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/sched/task.h>
35#include <linux/ethtool.h>
36#include <linux/rtnetlink.h>
37#include <linux/if_vlan.h>
38#include <linux/vmalloc.h>
39
40#include <xen/events.h>
41#include <asm/xen/hypercall.h>
42#include <xen/balloon.h>
43
44/* Number of bytes allowed on the internal guest Rx queue. */
45#define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE)
46
47/* This function is used to set SKBFL_ZEROCOPY_ENABLE as well as
48 * increasing the inflight counter. We need to increase the inflight
49 * counter because core driver calls into xenvif_zerocopy_callback
50 * which calls xenvif_skb_zerocopy_complete.
51 */
52void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
53 struct sk_buff *skb)
54{
55 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_ENABLE;
56 atomic_inc(&queue->inflight_packets);
57}
58
59void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
60{
61 atomic_dec(&queue->inflight_packets);
62
63 /* Wake the dealloc thread _after_ decrementing inflight_packets so
64 * that if kthread_stop() has already been called, the dealloc thread
65 * does not wait forever with nothing to wake it.
66 */
67 wake_up(&queue->dealloc_wq);
68}
69
70static int xenvif_schedulable(struct xenvif *vif)
71{
72 return netif_running(vif->dev) &&
73 test_bit(VIF_STATUS_CONNECTED, &vif->status) &&
74 !vif->disabled;
75}
76
77static bool xenvif_handle_tx_interrupt(struct xenvif_queue *queue)
78{
79 bool rc;
80
81 rc = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx);
82 if (rc)
83 napi_schedule(&queue->napi);
84 return rc;
85}
86
87static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
88{
89 struct xenvif_queue *queue = dev_id;
90 int old;
91
92 old = atomic_fetch_or(NETBK_TX_EOI, &queue->eoi_pending);
93 WARN(old & NETBK_TX_EOI, "Interrupt while EOI pending\n");
94
95 if (!xenvif_handle_tx_interrupt(queue)) {
96 atomic_andnot(NETBK_TX_EOI, &queue->eoi_pending);
97 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
98 }
99
100 return IRQ_HANDLED;
101}
102
103static int xenvif_poll(struct napi_struct *napi, int budget)
104{
105 struct xenvif_queue *queue =
106 container_of(napi, struct xenvif_queue, napi);
107 int work_done;
108
109 /* This vif is rogue, we pretend we've there is nothing to do
110 * for this vif to deschedule it from NAPI. But this interface
111 * will be turned off in thread context later.
112 */
113 if (unlikely(queue->vif->disabled)) {
114 napi_complete(napi);
115 return 0;
116 }
117
118 work_done = xenvif_tx_action(queue, budget);
119
120 if (work_done < budget) {
121 napi_complete_done(napi, work_done);
122 /* If the queue is rate-limited, it shall be
123 * rescheduled in the timer callback.
124 */
125 if (likely(!queue->rate_limited))
126 xenvif_napi_schedule_or_enable_events(queue);
127 }
128
129 return work_done;
130}
131
132static bool xenvif_handle_rx_interrupt(struct xenvif_queue *queue)
133{
134 bool rc;
135
136 rc = xenvif_have_rx_work(queue, false);
137 if (rc)
138 xenvif_kick_thread(queue);
139 return rc;
140}
141
142static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
143{
144 struct xenvif_queue *queue = dev_id;
145 int old;
146
147 old = atomic_fetch_or(NETBK_RX_EOI, &queue->eoi_pending);
148 WARN(old & NETBK_RX_EOI, "Interrupt while EOI pending\n");
149
150 if (!xenvif_handle_rx_interrupt(queue)) {
151 atomic_andnot(NETBK_RX_EOI, &queue->eoi_pending);
152 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
153 }
154
155 return IRQ_HANDLED;
156}
157
158irqreturn_t xenvif_interrupt(int irq, void *dev_id)
159{
160 struct xenvif_queue *queue = dev_id;
161 int old;
162 bool has_rx, has_tx;
163
164 old = atomic_fetch_or(NETBK_COMMON_EOI, &queue->eoi_pending);
165 WARN(old, "Interrupt while EOI pending\n");
166
167 has_tx = xenvif_handle_tx_interrupt(queue);
168 has_rx = xenvif_handle_rx_interrupt(queue);
169
170 if (!has_rx && !has_tx) {
171 atomic_andnot(NETBK_COMMON_EOI, &queue->eoi_pending);
172 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
173 }
174
175 return IRQ_HANDLED;
176}
177
178static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
179 struct net_device *sb_dev)
180{
181 struct xenvif *vif = netdev_priv(dev);
182 unsigned int size = vif->hash.size;
183 unsigned int num_queues;
184
185 /* If queues are not set up internally - always return 0
186 * as the packet going to be dropped anyway */
187 num_queues = READ_ONCE(vif->num_queues);
188 if (num_queues < 1)
189 return 0;
190
191 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
192 return netdev_pick_tx(dev, skb, NULL) %
193 dev->real_num_tx_queues;
194
195 xenvif_set_skb_hash(vif, skb);
196
197 if (size == 0)
198 return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
199
200 return vif->hash.mapping[vif->hash.mapping_sel]
201 [skb_get_hash_raw(skb) % size];
202}
203
204static netdev_tx_t
205xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
206{
207 struct xenvif *vif = netdev_priv(dev);
208 struct xenvif_queue *queue = NULL;
209 unsigned int num_queues;
210 u16 index;
211 struct xenvif_rx_cb *cb;
212
213 BUG_ON(skb->dev != dev);
214
215 /* Drop the packet if queues are not set up.
216 * This handler should be called inside an RCU read section
217 * so we don't need to enter it here explicitly.
218 */
219 num_queues = READ_ONCE(vif->num_queues);
220 if (num_queues < 1)
221 goto drop;
222
223 /* Obtain the queue to be used to transmit this packet */
224 index = skb_get_queue_mapping(skb);
225 if (index >= num_queues) {
226 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n",
227 index, vif->dev->name);
228 index %= num_queues;
229 }
230 queue = &vif->queues[index];
231
232 /* Drop the packet if queue is not ready */
233 if (queue->task == NULL ||
234 queue->dealloc_task == NULL ||
235 !xenvif_schedulable(vif))
236 goto drop;
237
238 if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
239 struct ethhdr *eth = (struct ethhdr *)skb->data;
240
241 if (!xenvif_mcast_match(vif, eth->h_dest))
242 goto drop;
243 }
244
245 cb = XENVIF_RX_CB(skb);
246 cb->expires = jiffies + vif->drain_timeout;
247
248 /* If there is no hash algorithm configured then make sure there
249 * is no hash information in the socket buffer otherwise it
250 * would be incorrectly forwarded to the frontend.
