software"), to deal in the Software without
14  * restriction, including without limitation the rights to use, copy, modify,
15  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
16  * and to permit persons to whom the Software is furnished to do so, subject to
17  * the following conditions:
18  *
19  * The above copyright notice and this permission notice shall be included in
20  * all copies or substantial portions of the Software.
21  *
22  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
23  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
24  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
25  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
26  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
27  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
28  * IN THE SOFTWARE.
29  */
30 
31 #include "common.h"
32 
33 #include <linux/kthread.h>
34 #include <linux/ethtool.h>
35 #include <linux/rtnetlink.h>
36 #include <linux/if_vlan.h>
37 #include <linux/vmalloc.h>
38 
39 #include <xen/events.h>
40 #include <asm/xen/hypercall.h>
41 
42 #define XENVIF_QUEUE_LENGTH 32
43 #define XENVIF_NAPI_WEIGHT  64
44 
45 int xenvif_schedulable(struct xenvif *vif)
46 {
47 	return netif_running(vif->dev) && netif_carrier_ok(vif->dev);
48 }
49 
50 static int xenvif_rx_schedulable(struct xenvif *vif)
51 {
52 	return xenvif_schedulable(vif) && !xenvif_rx_ring_full(vif);
53 }
54 
55 static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
56 {
57 	struct xenvif *vif = dev_id;
58 
59 	if (RING_HAS_UNCONSUMED_REQUESTS(&vif->tx))
60 		napi_schedule(&vif->napi);
61 
62 	return IRQ_HANDLED;
63 }
64 
65 static int xenvif_poll(struct napi_struct *napi, int budget)
66 {
67 	struct xenvif *vif = container_of(napi, struct xenvif, napi);
68 	int work_done;
69 
70 	work_done = xenvif_tx_action(vif, budget);
71 
72 	if (work_done < budget) {
73 		int more_to_do = 0;
74 		unsigned long flags;
75 
76 		/* It is necessary to disable IRQ before calling
77 		 * RING_HAS_UNCONSUMED_REQUESTS. Otherwise we might
78 		 * lose event from the frontend.
79 		 *
80 		 * Consider:
81 		 *   RING_HAS_UNCONSUMED_REQUESTS
82 		 *   <frontend generates event to trigger napi_schedule>
83 		 *   __napi_complete
84 		 *
85 		 * This handler is still in scheduled state so the
86 		 * event has no effect at all. After __napi_complete
87 		 * this handler is descheduled and cannot get
88 		 * scheduled again. We lose event in this case and the ring
89 		 * will be completely stalled.
90 		 */
91 
92 		local_irq_save(flags);
93 
94 		RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do);
95 		if (!more_to_do)
96 			__napi_complete(napi);
97 
98 		local_irq_restore(flags);
99 	}
100 
101 	return work_done;
102 }
103 
104 static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
105 {
106 	struct xenvif *vif = dev_id;
107 
108 	if (xenvif_rx_schedulable(vif))
109 		netif_wake_queue(vif->dev);
110 
111 	return IRQ_HANDLED;
112 }
113 
114 static irqreturn_t xenvif_interrupt(int irq, void *dev_id)
115 {
116 	xenvif_tx_interrupt(irq, dev_id);
117 	xenvif_rx_interrupt(irq, dev_id);
118 
119 	return IRQ_HANDLED;
120 }
121 
122 static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
123 {
