Loading...
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#include <xen/balloon.h>
42
43#define XENVIF_QUEUE_LENGTH 32
44#define XENVIF_NAPI_WEIGHT 64
45
46int xenvif_schedulable(struct xenvif *vif)
47{
48 return netif_running(vif->dev) && netif_carrier_ok(vif->dev);
49}
50
51static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
52{
53 struct xenvif *vif = dev_id;
54
55 if (RING_HAS_UNCONSUMED_REQUESTS(&vif->tx))
56 napi_schedule(&vif->napi);
57
58 return IRQ_HANDLED;
59}
60
61static int xenvif_poll(struct napi_struct *napi, int budget)
62{
63 struct xenvif *vif = container_of(napi, struct xenvif, napi);
64 int work_done;
65
66 /* This vif is rogue, we pretend we've there is nothing to do
67 * for this vif to deschedule it from NAPI. But this interface
68 * will be turned off in thread context later.
69 */
70 if (unlikely(vif->disabled)) {
71 napi_complete(napi);
72 return 0;
73 }
74
75 work_done = xenvif_tx_action(vif, budget);
76
77 if (work_done < budget) {
78 napi_complete(napi);
79 xenvif_napi_schedule_or_enable_events(vif);
80 }
81
82 return work_done;
83}
84
85static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
86{
87 struct xenvif *vif = dev_id;
88
89 xenvif_kick_thread(vif);
90
91 return IRQ_HANDLED;
92}
93
94static irqreturn_t xenvif_interrupt(int irq, void *dev_id)
95{
96 xenvif_tx_interrupt(irq, dev_id);
97 xenvif_rx_interrupt(irq, dev_id);
98
99 return IRQ_HANDLED;
100}
101
102static void xenvif_wake_queue(unsigned long data)
103{
104 struct xenvif *vif = (struct xenvif *)data;
105
106 if (netif_queue_stopped(vif->dev)) {
107 netdev_err(vif->dev, "draining TX queue\n");
108 vif->rx_queue_purge = true;
109 xenvif_kick_thread(vif);
110 netif_wake_queue(vif->dev);
111 }
112}
113
114static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
115{
116 struct xenvif *vif = netdev_priv(dev);
117 int min_slots_needed;
118
119 BUG_ON(skb->dev != dev);
120
121 /* Drop the packet if vif is not ready */
122 if (vif->task == NULL ||
123 vif->dealloc_task == NULL ||
124 !xenvif_schedulable(vif))
125 goto drop;
126
127 /* At best we'll need one slot for the header and one for each
128 * frag.
129 */
130 min_slots_needed = 1 + skb_shinfo(skb)->nr_frags;
131
132 /* If the skb is GSO then we'll also need an extra slot for the
133 * metadata.
134 */
135 if (skb_is_gso(skb))
136 min_slots_needed++;
137
138 /* If the skb can't possibly fit in the remaining slots
139 * then turn off the queue to give the ring a chance to
140 * drain.
141 */
142 if (!xenvif_rx_ring_slots_available(vif, min_slots_needed)) {
143 vif->wake_queue.function = xenvif_wake_queue;
144 vif->wake_queue.data = (unsigned long)vif;
145 xenvif_stop_queue(vif);
146 mod_timer(&vif->wake_queue,
147 jiffies + rx_drain_timeout_jiffies);
148 }
149
150 skb_queue_tail(&vif->rx_queue, skb);
151 xenvif_kick_thread(vif);
152
153 return NETDEV_TX_OK;
154
155 drop:
156 vif->dev->stats.tx_dropped++;
157 dev_kfree_skb(skb);
158 return NETDEV_TX_OK;
159}
160
161static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
162{
163 struct xenvif *vif = netdev_priv(dev);
164 return &vif->dev->stats;
165}
166
167static void xenvif_up(struct xenvif *vif)
168{
169 napi_enable(&vif->napi);
170 enable_irq(vif->tx_irq);
171 if (vif->tx_irq != vif->rx_irq)
172 enable_irq(vif->rx_irq);
173 xenvif_napi_schedule_or_enable_events(vif);
174}
175
176static void xenvif_down(struct xenvif *vif)
177{
178 napi_disable(&vif->napi);
179 disable_irq(vif->tx_irq);
180 if (vif->tx_irq != vif->rx_irq)
181 disable_irq(vif->rx_irq);
182 del_timer_sync(&vif->credit_timeout);
183}
184
185static int xenvif_open(struct net_device *dev)
186{
187 struct xenvif *vif = netdev_priv(dev);
188 if (netif_carrier_ok(dev))
189 xenvif_up(vif);
190 netif_start_queue(dev);
191 return 0;
192}
193
194static int xenvif_close(struct net_device *dev)
195{
196 struct xenvif *vif = netdev_priv(dev);
197 if (netif_carrier_ok(dev))
198 xenvif_down(vif);
199 netif_stop_queue(dev);
200 return 0;
201}
202
203static int xenvif_change_mtu(struct net_device *dev, int mtu)
204{
205 struct xenvif *vif = netdev_priv(dev);
206 int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
207
208 if (mtu > max)
209 return -EINVAL;
210 dev->mtu = mtu;
211 return 0;
212}
213
214static netdev_features_t xenvif_fix_features(struct net_device *dev,
215 netdev_features_t features)
216{
217 struct xenvif *vif = netdev_priv(dev);
218
219 if (!vif->can_sg)
220 features &= ~NETIF_F_SG;
221 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
222 features &= ~NETIF_F_TSO;
223 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6))
224 features &= ~NETIF_F_TSO6;
225 if (!vif->ip_csum)
226 features &= ~NETIF_F_IP_CSUM;
227 if (!vif->ipv6_csum)
228 features &= ~NETIF_F_IPV6_CSUM;
229
230 return features;
231}
232
233static const struct xenvif_stat {
234 char name[ETH_GSTRING_LEN];
235 u16 offset;
236} xenvif_stats[] = {
237 {
238 "rx_gso_checksum_fixup",
239 offsetof(struct xenvif, rx_gso_checksum_fixup)
240 },
241 /* If (sent != success + fail), there are probably packets never
242 * freed up properly!