251 */
252 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
253 skb_clear_hash(skb);
254
255 /* timestamp packet in software */
256 skb_tx_timestamp(skb);
257
258 if (!xenvif_rx_queue_tail(queue, skb))
259 goto drop;
260
261 xenvif_kick_thread(queue);
262
263 return NETDEV_TX_OK;
264
265 drop:
266 vif->dev->stats.tx_dropped++;
267 dev_kfree_skb_any(skb);
268 return NETDEV_TX_OK;
269}
270
271static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
272{
273 struct xenvif *vif = netdev_priv(dev);
274 struct xenvif_queue *queue = NULL;
275 unsigned int num_queues;
276 u64 rx_bytes = 0;
277 u64 rx_packets = 0;
278 u64 tx_bytes = 0;
279 u64 tx_packets = 0;
280 unsigned int index;
281
282 rcu_read_lock();
283 num_queues = READ_ONCE(vif->num_queues);
284
285 /* Aggregate tx and rx stats from each queue */
286 for (index = 0; index < num_queues; ++index) {
287 queue = &vif->queues[index];
288 rx_bytes += queue->stats.rx_bytes;
289 rx_packets += queue->stats.rx_packets;
290 tx_bytes += queue->stats.tx_bytes;
291 tx_packets += queue->stats.tx_packets;
292 }
293
294 rcu_read_unlock();
295
296 vif->dev->stats.rx_bytes = rx_bytes;
297 vif->dev->stats.rx_packets = rx_packets;
298 vif->dev->stats.tx_bytes = tx_bytes;
299 vif->dev->stats.tx_packets = tx_packets;
300
301 return &vif->dev->stats;
302}
303
304static void xenvif_up(struct xenvif *vif)
305{
306 struct xenvif_queue *queue = NULL;
307 unsigned int num_queues = vif->num_queues;
308 unsigned int queue_index;
309
310 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
311 queue = &vif->queues[queue_index];
312 napi_enable(&queue->napi);
313 enable_irq(queue->tx_irq);
314 if (queue->tx_irq != queue->rx_irq)
315 enable_irq(queue->rx_irq);
316 xenvif_napi_schedule_or_enable_events(queue);
317 }
318}
319
320static void xenvif_down(struct xenvif *vif)
321{
322 struct xenvif_queue *queue = NULL;
323 unsigned int num_queues = vif->num_queues;
324 unsigned int queue_index;
325
326 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
327 queue = &vif->queues[queue_index];
328 disable_irq(queue->tx_irq);
329 if (queue->tx_irq != queue->rx_irq)
330 disable_irq(queue->rx_irq);
331 napi_disable(&queue->napi);
332 del_timer_sync(&queue->credit_timeout);
333 }
334}
335
336static int xenvif_open(struct net_device *dev)
337{
338 struct xenvif *vif = netdev_priv(dev);
339 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
340 xenvif_up(vif);
341 netif_tx_start_all_queues(dev);
342 return 0;
343}
344
345static int xenvif_close(struct net_device *dev)
346{
347 struct xenvif *vif = netdev_priv(dev);
348 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
349 xenvif_down(vif);
350 netif_tx_stop_all_queues(dev);
351 return 0;
352}
353
354static int xenvif_change_mtu(struct net_device *dev, int mtu)
355{
356 struct xenvif *vif = netdev_priv(dev);
357 int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
358
359 if (mtu > max)
360 return -EINVAL;
361 dev->mtu = mtu;
362 return 0;
363}
364
365static netdev_features_t xenvif_fix_features(struct net_device *dev,
366 netdev_features_t features)
367{
368 struct xenvif *vif = netdev_priv(dev);
369
370 if (!vif->can_sg)
371 features &= ~NETIF_F_SG;
372 if (~(vif->gso_mask) & GSO_BIT(TCPV4))
373 features &= ~NETIF_F_TSO;
374 if (~(vif->gso_mask) & GSO_BIT(TCPV6))
375 features &= ~NETIF_F_TSO6;
376 if (!vif->ip_csum)
377 features &= ~NETIF_F_IP_CSUM;
378 if (!vif->ipv6_csum)
379 features &= ~NETIF_F_IPV6_CSUM;
380
381 return features;
382}
383
384static const struct xenvif_stat {
385 char name[ETH_GSTRING_LEN];
386 u16 offset;
387} xenvif_stats[] = {
388 {
389 "rx_gso_checksum_fixup",
390 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
391 },
392 /* If (sent != success + fail), there are probably packets never
393 * freed up properly!
394 */
395 {
396 "tx_zerocopy_sent",
397 offsetof(struct xenvif_stats, tx_zerocopy_sent),
398 },
399 {
400 "tx_zerocopy_success",
401 offsetof(struct xenvif_stats, tx_zerocopy_success),
402 },
403 {
404 "tx_zerocopy_fail",
405 offsetof(struct xenvif_stats, tx_zerocopy_fail)
406 },
407 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
408 * a guest with the same MAX_SKB_FRAG
409 */
410 {
411 "tx_frag_overflow",
412 offsetof(struct xenvif_stats, tx_frag_overflow)
413 },
414};
415
416static int xenvif_get_sset_count(struct net_device *dev, int string_set)
417{
418 switch (string_set) {
419 case ETH_SS_STATS:
420 return ARRAY_SIZE(xenvif_stats);
421 default:
422 return -EINVAL;
423 }
424}
425
426static void xenvif_get_ethtool_stats(struct net_device *dev,
427 struct ethtool_stats *stats, u64 * data)
428{
429 struct xenvif *vif = netdev_priv(dev);
430 unsigned int num_queues;
431 int i;
432 unsigned int queue_index;
433
434 rcu_read_lock();
435 num_queues = READ_ONCE(vif->num_queues);
436
437 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
438 unsigned long accum = 0;
439 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
440 void *vif_stats = &vif->queues[queue_index].stats;
441 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
442 }
443 data[i] = accum;
444 }
445
446 rcu_read_unlock();
447}
448
449static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
450{
451 int i;
452
453 switch (stringset) {
454 case ETH_SS_STATS:
455 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
456 memcpy(data + i * ETH_GSTRING_LEN,
457 xenvif_stats[i].name, ETH_GSTRING_LEN);
458 break;
459 }
460}
461
462static const struct ethtool_ops xenvif_ethtool_ops = {
463 .get_link = ethtool_op_get_link,
464 .get_ts_info = ethtool_op_get_ts_info,
465 .get_sset_count = xenvif_get_sset_count,
466 .get_ethtool_stats = xenvif_get_ethtool_stats,
467 .get_strings = xenvif_get_strings,
468};
469
470static const struct net_device_ops xenvif_netdev_ops = {
471 .ndo_select_queue = xenvif_select_queue,
472 .ndo_start_xmit = xenvif_start_xmit,
473 .ndo_get_stats = xenvif_get_stats,
474 .ndo_open = xenvif_open,
475 .ndo_stop = xenvif_close,
476 .ndo_change_mtu = xenvif_change_mtu,
477 .ndo_fix_features = xenvif_fix_features,
478 .ndo_set_mac_address = eth_mac_addr,
479 .ndo_validate_addr = eth_validate_addr,
480};
481
482struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
483 unsigned int handle)
484{
485 static const u8 dummy_addr[ETH_ALEN] = {
486 0xfe, 0xff, 0xff, 0xff, 0xff, 0xff,
487 };
488 int err;
489 struct net_device *dev;
490 struct xenvif *vif;
491 char name[IFNAMSIZ] = {};
492
493 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
494 /* Allocate a netdev with the max. supported number of queues.
