• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Virtual network driver for conversing with remote driver backends.
3  *
4  * Copyright (c) 2002-2005, K A Fraser
5  * Copyright (c) 2005, XenSource Ltd
6  *
7  * This program is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU General Public License version 2
9  * as published by the Free Software Foundation; or, when distributed
10  * separately from the Linux kernel or incorporated into other
11  * software packages, subject to the following license:
12  *
13  * Permission is hereby granted, free of charge, to any person obtaining a copy
14  * of this source file (the "Software"), to deal in the Software without
15  * restriction, including without limitation the rights to use, copy, modify,
16  * merge, publish, distribute, sublicense, and/or sell copies of the Software,
17  * and to permit persons to whom the Software is furnished to do so, subject to
18  * the following conditions:
19  *
20  * The above copyright notice and this permission notice shall be included in
21  * all copies or substantial portions of the Software.
22  *
23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
24  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
25  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
26  * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
27  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
28  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
29  * IN THE SOFTWARE.
30  */
31 
32 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
33 
34 #include <linux/module.h>
35 #include <linux/kernel.h>
36 #include <linux/netdevice.h>
37 #include <linux/etherdevice.h>
38 #include <linux/skbuff.h>
39 #include <linux/ethtool.h>
40 #include <linux/if_ether.h>
41 #include <net/tcp.h>
42 #include <linux/udp.h>
43 #include <linux/moduleparam.h>
44 #include <linux/mm.h>
45 #include <linux/slab.h>
46 #include <net/ip.h>
47 #include <linux/bpf.h>
48 #include <net/page_pool.h>
49 #include <linux/bpf_trace.h>
50 
51 #include <xen/xen.h>
52 #include <xen/xenbus.h>
53 #include <xen/events.h>
54 #include <xen/page.h>
55 #include <xen/platform_pci.h>
56 #include <xen/grant_table.h>
57 
58 #include <xen/interface/io/netif.h>
59 #include <xen/interface/memory.h>
60 #include <xen/interface/grant_table.h>
61 
62 /* Module parameters */
63 #define MAX_QUEUES_DEFAULT 8
64 static unsigned int xennet_max_queues;
65 module_param_named(max_queues, xennet_max_queues, uint, 0644);
66 MODULE_PARM_DESC(max_queues,
67 		 "Maximum number of queues per virtual interface");
68 
69 static bool __read_mostly xennet_trusted = true;
70 module_param_named(trusted, xennet_trusted, bool, 0644);
71 MODULE_PARM_DESC(trusted, "Is the backend trusted");
72 
73 #define XENNET_TIMEOUT  (5 * HZ)
74 
75 static const struct ethtool_ops xennet_ethtool_ops;
76 
77 struct netfront_cb {
78 	int pull_to;
79 };
80 
81 #define NETFRONT_SKB_CB(skb)	((struct netfront_cb *)((skb)->cb))
82 
83 #define RX_COPY_THRESHOLD 256
84 
85 #define GRANT_INVALID_REF	0
86 
87 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
88 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
89 
90 /* Minimum number of Rx slots (includes slot for GSO metadata). */
91 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
92 
93 /* Queue name is interface name with "-qNNN" appended */
94 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
95 
96 /* IRQ name is queue name with "-tx" or "-rx" appended */
97 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
98 
99 static DECLARE_WAIT_QUEUE_HEAD(module_wq);
100 
101 struct netfront_stats {
102 	u64			packets;
103 	u64			bytes;
104 	struct u64_stats_sync	syncp;
105 };
106 
107 struct netfront_info;
108 
109 struct netfront_queue {
110 	unsigned int id; /* Queue ID, 0-based */
111 	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
112 	struct netfront_info *info;
113 
114 	struct bpf_prog __rcu *xdp_prog;
115 
116 	struct napi_struct napi;
117 
118 	/* Split event channels support, tx_* == rx_* when using
119 	 * single event channel.
120 	 */
121 	unsigned int tx_evtchn, rx_evtchn;
122 	unsigned int tx_irq, rx_irq;
123 	/* Only used when split event channels support is enabled */
124 	char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
125 	char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
126 
127 	spinlock_t   tx_lock;
128 	struct xen_netif_tx_front_ring tx;
129 	int tx_ring_ref;
130 
131 	/*
132 	 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
133 	 * are linked from tx_skb_freelist through tx_link.
134 	 */
135 	struct sk_buff *tx_skbs[NET_TX_RING_SIZE];
136 	unsigned short tx_link[NET_TX_RING_SIZE];
137 #define TX_LINK_NONE 0xffff
138 #define TX_PENDING   0xfffe
139 	grant_ref_t gref_tx_head;
140 	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
141 	struct page *grant_tx_page[NET_TX_RING_SIZE];
142 	unsigned tx_skb_freelist;
143 	unsigned int tx_pend_queue;
144 
145 	spinlock_t   rx_lock ____cacheline_aligned_in_smp;
146 	struct xen_netif_rx_front_ring rx;
147 	int rx_ring_ref;
148 
149 	struct timer_list rx_refill_timer;
150 
151 	struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
152 	grant_ref_t gref_rx_head;
153 	grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
154 
155 	unsigned int rx_rsp_unconsumed;
156 	spinlock_t rx_cons_lock;
157 
158 	struct page_pool *page_pool;
159 	struct xdp_rxq_info xdp_rxq;
160 };
161 
162 struct netfront_info {
163 	struct list_head list;
164 	struct net_device *netdev;
165 
166 	struct xenbus_device *xbdev;
167 
168 	/* Multi-queue support */
169 	struct netfront_queue *queues;
170 
171 	/* Statistics */
172 	struct netfront_stats __percpu *rx_stats;
173 	struct netfront_stats __percpu *tx_stats;
174 
175 	/* XDP state */
176 	bool netback_has_xdp_headroom;
177 	bool netfront_xdp_enabled;
178 
179 	/* Is device behaving sane? */
180 	bool broken;
181 
182 	/* Should skbs be bounced into a zeroed buffer? */
183 	bool bounce;
184 
185 	atomic_t rx_gso_checksum_fixup;
186 };
187 
188 struct netfront_rx_info {
189 	struct xen_netif_rx_response rx;
190 	struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
191 };
192 
193 /*
194  * Access macros for acquiring freeing slots in tx_skbs[].
195  */
196 
add_id_to_list(unsigned * head,unsigned short * list,unsigned short id)197 static void add_id_to_list(unsigned *head, unsigned short *list,
198 			   unsigned short id)
199 {
200 	list[id] = *head;
201 	*head = id;
202 }
203 
get_id_from_list(unsigned * head,unsigned short * list)204 static unsigned short get_id_from_list(unsigned *head, unsigned short *list)
205 {
206 	unsigned int id = *head;
207 
208 	if (id != TX_LINK_NONE) {
209 		*head = list[id];
210 		list[id] = TX_LINK_NONE;
211 	}
212 	return id;
213 }
214 
xennet_rxidx(RING_IDX idx)215 static int xennet_rxidx(RING_IDX idx)
216 {
217 	return idx & (NET_RX_RING_SIZE - 1);
218 }
219 
xennet_get_rx_skb(struct netfront_queue * queue,RING_IDX ri)220 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
221 					 RING_IDX ri)
222 {
223 	int i = xennet_rxidx(ri);
224 	struct sk_buff *skb = queue->rx_skbs[i];
225 	queue->rx_skbs[i] = NULL;
226 	return skb;
227 }
228 
xennet_get_rx_ref(struct netfront_queue * queue,RING_IDX ri)229 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
230 					    RING_IDX ri)
231 {
232 	int i = xennet_rxidx(ri);
233 	grant_ref_t ref = queue->grant_rx_ref[i];
234 	queue->grant_rx_ref[i] = GRANT_INVALID_REF;
235 	return ref;
236 }
237 
238 #ifdef CONFIG_SYSFS
239 static const struct attribute_group xennet_dev_group;
240 #endif
241 
xennet_can_sg(struct net_device * dev)242 static bool xennet_can_sg(struct net_device *dev)
243 {
244 	return dev->features & NETIF_F_SG;
245 }
246 
247 
rx_refill_timeout(struct timer_list * t)248 static void rx_refill_timeout(struct timer_list *t)
249 {
250 	struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer);
251 	napi_schedule(&queue->napi);
252 }
253 
netfront_tx_slot_available(struct netfront_queue * queue)254 static int netfront_tx_slot_available(struct netfront_queue *queue)
255 {
256 	return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
257 		(NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
258 }
259 
xennet_maybe_wake_tx(struct netfront_queue * queue)260 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
261 {
262 	struct net_device *dev = queue->info->netdev;
263 	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
264 
265 	if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
266 	    netfront_tx_slot_available(queue) &&
267 	    likely(netif_running(dev)))
268 		netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
269 }
270 
271 
xennet_alloc_one_rx_buffer(struct netfront_queue * queue)272 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
273 {
274 	struct sk_buff *skb;
275 	struct page *page;
276 
277 	skb = __netdev_alloc_skb(queue->info->netdev,
278 				 RX_COPY_THRESHOLD + NET_IP_ALIGN,
279 				 GFP_ATOMIC | __GFP_NOWARN);
280 	if (unlikely(!skb))
281 		return NULL;
282 
283 	page = page_pool_alloc_pages(queue->page_pool,
284 				     GFP_ATOMIC | __GFP_NOWARN | __GFP_ZERO);
285 	if (unlikely(!page)) {
286 		kfree_skb(skb);
287 		return NULL;
288 	}
289 	skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
290 
291 	/* Align ip header to a 16 bytes boundary */
292 	skb_reserve(skb, NET_IP_ALIGN);
293 	skb->dev = queue->info->netdev;
294 
295 	return skb;
296 }
297 
298 
xennet_alloc_rx_buffers(struct netfront_queue * queue)299 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
300 {
301 	RING_IDX req_prod = queue->rx.req_prod_pvt;
302 	int notify;
303 	int err = 0;
304 
305 	if (unlikely(!netif_carrier_ok(queue->info->netdev)))
306 		return;
307 
308 	for (req_prod = queue->rx.req_prod_pvt;
309 	     req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
310 	     req_prod++) {
311 		struct sk_buff *skb;
312 		unsigned short id;
313 		grant_ref_t ref;
314 		struct page *page;
315 		struct xen_netif_rx_request *req;
316 
317 		skb = xennet_alloc_one_rx_buffer(queue);
318 		if (!skb) {
319 			err = -ENOMEM;
320 			break;
321 		}
322 
323 		id = xennet_rxidx(req_prod);
324 
325 		BUG_ON(queue->rx_skbs[id]);
326 		queue->rx_skbs[id] = skb;
327 
328 		ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
329 		WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
330 		queue->grant_rx_ref[id] = ref;
331 
332 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
333 
334 		req = RING_GET_REQUEST(&queue->rx, req_prod);
335 		gnttab_page_grant_foreign_access_ref_one(ref,
336 							 queue->info->xbdev->otherend_id,
337 							 page,
338 							 0);
339 		req->id = id;
340 		req->gref = ref;
341 	}
342 
343 	queue->rx.req_prod_pvt = req_prod;
344 
345 	/* Try again later if there are not enough requests or skb allocation
346 	 * failed.
