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