243 */
244 {
245 "tx_zerocopy_sent",
246 offsetof(struct xenvif, tx_zerocopy_sent),
247 },
248 {
249 "tx_zerocopy_success",
250 offsetof(struct xenvif, tx_zerocopy_success),
251 },
252 {
253 "tx_zerocopy_fail",
254 offsetof(struct xenvif, tx_zerocopy_fail)
255 },
256 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
257 * a guest with the same MAX_SKB_FRAG
258 */
259 {
260 "tx_frag_overflow",
261 offsetof(struct xenvif, tx_frag_overflow)
262 },
263};
264
265static int xenvif_get_sset_count(struct net_device *dev, int string_set)
266{
267 switch (string_set) {
268 case ETH_SS_STATS:
269 return ARRAY_SIZE(xenvif_stats);
270 default:
271 return -EINVAL;
272 }
273}
274
275static void xenvif_get_ethtool_stats(struct net_device *dev,
276 struct ethtool_stats *stats, u64 * data)
277{
278 void *vif = netdev_priv(dev);
279 int i;
280
281 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
282 data[i] = *(unsigned long *)(vif + xenvif_stats[i].offset);
283}
284
285static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
286{
287 int i;
288
289 switch (stringset) {
290 case ETH_SS_STATS:
291 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
292 memcpy(data + i * ETH_GSTRING_LEN,
293 xenvif_stats[i].name, ETH_GSTRING_LEN);
294 break;
295 }
296}
297
298static const struct ethtool_ops xenvif_ethtool_ops = {
299 .get_link = ethtool_op_get_link,
300
301 .get_sset_count = xenvif_get_sset_count,
302 .get_ethtool_stats = xenvif_get_ethtool_stats,
303 .get_strings = xenvif_get_strings,
304};
305
306static const struct net_device_ops xenvif_netdev_ops = {
307 .ndo_start_xmit = xenvif_start_xmit,
308 .ndo_get_stats = xenvif_get_stats,
309 .ndo_open = xenvif_open,
310 .ndo_stop = xenvif_close,
311 .ndo_change_mtu = xenvif_change_mtu,
312 .ndo_fix_features = xenvif_fix_features,
313 .ndo_set_mac_address = eth_mac_addr,
314 .ndo_validate_addr = eth_validate_addr,
315};
316
317struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
318 unsigned int handle)
319{
320 int err;
321 struct net_device *dev;
322 struct xenvif *vif;
323 char name[IFNAMSIZ] = {};
324 int i;
325
326 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
327 dev = alloc_netdev(sizeof(struct xenvif), name, ether_setup);
328 if (dev == NULL) {
329 pr_warn("Could not allocate netdev for %s\n", name);
330 return ERR_PTR(-ENOMEM);
331 }
332
333 SET_NETDEV_DEV(dev, parent);
334
335 vif = netdev_priv(dev);
336
337 vif->grant_copy_op = vmalloc(sizeof(struct gnttab_copy) *
338 MAX_GRANT_COPY_OPS);
339 if (vif->grant_copy_op == NULL) {
340 pr_warn("Could not allocate grant copy space for %s\n", name);
341 free_netdev(dev);
342 return ERR_PTR(-ENOMEM);
343 }
344
345 vif->domid = domid;
346 vif->handle = handle;
347 vif->can_sg = 1;
348 vif->ip_csum = 1;
349 vif->dev = dev;
350
351 vif->disabled = false;
352
353 vif->credit_bytes = vif->remaining_credit = ~0UL;
354 vif->credit_usec = 0UL;
355 init_timer(&vif->credit_timeout);
356 vif->credit_window_start = get_jiffies_64();
357
358 init_timer(&vif->wake_queue);
359
360 dev->netdev_ops = &xenvif_netdev_ops;
361 dev->hw_features = NETIF_F_SG |
362 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
363 NETIF_F_TSO | NETIF_F_TSO6;
364 dev->features = dev->hw_features | NETIF_F_RXCSUM;
365 SET_ETHTOOL_OPS(dev, &xenvif_ethtool_ops);
366
367 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
368
369 skb_queue_head_init(&vif->rx_queue);
370 skb_queue_head_init(&vif->tx_queue);
371
372 vif->pending_cons = 0;
373 vif->pending_prod = MAX_PENDING_REQS;
374 for (i = 0; i < MAX_PENDING_REQS; i++)
375 vif->pending_ring[i] = i;
376 spin_lock_init(&vif->callback_lock);
377 spin_lock_init(&vif->response_lock);
378 /* If ballooning is disabled, this will consume real memory, so you
379 * better enable it. The long term solution would be to use just a