347 	 * Enough requests is quantified as the sum of newly created slots and
348 	 * the unconsumed slots at the backend.
349 	 */
350 	if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
351 	    unlikely(err)) {
352 		mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
353 		return;
354 	}
355 
356 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
357 	if (notify)
358 		notify_remote_via_irq(queue->rx_irq);
359 }
360 
xennet_open(struct net_device * dev)361 static int xennet_open(struct net_device *dev)
362 {
363 	struct netfront_info *np = netdev_priv(dev);
364 	unsigned int num_queues = dev->real_num_tx_queues;
365 	unsigned int i = 0;
366 	struct netfront_queue *queue = NULL;
367 
368 	if (!np->queues || np->broken)
369 		return -ENODEV;
370 
371 	for (i = 0; i < num_queues; ++i) {
372 		queue = &np->queues[i];
373 		napi_enable(&queue->napi);
374 
375 		spin_lock_bh(&queue->rx_lock);
376 		if (netif_carrier_ok(dev)) {
377 			xennet_alloc_rx_buffers(queue);
378 			queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
379 			if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
380 				napi_schedule(&queue->napi);
381 		}
382 		spin_unlock_bh(&queue->rx_lock);
383 	}
384 
385 	netif_tx_start_all_queues(dev);
386 
387 	return 0;
388 }
389 
xennet_tx_buf_gc(struct netfront_queue * queue)390 static bool xennet_tx_buf_gc(struct netfront_queue *queue)
391 {
392 	RING_IDX cons, prod;
393 	unsigned short id;
394 	struct sk_buff *skb;
395 	bool more_to_do;
396 	bool work_done = false;
397 	const struct device *dev = &queue->info->netdev->dev;
398 
399 	BUG_ON(!netif_carrier_ok(queue->info->netdev));
400 
401 	do {
402 		prod = queue->tx.sring->rsp_prod;
403 		if (RING_RESPONSE_PROD_OVERFLOW(&queue->tx, prod)) {
404 			dev_alert(dev, "Illegal number of responses %u\n",
405 				  prod - queue->tx.rsp_cons);
406 			goto err;
407 		}
408 		rmb(); /* Ensure we see responses up to 'rp'. */
409 
410 		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
411 			struct xen_netif_tx_response txrsp;
412 
413 			work_done = true;
414 
415 			RING_COPY_RESPONSE(&queue->tx, cons, &txrsp);
416 			if (txrsp.status == XEN_NETIF_RSP_NULL)
417 				continue;
418 
419 			id = txrsp.id;
420 			if (id >= RING_SIZE(&queue->tx)) {
421 				dev_alert(dev,
422 					  "Response has incorrect id (%u)\n",
423 					  id);
424 				goto err;
425 			}
426 			if (queue->tx_link[id] != TX_PENDING) {
427 				dev_alert(dev,
428 					  "Response for inactive request\n");
429 				goto err;
430 			}
431 
432 			queue->tx_link[id] = TX_LINK_NONE;
433 			skb = queue->tx_skbs[id];
434 			queue->tx_skbs[id] = NULL;
435 			if (unlikely(!gnttab_end_foreign_access_ref(
436 				queue->grant_tx_ref[id], GNTMAP_readonly))) {
437 				dev_alert(dev,
438 					  "Grant still in use by backend domain\n");
439 				goto err;
440 			}
441 			gnttab_release_grant_reference(
442 				&queue->gref_tx_head, queue->grant_tx_ref[id]);
443 			queue->grant_tx_ref[id] = GRANT_INVALID_REF;
444 			queue->grant_tx_page[id] = NULL;
445 			add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, id);
446 			dev_kfree_skb_irq(skb);
447 		}
448 
449 		queue->tx.rsp_cons = prod;
450 
451 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
452 	} while (more_to_do);
453 
454 	xennet_maybe_wake_tx(queue);
455 
456 	return work_done;
457 
458  err:
459 	queue->info->broken = true;
460 	dev_alert(dev, "Disabled for further use\n");
461 
462 	return work_done;
463 }
464 
465 struct xennet_gnttab_make_txreq {
466 	struct netfront_queue *queue;
467 	struct sk_buff *skb;
468 	struct page *page;
469 	struct xen_netif_tx_request *tx;      /* Last request on ring page */
470 	struct xen_netif_tx_request tx_local; /* Last request local copy*/
471 	unsigned int size;
472 };
473 
xennet_tx_setup_grant(unsigned long gfn,unsigned int offset,unsigned int len,void * data)474 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
475 				  unsigned int len, void *data)
476 {
477 	struct xennet_gnttab_make_txreq *info = data;
478 	unsigned int id;
479 	struct xen_netif_tx_request *tx;
480 	grant_ref_t ref;
481 	/* convenient aliases */
482 	struct page *page = info->page;
483 	struct netfront_queue *queue = info->queue;
484 	struct sk_buff *skb = info->skb;
485 
486 	id = get_id_from_list(&queue->tx_skb_freelist, queue->tx_link);
487 	tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
488 	ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
489 	WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
490 
491 	gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
492 					gfn, GNTMAP_readonly);
493 
494 	queue->tx_skbs[id] = skb;
495 	queue->grant_tx_page[id] = page;
496 	queue->grant_tx_ref[id] = ref;
497 
498 	info->tx_local.id = id;
499 	info->tx_local.gref = ref;
500 	info->tx_local.offset = offset;
501 	info->tx_local.size = len;
502 	info->tx_local.flags = 0;
503 
504 	*tx = info->tx_local;
505 
506 	/*
507 	 * Put the request in the pending queue, it will be set to be pending
508 	 * when the producer index is about to be raised.
509 	 */
510 	add_id_to_list(&queue->tx_pend_queue, queue->tx_link, id);
511 
512 	info->tx = tx;
513 	info->size += info->tx_local.size;
514 }
515 
xennet_make_first_txreq(struct xennet_gnttab_make_txreq * info,unsigned int offset,unsigned int len)516 static struct xen_netif_tx_request *xennet_make_first_txreq(
517 	struct xennet_gnttab_make_txreq *info,
518 	unsigned int offset, unsigned int len)
519 {
520 	info->size = 0;
521 
522 	gnttab_for_one_grant(info->page, offset, len, xennet_tx_setup_grant, info);
523 
524 	return info->tx;
525 }
526 
xennet_make_one_txreq(unsigned long gfn,unsigned int offset,unsigned int len,void * data)527 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
528 				  unsigned int len, void *data)
529 {
530 	struct xennet_gnttab_make_txreq *info = data;
531 
532 	info->tx->flags |= XEN_NETTXF_more_data;
533 	skb_get(info->skb);
534 	xennet_tx_setup_grant(gfn, offset, len, data);
535 }
536 
xennet_make_txreqs(struct xennet_gnttab_make_txreq * info,struct page * page,unsigned int offset,unsigned int len)537 static void xennet_make_txreqs(
538 	struct xennet_gnttab_make_txreq *info,
539 	struct page *page,
540 	unsigned int offset, unsigned int len)
541 {
542 	/* Skip unused frames from start of page */
543 	page += offset >> PAGE_SHIFT;
544 	offset &= ~PAGE_MASK;
545 
546 	while (len) {
547 		info->page = page;
548 		info->size = 0;
549 
550 		gnttab_foreach_grant_in_range(page, offset, len,
551 					      xennet_make_one_txreq,
552 					      info);
553 
554 		page++;
555 		offset = 0;
556 		len -= info->size;
557 	}
558 }
559 
560 /*
561  * Count how many ring slots are required to send this skb. Each frag
562  * might be a compound page.