495 * When the guest selects the desired number, it will be updated
496 * via netif_set_real_num_*_queues().
497 */
498 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
499 ether_setup, xenvif_max_queues);
500 if (dev == NULL) {
501 pr_warn("Could not allocate netdev for %s\n", name);
502 return ERR_PTR(-ENOMEM);
503 }
504
505 SET_NETDEV_DEV(dev, parent);
506
507 vif = netdev_priv(dev);
508
509 vif->domid = domid;
510 vif->handle = handle;
511 vif->can_sg = 1;
512 vif->ip_csum = 1;
513 vif->dev = dev;
514 vif->disabled = false;
515 vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
516 vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
517
518 /* Start out with no queues. */
519 vif->queues = NULL;
520 vif->num_queues = 0;
521
522 vif->xdp_headroom = 0;
523
524 spin_lock_init(&vif->lock);
525 INIT_LIST_HEAD(&vif->fe_mcast_addr);
526
527 dev->netdev_ops = &xenvif_netdev_ops;
528 dev->hw_features = NETIF_F_SG |
529 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
530 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
531 dev->features = dev->hw_features | NETIF_F_RXCSUM;
532 dev->ethtool_ops = &xenvif_ethtool_ops;
533
534 dev->min_mtu = ETH_MIN_MTU;
535 dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
536
537 /*
538 * Initialise a dummy MAC address. We choose the numerically
539 * largest non-broadcast address to prevent the address getting
540 * stolen by an Ethernet bridge for STP purposes.
541 * (FE:FF:FF:FF:FF:FF)
542 */
543 eth_hw_addr_set(dev, dummy_addr);
544
545 netif_carrier_off(dev);
546
547 err = register_netdev(dev);
548 if (err) {
549 netdev_warn(dev, "Could not register device: err=%d\n", err);
550 free_netdev(dev);
551 return ERR_PTR(err);
552 }
553
554 netdev_dbg(dev, "Successfully created xenvif\n");
555
556 __module_get(THIS_MODULE);
557
558 return vif;
559}
560
561int xenvif_init_queue(struct xenvif_queue *queue)
562{
563 int err, i;
564
565 queue->credit_bytes = queue->remaining_credit = ~0UL;
566 queue->credit_usec = 0UL;
567 timer_setup(&queue->credit_timeout, xenvif_tx_credit_callback, 0);
568 queue->credit_window_start = get_jiffies_64();
569
570 queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
571
572 skb_queue_head_init(&queue->rx_queue);
573 skb_queue_head_init(&queue->tx_queue);
574
575 queue->pending_cons = 0;
576 queue->pending_prod = MAX_PENDING_REQS;
577 for (i = 0; i < MAX_PENDING_REQS; ++i)
578 queue->pending_ring[i] = i;
579
580 spin_lock_init(&queue->callback_lock);
581 spin_lock_init(&queue->response_lock);
582
583 /* If ballooning is disabled, this will consume real memory, so you
584 * better enable it. The long term solution would be to use just a
585 * bunch of valid page descriptors, without dependency on ballooning
586 */
587 err = gnttab_alloc_pages(MAX_PENDING_REQS,
588 queue->mmap_pages);
589 if (err) {
590 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
591 return -ENOMEM;
592 }
593
594 for (i = 0; i < MAX_PENDING_REQS; i++) {
595 queue->pending_tx_info[i].callback_struct = (struct ubuf_info_msgzc)
596 { { .callback = xenvif_zerocopy_callback },
597 { { .ctx = NULL,
598 .desc = i } } };
599 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
600 }
601
602 return 0;
603}
604
605void xenvif_carrier_on(struct xenvif *vif)
606{
607 rtnl_lock();
608 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
609 dev_set_mtu(vif->dev, ETH_DATA_LEN);
610 netdev_update_features(vif->dev);
611 set_bit(VIF_STATUS_CONNECTED, &vif->status);
612 if (netif_running(vif->dev))
613 xenvif_up(vif);
614 rtnl_unlock();
615}
616
617int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
618 unsigned int evtchn)
619{
620 struct net_device *dev = vif->dev;
621 struct xenbus_device *xendev = xenvif_to_xenbus_device(vif);
622 void *addr;
623 struct xen_netif_ctrl_sring *shared;
624 RING_IDX rsp_prod, req_prod;
625 int err;
626
627 err = xenbus_map_ring_valloc(xendev, &ring_ref, 1, &addr);
628 if (err)
629 goto err;
630
631 shared = (struct xen_netif_ctrl_sring *)addr;
632 rsp_prod = READ_ONCE(shared->rsp_prod);
633 req_prod = READ_ONCE(shared->req_prod);
634
635 BACK_RING_ATTACH(&vif->ctrl, shared, rsp_prod, XEN_PAGE_SIZE);
636
637 err = -EIO;
638 if (req_prod - rsp_prod > RING_SIZE(&vif->ctrl))
639 goto err_unmap;
640
641 err = bind_interdomain_evtchn_to_irq_lateeoi(xendev, evtchn);
642 if (err < 0)
643 goto err_unmap;
644
645 vif->ctrl_irq = err;
646
647 xenvif_init_hash(vif);
648
649 err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
650 IRQF_ONESHOT, "xen-netback-ctrl", vif);
651 if (err) {
652 pr_warn("Could not setup irq handler for %s\n", dev->name);
653 goto err_deinit;
654 }
655
656 return 0;
657
658err_deinit:
659 xenvif_deinit_hash(vif);
660 unbind_from_irqhandler(vif->ctrl_irq, vif);
661 vif->ctrl_irq = 0;
662
663err_unmap:
664 xenbus_unmap_ring_vfree(xendev, vif->ctrl.sring);
665 vif->ctrl.sring = NULL;
666
667err:
668 return err;
669}
670
671static void xenvif_disconnect_queue(struct xenvif_queue *queue)
672{
673 if (queue->task) {
674 kthread_stop_put(queue->task);
675 queue->task = NULL;
676 }
677
678 if (queue->dealloc_task) {
679 kthread_stop(queue->dealloc_task);
680 queue->dealloc_task = NULL;
681 }
682
683 if (queue->napi.poll) {
684 netif_napi_del(&queue->napi);
685 queue->napi.poll = NULL;
686 }
687
688 if (queue->tx_irq) {
689 unbind_from_irqhandler(queue->tx_irq, queue);
690 if (queue->tx_irq == queue->rx_irq)
691 queue->rx_irq = 0;
692 queue->tx_irq = 0;
693 }
694
695 if (queue->rx_irq) {
696 unbind_from_irqhandler(queue->rx_irq, queue);
697 queue->rx_irq = 0;
698 }
699
700 xenvif_unmap_frontend_data_rings(queue);
701}
702
703int xenvif_connect_data(struct xenvif_queue *queue,
704 unsigned long tx_ring_ref,
705 unsigned long rx_ring_ref,
706 unsigned int tx_evtchn,
707 unsigned int rx_evtchn)
708{
709 struct xenbus_device *dev = xenvif_to_xenbus_device(queue->vif);
710 struct task_struct *task;
711 int err;
712
713 BUG_ON(queue->tx_irq);
714 BUG_ON(queue->task);
715 BUG_ON(queue->dealloc_task);
716
717 err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
718 rx_ring_ref);
719 if (err < 0)
720 goto err;
721
722 init_waitqueue_head(&queue->wq);
723 init_waitqueue_head(&queue->dealloc_wq);
724 atomic_set(&queue->inflight_packets, 0);
725
726 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll);
727
728 queue->stalled = true;
729
730 task = kthread_run(xenvif_kthread_guest_rx, queue,
731 "%s-guest-rx", queue->name);
732 if (IS_ERR(task))
733 goto kthread_err;
734 queue->task = task;
735 /*
736 * Take a reference to the task in order to prevent it from being freed
737 * if the thread function returns before kthread_stop is called.