124 	struct xenvif *vif = netdev_priv(dev);
125 
126 	BUG_ON(skb->dev != dev);
127 
128 	/* Drop the packet if vif is not ready */
129 	if (vif->task == NULL)
130 		goto drop;
131 
132 	/* Drop the packet if the target domain has no receive buffers. */
133 	if (!xenvif_rx_schedulable(vif))
134 		goto drop;
135 
136 	/* Reserve ring slots for the worst-case number of fragments. */
137 	vif->rx_req_cons_peek += xenvif_count_skb_slots(vif, skb);
138 
139 	if (vif->can_queue && xenvif_must_stop_queue(vif))
140 		netif_stop_queue(dev);
141 
142 	xenvif_queue_tx_skb(vif, skb);
143 
144 	return NETDEV_TX_OK;
145 
146  drop:
147 	vif->dev->stats.tx_dropped++;
148 	dev_kfree_skb(skb);
149 	return NETDEV_TX_OK;
150 }
151 
152 void xenvif_notify_tx_completion(struct xenvif *vif)
153 {
154 	if (netif_queue_stopped(vif->dev) && xenvif_rx_schedulable(vif))
155 		netif_wake_queue(vif->dev);
156 }
157 
158 static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
159 {
160 	struct xenvif *vif = netdev_priv(dev);
161 	return &vif->dev->stats;
162 }
163 
164 static void xenvif_up(struct xenvif *vif)
165 {
166 	napi_enable(&vif->napi);
167 	enable_irq(vif->tx_irq);
168 	if (vif->tx_irq != vif->rx_irq)
169 		enable_irq(vif->rx_irq);
170 	xenvif_check_rx_xenvif(vif);
171 }
172 
173 static void xenvif_down(struct xenvif *vif)
174 {
175 	napi_disable(&vif->napi);
176 	disable_irq(vif->tx_irq);
177 	if (vif->tx_irq != vif->rx_irq)
178 		disable_irq(vif->rx_irq);
179 	del_timer_sync(&vif->credit_timeout);
180 }
181 
182 static int xenvif_open(struct net_device *dev)
183 {
184 	struct xenvif *vif = netdev_priv(dev);
185 	if (netif_carrier_ok(dev))
186 		xenvif_up(vif);
187 	netif_start_queue(dev);
188 	return 0;
189 }
190 
191 static int xenvif_close(struct net_device *dev)
192 {
193 	struct xenvif *vif = netdev_priv(dev);
194 	if (netif_carrier_ok(dev))
195 		xenvif_down(vif);
196 	netif_stop_queue(dev);
197 	return 0;
198 }
199 
200 static int xenvif_change_mtu(struct net_device *dev, int mtu)
201 {
202 	struct xenvif *vif = netdev_priv(dev);
203 	int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
204 
205 	if (mtu > max)
206 		return -EINVAL;
207 	dev->mtu = mtu;
208 	return 0;
209 }
210 
211 static netdev_features_t xenvif_fix_features(struct net_device *dev,
212 	netdev_features_t features)
213 {
214 	struct xenvif *vif = netdev_priv(dev);
215 
216 	if (!vif->can_sg)
217 		features &= ~NETIF_F_SG;
218 	if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
219 		features &= ~NETIF_F_TSO;
220 	if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6))
221 		features &= ~NETIF_F_TSO6;
222 	if (!vif->ip_csum)
223 		features &= ~NETIF_F_IP_CSUM;
224 	if (!vif->ipv6_csum)
225 		features &= ~NETIF_F_IPV6_CSUM;
226 
227 	return features;
228 }
229 
230 static const struct xenvif_stat {
231 	char name[ETH_GSTRING_LEN];
232 	u16 offset;
233 } xenvif_stats[] = {
234 	{
235 		"rx_gso_checksum_fixup",
236 		offsetof(struct xenvif, rx_gso_checksum_fixup)
237 	},
238 };
239 
240 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
241 {
242 	switch (string_set) {
243 	case ETH_SS_STATS:
244 		return ARRAY_SIZE(xenvif_stats);
245 	default:
246 		return -EINVAL;
247 	}
248 }
249 
250 static void xenvif_get_ethtool_stats(struct net_device *dev,