563  */
xennet_count_skb_slots(struct sk_buff * skb)564 static int xennet_count_skb_slots(struct sk_buff *skb)
565 {
566 	int i, frags = skb_shinfo(skb)->nr_frags;
567 	int slots;
568 
569 	slots = gnttab_count_grant(offset_in_page(skb->data),
570 				   skb_headlen(skb));
571 
572 	for (i = 0; i < frags; i++) {
573 		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
574 		unsigned long size = skb_frag_size(frag);
575 		unsigned long offset = skb_frag_off(frag);
576 
577 		/* Skip unused frames from start of page */
578 		offset &= ~PAGE_MASK;
579 
580 		slots += gnttab_count_grant(offset, size);
581 	}
582 
583 	return slots;
584 }
585 
xennet_select_queue(struct net_device * dev,struct sk_buff * skb,struct net_device * sb_dev)586 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
587 			       struct net_device *sb_dev)
588 {
589 	unsigned int num_queues = dev->real_num_tx_queues;
590 	u32 hash;
591 	u16 queue_idx;
592 
593 	/* First, check if there is only one queue */
594 	if (num_queues == 1) {
595 		queue_idx = 0;
596 	} else {
597 		hash = skb_get_hash(skb);
598 		queue_idx = hash % num_queues;
599 	}
600 
601 	return queue_idx;
602 }
603 
xennet_mark_tx_pending(struct netfront_queue * queue)604 static void xennet_mark_tx_pending(struct netfront_queue *queue)
605 {
606 	unsigned int i;
607 
608 	while ((i = get_id_from_list(&queue->tx_pend_queue, queue->tx_link)) !=
609 	       TX_LINK_NONE)
610 		queue->tx_link[i] = TX_PENDING;
611 }
612 
xennet_xdp_xmit_one(struct net_device * dev,struct netfront_queue * queue,struct xdp_frame * xdpf)613 static int xennet_xdp_xmit_one(struct net_device *dev,
614 			       struct netfront_queue *queue,
615 			       struct xdp_frame *xdpf)
616 {
617 	struct netfront_info *np = netdev_priv(dev);
618 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
619 	struct xennet_gnttab_make_txreq info = {
620 		.queue = queue,
621 		.skb = NULL,
622 		.page = virt_to_page(xdpf->data),
623 	};
624 	int notify;
625 
626 	xennet_make_first_txreq(&info,
627 				offset_in_page(xdpf->data),
628 				xdpf->len);
629 
630 	xennet_mark_tx_pending(queue);
631 
632 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
633 	if (notify)
634 		notify_remote_via_irq(queue->tx_irq);
635 
636 	u64_stats_update_begin(&tx_stats->syncp);
637 	tx_stats->bytes += xdpf->len;
638 	tx_stats->packets++;
639 	u64_stats_update_end(&tx_stats->syncp);
640 
641 	xennet_tx_buf_gc(queue);
642 
643 	return 0;
644 }
645 
xennet_xdp_xmit(struct net_device * dev,int n,struct xdp_frame ** frames,u32 flags)646 static int xennet_xdp_xmit(struct net_device *dev, int n,
647 			   struct xdp_frame **frames, u32 flags)
648 {
649 	unsigned int num_queues = dev->real_num_tx_queues;
650 	struct netfront_info *np = netdev_priv(dev);
651 	struct netfront_queue *queue = NULL;
652 	unsigned long irq_flags;
653 	int drops = 0;
654 	int i, err;
655 
656 	if (unlikely(np->broken))
657 		return -ENODEV;
658 	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
659 		return -EINVAL;
660 
661 	queue = &np->queues[smp_processor_id() % num_queues];
662 
663 	spin_lock_irqsave(&queue->tx_lock, irq_flags);
664 	for (i = 0; i < n; i++) {
665 		struct xdp_frame *xdpf = frames[i];
666 
667 		if (!xdpf)
668 			continue;
669 		err = xennet_xdp_xmit_one(dev, queue, xdpf);
670 		if (err) {
671 			xdp_return_frame_rx_napi(xdpf);
672 			drops++;
673 		}
674 	}
675 	spin_unlock_irqrestore(&queue->tx_lock, irq_flags);
676 
677 	return n - drops;
678 }
679 
bounce_skb(const struct sk_buff * skb)680 struct sk_buff *bounce_skb(const struct sk_buff *skb)
681 {
682 	unsigned int headerlen = skb_headroom(skb);
683 	/* Align size to allocate full pages and avoid contiguous data leaks */
684 	unsigned int size = ALIGN(skb_end_offset(skb) + skb->data_len,
685 				  XEN_PAGE_SIZE);
686 	struct sk_buff *n = alloc_skb(size, GFP_ATOMIC | __GFP_ZERO);
687 
688 	if (!n)
689 		return NULL;
690 
691 	if (!IS_ALIGNED((uintptr_t)n->head, XEN_PAGE_SIZE)) {
692 		WARN_ONCE(1, "misaligned skb allocated\n");
693 		kfree_skb(n);
694 		return NULL;
695 	}
696 
697 	/* Set the data pointer */
698 	skb_reserve(n, headerlen);
699 	/* Set the tail pointer and length */
700 	skb_put(n, skb->len);
701 
702 	BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
703 
704 	skb_copy_header(n, skb);
705 	return n;
706 }
707 
708 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
709 
xennet_start_xmit(struct sk_buff * skb,struct net_device * dev)710 static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
711 {
712 	struct netfront_info *np = netdev_priv(dev);
713 	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
714 	struct xen_netif_tx_request *first_tx;
715 	unsigned int i;
716 	int notify;
717 	int slots;
718 	struct page *page;
719 	unsigned int offset;
720 	unsigned int len;
721 	unsigned long flags;
722 	struct netfront_queue *queue = NULL;
723 	struct xennet_gnttab_make_txreq info = { };
724 	unsigned int num_queues = dev->real_num_tx_queues;
725 	u16 queue_index;
726 	struct sk_buff *nskb;
727 
728 	/* Drop the packet if no queues are set up */
729 	if (num_queues < 1)
730 		goto drop;
731 	if (unlikely(np->broken))
732 		goto drop;
733 	/* Determine which queue to transmit this SKB on */
734 	queue_index = skb_get_queue_mapping(skb);
735 	queue = &np->queues[queue_index];
736 
737 	/* If skb->len is too big for wire format, drop skb and alert
738 	 * user about misconfiguration.
739 	 */
740 	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
741 		net_alert_ratelimited(
742 			"xennet: skb->len = %u, too big for wire format\n",
743 			skb->len);
744 		goto drop;
745 	}
746 
747 	slots = xennet_count_skb_slots(skb);
748 	if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
749 		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
750 				    slots, skb->len);
751 		if (skb_linearize(skb))
752 			goto drop;
753 	}
754 
755 	page = virt_to_page(skb->data);
756 	offset = offset_in_page(skb->data);
757 
758 	/* The first req should be at least ETH_HLEN size or the packet will be
759 	 * dropped by netback.
760 	 *
761 	 * If the backend is not trusted bounce all data to zeroed pages to
762 	 * avoid exposing contiguous data on the granted page not belonging to
763 	 * the skb.
764 	 */
765 	if (np->bounce || unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
766 		nskb = bounce_skb(skb);
767 		if (!nskb)
768 			goto drop;
769 		dev_consume_skb_any(skb);
770 		skb = nskb;
771 		page = virt_to_page(skb->data);
772 		offset = offset_in_page(skb->data);
773 	}
774 
775 	len = skb_headlen(skb);
776 
777 	spin_lock_irqsave(&queue->tx_lock, flags);
778 
779 	if (unlikely(!netif_carrier_ok(dev) ||
780 		     (slots > 1 && !xennet_can_sg(dev)) ||
781 		     netif_needs_gso(skb, netif_skb_features(skb)))) {
782 		spin_unlock_irqrestore(&queue->tx_lock, flags);
783 		goto drop;
784 	}
785 
786 	/* First request for the linear area. */
787 	info.queue = queue;
788 	info.skb = skb;
789 	info.page = page;
790 	first_tx = xennet_make_first_txreq(&info, offset, len);
791 	offset += info.tx_local.size;
792 	if (offset == PAGE_SIZE) {
793 		page++;
794 		offset = 0;
795 	}
796 	len -= info.tx_local.size;
797 
798 	if (skb->ip_summed == CHECKSUM_PARTIAL)
799 		/* local packet? */
800 		first_tx->flags |= XEN_NETTXF_csum_blank |
801 				   XEN_NETTXF_data_validated;
802 	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
803 		/* remote but checksummed. */
804 		first_tx->flags |= XEN_NETTXF_data_validated;
805 
806 	/* Optional extra info after the first request. */
807 	if (skb_shinfo(skb)->gso_size) {
808 		struct xen_netif_extra_info *gso;
809 
810 		gso = (struct xen_netif_extra_info *)
811 			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
812 
813 		first_tx->flags |= XEN_NETTXF_extra_info;
814 
815 		gso->u.gso.size = skb_shinfo(skb)->gso_size;
816 		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
817 			XEN_NETIF_GSO_TYPE_TCPV6 :
818 			XEN_NETIF_GSO_TYPE_TCPV4;
819 		gso->u.gso.pad = 0;