738 */
739 get_task_struct(task);
740
741 task = kthread_run(xenvif_dealloc_kthread, queue,
742 "%s-dealloc", queue->name);
743 if (IS_ERR(task))
744 goto kthread_err;
745 queue->dealloc_task = task;
746
747 if (tx_evtchn == rx_evtchn) {
748 /* feature-split-event-channels == 0 */
749 err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
750 dev, tx_evtchn, xenvif_interrupt, 0,
751 queue->name, queue);
752 if (err < 0)
753 goto err;
754 queue->tx_irq = queue->rx_irq = err;
755 disable_irq(queue->tx_irq);
756 } else {
757 /* feature-split-event-channels == 1 */
758 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
759 "%s-tx", queue->name);
760 err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
761 dev, tx_evtchn, xenvif_tx_interrupt, 0,
762 queue->tx_irq_name, queue);
763 if (err < 0)
764 goto err;
765 queue->tx_irq = err;
766 disable_irq(queue->tx_irq);
767
768 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
769 "%s-rx", queue->name);
770 err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
771 dev, rx_evtchn, xenvif_rx_interrupt, 0,
772 queue->rx_irq_name, queue);
773 if (err < 0)
774 goto err;
775 queue->rx_irq = err;
776 disable_irq(queue->rx_irq);
777 }
778
779 return 0;
780
781kthread_err:
782 pr_warn("Could not allocate kthread for %s\n", queue->name);
783 err = PTR_ERR(task);
784err:
785 xenvif_disconnect_queue(queue);
786 return err;
787}
788
789void xenvif_carrier_off(struct xenvif *vif)
790{
791 struct net_device *dev = vif->dev;
792
793 rtnl_lock();
794 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
795 netif_carrier_off(dev); /* discard queued packets */
796 if (netif_running(dev))
797 xenvif_down(vif);
798 }
799 rtnl_unlock();
800}
801
802void xenvif_disconnect_data(struct xenvif *vif)
803{
804 struct xenvif_queue *queue = NULL;
805 unsigned int num_queues = vif->num_queues;
806 unsigned int queue_index;
807
808 xenvif_carrier_off(vif);
809
810 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
811 queue = &vif->queues[queue_index];
812
813 xenvif_disconnect_queue(queue);
814 }
815
816 xenvif_mcast_addr_list_free(vif);
817}
818
819void xenvif_disconnect_ctrl(struct xenvif *vif)
820{
821 if (vif->ctrl_irq) {
822 xenvif_deinit_hash(vif);
823 unbind_from_irqhandler(vif->ctrl_irq, vif);
824 vif->ctrl_irq = 0;
825 }
826
827 if (vif->ctrl.sring) {
828 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
829 vif->ctrl.sring);
830 vif->ctrl.sring = NULL;
831 }
832}
833
834/* Reverse the relevant parts of xenvif_init_queue().
835 * Used for queue teardown from xenvif_free(), and on the
836 * error handling paths in xenbus.c:connect().
837 */
838void xenvif_deinit_queue(struct xenvif_queue *queue)
839{
840 gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
841}
842
843void xenvif_free(struct xenvif *vif)
844{
845 struct xenvif_queue *queues = vif->queues;
846 unsigned int num_queues = vif->num_queues;
847 unsigned int queue_index;
848
849 unregister_netdev(vif->dev);
850 free_netdev(vif->dev);
851
852 for (queue_index = 0; queue_index < num_queues; ++queue_index)
853 xenvif_deinit_queue(&queues[queue_index]);
854 vfree(queues);
855
856 module_put(THIS_MODULE);
857}
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/sched/task.h>
35#include <linux/ethtool.h>
36#include <linux/rtnetlink.h>
37#include <linux/if_vlan.h>
38#include <linux/vmalloc.h>
39
40#include <xen/events.h>
41#include <asm/xen/hypercall.h>
42#include <xen/balloon.h>
43
44#define XENVIF_QUEUE_LENGTH 32
45#define XENVIF_NAPI_WEIGHT 64
46
47/* Number of bytes allowed on the internal guest Rx queue. */
48#define XENVIF_RX_QUEUE_BYTES (XEN_NETIF_RX_RING_SIZE/2 * PAGE_SIZE)
49
50/* This function is used to set SKBFL_ZEROCOPY_ENABLE as well as
51 * increasing the inflight counter. We need to increase the inflight
52 * counter because core driver calls into xenvif_zerocopy_callback
53 * which calls xenvif_skb_zerocopy_complete.
54 */
55void xenvif_skb_zerocopy_prepare(struct xenvif_queue *queue,
56 struct sk_buff *skb)
57{
58 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_ENABLE;
59 atomic_inc(&queue->inflight_packets);
60}
61
62void xenvif_skb_zerocopy_complete(struct xenvif_queue *queue)
63{
64 atomic_dec(&queue->inflight_packets);
65
66 /* Wake the dealloc thread _after_ decrementing inflight_packets so
67 * that if kthread_stop() has already been called, the dealloc thread
68 * does not wait forever with nothing to wake it.