380 * bunch of valid page descriptors, without dependency on ballooning
381 */
382 err = alloc_xenballooned_pages(MAX_PENDING_REQS,
383 vif->mmap_pages,
384 false);
385 if (err) {
386 netdev_err(dev, "Could not reserve mmap_pages\n");
387 return ERR_PTR(-ENOMEM);
388 }
389 for (i = 0; i < MAX_PENDING_REQS; i++) {
390 vif->pending_tx_info[i].callback_struct = (struct ubuf_info)
391 { .callback = xenvif_zerocopy_callback,
392 .ctx = NULL,
393 .desc = i };
394 vif->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
395 }
396
397 /*
398 * Initialise a dummy MAC address. We choose the numerically
399 * largest non-broadcast address to prevent the address getting
400 * stolen by an Ethernet bridge for STP purposes.
401 * (FE:FF:FF:FF:FF:FF)
402 */
403 memset(dev->dev_addr, 0xFF, ETH_ALEN);
404 dev->dev_addr[0] &= ~0x01;
405
406 netif_napi_add(dev, &vif->napi, xenvif_poll, XENVIF_NAPI_WEIGHT);
407
408 netif_carrier_off(dev);
409
410 err = register_netdev(dev);
411 if (err) {
412 netdev_warn(dev, "Could not register device: err=%d\n", err);
413 free_netdev(dev);
414 return ERR_PTR(err);
415 }
416
417 netdev_dbg(dev, "Successfully created xenvif\n");
418
419 __module_get(THIS_MODULE);
420
421 return vif;
422}
423
424int xenvif_connect(struct xenvif *vif, unsigned long tx_ring_ref,
425 unsigned long rx_ring_ref, unsigned int tx_evtchn,
426 unsigned int rx_evtchn)
427{
428 struct task_struct *task;
429 int err = -ENOMEM;
430
431 BUG_ON(vif->tx_irq);
432 BUG_ON(vif->task);
433 BUG_ON(vif->dealloc_task);
434
435 err = xenvif_map_frontend_rings(vif, tx_ring_ref, rx_ring_ref);
436 if (err < 0)
437 goto err;
438
439 init_waitqueue_head(&vif->wq);
440 init_waitqueue_head(&vif->dealloc_wq);
441
442 if (tx_evtchn == rx_evtchn) {
443 /* feature-split-event-channels == 0 */
444 err = bind_interdomain_evtchn_to_irqhandler(
445 vif->domid, tx_evtchn, xenvif_interrupt, 0,
446 vif->dev->name, vif);
447 if (err < 0)
448 goto err_unmap;
449 vif->tx_irq = vif->rx_irq = err;
450 disable_irq(vif->tx_irq);
451 } else {
452 /* feature-split-event-channels == 1 */
453 snprintf(vif->tx_irq_name, sizeof(vif->tx_irq_name),
454 "%s-tx", vif->dev->name);
455 err = bind_interdomain_evtchn_to_irqhandler(
456 vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
457 vif->tx_irq_name, vif);
458 if (err < 0)
459 goto err_unmap;
460 vif->tx_irq = err;
461 disable_irq(vif->tx_irq);
462
463 snprintf(vif->rx_irq_name, sizeof(vif->rx_irq_name),
464 "%s-rx", vif->dev->name);
465 err = bind_interdomain_evtchn_to_irqhandler(
466 vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
467 vif->rx_irq_name, vif);
468 if (err < 0)
469 goto err_tx_unbind;
470 vif->rx_irq = err;
471 disable_irq(vif->rx_irq);
472 }
473
474 task = kthread_create(xenvif_kthread_guest_rx,
475 (void *)vif, "%s-guest-rx", vif->dev->name);
476 if (IS_ERR(task)) {
477 pr_warn("Could not allocate kthread for %s\n", vif->dev->name);
478 err = PTR_ERR(task);
479 goto err_rx_unbind;
480 }
481
482 vif->task = task;
483
484 task = kthread_create(xenvif_dealloc_kthread,
485 (void *)vif, "%s-dealloc", vif->dev->name);
486 if (IS_ERR(task)) {
487 pr_warn("Could not allocate kthread for %s\n", vif->dev->name);
488 err = PTR_ERR(task);
489 goto err_rx_unbind;
490 }
491
492 vif->dealloc_task = task;
493
494 rtnl_lock();
495 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
496 dev_set_mtu(vif->dev, ETH_DATA_LEN);
497 netdev_update_features(vif->dev);
498 netif_carrier_on(vif->dev);
499 if (netif_running(vif->dev))
500 xenvif_up(vif);
501 rtnl_unlock();
502
503 wake_up_process(vif->task);
504 wake_up_process(vif->dealloc_task);
505
506 return 0;
507
508err_rx_unbind:
509 unbind_from_irqhandler(vif->rx_irq, vif);
510 vif->rx_irq = 0;
511err_tx_unbind:
512 unbind_from_irqhandler(vif->tx_irq, vif);
513 vif->tx_irq = 0;
514err_unmap:
