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