69 */
70 wake_up(&queue->dealloc_wq);
71}
72
73int xenvif_schedulable(struct xenvif *vif)
74{
75 return netif_running(vif->dev) &&
76 test_bit(VIF_STATUS_CONNECTED, &vif->status) &&
77 !vif->disabled;
78}
79
80static bool xenvif_handle_tx_interrupt(struct xenvif_queue *queue)
81{
82 bool rc;
83
84 rc = RING_HAS_UNCONSUMED_REQUESTS(&queue->tx);
85 if (rc)
86 napi_schedule(&queue->napi);
87 return rc;
88}
89
90static irqreturn_t xenvif_tx_interrupt(int irq, void *dev_id)
91{
92 struct xenvif_queue *queue = dev_id;
93 int old;
94
95 old = atomic_fetch_or(NETBK_TX_EOI, &queue->eoi_pending);
96 WARN(old & NETBK_TX_EOI, "Interrupt while EOI pending\n");
97
98 if (!xenvif_handle_tx_interrupt(queue)) {
99 atomic_andnot(NETBK_TX_EOI, &queue->eoi_pending);
100 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
101 }
102
103 return IRQ_HANDLED;
104}
105
106static int xenvif_poll(struct napi_struct *napi, int budget)
107{
108 struct xenvif_queue *queue =
109 container_of(napi, struct xenvif_queue, napi);
110 int work_done;
111
112 /* This vif is rogue, we pretend we've there is nothing to do
113 * for this vif to deschedule it from NAPI. But this interface
114 * will be turned off in thread context later.
115 */
116 if (unlikely(queue->vif->disabled)) {
117 napi_complete(napi);
118 return 0;
119 }
120
121 work_done = xenvif_tx_action(queue, budget);
122
123 if (work_done < budget) {
124 napi_complete_done(napi, work_done);
125 /* If the queue is rate-limited, it shall be
126 * rescheduled in the timer callback.
127 */
128 if (likely(!queue->rate_limited))
129 xenvif_napi_schedule_or_enable_events(queue);
130 }
131
132 return work_done;
133}
134
135static bool xenvif_handle_rx_interrupt(struct xenvif_queue *queue)
136{
137 bool rc;
138
139 rc = xenvif_have_rx_work(queue, false);
140 if (rc)
141 xenvif_kick_thread(queue);
142 return rc;
143}
144
145static irqreturn_t xenvif_rx_interrupt(int irq, void *dev_id)
146{
147 struct xenvif_queue *queue = dev_id;
148 int old;
149
150 old = atomic_fetch_or(NETBK_RX_EOI, &queue->eoi_pending);
151 WARN(old & NETBK_RX_EOI, "Interrupt while EOI pending\n");
152
153 if (!xenvif_handle_rx_interrupt(queue)) {
154 atomic_andnot(NETBK_RX_EOI, &queue->eoi_pending);
155 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
156 }
157
158 return IRQ_HANDLED;
159}
160
161irqreturn_t xenvif_interrupt(int irq, void *dev_id)
162{
163 struct xenvif_queue *queue = dev_id;
164 int old;
165 bool has_rx, has_tx;
166
167 old = atomic_fetch_or(NETBK_COMMON_EOI, &queue->eoi_pending);
168 WARN(old, "Interrupt while EOI pending\n");
169
170 has_tx = xenvif_handle_tx_interrupt(queue);
171 has_rx = xenvif_handle_rx_interrupt(queue);
172
173 if (!has_rx && !has_tx) {
174 atomic_andnot(NETBK_COMMON_EOI, &queue->eoi_pending);
175 xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS);
176 }
177
178 return IRQ_HANDLED;
179}
180
181int xenvif_queue_stopped(struct xenvif_queue *queue)
182{
183 struct net_device *dev = queue->vif->dev;
184 unsigned int id = queue->id;
185 return netif_tx_queue_stopped(netdev_get_tx_queue(dev, id));
186}
187
188void xenvif_wake_queue(struct xenvif_queue *queue)
189{
190 struct net_device *dev = queue->vif->dev;
191 unsigned int id = queue->id;
192 netif_tx_wake_queue(netdev_get_tx_queue(dev, id));
193}
194
195static u16 xenvif_select_queue(struct net_device *dev, struct sk_buff *skb,
196 struct net_device *sb_dev)
197{
198 struct xenvif *vif = netdev_priv(dev);
199 unsigned int size = vif->hash.size;
200 unsigned int num_queues;
201
202 /* If queues are not set up internally - always return 0
203 * as the packet going to be dropped anyway */
204 num_queues = READ_ONCE(vif->num_queues);
205 if (num_queues < 1)
206 return 0;
207
208 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
209 return netdev_pick_tx(dev, skb, NULL) %
210 dev->real_num_tx_queues;
211
212 xenvif_set_skb_hash(vif, skb);
213
214 if (size == 0)
215 return skb_get_hash_raw(skb) % dev->real_num_tx_queues;
216
217 return vif->hash.mapping[vif->hash.mapping_sel]
218 [skb_get_hash_raw(skb) % size];
219}
220
221static netdev_tx_t
222xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
223{
224 struct xenvif *vif = netdev_priv(dev);
225 struct xenvif_queue *queue = NULL;
226 unsigned int num_queues;
227 u16 index;
228 struct xenvif_rx_cb *cb;
229
230 BUG_ON(skb->dev != dev);
231
232 /* Drop the packet if queues are not set up.
233 * This handler should be called inside an RCU read section
234 * so we don't need to enter it here explicitly.
235 */
236 num_queues = READ_ONCE(vif->num_queues);
237 if (num_queues < 1)
238 goto drop;
239
240 /* Obtain the queue to be used to transmit this packet */
241 index = skb_get_queue_mapping(skb);
242 if (index >= num_queues) {
243 pr_warn_ratelimited("Invalid queue %hu for packet on interface %s\n",
244 index, vif->dev->name);
245 index %= num_queues;
246 }
247 queue = &vif->queues[index];
248
249 /* Drop the packet if queue is not ready */
250 if (queue->task == NULL ||
251 queue->dealloc_task == NULL ||
252 !xenvif_schedulable(vif))
253 goto drop;
254
255 if (vif->multicast_control && skb->pkt_type == PACKET_MULTICAST) {
256 struct ethhdr *eth = (struct ethhdr *)skb->data;
257
258 if (!xenvif_mcast_match(vif, eth->h_dest))
259 goto drop;
260 }
261
262 cb = XENVIF_RX_CB(skb);
263 cb->expires = jiffies + vif->drain_timeout;
264
265 /* If there is no hash algorithm configured then make sure there
266 * is no hash information in the socket buffer otherwise it
267 * would be incorrectly forwarded to the frontend.