515 xenvif_unmap_frontend_rings(vif);
516err:
517 module_put(THIS_MODULE);
518 return err;
519}
520
521void xenvif_carrier_off(struct xenvif *vif)
522{
523 struct net_device *dev = vif->dev;
524
525 rtnl_lock();
526 netif_carrier_off(dev); /* discard queued packets */
527 if (netif_running(dev))
528 xenvif_down(vif);
529 rtnl_unlock();
530}
531
532void xenvif_disconnect(struct xenvif *vif)
533{
534 if (netif_carrier_ok(vif->dev))
535 xenvif_carrier_off(vif);
536
537 if (vif->task) {
538 del_timer_sync(&vif->wake_queue);
539 kthread_stop(vif->task);
540 vif->task = NULL;
541 }
542
543 if (vif->dealloc_task) {
544 kthread_stop(vif->dealloc_task);
545 vif->dealloc_task = NULL;
546 }
547
548 if (vif->tx_irq) {
549 if (vif->tx_irq == vif->rx_irq)
550 unbind_from_irqhandler(vif->tx_irq, vif);
551 else {
552 unbind_from_irqhandler(vif->tx_irq, vif);
553 unbind_from_irqhandler(vif->rx_irq, vif);
554 }
555 vif->tx_irq = 0;
556 }
557
558 xenvif_unmap_frontend_rings(vif);
559}
560
561void xenvif_free(struct xenvif *vif)
562{
563 int i, unmap_timeout = 0;
564 /* Here we want to avoid timeout messages if an skb can be legitimately
565 * stuck somewhere else. Realistically this could be an another vif's
566 * internal or QDisc queue. That another vif also has this
567 * rx_drain_timeout_msecs timeout, but the timer only ditches the
568 * internal queue. After that, the QDisc queue can put in worst case
569 * XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS skbs into that another vif's
570 * internal queue, so we need several rounds of such timeouts until we
571 * can be sure that no another vif should have skb's from us. We are
572 * not sending more skb's, so newly stuck packets are not interesting
573 * for us here.
574 */
575 unsigned int worst_case_skb_lifetime = (rx_drain_timeout_msecs/1000) *
576 DIV_ROUND_UP(XENVIF_QUEUE_LENGTH, (XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS));
577
578 for (i = 0; i < MAX_PENDING_REQS; ++i) {
579 if (vif->grant_tx_handle[i] != NETBACK_INVALID_HANDLE) {
580 unmap_timeout++;
581 schedule_timeout(msecs_to_jiffies(1000));
582 if (unmap_timeout > worst_case_skb_lifetime &&
583 net_ratelimit())
584 netdev_err(vif->dev,
585 "Page still granted! Index: %x\n",
586 i);
587 /* If there are still unmapped pages, reset the loop to
588 * start checking again. We shouldn't exit here until
589 * dealloc thread and NAPI instance release all the
590 * pages. If a kernel bug causes the skbs to stall
591 * somewhere, the interface cannot be brought down
592 * properly.
593 */
594 i = -1;
595 }
596 }
597
598 free_xenballooned_pages(MAX_PENDING_REQS, vif->mmap_pages);
599
600 netif_napi_del(&vif->napi);
601
602 unregister_netdev(vif->dev);
603
604 vfree(vif->grant_copy_op);
605 free_netdev(vif->dev);
606
607 module_put(THIS_MODULE);
608}
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#include <xen/balloon.h>
42
43#define XENVIF_QUEUE_LENGTH 32
44#define XENVIF_NAPI_WEIGHT 64
45
46/* Number of bytes allowed on the internal guest Rx queue. */
47#define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE)
48
49/* This function is used to set SKBTX_DEV_ZEROCOPY as well as
50 * increasing the inflight counter. We need to increase the inflight
51 * counter because core driver calls into xenvif_zerocopy_callback
52 * which calls xenvif_skb_zerocopy_complete.
53 */
54void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
55 struct sk_buff *skb)
56{
57 skb_shinfo(skb)->tx_flags |= SKBTX_DEV_ZEROCOPY;
58 atomic_inc(&queue->inflight_packets);
59}
60
61void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
62{
63 atomic_dec(&queue->inflight_packets);
64
65 /* Wake the dealloc thread _after_ decrementing inflight_packets so
66 * that if kthread_stop() has already been called, the dealloc thread
67 * does not wait forever with nothing to wake it.