251 				     struct ethtool_stats *stats, u64 * data)
252 {
253 	void *vif = netdev_priv(dev);
254 	int i;
255 
256 	for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
257 		data[i] = *(unsigned long *)(vif + xenvif_stats[i].offset);
258 }
259 
260 static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
261 {
262 	int i;
263 
264 	switch (stringset) {
265 	case ETH_SS_STATS:
266 		for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
267 			memcpy(data + i * ETH_GSTRING_LEN,
268 			       xenvif_stats[i].name, ETH_GSTRING_LEN);
269 		break;
270 	}
271 }
272 
273 static const struct ethtool_ops xenvif_ethtool_ops = {
274 	.get_link	= ethtool_op_get_link,
275 
276 	.get_sset_count = xenvif_get_sset_count,
277 	.get_ethtool_stats = xenvif_get_ethtool_stats,
278 	.get_strings = xenvif_get_strings,
279 };
280 
281 static const struct net_device_ops xenvif_netdev_ops = {
282 	.ndo_start_xmit	= xenvif_start_xmit,
283 	.ndo_get_stats	= xenvif_get_stats,
284 	.ndo_open	= xenvif_open,
285 	.ndo_stop	= xenvif_close,
286 	.ndo_change_mtu	= xenvif_change_mtu,
287 	.ndo_fix_features = xenvif_fix_features,
288 	.ndo_set_mac_address = eth_mac_addr,
289 	.ndo_validate_addr   = eth_validate_addr,
290 };
291 
292 struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
293 			    unsigned int handle)
294 {
295 	int err;
296 	struct net_device *dev;
297 	struct xenvif *vif;
298 	char name[IFNAMSIZ] = {};
299 	int i;
300 
301 	snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
302 	dev = alloc_netdev(sizeof(struct xenvif), name, ether_setup);
303 	if (dev == NULL) {
304 		pr_warn("Could not allocate netdev for %s\n", name);
305 		return ERR_PTR(-ENOMEM);
306 	}
307 
308 	SET_NETDEV_DEV(dev, parent);
309 
310 	vif = netdev_priv(dev);
311 
312 	vif->grant_copy_op = vmalloc(sizeof(struct gnttab_copy) *
313 				     MAX_GRANT_COPY_OPS);
314 	if (vif->grant_copy_op == NULL) {
315 		pr_warn("Could not allocate grant copy space for %s\n", name);
316 		free_netdev(dev);
317 		return ERR_PTR(-ENOMEM);
318 	}
319 
320 	vif->domid  = domid;
321 	vif->handle = handle;
322 	vif->can_sg = 1;
323 	vif->ip_csum = 1;
324 	vif->dev = dev;
325 
326 	vif->credit_bytes = vif->remaining_credit = ~0UL;
327 	vif->credit_usec  = 0UL;
328 	init_timer(&vif->credit_timeout);
329 	vif->credit_window_start = get_jiffies_64();
330 
331 	dev->netdev_ops	= &xenvif_netdev_ops;
332 	dev->hw_features = NETIF_F_SG |
333 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
334 		NETIF_F_TSO | NETIF_F_TSO6;
335 	dev->features = dev->hw_features | NETIF_F_RXCSUM;
336 	SET_ETHTOOL_OPS(dev, &xenvif_ethtool_ops);
337 
338 	dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
339 
340 	skb_queue_head_init(&vif->rx_queue);
341 	skb_queue_head_init(&vif->tx_queue);
342 
343 	vif->pending_cons = 0;
344 	vif->pending_prod = MAX_PENDING_REQS;
345 	for (i = 0; i < MAX_PENDING_REQS; i++)
346 		vif->pending_ring[i] = i;
347 	for (i = 0; i < MAX_PENDING_REQS; i++)
348 		vif->mmap_pages[i] = NULL;
349 
350 	/*
351 	 * Initialise a dummy MAC address. We choose the numerically
352 	 * largest non-broadcast address to prevent the address getting
353 	 * stolen by an Ethernet bridge for STP purposes.