268 */
269 if (vif->hash.alg == XEN_NETIF_CTRL_HASH_ALGORITHM_NONE)
270 skb_clear_hash(skb);
271
272 xenvif_rx_queue_tail(queue, skb);
273 xenvif_kick_thread(queue);
274
275 return NETDEV_TX_OK;
276
277 drop:
278 vif->dev->stats.tx_dropped++;
279 dev_kfree_skb(skb);
280 return NETDEV_TX_OK;
281}
282
283static struct net_device_stats *xenvif_get_stats(struct net_device *dev)
284{
285 struct xenvif *vif = netdev_priv(dev);
286 struct xenvif_queue *queue = NULL;
287 unsigned int num_queues;
288 u64 rx_bytes = 0;
289 u64 rx_packets = 0;
290 u64 tx_bytes = 0;
291 u64 tx_packets = 0;
292 unsigned int index;
293
294 rcu_read_lock();
295 num_queues = READ_ONCE(vif->num_queues);
296
297 /* Aggregate tx and rx stats from each queue */
298 for (index = 0; index < num_queues; ++index) {
299 queue = &vif->queues[index];
300 rx_bytes += queue->stats.rx_bytes;
301 rx_packets += queue->stats.rx_packets;
302 tx_bytes += queue->stats.tx_bytes;
303 tx_packets += queue->stats.tx_packets;
304 }
305
306 rcu_read_unlock();
307
308 vif->dev->stats.rx_bytes = rx_bytes;
309 vif->dev->stats.rx_packets = rx_packets;
310 vif->dev->stats.tx_bytes = tx_bytes;
311 vif->dev->stats.tx_packets = tx_packets;
312
313 return &vif->dev->stats;
314}
315
316static void xenvif_up(struct xenvif *vif)
317{
318 struct xenvif_queue *queue = NULL;
319 unsigned int num_queues = vif->num_queues;
320 unsigned int queue_index;
321
322 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
323 queue = &vif->queues[queue_index];
324 napi_enable(&queue->napi);
325 enable_irq(queue->tx_irq);
326 if (queue->tx_irq != queue->rx_irq)
327 enable_irq(queue->rx_irq);
328 xenvif_napi_schedule_or_enable_events(queue);
329 }
330}
331
332static void xenvif_down(struct xenvif *vif)
333{
334 struct xenvif_queue *queue = NULL;
335 unsigned int num_queues = vif->num_queues;
336 unsigned int queue_index;
337
338 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
339 queue = &vif->queues[queue_index];
340 disable_irq(queue->tx_irq);
341 if (queue->tx_irq != queue->rx_irq)
342 disable_irq(queue->rx_irq);
343 napi_disable(&queue->napi);
344 del_timer_sync(&queue->credit_timeout);
345 }
346}
347
348static int xenvif_open(struct net_device *dev)
349{
350 struct xenvif *vif = netdev_priv(dev);
351 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
352 xenvif_up(vif);
353 netif_tx_start_all_queues(dev);
354 return 0;
355}
356
357static int xenvif_close(struct net_device *dev)
358{
359 struct xenvif *vif = netdev_priv(dev);
360 if (test_bit(VIF_STATUS_CONNECTED, &vif->status))
361 xenvif_down(vif);
362 netif_tx_stop_all_queues(dev);
363 return 0;
364}
365
366static int xenvif_change_mtu(struct net_device *dev, int mtu)
367{
368 struct xenvif *vif = netdev_priv(dev);
369 int max = vif->can_sg ? ETH_MAX_MTU - VLAN_ETH_HLEN : ETH_DATA_LEN;
370
371 if (mtu > max)
372 return -EINVAL;
373 dev->mtu = mtu;
374 return 0;
375}
376
377static netdev_features_t xenvif_fix_features(struct net_device *dev,
378 netdev_features_t features)
379{
380 struct xenvif *vif = netdev_priv(dev);
381
382 if (!vif->can_sg)
383 features &= ~NETIF_F_SG;
384 if (~(vif->gso_mask) & GSO_BIT(TCPV4))
385 features &= ~NETIF_F_TSO;
386 if (~(vif->gso_mask) & GSO_BIT(TCPV6))
387 features &= ~NETIF_F_TSO6;
388 if (!vif->ip_csum)
389 features &= ~NETIF_F_IP_CSUM;
390 if (!vif->ipv6_csum)
391 features &= ~NETIF_F_IPV6_CSUM;
392
393 return features;
394}
395
396static const struct xenvif_stat {
397 char name[ETH_GSTRING_LEN];
398 u16 offset;
399} xenvif_stats[] = {
400 {
401 "rx_gso_checksum_fixup",
402 offsetof(struct xenvif_stats, rx_gso_checksum_fixup)
403 },
404 /* If (sent != success + fail), there are probably packets never
405 * freed up properly!
406 */
407 {
408 "tx_zerocopy_sent",
409 offsetof(struct xenvif_stats, tx_zerocopy_sent),
410 },
411 {
412 "tx_zerocopy_success",
413 offsetof(struct xenvif_stats, tx_zerocopy_success),
414 },
415 {
416 "tx_zerocopy_fail",
417 offsetof(struct xenvif_stats, tx_zerocopy_fail)
418 },
419 /* Number of packets exceeding MAX_SKB_FRAG slots. You should use
420 * a guest with the same MAX_SKB_FRAG
421 */
422 {
423 "tx_frag_overflow",
424 offsetof(struct xenvif_stats, tx_frag_overflow)
425 },
426};
427
428static int xenvif_get_sset_count(struct net_device *dev, int string_set)
429{
430 switch (string_set) {
431 case ETH_SS_STATS:
432 return ARRAY_SIZE(xenvif_stats);
433 default:
434 return -EINVAL;
435 }
436}
437
438static void xenvif_get_ethtool_stats(struct net_device *dev,
439 struct ethtool_stats *stats, u64 * data)
440{
441 struct xenvif *vif = netdev_priv(dev);
442 unsigned int num_queues;
443 int i;
444 unsigned int queue_index;
445
446 rcu_read_lock();
447 num_queues = READ_ONCE(vif->num_queues);
448
449 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++) {
450 unsigned long accum = 0;
451 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
452 void *vif_stats = &vif->queues[queue_index].stats;
453 accum += *(unsigned long *)(vif_stats + xenvif_stats[i].offset);
454 }
455 data[i] = accum;
456 }
457
458 rcu_read_unlock();
459}
460
461static void xenvif_get_strings(struct net_device *dev, u32 stringset, u8 * data)
462{
463 int i;
464
465 switch (stringset) {
466 case ETH_SS_STATS:
467 for (i = 0; i < ARRAY_SIZE(xenvif_stats); i++)
468 memcpy(data + i * ETH_GSTRING_LEN,
469 xenvif_stats[i].name, ETH_GSTRING_LEN);
470 break;
471 }
472}
473
474static const struct ethtool_ops xenvif_ethtool_ops = {
475 .get_link = ethtool_op_get_link,
476
477 .get_sset_count = xenvif_get_sset_count,
478 .get_ethtool_stats = xenvif_get_ethtool_stats,
479 .get_strings = xenvif_get_strings,
480};
481
482static const struct net_device_ops xenvif_netdev_ops = {
483 .ndo_select_queue = xenvif_select_queue,
484 .ndo_start_xmit = xenvif_start_xmit,
485 .ndo_get_stats = xenvif_get_stats,
486 .ndo_open = xenvif_open,
487 .ndo_stop = xenvif_close,
488 .ndo_change_mtu = xenvif_change_mtu,
489 .ndo_fix_features = xenvif_fix_features,
490 .ndo_set_mac_address = eth_mac_addr,
491 .ndo_validate_addr = eth_validate_addr,
492};
493
494struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
495 unsigned int handle)
496{
497 int err;
498 struct net_device *dev;
499 struct xenvif *vif;
500 char name[IFNAMSIZ] = {};
501
502 snprintf(name, IFNAMSIZ - 1, "vif%u.%u", domid, handle);
503 /* Allocate a netdev with the max. supported number of queues.