820 		gso->u.gso.features = 0;
821 
822 		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
823 		gso->flags = 0;
824 	}
825 
826 	/* Requests for the rest of the linear area. */
827 	xennet_make_txreqs(&info, page, offset, len);
828 
829 	/* Requests for all the frags. */
830 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
831 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
832 		xennet_make_txreqs(&info, skb_frag_page(frag),
833 					skb_frag_off(frag),
834 					skb_frag_size(frag));
835 	}
836 
837 	/* First request has the packet length. */
838 	first_tx->size = skb->len;
839 
840 	/* timestamp packet in software */
841 	skb_tx_timestamp(skb);
842 
843 	xennet_mark_tx_pending(queue);
844 
845 	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
846 	if (notify)
847 		notify_remote_via_irq(queue->tx_irq);
848 
849 	u64_stats_update_begin(&tx_stats->syncp);
850 	tx_stats->bytes += skb->len;
851 	tx_stats->packets++;
852 	u64_stats_update_end(&tx_stats->syncp);
853 
854 	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
855 	xennet_tx_buf_gc(queue);
856 
857 	if (!netfront_tx_slot_available(queue))
858 		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
859 
860 	spin_unlock_irqrestore(&queue->tx_lock, flags);
861 
862 	return NETDEV_TX_OK;
863 
864  drop:
865 	dev->stats.tx_dropped++;
866 	dev_kfree_skb_any(skb);
867 	return NETDEV_TX_OK;
868 }
869 
xennet_close(struct net_device * dev)870 static int xennet_close(struct net_device *dev)
871 {
872 	struct netfront_info *np = netdev_priv(dev);
873 	unsigned int num_queues = dev->real_num_tx_queues;
874 	unsigned int i;
875 	struct netfront_queue *queue;
876 	netif_tx_stop_all_queues(np->netdev);
877 	for (i = 0; i < num_queues; ++i) {
878 		queue = &np->queues[i];
879 		napi_disable(&queue->napi);
880 	}
881 	return 0;
882 }
883 
xennet_destroy_queues(struct netfront_info * info)884 static void xennet_destroy_queues(struct netfront_info *info)
885 {
886 	unsigned int i;
887 
888 	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
889 		struct netfront_queue *queue = &info->queues[i];
890 
891 		if (netif_running(info->netdev))
892 			napi_disable(&queue->napi);
893 		netif_napi_del(&queue->napi);
894 	}
895 
896 	kfree(info->queues);
897 	info->queues = NULL;
898 }
899 
xennet_uninit(struct net_device * dev)900 static void xennet_uninit(struct net_device *dev)
901 {
902 	struct netfront_info *np = netdev_priv(dev);
903 	xennet_destroy_queues(np);
904 }
905 
xennet_set_rx_rsp_cons(struct netfront_queue * queue,RING_IDX val)906 static void xennet_set_rx_rsp_cons(struct netfront_queue *queue, RING_IDX val)
907 {
908 	unsigned long flags;
909 
910 	spin_lock_irqsave(&queue->rx_cons_lock, flags);
911 	queue->rx.rsp_cons = val;
912 	queue->rx_rsp_unconsumed = RING_HAS_UNCONSUMED_RESPONSES(&queue->rx);
913 	spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
914 }
915 
xennet_move_rx_slot(struct netfront_queue * queue,struct sk_buff * skb,grant_ref_t ref)916 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
917 				grant_ref_t ref)
918 {
919 	int new = xennet_rxidx(queue->rx.req_prod_pvt);
920 
921 	BUG_ON(queue->rx_skbs[new]);
922 	queue->rx_skbs[new] = skb;
923 	queue->grant_rx_ref[new] = ref;
924 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
925 	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
926 	queue->rx.req_prod_pvt++;
927 }
928 
xennet_get_extras(struct netfront_queue * queue,struct xen_netif_extra_info * extras,RING_IDX rp)929 static int xennet_get_extras(struct netfront_queue *queue,
930 			     struct xen_netif_extra_info *extras,
931 			     RING_IDX rp)
932 
933 {
934 	struct xen_netif_extra_info extra;
935 	struct device *dev = &queue->info->netdev->dev;
936 	RING_IDX cons = queue->rx.rsp_cons;
937 	int err = 0;
938 
939 	do {
940 		struct sk_buff *skb;
941 		grant_ref_t ref;
942 
943 		if (unlikely(cons + 1 == rp)) {
944 			if (net_ratelimit())
945 				dev_warn(dev, "Missing extra info\n");
946 			err = -EBADR;
947 			break;
948 		}
949 
950 		RING_COPY_RESPONSE(&queue->rx, ++cons, &extra);
951 
952 		if (unlikely(!extra.type ||
953 			     extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
954 			if (net_ratelimit())
955 				dev_warn(dev, "Invalid extra type: %d\n",
956 					 extra.type);
957 			err = -EINVAL;
958 		} else {
959 			extras[extra.type - 1] = extra;
960 		}
961 
962 		skb = xennet_get_rx_skb(queue, cons);
963 		ref = xennet_get_rx_ref(queue, cons);
964 		xennet_move_rx_slot(queue, skb, ref);
965 	} while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
966 
967 	xennet_set_rx_rsp_cons(queue, cons);
968 	return err;
969 }
970 
xennet_run_xdp(struct netfront_queue * queue,struct page * pdata,struct xen_netif_rx_response * rx,struct bpf_prog * prog,struct xdp_buff * xdp,bool * need_xdp_flush)971 static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
972 		   struct xen_netif_rx_response *rx, struct bpf_prog *prog,
973 		   struct xdp_buff *xdp, bool *need_xdp_flush)
974 {
975 	struct xdp_frame *xdpf;
976 	u32 len = rx->status;
977 	u32 act;
978 	int err;
979 
980 	xdp->data_hard_start = page_address(pdata);
981 	xdp->data = xdp->data_hard_start + XDP_PACKET_HEADROOM;
982 	xdp_set_data_meta_invalid(xdp);
983 	xdp->data_end = xdp->data + len;
984 	xdp->rxq = &queue->xdp_rxq;
985 	xdp->frame_sz = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
986 
987 	act = bpf_prog_run_xdp(prog, xdp);
988 	switch (act) {
989 	case XDP_TX:
990 		get_page(pdata);
991 		xdpf = xdp_convert_buff_to_frame(xdp);
992 		err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0);
993 		if (unlikely(err < 0))
994 			trace_xdp_exception(queue->info->netdev, prog, act);
995 		break;
996 	case XDP_REDIRECT:
997 		get_page(pdata);
998 		err = xdp_do_redirect(queue->info->netdev, xdp, prog);
999 		*need_xdp_flush = true;
1000 		if (unlikely(err))
1001 			trace_xdp_exception(queue->info->netdev, prog, act);
1002 		break;
1003 	case XDP_PASS:
1004 	case XDP_DROP:
1005 		break;
1006 
1007 	case XDP_ABORTED:
1008 		trace_xdp_exception(queue->info->netdev, prog, act);
1009 		break;
1010 
1011 	default:
1012 		bpf_warn_invalid_xdp_action(act);
1013 	}
1014 
1015 	return act;
1016 }
1017 
xennet_get_responses(struct netfront_queue * queue,struct netfront_rx_info * rinfo,RING_IDX rp,struct sk_buff_head * list,bool * need_xdp_flush)1018 static int xennet_get_responses(struct netfront_queue *queue,
1019 				struct netfront_rx_info *rinfo, RING_IDX rp,
1020 				struct sk_buff_head *list,
1021 				bool *need_xdp_flush)
1022 {
1023 	struct xen_netif_rx_response *rx = &rinfo->rx, rx_local;
1024 	int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
1025 	RING_IDX cons = queue->rx.rsp_cons;
1026 	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
1027 	struct xen_netif_extra_info *extras = rinfo->extras;
1028 	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
1029 	struct device *dev = &queue->info->netdev->dev;
1030 	struct bpf_prog *xdp_prog;
1031 	struct xdp_buff xdp;
1032 	int slots = 1;
1033 	int err = 0;
1034 	u32 verdict;
1035 
1036 	if (rx->flags & XEN_NETRXF_extra_info) {
1037 		err = xennet_get_extras(queue, extras, rp);
1038 		if (!err) {
1039 			if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) {
1040 				struct xen_netif_extra_info *xdp;
1041 
1042 				xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1];
1043 				rx->offset = xdp->u.xdp.headroom;
1044 			}
1045 		}
1046 		cons = queue->rx.rsp_cons;
1047 	}
1048 
1049 	for (;;) {
1050 		if (unlikely(rx->status < 0 ||
1051 			     rx->offset + rx->status > XEN_PAGE_SIZE)) {
1052 			if (net_ratelimit())
1053 				dev_warn(dev, "rx->offset: %u, size: %d\n",
1054 					 rx->offset, rx->status);
1055 			xennet_move_rx_slot(queue, skb, ref);
1056 			err = -EINVAL;
1057 			goto next;
1058 		}
1059 
1060 		/*
1061 		 * This definitely indicates a bug, either in this driver or in
1062 		 * the backend driver. In future this should flag the bad
1063 		 * situation to the system controller to reboot the backend.