68 */
69 wake_up(&queue->dealloc_wq);
70}
71
72int xenvif_schedulable(struct xenvif *vif)
73{
74 return netif_running(vif->dev) &&
75 test_bit(VIF_STATUS_CONNECTED, &vif->status) &&
76 !vif->disabled;
77}
78
79static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
80{
81 struct xenvif_queue *queue = dev_id;
82
83 if (RING_HAS_UNCONSUMED_REQUESTS(&queue->tx))
84 napi_schedule(&queue->napi);
85
86 return IRQ_HANDLED;
87}
88
89static int xenvif_poll(struct napi_struct *napi, int budget)
90{
91 struct xenvif_queue *queue =
92 container_of(napi, struct xenvif_queue, napi);
93 int work_done;
94
95 /* This vif is rogue, we pretend we've there is nothing to do
96 * for this vif to deschedule it from NAPI. But this interface
97 * will be turned off in thread context later.
98 */
99 if (unlikely(queue->vif->disabled)) {
100 napi_complete(napi);
101 return 0;
102 }
103
104 work_done = xenvif_tx_action(queue, budget);
105
106 if (work_done < budget) {
107 napi_complete(napi);
108 xenvif_napi_schedule_or_enable_events(queue);
109 }
110
111 return work_done;
112}
113
114static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
115{
116 struct xenvif_queue *queue = dev_id;
117
118 xenvif_kick_thread(queue);
119
120 return IRQ_HANDLED;
121}
122
123irqreturn_t xenvif_interrupt(int irq, void *dev_id)
124{
125 xenvif_tx_interrupt(irq, dev_id);
126 xenvif_rx_interrupt(irq, dev_id);
127
128 return IRQ_HANDLED;
129}
130
131int xenvif_queue_stopped(struct xenvif_queue *queue)
132{
133 struct net_device *dev = queue->vif->dev;
134 unsigned int id = queue->id;
135 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
136}
137
138void xenvif_wake_queue(struct xenvif_queue *queue)
139{
140 struct net_device *dev = queue->vif->dev;
141 unsigned int id = queue->id;
142 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
143}
144
145static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
146{
147 struct xenvif *vif = netdev_priv(dev);
148 struct xenvif_queue *queue = NULL;
149 unsigned int num_queues = vif->num_queues;
150 u16 index;
151 struct xenvif_rx_cb *cb;
152
153 BUG_ON(skb->dev != dev);
154
155 /* Drop the packet if queues are not set up */
156 if (num_queues < 1)
157 goto drop;
158
159 /* Obtain the queue to be used to transmit this packet */
160 index = skb_get_queue_mapping(skb);
161 if (index >= num_queues) {
162 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n.",
163 index, vif->dev->name);
164 index %= num_queues;
165 }
166 queue = &vif->queues[index];
167
168 /* Drop the packet if queue is not ready */
169 if (queue->task == NULL ||
170 queue->dealloc_task == NULL ||
171 !xenvif_schedulable(vif))
172 goto drop;
173
174 if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
175 struct ethhdr *eth = (struct ethhdr *)skb->data;
176
177 if (!xenvif_mcast_match(vif, eth->h_dest))
178 goto drop;
179 }
180
181 cb = XENVIF_RX_CB(skb);
182 cb->expires = jiffies + vif->drain_timeout;
183
184 xenvif_rx_queue_tail(queue, skb);
185 xenvif_kick_thread(queue);
186
187 return NETDEV_TX_OK;
188
189 drop:
190 vif->dev->stats.tx_dropped++;
191 dev_kfree_skb(skb);
192 return NETDEV_TX_OK;
193}
194
195static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
196{
197 struct xenvif *vif = netdev_priv(dev);
198 struct xenvif_queue *queue = NULL;
199 unsigned int num_queues = vif->num_queues;
200 unsigned long rx_bytes = 0;
201 unsigned long rx_packets = 0;
202 unsigned long tx_bytes = 0;
203 unsigned long tx_packets = 0;
204 unsigned int index;
205
206 if (vif->queues == NULL)
207 goto out;
208
209 /* Aggregate tx and rx stats from each queue */
210 for (index = 0; index < num_queues; ++index) {
211 queue = &vif->queues[index];
212 rx_bytes += queue->stats.rx_bytes;
213 rx_packets += queue->stats.rx_packets;
214 tx_bytes += queue->stats.tx_bytes;
215 tx_packets += queue->stats.tx_packets;
216 }
217
218out:
219 vif->dev->stats.rx_bytes = rx_bytes;
220 vif->dev->stats.rx_packets = rx_packets;
221 vif->dev->stats.tx_bytes = tx_bytes;
222 vif->dev->stats.tx_packets = tx_packets;
223
224 return &vif->dev->stats;
225}
226
227static void xenvif_up(struct xenvif *vif)
228{
229 struct xenvif_queue *queue = NULL;
230 unsigned int num_queues = vif->num_queues;
231 unsigned int queue_index;
232
233 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
234 queue = &vif->queues[queue_index];
235 napi_enable(&queue->napi);
236 enable_irq(queue->tx_irq);
237 if (queue->tx_irq != queue->rx_irq)
238 enable_irq(queue->rx_irq);
239 xenvif_napi_schedule_or_enable_events(queue);
240 }
241}
242