504 * When the guest selects the desired number, it will be updated
505 * via netif_set_real_num_*_queues().
506 */
507 dev = alloc_netdev_mq(sizeof(struct xenvif), name, NET_NAME_UNKNOWN,
508 ether_setup, xenvif_max_queues);
509 if (dev == NULL) {
510 pr_warn("Could not allocate netdev for %s\n", name);
511 return ERR_PTR(-ENOMEM);
512 }
513
514 SET_NETDEV_DEV(dev, parent);
515
516 vif = netdev_priv(dev);
517
518 vif->domid = domid;
519 vif->handle = handle;
520 vif->can_sg = 1;
521 vif->ip_csum = 1;
522 vif->dev = dev;
523 vif->disabled = false;
524 vif->drain_timeout = msecs_to_jiffies(rx_drain_timeout_msecs);
525 vif->stall_timeout = msecs_to_jiffies(rx_stall_timeout_msecs);
526
527 /* Start out with no queues. */
528 vif->queues = NULL;
529 vif->num_queues = 0;
530
531 vif->xdp_headroom = 0;
532
533 spin_lock_init(&vif->lock);
534 INIT_LIST_HEAD(&vif->fe_mcast_addr);
535
536 dev->netdev_ops = &xenvif_netdev_ops;
537 dev->hw_features = NETIF_F_SG |
538 NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
539 NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_FRAGLIST;
540 dev->features = dev->hw_features | NETIF_F_RXCSUM;
541 dev->ethtool_ops = &xenvif_ethtool_ops;
542
543 dev->tx_queue_len = XENVIF_QUEUE_LENGTH;
544
545 dev->min_mtu = ETH_MIN_MTU;
546 dev->max_mtu = ETH_MAX_MTU - VLAN_ETH_HLEN;
547
548 /*
549 * Initialise a dummy MAC address. We choose the numerically
550 * largest non-broadcast address to prevent the address getting
551 * stolen by an Ethernet bridge for STP purposes.
552 * (FE:FF:FF:FF:FF:FF)
553 */
554 eth_broadcast_addr(dev->dev_addr);
555 dev->dev_addr[0] &= ~0x01;
556
557 netif_carrier_off(dev);
558
559 err = register_netdev(dev);
560 if (err) {
561 netdev_warn(dev, "Could not register device: err=%d\n", err);
562 free_netdev(dev);
563 return ERR_PTR(err);
564 }
565
566 netdev_dbg(dev, "Successfully created xenvif\n");
567
568 __module_get(THIS_MODULE);
569
570 return vif;
571}
572
573int xenvif_init_queue(struct xenvif_queue *queue)
574{
575 int err, i;
576
577 queue->credit_bytes = queue->remaining_credit = ~0UL;
578 queue->credit_usec = 0UL;
579 timer_setup(&queue->credit_timeout, xenvif_tx_credit_callback, 0);
580 queue->credit_window_start = get_jiffies_64();
581
582 queue->rx_queue_max = XENVIF_RX_QUEUE_BYTES;
583
584 skb_queue_head_init(&queue->rx_queue);
585 skb_queue_head_init(&queue->tx_queue);
586
587 queue->pending_cons = 0;
588 queue->pending_prod = MAX_PENDING_REQS;
589 for (i = 0; i < MAX_PENDING_REQS; ++i)
590 queue->pending_ring[i] = i;
591
592 spin_lock_init(&queue->callback_lock);
593 spin_lock_init(&queue->response_lock);
594
595 /* If ballooning is disabled, this will consume real memory, so you
596 * better enable it. The long term solution would be to use just a
597 * bunch of valid page descriptors, without dependency on ballooning
598 */
599 err = gnttab_alloc_pages(MAX_PENDING_REQS,
600 queue->mmap_pages);
601 if (err) {
602 netdev_err(queue->vif->dev, "Could not reserve mmap_pages\n");
603 return -ENOMEM;
604 }
605
606 for (i = 0; i < MAX_PENDING_REQS; i++) {
607 queue->pending_tx_info[i].callback_struct = (struct ubuf_info)
608 { .callback = xenvif_zerocopy_callback,
609 { { .ctx = NULL,
610 .desc = i } } };
611 queue->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
612 }
613
614 return 0;
615}
616
617void xenvif_carrier_on(struct xenvif *vif)
618{
619 rtnl_lock();
620 if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
621 dev_set_mtu(vif->dev, ETH_DATA_LEN);
622 netdev_update_features(vif->dev);
623 set_bit(VIF_STATUS_CONNECTED, &vif->status);
624 if (netif_running(vif->dev))
625 xenvif_up(vif);
626 rtnl_unlock();
627}
628
629int xenvif_connect_ctrl(struct xenvif *vif, grant_ref_t ring_ref,
630 unsigned int evtchn)
631{
632 struct net_device *dev = vif->dev;
633 struct xenbus_device *xendev = xenvif_to_xenbus_device(vif);
634 void *addr;
635 struct xen_netif_ctrl_sring *shared;
636 RING_IDX rsp_prod, req_prod;
637 int err;
638
639 err = xenbus_map_ring_valloc(xendev, &ring_ref, 1, &addr);
640 if (err)
641 goto err;
642
643 shared = (struct xen_netif_ctrl_sring *)addr;
644 rsp_prod = READ_ONCE(shared->rsp_prod);
645 req_prod = READ_ONCE(shared->req_prod);
646
647 BACK_RING_ATTACH(&vif->ctrl, shared, rsp_prod, XEN_PAGE_SIZE);
648
649 err = -EIO;
650 if (req_prod - rsp_prod > RING_SIZE(&vif->ctrl))
651 goto err_unmap;
652
653 err = bind_interdomain_evtchn_to_irq_lateeoi(xendev, evtchn);
654 if (err < 0)
655 goto err_unmap;
656
657 vif->ctrl_irq = err;
658
659 xenvif_init_hash(vif);
660
661 err = request_threaded_irq(vif->ctrl_irq, NULL, xenvif_ctrl_irq_fn,
662 IRQF_ONESHOT, "xen-netback-ctrl", vif);
663 if (err) {
664 pr_warn("Could not setup irq handler for %s\n", dev->name);
665 goto err_deinit;
666 }
667
668 return 0;
669
670err_deinit:
671 xenvif_deinit_hash(vif);
672 unbind_from_irqhandler(vif->ctrl_irq, vif);
673 vif->ctrl_irq = 0;
674
675err_unmap:
676 xenbus_unmap_ring_vfree(xendev, vif->ctrl.sring);
677 vif->ctrl.sring = NULL;
678