1064 		 */
1065 		if (ref == GRANT_INVALID_REF) {
1066 			if (net_ratelimit())
1067 				dev_warn(dev, "Bad rx response id %d.\n",
1068 					 rx->id);
1069 			err = -EINVAL;
1070 			goto next;
1071 		}
1072 
1073 		if (!gnttab_end_foreign_access_ref(ref, 0)) {
1074 			dev_alert(dev,
1075 				  "Grant still in use by backend domain\n");
1076 			queue->info->broken = true;
1077 			dev_alert(dev, "Disabled for further use\n");
1078 			return -EINVAL;
1079 		}
1080 
1081 		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
1082 
1083 		rcu_read_lock();
1084 		xdp_prog = rcu_dereference(queue->xdp_prog);
1085 		if (xdp_prog) {
1086 			if (!(rx->flags & XEN_NETRXF_more_data)) {
1087 				/* currently only a single page contains data */
1088 				verdict = xennet_run_xdp(queue,
1089 							 skb_frag_page(&skb_shinfo(skb)->frags[0]),
1090 							 rx, xdp_prog, &xdp, need_xdp_flush);
1091 				if (verdict != XDP_PASS)
1092 					err = -EINVAL;
1093 			} else {
1094 				/* drop the frame */
1095 				err = -EINVAL;
1096 			}
1097 		}
1098 		rcu_read_unlock();
1099 
1100 		__skb_queue_tail(list, skb);
1101 
1102 next:
1103 		if (!(rx->flags & XEN_NETRXF_more_data))
1104 			break;
1105 
1106 		if (cons + slots == rp) {
1107 			if (net_ratelimit())
1108 				dev_warn(dev, "Need more slots\n");
1109 			err = -ENOENT;
1110 			break;
1111 		}
1112 
1113 		RING_COPY_RESPONSE(&queue->rx, cons + slots, &rx_local);
1114 		rx = &rx_local;
1115 		skb = xennet_get_rx_skb(queue, cons + slots);
1116 		ref = xennet_get_rx_ref(queue, cons + slots);
1117 		slots++;
1118 	}
1119 
1120 	if (unlikely(slots > max)) {
1121 		if (net_ratelimit())
1122 			dev_warn(dev, "Too many slots\n");
1123 		err = -E2BIG;
1124 	}
1125 
1126 	if (unlikely(err))
1127 		xennet_set_rx_rsp_cons(queue, cons + slots);
1128 
1129 	return err;
1130 }
1131 
xennet_set_skb_gso(struct sk_buff * skb,struct xen_netif_extra_info * gso)1132 static int xennet_set_skb_gso(struct sk_buff *skb,
1133 			      struct xen_netif_extra_info *gso)
1134 {
1135 	if (!gso->u.gso.size) {
1136 		if (net_ratelimit())
1137 			pr_warn("GSO size must not be zero\n");
1138 		return -EINVAL;
1139 	}
1140 
1141 	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
1142 	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
1143 		if (net_ratelimit())
1144 			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
1145 		return -EINVAL;
1146 	}
1147 
1148 	skb_shinfo(skb)->gso_size = gso->u.gso.size;
1149 	skb_shinfo(skb)->gso_type =
1150 		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
1151 		SKB_GSO_TCPV4 :
1152 		SKB_GSO_TCPV6;
1153 
1154 	/* Header must be checked, and gso_segs computed. */
1155 	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1156 	skb_shinfo(skb)->gso_segs = 0;
1157 
1158 	return 0;
1159 }
1160 
xennet_fill_frags(struct netfront_queue * queue,struct sk_buff * skb,struct sk_buff_head * list)1161 static int xennet_fill_frags(struct netfront_queue *queue,
1162 			     struct sk_buff *skb,
1163 			     struct sk_buff_head *list)
1164 {
1165 	RING_IDX cons = queue->rx.rsp_cons;
1166 	struct sk_buff *nskb;
1167 
1168 	while ((nskb = __skb_dequeue(list))) {
1169 		struct xen_netif_rx_response rx;
1170 		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
1171 
1172 		RING_COPY_RESPONSE(&queue->rx, ++cons, &rx);
1173 
1174 		if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
1175 			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1176 
1177 			BUG_ON(pull_to < skb_headlen(skb));
1178 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1179 		}
1180 		if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
1181 			xennet_set_rx_rsp_cons(queue,
1182 					       ++cons + skb_queue_len(list));
1183 			kfree_skb(nskb);
1184 			return -ENOENT;
1185 		}
1186 
1187 		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
1188 				skb_frag_page(nfrag),
1189 				rx.offset, rx.status, PAGE_SIZE);
1190 
1191 		skb_shinfo(nskb)->nr_frags = 0;
1192 		kfree_skb(nskb);
1193 	}
1194 
1195 	xennet_set_rx_rsp_cons(queue, cons);
1196 
1197 	return 0;
1198 }
1199 
checksum_setup(struct net_device * dev,struct sk_buff * skb)1200 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
1201 {
1202 	bool recalculate_partial_csum = false;
1203 
1204 	/*
1205 	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1206 	 * peers can fail to set NETRXF_csum_blank when sending a GSO
1207 	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1208 	 * recalculate the partial checksum.
1209 	 */
1210 	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1211 		struct netfront_info *np = netdev_priv(dev);
1212 		atomic_inc(&np->rx_gso_checksum_fixup);
1213 		skb->ip_summed = CHECKSUM_PARTIAL;
1214 		recalculate_partial_csum = true;
1215 	}
1216 
1217 	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1218 	if (skb->ip_summed != CHECKSUM_PARTIAL)
1219 		return 0;
1220 
1221 	return skb_checksum_setup(skb, recalculate_partial_csum);
1222 }
1223 
handle_incoming_queue(struct netfront_queue * queue,struct sk_buff_head * rxq)1224 static int handle_incoming_queue(struct netfront_queue *queue,
1225 				 struct sk_buff_head *rxq)
1226 {
1227 	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
1228 	int packets_dropped = 0;
1229 	struct sk_buff *skb;
1230 
1231 	while ((skb = __skb_dequeue(rxq)) != NULL) {
1232 		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1233 
1234 		if (pull_to > skb_headlen(skb))
1235 			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1236 
1237 		/* Ethernet work: Delayed to here as it peeks the header. */
1238 		skb->protocol = eth_type_trans(skb, queue->info->netdev);
1239 		skb_reset_network_header(skb);
1240 
1241 		if (checksum_setup(queue->info->netdev, skb)) {
1242 			kfree_skb(skb);
1243 			packets_dropped++;
1244 			queue->info->netdev->stats.rx_errors++;
1245 			continue;
1246 		}
1247 
1248 		u64_stats_update_begin(&rx_stats->syncp);
1249 		rx_stats->packets++;
1250 		rx_stats->bytes += skb->len;
1251 		u64_stats_update_end(&rx_stats->syncp);
1252 
1253 		/* Pass it up. */
1254 		napi_gro_receive(&queue->napi, skb);
1255 	}
1256 
1257 	return packets_dropped;
1258 }
1259 
xennet_poll(struct napi_struct * napi,int budget)1260 static int xennet_poll(struct napi_struct *napi, int budget)
1261 {
1262 	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
1263 	struct net_device *dev = queue->info->netdev;
1264 	struct sk_buff *skb;
1265 	struct netfront_rx_info rinfo;
1266 	struct xen_netif_rx_response *rx = &rinfo.rx;
1267 	struct xen_netif_extra_info *extras = rinfo.extras;
1268 	RING_IDX i, rp;
1269 	int work_done;
1270 	struct sk_buff_head rxq;
1271 	struct sk_buff_head errq;
1272 	struct sk_buff_head tmpq;
1273 	int err;
1274 	bool need_xdp_flush = false;
1275 
1276 	spin_lock(&queue->rx_lock);
1277 
1278 	skb_queue_head_init(&rxq);
1279 	skb_queue_head_init(&errq);
1280 	skb_queue_head_init(&tmpq);
1281 
1282 	rp = queue->rx.sring->rsp_prod;
1283 	if (RING_RESPONSE_PROD_OVERFLOW(&queue->rx, rp)) {
1284 		dev_alert(&dev->dev, "Illegal number of responses %u\n",
1285 			  rp - queue->rx.rsp_cons);
1286 		queue->info->broken = true;
1287 		spin_unlock(&queue->rx_lock);
1288 		return 0;
1289 	}
1290 	rmb(); /* Ensure we see queued responses up to 'rp'. */
1291 
1292 	i = queue->rx.rsp_cons;
1293 	work_done = 0;
1294 	while ((i != rp) && (work_done < budget)) {
1295 		RING_COPY_RESPONSE(&queue->rx, i, rx);
1296 		memset(extras, 0, sizeof(rinfo.extras));
1297 
1298 		err = xennet_get_responses(queue, &rinfo, rp, &tmpq,
1299 					   &need_xdp_flush);
1300 
1301 		if (unlikely(err)) {
1302 			if (queue->info->broken) {
1303 				spin_unlock(&queue->rx_lock);
1304 				return 0;
1305 			}
1306 err:
1307 			while ((skb = __skb_dequeue(&tmpq)))
1308 				__skb_queue_tail(&errq, skb);
1309 			dev->stats.rx_errors++;
1310 			i = queue->rx.rsp_cons;
1311 			continue;
1312 		}
1313 
1314 		skb = __skb_dequeue(&tmpq);
1315 
1316 		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1317 			struct xen_netif_extra_info *gso;
1318 			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1319 
1320 			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1321 				__skb_queue_head(&tmpq, skb);
1322 				xennet_set_rx_rsp_cons(queue,
1323 						       queue->rx.rsp_cons +
1324 						       skb_queue_len(&tmpq));
1325 				goto err;
1326 			}
1327 		}
1328 
1329 		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1330 		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1331 			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1332 
1333 		skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset);
1334 		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1335 		skb->data_len = rx->status;
1336 		skb->len += rx->status;
1337 
1338 		if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
1339 			goto err;
1340 
1341 		if (rx->flags & XEN_NETRXF_csum_blank)
1342 			skb->ip_summed = CHECKSUM_PARTIAL;
1343 		else if (rx->flags & XEN_NETRXF_data_validated)
1344 			skb->ip_summed = CHECKSUM_UNNECESSARY;
1345 
1346 		__skb_queue_tail(&rxq, skb);
1347 
1348 		i = queue->rx.rsp_cons + 1;
1349 		xennet_set_rx_rsp_cons(queue, i);
1350 		work_done++;
1351 	}
1352 	if (need_xdp_flush)
1353 		xdp_do_flush();
1354 
1355 	__skb_queue_purge(&errq);
1356 
1357 	work_done -= handle_incoming_queue(queue, &rxq);
1358 
1359 	xennet_alloc_rx_buffers(queue);
1360 
1361 	if (work_done < budget) {
1362 		int more_to_do = 0;
1363 
1364 		napi_complete_done(napi, work_done);
1365 
1366 		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1367 		if (more_to_do)
1368 			napi_schedule(napi);
1369 	}
1370 
1371 	spin_unlock(&queue->rx_lock);
1372 
1373 	return work_done;
1374 }
1375 
xennet_change_mtu(struct net_device * dev,int mtu)1376 static int xennet_change_mtu(struct net_device *dev, int mtu)
1377 {
1378 	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1379 
1380 	if (mtu > max)
1381 		return -EINVAL;
1382 	dev->mtu = mtu;
1383 	return 0;
1384 }
1385 
xennet_get_stats64(struct net_device * dev,struct rtnl_link_stats64 * tot)1386 static void xennet_get_stats64(struct net_device *dev,
1387 			       struct rtnl_link_stats64 *tot)
1388 {
1389 	struct netfront_info *np = netdev_priv(dev);
1390 	int cpu;
1391 
1392 	for_each_possible_cpu(cpu) {
1393 		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1394 		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1395 		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1396 		unsigned int start;
1397 
1398 		do {
1399 			start = u64_stats_fetch_begin_irq(&tx_stats->syncp);
1400 			tx_packets = tx_stats->packets;
1401 			tx_bytes = tx_stats->bytes;
1402 		} while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start));
1403 
1404 		do {
1405 			start = u64_stats_fetch_begin_irq(&rx_stats->syncp);
1406 			rx_packets = rx_stats->packets;
1407 			rx_bytes = rx_stats->bytes;
1408 		} while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start));
1409 
1410 		tot->rx_packets += rx_packets;
1411 		tot->tx_packets += tx_packets;
1412 		tot->rx_bytes   += rx_bytes;
1413 		tot->tx_bytes   += tx_bytes;
1414 	}
1415 
1416 	tot->rx_errors  = dev->stats.rx_errors;
1417 	tot->tx_dropped = dev->stats.tx_dropped;
1418 }
1419 
xennet_release_tx_bufs(struct netfront_queue * queue)1420 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1421 {
1422 	struct sk_buff *skb;
1423 	int i;
1424 
1425 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1426 		/* Skip over entries which are actually freelist references */
1427 		if (!queue->tx_skbs[i])
1428 			continue;
1429 
1430 		skb = queue->tx_skbs[i];
1431 		queue->tx_skbs[i] = NULL;
1432 		get_page(queue->grant_tx_page[i]);
1433 		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1434 					  GNTMAP_readonly,
1435 					  (unsigned long)page_address(queue->grant_tx_page[i]));
1436 		queue->grant_tx_page[i] = NULL;
1437 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
1438 		add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, i);
1439 		dev_kfree_skb_irq(skb);
1440 	}
1441 }
1442 
xennet_release_rx_bufs(struct netfront_queue * queue)1443 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1444 {
1445 	int id, ref;
1446 
1447 	spin_lock_bh(&queue->rx_lock);
1448 
1449 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1450 		struct sk_buff *skb;
1451 		struct page *page;
1452 
1453 		skb = queue->rx_skbs[id];
1454 		if (!skb)
1455 			continue;
1456 
1457 		ref = queue->grant_rx_ref[id];
1458 		if (ref == GRANT_INVALID_REF)
1459 			continue;
1460 
1461 		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1462 
1463 		/* gnttab_end_foreign_access() needs a page ref until
1464 		 * foreign access is ended (which may be deferred).