243static void xenvif_down(struct xenvif *vif)
244{
245 struct xenvif_queue *queue = NULL;
246 unsigned int num_queues = vif->num_queues;
247 unsigned int queue_index;
248
249 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
250 queue = &vif->queues[queue_index];
251 disable_irq(queue->tx_irq);
252 if (queue->tx_irq != queue->rx_irq)
253 disable_irq(queue->rx_irq);
254 napi_disable(&queue->napi);
255 del_timer_sync(&queue->credit_timeout);
256 }
257}
258
259static int xenvif_open(struct net_device *dev)
260{
261 struct xenvif *vif = netdev_priv(dev);
262 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
263 xenvif_up(vif);
264 netif_tx_start_all_queues(dev);
265 return 0;
266}
267
268static int xenvif_close(struct net_device *dev)
269{
270 struct xenvif *vif = netdev_priv(dev);
271 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
272 xenvif_down(vif);
273 netif_tx_stop_all_queues(dev);
274 return 0;
275}
276
277static int xenvif_change_mtu(struct net_device *dev, int mtu)
278{
279 struct xenvif *vif = netdev_priv(dev);
280 int max = vif->can_sg ? 65535 - VLAN_ETH_HLEN : ETH_DATA_LEN;
281
282 if (mtu > max)
283 return -EINVAL;
284 dev->mtu = mtu;
285 return 0;
286}
287
288static netdev_features_t xenvif_fix_features(struct net_device *dev,
289 netdev_features_t features)
290{
291 struct xenvif *vif = netdev_priv(dev);
292
293 if (!vif->can_sg)
294 features &= ~NETIF_F_SG;
295 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV4))
296 features &= ~NETIF_F_TSO;
297 if (~(vif->gso_mask | vif->gso_prefix_mask) & GSO_BIT(TCPV6))
298 features &= ~NETIF_F_TSO6;
299 if (!vif->ip_csum)
300 features &= ~NETIF_F_IP_CSUM;
301 if (!vif->ipv6_csum)
302 features &= ~NETIF_F_IPV6_CSUM;
303
304 return features;
305}
306
307static const struct xenvif_stat {
308 char name[ETH_GSTRING_LEN];
309 u16 offset;
310} xenvif_stats[] = {
311 {
312 "rx_gso_checksum_fixup",
313 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
314 },
315 /* If (sent != success + fail), there are probably packets never
316 * freed up properly!
317 */
318 {
319 "tx_zerocopy_sent",
320 offsetof(struct xenvif_stats, tx_zerocopy_sent),
321 },
322 {
323 "tx_zerocopy_success",
324 offsetof(struct xenvif_stats, tx_zerocopy_success),
325 },
326 {
327 "tx_zerocopy_fail",
328 offsetof(struct xenvif_stats, tx_zerocopy_fail)
329 },
330 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
331 * a guest with the same MAX_SKB_FRAG
332 */
333 {
334 "tx_frag_overflow",
335 offsetof(struct xenvif_stats, tx_frag_overflow)
336 },
337};
338
339static int xenvif_get_sset_count(struct net_device *dev, int string_set)
340{
341 switch (string_set) {
342 case ETH_SS_STATS:
343 return ARRAY_SIZE(xenvif_stats);
344 default:
345 return -EINVAL;
346 }
347}
348
349static void xenvif_get_ethtool_stats(struct net_device *dev,
350 struct ethtool_stats *stats, u64 * data)
351{
352 struct xenvif *vif = netdev_priv(dev);
353 unsigned int num_queues = vif->num_queues;
354 int i;
355 unsigned int queue_index;
356
357 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
358 unsigned long accum = 0;
359 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
360 void *vif_stats = &vif->queues[queue_index].stats;
361 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
362 }
363 data[i] = accum;
364 }
365}
366
367static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
368{
369 int i;
370
371 switch (stringset) {
372 case ETH_SS_STATS:
373 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
374 memcpy(data + i * ETH_GSTRING_LEN,
375 xenvif_stats[i].name, ETH_GSTRING_LEN);
376 break;
377 }
378}
379
380static const struct ethtool_ops xenvif_ethtool_ops = {
381 .get_link = ethtool_op_get_link,
382
383 .get_sset_count = xenvif_get_sset_count,
384 .get_ethtool_stats = xenvif_get_ethtool_stats,
385 .get_strings = xenvif_get_strings,
386};
387
388static const struct net_device_ops xenvif_netdev_ops = {
389 .ndo_start_xmit = xenvif_start_xmit,
390 .ndo_get_stats = xenvif_get_stats,
391 .ndo_open = xenvif_open,
392 .ndo_stop = xenvif_close,
393 .ndo_change_mtu = xenvif_change_mtu,
394 .ndo_fix_features = xenvif_fix_features,
395 .ndo_set_mac_address = eth_mac_addr,
396 .ndo_validate_addr = eth_validate_addr,
397};
398
399struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
400 unsigned int handle)
401{
402 int err;
403 struct net_device *dev;
404 struct xenvif *vif;
405 char name[IFNAMSIZ] = {};
406
407 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
408 /* Allocate a netdev with the max. supported number of queues.