679err:
680 return err;
681}
682
683static void xenvif_disconnect_queue(struct xenvif_queue *queue)
684{
685 if (queue->task) {
686 kthread_stop(queue->task);
687 put_task_struct(queue->task);
688 queue->task = NULL;
689 }
690
691 if (queue->dealloc_task) {
692 kthread_stop(queue->dealloc_task);
693 queue->dealloc_task = NULL;
694 }
695
696 if (queue->napi.poll) {
697 netif_napi_del(&queue->napi);
698 queue->napi.poll = NULL;
699 }
700
701 if (queue->tx_irq) {
702 unbind_from_irqhandler(queue->tx_irq, queue);
703 if (queue->tx_irq == queue->rx_irq)
704 queue->rx_irq = 0;
705 queue->tx_irq = 0;
706 }
707
708 if (queue->rx_irq) {
709 unbind_from_irqhandler(queue->rx_irq, queue);
710 queue->rx_irq = 0;
711 }
712
713 xenvif_unmap_frontend_data_rings(queue);
714}
715
716int xenvif_connect_data(struct xenvif_queue *queue,
717 unsigned long tx_ring_ref,
718 unsigned long rx_ring_ref,
719 unsigned int tx_evtchn,
720 unsigned int rx_evtchn)
721{
722 struct xenbus_device *dev = xenvif_to_xenbus_device(queue->vif);
723 struct task_struct *task;
724 int err;
725
726 BUG_ON(queue->tx_irq);
727 BUG_ON(queue->task);
728 BUG_ON(queue->dealloc_task);
729
730 err = xenvif_map_frontend_data_rings(queue, tx_ring_ref,
731 rx_ring_ref);
732 if (err < 0)
733 goto err;
734
735 init_waitqueue_head(&queue->wq);
736 init_waitqueue_head(&queue->dealloc_wq);
737 atomic_set(&queue->inflight_packets, 0);
738
739 netif_napi_add(queue->vif->dev, &queue->napi, xenvif_poll,
740 XENVIF_NAPI_WEIGHT);
741
742 queue->stalled = true;
743
744 task = kthread_run(xenvif_kthread_guest_rx, queue,
745 "%s-guest-rx", queue->name);
746 if (IS_ERR(task))
747 goto kthread_err;
748 queue->task = task;
749 /*
750 * Take a reference to the task in order to prevent it from being freed
751 * if the thread function returns before kthread_stop is called.
752 */
753 get_task_struct(task);
754
755 task = kthread_run(xenvif_dealloc_kthread, queue,
756 "%s-dealloc", queue->name);
757 if (IS_ERR(task))
758 goto kthread_err;
759 queue->dealloc_task = task;
760
761 if (tx_evtchn == rx_evtchn) {
762 /* feature-split-event-channels == 0 */
763 err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
764 dev, tx_evtchn, xenvif_interrupt, 0,
765 queue->name, queue);
766 if (err < 0)
767 goto err;
768 queue->tx_irq = queue->rx_irq = err;
769 disable_irq(queue->tx_irq);
770 } else {
771 /* feature-split-event-channels == 1 */
772 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
773 "%s-tx", queue->name);
774 err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
775 dev, tx_evtchn, xenvif_tx_interrupt, 0,
776 queue->tx_irq_name, queue);
777 if (err < 0)
778 goto err;
779 queue->tx_irq = err;
780 disable_irq(queue->tx_irq);
781
782 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
783 "%s-rx", queue->name);
784 err = bind_interdomain_evtchn_to_irqhandler_lateeoi(
785 dev, rx_evtchn, xenvif_rx_interrupt, 0,
786 queue->rx_irq_name, queue);
787 if (err < 0)
788 goto err;
789 queue->rx_irq = err;
790 disable_irq(queue->rx_irq);
791 }
792
793 return 0;
794
795kthread_err:
796 pr_warn("Could not allocate kthread for %s\n", queue->name);
797 err = PTR_ERR(task);
798err:
799 xenvif_disconnect_queue(queue);
800 return err;
801}
802
803void xenvif_carrier_off(struct xenvif *vif)
804{
805 struct net_device *dev = vif->dev;
806
807 rtnl_lock();
808 if (test_and_clear_bit(VIF_STATUS_CONNECTED, &vif->status)) {
809 netif_carrier_off(dev); /* discard queued packets */
810 if (netif_running(dev))
811 xenvif_down(vif);
812 }
813 rtnl_unlock();
814}
815
816void xenvif_disconnect_data(struct xenvif *vif)
817{
818 struct xenvif_queue *queue = NULL;
819 unsigned int num_queues = vif->num_queues;
820 unsigned int queue_index;
821
822 xenvif_carrier_off(vif);
823
824 for (queue_index = 0; queue_index < num_queues; ++queue_index) {
825 queue = &vif->queues[queue_index];
826
827 xenvif_disconnect_queue(queue);
828 }
829
830 xenvif_mcast_addr_list_free(vif);
831}
832
833void xenvif_disconnect_ctrl(struct xenvif *vif)
834{
835 if (vif->ctrl_irq) {
836 xenvif_deinit_hash(vif);
837 unbind_from_irqhandler(vif->ctrl_irq, vif);
838 vif->ctrl_irq = 0;
839 }
840
841 if (vif->ctrl.sring) {
842 xenbus_unmap_ring_vfree(xenvif_to_xenbus_device(vif),
843 vif->ctrl.sring);
844 vif->ctrl.sring = NULL;
845 }
846}
847
848/* Reverse the relevant parts of xenvif_init_queue().
849 * Used for queue teardown from xenvif_free(), and on the
850 * error handling paths in xenbus.c:connect().
851 */
852void xenvif_deinit_queue(struct xenvif_queue *queue)
853{
854 gnttab_free_pages(MAX_PENDING_REQS, queue->mmap_pages);
855}
856
857void xenvif_free(struct xenvif *vif)
858{
859 struct xenvif_queue *queues = vif->queues;
860 unsigned int num_queues = vif->num_queues;
861 unsigned int queue_index;
862
863 unregister_netdev(vif->dev);
864 free_netdev(vif->dev);
865
866 for (queue_index = 0; queue_index < num_queues; ++queue_index)
867 xenvif_deinit_queue(&queues[queue_index]);
868 vfree(queues);
869
870 module_put(THIS_MODULE);
871}