1465 		 */
1466 		get_page(page);
1467 		gnttab_end_foreign_access(ref, 0,
1468 					  (unsigned long)page_address(page));
1469 		queue->grant_rx_ref[id] = GRANT_INVALID_REF;
1470 
1471 		kfree_skb(skb);
1472 	}
1473 
1474 	spin_unlock_bh(&queue->rx_lock);
1475 }
1476 
xennet_fix_features(struct net_device * dev,netdev_features_t features)1477 static netdev_features_t xennet_fix_features(struct net_device *dev,
1478 	netdev_features_t features)
1479 {
1480 	struct netfront_info *np = netdev_priv(dev);
1481 
1482 	if (features & NETIF_F_SG &&
1483 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1484 		features &= ~NETIF_F_SG;
1485 
1486 	if (features & NETIF_F_IPV6_CSUM &&
1487 	    !xenbus_read_unsigned(np->xbdev->otherend,
1488 				  "feature-ipv6-csum-offload", 0))
1489 		features &= ~NETIF_F_IPV6_CSUM;
1490 
1491 	if (features & NETIF_F_TSO &&
1492 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1493 		features &= ~NETIF_F_TSO;
1494 
1495 	if (features & NETIF_F_TSO6 &&
1496 	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1497 		features &= ~NETIF_F_TSO6;
1498 
1499 	return features;
1500 }
1501 
xennet_set_features(struct net_device * dev,netdev_features_t features)1502 static int xennet_set_features(struct net_device *dev,
1503 	netdev_features_t features)
1504 {
1505 	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1506 		netdev_info(dev, "Reducing MTU because no SG offload");
1507 		dev->mtu = ETH_DATA_LEN;
1508 	}
1509 
1510 	return 0;
1511 }
1512 
xennet_handle_tx(struct netfront_queue * queue,unsigned int * eoi)1513 static bool xennet_handle_tx(struct netfront_queue *queue, unsigned int *eoi)
1514 {
1515 	unsigned long flags;
1516 
1517 	if (unlikely(queue->info->broken))
1518 		return false;
1519 
1520 	spin_lock_irqsave(&queue->tx_lock, flags);
1521 	if (xennet_tx_buf_gc(queue))
1522 		*eoi = 0;
1523 	spin_unlock_irqrestore(&queue->tx_lock, flags);
1524 
1525 	return true;
1526 }
1527 
xennet_tx_interrupt(int irq,void * dev_id)1528 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1529 {
1530 	unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1531 
1532 	if (likely(xennet_handle_tx(dev_id, &eoiflag)))
1533 		xen_irq_lateeoi(irq, eoiflag);
1534 
1535 	return IRQ_HANDLED;
1536 }
1537 
xennet_handle_rx(struct netfront_queue * queue,unsigned int * eoi)1538 static bool xennet_handle_rx(struct netfront_queue *queue, unsigned int *eoi)
1539 {
1540 	unsigned int work_queued;
1541 	unsigned long flags;
1542 
1543 	if (unlikely(queue->info->broken))
1544 		return false;
1545 
1546 	spin_lock_irqsave(&queue->rx_cons_lock, flags);
1547 	work_queued = RING_HAS_UNCONSUMED_RESPONSES(&queue->rx);
1548 	if (work_queued > queue->rx_rsp_unconsumed) {
1549 		queue->rx_rsp_unconsumed = work_queued;
1550 		*eoi = 0;
1551 	} else if (unlikely(work_queued < queue->rx_rsp_unconsumed)) {
1552 		const struct device *dev = &queue->info->netdev->dev;
1553 
1554 		spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
1555 		dev_alert(dev, "RX producer index going backwards\n");
1556 		dev_alert(dev, "Disabled for further use\n");
1557 		queue->info->broken = true;
1558 		return false;
1559 	}
1560 	spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
1561 
1562 	if (likely(netif_carrier_ok(queue->info->netdev) && work_queued))
1563 		napi_schedule(&queue->napi);
1564 
1565 	return true;
1566 }
1567 
xennet_rx_interrupt(int irq,void * dev_id)1568 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1569 {
1570 	unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1571 
1572 	if (likely(xennet_handle_rx(dev_id, &eoiflag)))
1573 		xen_irq_lateeoi(irq, eoiflag);
1574 
1575 	return IRQ_HANDLED;
1576 }
1577 
xennet_interrupt(int irq,void * dev_id)1578 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1579 {
1580 	unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1581 
1582 	if (xennet_handle_tx(dev_id, &eoiflag) &&
1583 	    xennet_handle_rx(dev_id, &eoiflag))
1584 		xen_irq_lateeoi(irq, eoiflag);
1585 
1586 	return IRQ_HANDLED;
1587 }
1588 
1589 #ifdef CONFIG_NET_POLL_CONTROLLER
xennet_poll_controller(struct net_device * dev)1590 static void xennet_poll_controller(struct net_device *dev)
1591 {
1592 	/* Poll each queue */
1593 	struct netfront_info *info = netdev_priv(dev);
1594 	unsigned int num_queues = dev->real_num_tx_queues;
1595 	unsigned int i;
1596 
1597 	if (info->broken)
1598 		return;
1599 
1600 	for (i = 0; i < num_queues; ++i)
1601 		xennet_interrupt(0, &info->queues[i]);
1602 }
1603 #endif
1604 
1605 #define NETBACK_XDP_HEADROOM_DISABLE	0
1606 #define NETBACK_XDP_HEADROOM_ENABLE	1
1607 
talk_to_netback_xdp(struct netfront_info * np,int xdp)1608 static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
1609 {
1610 	int err;
1611 	unsigned short headroom;
1612 
1613 	headroom = xdp ? XDP_PACKET_HEADROOM : 0;
1614 	err = xenbus_printf(XBT_NIL, np->xbdev->nodename,
1615 			    "xdp-headroom", "%hu",
1616 			    headroom);
1617 	if (err)
1618 		pr_warn("Error writing xdp-headroom\n");
1619 
1620 	return err;
1621 }
1622 
xennet_xdp_set(struct net_device * dev,struct bpf_prog * prog,struct netlink_ext_ack * extack)1623 static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1624 			  struct netlink_ext_ack *extack)
1625 {
1626 	unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
1627 	struct netfront_info *np = netdev_priv(dev);
1628 	struct bpf_prog *old_prog;
1629 	unsigned int i, err;
1630 
1631 	if (dev->mtu > max_mtu) {
1632 		netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu);
1633 		return -EINVAL;
1634 	}
1635 
1636 	if (!np->netback_has_xdp_headroom)
1637 		return 0;
1638 
1639 	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);
1640 
1641 	err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE :
1642 				  NETBACK_XDP_HEADROOM_DISABLE);
1643 	if (err)
1644 		return err;
1645 
1646 	/* avoid the race with XDP headroom adjustment */
1647 	wait_event(module_wq,
1648 		   xenbus_read_driver_state(np->xbdev->otherend) ==
1649 		   XenbusStateReconfigured);
1650 	np->netfront_xdp_enabled = true;
1651 
1652 	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
1653 
1654 	if (prog)
1655 		bpf_prog_add(prog, dev->real_num_tx_queues);
1656 
1657 	for (i = 0; i < dev->real_num_tx_queues; ++i)
1658 		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
1659 
1660 	if (old_prog)
1661 		for (i = 0; i < dev->real_num_tx_queues; ++i)
1662 			bpf_prog_put(old_prog);
1663 
1664 	xenbus_switch_state(np->xbdev, XenbusStateConnected);
1665 
1666 	return 0;
1667 }
1668 
xennet_xdp(struct net_device * dev,struct netdev_bpf * xdp)1669 static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1670 {
1671 	struct netfront_info *np = netdev_priv(dev);
1672 
1673 	if (np->broken)
1674 		return -ENODEV;
1675 
1676 	switch (xdp->command) {
1677 	case XDP_SETUP_PROG:
1678 		return xennet_xdp_set(dev, xdp->prog, xdp->extack);
1679 	default:
1680 		return -EINVAL;
1681 	}
1682 }
1683 
1684 static const struct net_device_ops xennet_netdev_ops = {
1685 	.ndo_uninit          = xennet_uninit,
1686 	.ndo_open            = xennet_open,
1687 	.ndo_stop            = xennet_close,
1688 	.ndo_start_xmit      = xennet_start_xmit,
1689 	.ndo_change_mtu	     = xennet_change_mtu,
1690 	.ndo_get_stats64     = xennet_get_stats64,
1691 	.ndo_set_mac_address = eth_mac_addr,
1692 	.ndo_validate_addr   = eth_validate_addr,
1693 	.ndo_fix_features    = xennet_fix_features,
1694 	.ndo_set_features    = xennet_set_features,
1695 	.ndo_select_queue    = xennet_select_queue,
1696 	.ndo_bpf            = xennet_xdp,
1697 	.ndo_xdp_xmit	    = xennet_xdp_xmit,
1698 #ifdef CONFIG_NET_POLL_CONTROLLER
1699 	.ndo_poll_controller = xennet_poll_controller,
1700 #endif
1701 };
1702 
xennet_free_netdev(struct net_device * netdev)1703 static void xennet_free_netdev(struct net_device *netdev)
1704 {
1705 	struct netfront_info *np = netdev_priv(netdev);
1706 
1707 	free_percpu(np->rx_stats);
1708 	free_percpu(np->tx_stats);
1709 	free_netdev(netdev);
1710 }
1711 
xennet_create_dev(struct xenbus_device * dev)1712 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1713 {
1714 	int err;
1715 	struct net_device *netdev;
1716 	struct netfront_info *np;
1717 
1718 	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1719 	if (!netdev)
1720 		return ERR_PTR(-ENOMEM);
1721 
1722 	np                   = netdev_priv(netdev);
1723 	np->xbdev            = dev;
1724 
1725 	np->queues = NULL;
1726 
1727 	err = -ENOMEM;
1728 	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1729 	if (np->rx_stats == NULL)
1730 		goto exit;
1731 	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1732 	if (np->tx_stats == NULL)
1733 		goto exit;
1734 
1735 	netdev->netdev_ops	= &xennet_netdev_ops;
1736 
1737 	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1738 				  NETIF_F_GSO_ROBUST;
1739 	netdev->hw_features	= NETIF_F_SG |
1740 				  NETIF_F_IPV6_CSUM |
1741 				  NETIF_F_TSO | NETIF_F_TSO6;
1742 
1743 	/*
1744          * Assume that all hw features are available for now. This set
1745          * will be adjusted by the call to netdev_update_features() in
1746          * xennet_connect() which is the earliest point where we can
1747          * negotiate with the backend regarding supported features.