409 * When the guest selects the desired number, it will be updated
410 * via netif_set_real_num_*_queues().
411 */
412 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
413 ether_setup, xenvif_max_queues);
414 if (dev == NULL) {
415 pr_warn("Could not allocate netdev for %s\n", name);
416 return ERR_PTR(-ENOMEM);
417 }
418
419 SET_NETDEV_DEV(dev, parent);
420
421 vif = netdev_priv(dev);
422
423 vif->domid = domid;
424 vif->handle = handle;
425 vif->can_sg = 1;
426 vif->ip_csum = 1;
427 vif->dev = dev;
428 vif->disabled = false;
429 vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
430 vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
431
432 /* Start out with no queues. */
433 vif->queues = NULL;
434 vif->num_queues = 0;
435
436 spin_lock_init(&vif->lock);
437 INIT_LIST_HEAD(&vif->fe_mcast_addr);
438
439 dev->netdev_ops = &xenvif_netdev_ops;
440 dev->hw_features = NETIF_F_SG |
441 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
442 NETIF_F_TSO | NETIF_F_TSO6;
443 dev->features = dev->hw_features | NETIF_F_RXCSUM;
444 dev->ethtool_ops = &xenvif_ethtool_ops;
445
446 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
447
448 /*
449 * Initialise a dummy MAC address. We choose the numerically
450 * largest non-broadcast address to prevent the address getting
451 * stolen by an Ethernet bridge for STP purposes.
452 * (FE:FF:FF:FF:FF:FF)
453 */
454 eth_broadcast_addr(dev->dev_addr);
455 dev->dev_addr[0] &= ~0x01;
456
457 netif_carrier_off(dev);
458
459 err = register_netdev(dev);
460 if (err) {
461 netdev_warn(dev, "Could not register device: err=%d\n", err);
462 free_netdev(dev);
463 return ERR_PTR(err);
464 }
465
466 netdev_dbg(dev, "Successfully created xenvif\n");
467
468 __module_get(THIS_MODULE);
469
470 return vif;
471}
472
473int xenvif_init_queue(struct xenvif_queue *queue)
474{
475 int err, i;
476
477 queue->credit_bytes = queue->remaining_credit = ~0UL;
478 queue->credit_usec = 0UL;
479 init_timer(&queue->credit_timeout);
480 queue->credit_timeout.function = xenvif_tx_credit_callback;
481 queue->credit_window_start = get_jiffies_64();
482
483 queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
484
485 skb_queue_head_init(&queue->rx_queue);
486 skb_queue_head_init(&queue->tx_queue);
487
488 queue->pending_cons = 0;
489 queue->pending_prod = MAX_PENDING_REQS;
490 for (i = 0; i < MAX_PENDING_REQS; ++i)
491 queue->pending_ring[i] = i;
492
493 spin_lock_init(&queue->callback_lock);
494 spin_lock_init(&queue->response_lock);
495
496 /* If ballooning is disabled, this will consume real memory, so you
497 * better enable it. The long term solution would be to use just a
498 * bunch of valid page descriptors, without dependency on ballooning
499 */
500 err = gnttab_alloc_pages(MAX_PENDING_REQS,
501 queue->mmap_pages);
502 if (err) {
503 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
504 return -ENOMEM;
505 }
506
507 for (i = 0; i < MAX_PENDING_REQS; i++) {
508 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
509 { .callback = xenvif_zerocopy_callback,
510 .ctx = NULL,
511 .desc = i };
512 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
513 }
514
515 return 0;
516}
517
518void xenvif_carrier_on(struct xenvif *vif)
519{
520 rtnl_lock();
521 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
522 dev_set_mtu(vif->dev, ETH_DATA_LEN);
523 netdev_update_features(vif->dev);
524 set_bit(VIF_STATUS_CONNECTED, &vif->status);
525 if (netif_running(vif->dev))
526 xenvif_up(vif);
527 rtnl_unlock();
528}
529
530int xenvif_connect(struct xenvif_queue *queue, unsigned long tx_ring_ref,
531 unsigned long rx_ring_ref, unsigned int tx_evtchn,
532 unsigned int rx_evtchn)
533{
534 struct task_struct *task;
535 int err = -ENOMEM;
536
537 BUG_ON(queue->tx_irq);
538 BUG_ON(queue->task);
539 BUG_ON(queue->dealloc_task);
540
541 err = xenvif_map_frontend_rings(queue, tx_ring_ref, rx_ring_ref);
542 if (err < 0)
543 goto err;
544
545 init_waitqueue_head(&queue->wq);
546 init_waitqueue_head(&queue->dealloc_wq);
547 atomic_set(&queue->inflight_packets, 0);
548
549 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