354 	 * (FE:FF:FF:FF:FF:FF)
355 	 */
356 	memset(dev->dev_addr, 0xFF, ETH_ALEN);
357 	dev->dev_addr[0] &= ~0x01;
358 
359 	netif_napi_add(dev, &vif->napi, xenvif_poll, XENVIF_NAPI_WEIGHT);
360 
361 	netif_carrier_off(dev);
362 
363 	err = register_netdev(dev);
364 	if (err) {
365 		netdev_warn(dev, "Could not register device: err=%d\n", err);
366 		free_netdev(dev);
367 		return ERR_PTR(err);
368 	}
369 
370 	netdev_dbg(dev, "Successfully created xenvif\n");
371 
372 	__module_get(THIS_MODULE);
373 
374 	return vif;
375 }
376 
377 int xenvif_connect(struct xenvif *vif, unsigned long tx_ring_ref,
378 		   unsigned long rx_ring_ref, unsigned int tx_evtchn,
379 		   unsigned int rx_evtchn)
380 {
381 	struct task_struct *task;
382 	int err = -ENOMEM;
383 
384 	BUG_ON(vif->tx_irq);
385 	BUG_ON(vif->task);
386 
387 	err = xenvif_map_frontend_rings(vif, tx_ring_ref, rx_ring_ref);
388 	if (err < 0)
389 		goto err;
390 
391 	if (tx_evtchn == rx_evtchn) {
392 		/* feature-split-event-channels == 0 */
393 		err = bind_interdomain_evtchn_to_irqhandler(
394 			vif->domid, tx_evtchn, xenvif_interrupt, 0,
395 			vif->dev->name, vif);
396 		if (err < 0)
397 			goto err_unmap;
398 		vif->tx_irq = vif->rx_irq = err;
399 		disable_irq(vif->tx_irq);
400 	} else {
401 		/* feature-split-event-channels == 1 */
402 		snprintf(vif->tx_irq_name, sizeof(vif->tx_irq_name),
403 			 "%s-tx", vif->dev->name);
404 		err = bind_interdomain_evtchn_to_irqhandler(
405 			vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
406 			vif->tx_irq_name, vif);
407 		if (err < 0)
408 			goto err_unmap;
409 		vif->tx_irq = err;
410 		disable_irq(vif->tx_irq);
411 
412 		snprintf(vif->rx_irq_name, sizeof(vif->rx_irq_name),
413 			 "%s-rx", vif->dev->name);
414 		err = bind_interdomain_evtchn_to_irqhandler(
415 			vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
416 			vif->rx_irq_name, vif);
417 		if (err < 0)
418 			goto err_tx_unbind;
419 		vif->rx_irq = err;
420 		disable_irq(vif->rx_irq);
421 	}
422 
423 	init_waitqueue_head(&vif->wq);
424 	task = kthread_create(xenvif_kthread,
425 			      (void *)vif, "%s", vif->dev->name);
426 	if (IS_ERR(task)) {
427 		pr_warn("Could not allocate kthread for %s\n", vif->dev->name);
428 		err = PTR_ERR(task);
429 		goto err_rx_unbind;
430 	}
431 
432 	vif->task = task;
433 
434 	rtnl_lock();
435 	if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
436 		dev_set_mtu(vif->dev, ETH_DATA_LEN);
437 	netdev_update_features(vif->dev);
438 	netif_carrier_on(vif->dev);
439 	if (netif_running(vif->dev))
440 		xenvif_up(vif);
441 	rtnl_unlock();
442 
443 	wake_up_process(vif->task);
444 
445 	return 0;
446 
447 err_rx_unbind:
448 	unbind_from_irqhandler(vif->rx_irq, vif);
449 	vif->rx_irq = 0;
450 err_tx_unbind:
451 	unbind_from_irqhandler(vif->tx_irq, vif);
452 	vif->tx_irq = 0;
453 err_unmap:
454 	xenvif_unmap_frontend_rings(vif);
455 err:
456 	module_put(THIS_MODULE);
457 	return err;
458 }
459 
460 void xenvif_carrier_off(struct xenvif *vif)
461 {
462 	struct net_device *dev = vif->dev;
463 
464 	rtnl_lock();
465 	netif_carrier_off(dev); /* discard queued packets */
466 	if (netif_running(dev))
467 		xenvif_down(vif);
468 	rtnl_unlock();
469 }
470 
471 void xenvif_disconnect(struct xenvif *vif)
472 {
473 	if (netif_carrier_ok(vif->dev))
474 		xenvif_carrier_off(vif);
475 
476 	if (vif->task) {
477 		kthread_stop(vif->task);
478 		vif->task = NULL;
479 	}
480 
481 	if (vif->tx_irq) {
482 		if (vif->tx_irq == vif->rx_irq)
483 			unbind_from_irqhandler(vif->tx_irq, vif);
484 		else {
485 			unbind_from_irqhandler(vif->tx_irq, vif);
486 			unbind_from_irqhandler(vif->rx_irq, vif);
487 		}
488 		vif->tx_irq = 0;
489 	}
490 
491 	xenvif_unmap_frontend_rings(vif);
492 }
493 
494 void xenvif_free(struct xenvif *vif)
495 {
496 	netif_napi_del(&vif->napi);
497 
498 	unregister_netdev(vif->dev);
499 
500 	vfree(vif->grant_copy_op);
501 	free_netdev(vif->dev);
502 
503 	module_put(THIS_MODULE);
504 }
505