1748          */
1749 	netdev->features |= netdev->hw_features;
1750 
1751 	netdev->ethtool_ops = &xennet_ethtool_ops;
1752 	netdev->min_mtu = ETH_MIN_MTU;
1753 	netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1754 	SET_NETDEV_DEV(netdev, &dev->dev);
1755 
1756 	np->netdev = netdev;
1757 	np->netfront_xdp_enabled = false;
1758 
1759 	netif_carrier_off(netdev);
1760 
1761 	do {
1762 		xenbus_switch_state(dev, XenbusStateInitialising);
1763 		err = wait_event_timeout(module_wq,
1764 				 xenbus_read_driver_state(dev->otherend) !=
1765 				 XenbusStateClosed &&
1766 				 xenbus_read_driver_state(dev->otherend) !=
1767 				 XenbusStateUnknown, XENNET_TIMEOUT);
1768 	} while (!err);
1769 
1770 	return netdev;
1771 
1772  exit:
1773 	xennet_free_netdev(netdev);
1774 	return ERR_PTR(err);
1775 }
1776 
1777 /**
1778  * Entry point to this code when a new device is created.  Allocate the basic
1779  * structures and the ring buffers for communication with the backend, and
1780  * inform the backend of the appropriate details for those.
1781  */
netfront_probe(struct xenbus_device * dev,const struct xenbus_device_id * id)1782 static int netfront_probe(struct xenbus_device *dev,
1783 			  const struct xenbus_device_id *id)
1784 {
1785 	int err;
1786 	struct net_device *netdev;
1787 	struct netfront_info *info;
1788 
1789 	netdev = xennet_create_dev(dev);
1790 	if (IS_ERR(netdev)) {
1791 		err = PTR_ERR(netdev);
1792 		xenbus_dev_fatal(dev, err, "creating netdev");
1793 		return err;
1794 	}
1795 
1796 	info = netdev_priv(netdev);
1797 	dev_set_drvdata(&dev->dev, info);
1798 #ifdef CONFIG_SYSFS
1799 	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1800 #endif
1801 
1802 	return 0;
1803 }
1804 
xennet_end_access(int ref,void * page)1805 static void xennet_end_access(int ref, void *page)
1806 {
1807 	/* This frees the page as a side-effect */
1808 	if (ref != GRANT_INVALID_REF)
1809 		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1810 }
1811 
xennet_disconnect_backend(struct netfront_info * info)1812 static void xennet_disconnect_backend(struct netfront_info *info)
1813 {
1814 	unsigned int i = 0;
1815 	unsigned int num_queues = info->netdev->real_num_tx_queues;
1816 
1817 	netif_carrier_off(info->netdev);
1818 
1819 	for (i = 0; i < num_queues && info->queues; ++i) {
1820 		struct netfront_queue *queue = &info->queues[i];
1821 
1822 		del_timer_sync(&queue->rx_refill_timer);
1823 
1824 		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1825 			unbind_from_irqhandler(queue->tx_irq, queue);
1826 		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1827 			unbind_from_irqhandler(queue->tx_irq, queue);
1828 			unbind_from_irqhandler(queue->rx_irq, queue);
1829 		}
1830 		queue->tx_evtchn = queue->rx_evtchn = 0;
1831 		queue->tx_irq = queue->rx_irq = 0;
1832 
1833 		if (netif_running(info->netdev))
1834 			napi_synchronize(&queue->napi);
1835 
1836 		xennet_release_tx_bufs(queue);
1837 		xennet_release_rx_bufs(queue);
1838 		gnttab_free_grant_references(queue->gref_tx_head);
1839 		gnttab_free_grant_references(queue->gref_rx_head);
1840 
1841 		/* End access and free the pages */
1842 		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1843 		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1844 
1845 		queue->tx_ring_ref = GRANT_INVALID_REF;
1846 		queue->rx_ring_ref = GRANT_INVALID_REF;
1847 		queue->tx.sring = NULL;
1848 		queue->rx.sring = NULL;
1849 
1850 		page_pool_destroy(queue->page_pool);
1851 	}
1852 }
1853 
1854 /**
1855  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1856  * driver restart.  We tear down our netif structure and recreate it, but
1857  * leave the device-layer structures intact so that this is transparent to the
1858  * rest of the kernel.
1859  */
netfront_resume(struct xenbus_device * dev)1860 static int netfront_resume(struct xenbus_device *dev)
1861 {
1862 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1863 
1864 	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1865 
1866 	netif_tx_lock_bh(info->netdev);
1867 	netif_device_detach(info->netdev);
1868 	netif_tx_unlock_bh(info->netdev);
1869 
1870 	xennet_disconnect_backend(info);
1871 
1872 	rtnl_lock();
1873 	if (info->queues)
1874 		xennet_destroy_queues(info);
1875 	rtnl_unlock();
1876 
1877 	return 0;
1878 }
1879 
xen_net_read_mac(struct xenbus_device * dev,u8 mac[])1880 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1881 {
1882 	char *s, *e, *macstr;
1883 	int i;
1884 
1885 	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1886 	if (IS_ERR(macstr))
1887 		return PTR_ERR(macstr);
1888 
1889 	for (i = 0; i < ETH_ALEN; i++) {
1890 		mac[i] = simple_strtoul(s, &e, 16);
1891 		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1892 			kfree(macstr);
1893 			return -ENOENT;
1894 		}
1895 		s = e+1;
1896 	}
1897 
1898 	kfree(macstr);
1899 	return 0;
1900 }
1901 
setup_netfront_single(struct netfront_queue * queue)1902 static int setup_netfront_single(struct netfront_queue *queue)
1903 {
1904 	int err;
1905 
1906 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1907 	if (err < 0)
1908 		goto fail;
1909 
1910 	err = bind_evtchn_to_irqhandler_lateeoi(queue->tx_evtchn,
1911 						xennet_interrupt, 0,
1912 						queue->info->netdev->name,
1913 						queue);
1914 	if (err < 0)
1915 		goto bind_fail;
1916 	queue->rx_evtchn = queue->tx_evtchn;
1917 	queue->rx_irq = queue->tx_irq = err;
1918 
1919 	return 0;
1920 
1921 bind_fail:
1922 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1923 	queue->tx_evtchn = 0;
1924 fail:
1925 	return err;
1926 }
1927 
setup_netfront_split(struct netfront_queue * queue)1928 static int setup_netfront_split(struct netfront_queue *queue)
1929 {
1930 	int err;
1931 
1932 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1933 	if (err < 0)
1934 		goto fail;
1935 	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1936 	if (err < 0)
1937 		goto alloc_rx_evtchn_fail;
1938 
1939 	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1940 		 "%s-tx", queue->name);
1941 	err = bind_evtchn_to_irqhandler_lateeoi(queue->tx_evtchn,
1942 						xennet_tx_interrupt, 0,
1943 						queue->tx_irq_name, queue);
1944 	if (err < 0)
1945 		goto bind_tx_fail;
1946 	queue->tx_irq = err;
1947 
1948 	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1949 		 "%s-rx", queue->name);
1950 	err = bind_evtchn_to_irqhandler_lateeoi(queue->rx_evtchn,
1951 						xennet_rx_interrupt, 0,
1952 						queue->rx_irq_name, queue);
1953 	if (err < 0)
1954 		goto bind_rx_fail;
1955 	queue->rx_irq = err;
1956 
1957 	return 0;
1958 
1959 bind_rx_fail:
1960 	unbind_from_irqhandler(queue->tx_irq, queue);
1961 	queue->tx_irq = 0;
1962 bind_tx_fail:
1963 	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1964 	queue->rx_evtchn = 0;
1965 alloc_rx_evtchn_fail:
1966 	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1967 	queue->tx_evtchn = 0;
1968 fail:
1969 	return err;
1970 }
1971 
setup_netfront(struct xenbus_device * dev,struct netfront_queue * queue,unsigned int feature_split_evtchn)1972 static int setup_netfront(struct xenbus_device *dev,
1973 			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1974 {
1975 	struct xen_netif_tx_sring *txs;
1976 	struct xen_netif_rx_sring *rxs = NULL;
1977 	grant_ref_t gref;
1978 	int err;
1979 
1980 	queue->tx_ring_ref = GRANT_INVALID_REF;
1981 	queue->rx_ring_ref = GRANT_INVALID_REF;
1982 	queue->rx.sring = NULL;
1983 	queue->tx.sring = NULL;
1984 
1985 	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1986 	if (!txs) {
1987 		err = -ENOMEM;
1988 		xenbus_dev_fatal(dev, err, "allocating tx ring page");
1989 		goto fail;
1990 	}
1991 	SHARED_RING_INIT(txs);
1992 	FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1993 
1994 	err = xenbus_grant_ring(dev, txs, 1, &gref);
1995 	if (err < 0)
1996 		goto fail;
1997 	queue->tx_ring_ref = gref;
1998 
1999 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
2000 	if (!rxs) {
2001 		err = -ENOMEM;
2002 		xenbus_dev_fatal(dev, err, "allocating rx ring page");
2003 		goto fail;
2004 	}
2005 	SHARED_RING_INIT(rxs);
2006 	FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
2007 
2008 	err = xenbus_grant_ring(dev, rxs, 1, &gref);
2009 	if (err < 0)
2010 		goto fail;
2011 	queue->rx_ring_ref = gref;
2012 
2013 	if (feature_split_evtchn)
2014 		err = setup_netfront_split(queue);
2015 	/* setup single event channel if
2016 	 *  a) feature-split-event-channels == 0
2017 	 *  b) feature-split-event-channels == 1 but failed to setup
2018 	 */
2019 	if (!feature_split_evtchn || (feature_split_evtchn && err))
2020 		err = setup_netfront_single(queue);
2021 
2022 	if (err)
2023 		goto fail;
2024 
2025 	return 0;
2026 
2027 	/* If we fail to setup netfront, it is safe to just revoke access to
2028 	 * granted pages because backend is not accessing it at this point.