550 XENVIF_NAPI_WEIGHT);
551
552 if (tx_evtchn == rx_evtchn) {
553 /* feature-split-event-channels == 0 */
554 err = bind_interdomain_evtchn_to_irqhandler(
555 queue->vif->domid, tx_evtchn, xenvif_interrupt, 0,
556 queue->name, queue);
557 if (err < 0)
558 goto err_unmap;
559 queue->tx_irq = queue->rx_irq = err;
560 disable_irq(queue->tx_irq);
561 } else {
562 /* feature-split-event-channels == 1 */
563 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
564 "%s-tx", queue->name);
565 err = bind_interdomain_evtchn_to_irqhandler(
566 queue->vif->domid, tx_evtchn, xenvif_tx_interrupt, 0,
567 queue->tx_irq_name, queue);
568 if (err < 0)
569 goto err_unmap;
570 queue->tx_irq = err;
571 disable_irq(queue->tx_irq);
572
573 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
574 "%s-rx", queue->name);
575 err = bind_interdomain_evtchn_to_irqhandler(
576 queue->vif->domid, rx_evtchn, xenvif_rx_interrupt, 0,
577 queue->rx_irq_name, queue);
578 if (err < 0)
579 goto err_tx_unbind;
580 queue->rx_irq = err;
581 disable_irq(queue->rx_irq);
582 }
583
584 queue->stalled = true;
585
586 task = kthread_create(xenvif_kthread_guest_rx,
587 (void *)queue, "%s-guest-rx", queue->name);
588 if (IS_ERR(task)) {
589 pr_warn("Could not allocate kthread for %s\n", queue->name);
590 err = PTR_ERR(task);
591 goto err_rx_unbind;
592 }
593 queue->task = task;
594 get_task_struct(task);
595
596 task = kthread_create(xenvif_dealloc_kthread,
597 (void *)queue, "%s-dealloc", queue->name);
598 if (IS_ERR(task)) {
599 pr_warn("Could not allocate kthread for %s\n", queue->name);
600 err = PTR_ERR(task);
601 goto err_rx_unbind;
602 }
603 queue->dealloc_task = task;
604
605 wake_up_process(queue->task);
606 wake_up_process(queue->dealloc_task);
607
608 return 0;
609
610err_rx_unbind:
611 unbind_from_irqhandler(queue->rx_irq, queue);
612 queue->rx_irq = 0;
613err_tx_unbind:
614 unbind_from_irqhandler(queue->tx_irq, queue);
615 queue->tx_irq = 0;
616err_unmap:
617 xenvif_unmap_frontend_rings(queue);
618 netif_napi_del(&queue->napi);
619err:
620 module_put(THIS_MODULE);
621 return err;
622}
623
624void xenvif_carrier_off(struct xenvif *vif)
625{
626 struct net_device *dev = vif->dev;
627
628 rtnl_lock();
629 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
630 netif_carrier_off(dev); /* discard queued packets */
631 if (netif_running(dev))
632 xenvif_down(vif);
633 }
634 rtnl_unlock();
635}
636
637void xenvif_disconnect(struct xenvif *vif)
638{
639 struct xenvif_queue *queue = NULL;
640 unsigned int num_queues = vif->num_queues;
641 unsigned int queue_index;
642
643 xenvif_carrier_off(vif);
644
645 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
646 queue = &vif->queues[queue_index];
647
648 netif_napi_del(&queue->napi);
649
650 if (queue->task) {
651 kthread_stop(queue->task);
652 put_task_struct(queue->task);
653 queue->task = NULL;
654 }
655
656 if (queue->dealloc_task) {
657 kthread_stop(queue->dealloc_task);
658 queue->dealloc_task = NULL;
659 }
660
661 if (queue->tx_irq) {
662 if (queue->tx_irq == queue->rx_irq)
663 unbind_from_irqhandler(queue->tx_irq, queue);
664 else {
665 unbind_from_irqhandler(queue->tx_irq, queue);
666 unbind_from_irqhandler(queue->rx_irq, queue);
667 }
668 queue->tx_irq = 0;
669 }
670
671 xenvif_unmap_frontend_rings(queue);
672 }
673
674 xenvif_mcast_addr_list_free(vif);
675}
676
677/* Reverse the relevant parts of xenvif_init_queue().
678 * Used for queue teardown from xenvif_free(), and on the
679 * error handling paths in xenbus.c:connect().
680 */
681void xenvif_deinit_queue(struct xenvif_queue *queue)
682{
683 gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
684}
685
686void xenvif_free(struct xenvif *vif)
687{
688 struct xenvif_queue *queues = vif->queues;
689 unsigned int num_queues = vif->num_queues;
690 unsigned int queue_index;
691
692 unregister_netdev(vif->dev);
693 free_netdev(vif->dev);
694
695 for (queue_index = 0; queue_index < num_queues; ++queue_index)
696 xenvif_deinit_queue(&queues[queue_index]);
697 vfree(queues);
698
699 module_put(THIS_MODULE);
700}