2029 	 */
2030  fail:
2031 	if (queue->rx_ring_ref != GRANT_INVALID_REF) {
2032 		gnttab_end_foreign_access(queue->rx_ring_ref, 0,
2033 					  (unsigned long)rxs);
2034 		queue->rx_ring_ref = GRANT_INVALID_REF;
2035 	} else {
2036 		free_page((unsigned long)rxs);
2037 	}
2038 	if (queue->tx_ring_ref != GRANT_INVALID_REF) {
2039 		gnttab_end_foreign_access(queue->tx_ring_ref, 0,
2040 					  (unsigned long)txs);
2041 		queue->tx_ring_ref = GRANT_INVALID_REF;
2042 	} else {
2043 		free_page((unsigned long)txs);
2044 	}
2045 	return err;
2046 }
2047 
2048 /* Queue-specific initialisation
2049  * This used to be done in xennet_create_dev() but must now
2050  * be run per-queue.
2051  */
xennet_init_queue(struct netfront_queue * queue)2052 static int xennet_init_queue(struct netfront_queue *queue)
2053 {
2054 	unsigned short i;
2055 	int err = 0;
2056 	char *devid;
2057 
2058 	spin_lock_init(&queue->tx_lock);
2059 	spin_lock_init(&queue->rx_lock);
2060 	spin_lock_init(&queue->rx_cons_lock);
2061 
2062 	timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
2063 
2064 	devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
2065 	snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
2066 		 devid, queue->id);
2067 
2068 	/* Initialise tx_skb_freelist as a free chain containing every entry. */
2069 	queue->tx_skb_freelist = 0;
2070 	queue->tx_pend_queue = TX_LINK_NONE;
2071 	for (i = 0; i < NET_TX_RING_SIZE; i++) {
2072 		queue->tx_link[i] = i + 1;
2073 		queue->grant_tx_ref[i] = GRANT_INVALID_REF;
2074 		queue->grant_tx_page[i] = NULL;
2075 	}
2076 	queue->tx_link[NET_TX_RING_SIZE - 1] = TX_LINK_NONE;
2077 
2078 	/* Clear out rx_skbs */
2079 	for (i = 0; i < NET_RX_RING_SIZE; i++) {
2080 		queue->rx_skbs[i] = NULL;
2081 		queue->grant_rx_ref[i] = GRANT_INVALID_REF;
2082 	}
2083 
2084 	/* A grant for every tx ring slot */
2085 	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2086 					  &queue->gref_tx_head) < 0) {
2087 		pr_alert("can't alloc tx grant refs\n");
2088 		err = -ENOMEM;
2089 		goto exit;
2090 	}
2091 
2092 	/* A grant for every rx ring slot */
2093 	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
2094 					  &queue->gref_rx_head) < 0) {
2095 		pr_alert("can't alloc rx grant refs\n");
2096 		err = -ENOMEM;
2097 		goto exit_free_tx;
2098 	}
2099 
2100 	return 0;
2101 
2102  exit_free_tx:
2103 	gnttab_free_grant_references(queue->gref_tx_head);
2104  exit:
2105 	return err;
2106 }
2107 
write_queue_xenstore_keys(struct netfront_queue * queue,struct xenbus_transaction * xbt,int write_hierarchical)2108 static int write_queue_xenstore_keys(struct netfront_queue *queue,
2109 			   struct xenbus_transaction *xbt, int write_hierarchical)
2110 {
2111 	/* Write the queue-specific keys into XenStore in the traditional
2112 	 * way for a single queue, or in a queue subkeys for multiple
2113 	 * queues.
2114 	 */
2115 	struct xenbus_device *dev = queue->info->xbdev;
2116 	int err;
2117 	const char *message;
2118 	char *path;
2119 	size_t pathsize;
2120 
2121 	/* Choose the correct place to write the keys */
2122 	if (write_hierarchical) {
2123 		pathsize = strlen(dev->nodename) + 10;
2124 		path = kzalloc(pathsize, GFP_KERNEL);
2125 		if (!path) {
2126 			err = -ENOMEM;
2127 			message = "out of memory while writing ring references";
2128 			goto error;
2129 		}
2130 		snprintf(path, pathsize, "%s/queue-%u",
2131 				dev->nodename, queue->id);
2132 	} else {
2133 		path = (char *)dev->nodename;
2134 	}
2135 
2136 	/* Write ring references */
2137 	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
2138 			queue->tx_ring_ref);
2139 	if (err) {
2140 		message = "writing tx-ring-ref";
2141 		goto error;
2142 	}
2143 
2144 	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
2145 			queue->rx_ring_ref);
2146 	if (err) {
2147 		message = "writing rx-ring-ref";
2148 		goto error;
2149 	}
2150 
2151 	/* Write event channels; taking into account both shared
2152 	 * and split event channel scenarios.
2153 	 */
2154 	if (queue->tx_evtchn == queue->rx_evtchn) {
2155 		/* Shared event channel */
2156 		err = xenbus_printf(*xbt, path,
2157 				"event-channel", "%u", queue->tx_evtchn);
2158 		if (err) {
2159 			message = "writing event-channel";
2160 			goto error;
2161 		}
2162 	} else {
2163 		/* Split event channels */
2164 		err = xenbus_printf(*xbt, path,
2165 				"event-channel-tx", "%u", queue->tx_evtchn);
2166 		if (err) {
2167 			message = "writing event-channel-tx";
2168 			goto error;
2169 		}
2170 
2171 		err = xenbus_printf(*xbt, path,
2172 				"event-channel-rx", "%u", queue->rx_evtchn);
2173 		if (err) {
2174 			message = "writing event-channel-rx";
2175 			goto error;
2176 		}
2177 	}
2178 
2179 	if (write_hierarchical)
2180 		kfree(path);
2181 	return 0;
2182 
2183 error:
2184 	if (write_hierarchical)
2185 		kfree(path);
2186 	xenbus_dev_fatal(dev, err, "%s", message);
2187 	return err;
2188 }
2189 
2190 
2191 
xennet_create_page_pool(struct netfront_queue * queue)2192 static int xennet_create_page_pool(struct netfront_queue *queue)
2193 {
2194 	int err;
2195 	struct page_pool_params pp_params = {
2196 		.order = 0,
2197 		.flags = 0,
2198 		.pool_size = NET_RX_RING_SIZE,
2199 		.nid = NUMA_NO_NODE,
2200 		.dev = &queue->info->netdev->dev,
2201 		.offset = XDP_PACKET_HEADROOM,
2202 		.max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
2203 	};
2204 
2205 	queue->page_pool = page_pool_create(&pp_params);
2206 	if (IS_ERR(queue->page_pool)) {
2207 		err = PTR_ERR(queue->page_pool);
2208 		queue->page_pool = NULL;
2209 		return err;
2210 	}
2211 
2212 	err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev,
2213 			       queue->id);
2214 	if (err) {
2215 		netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n");
2216 		goto err_free_pp;
2217 	}
2218 
2219 	err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq,
2220 					 MEM_TYPE_PAGE_POOL, queue->page_pool);
2221 	if (err) {
2222 		netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n");
2223 		goto err_unregister_rxq;
2224 	}
2225 	return 0;
2226 
2227 err_unregister_rxq:
2228 	xdp_rxq_info_unreg(&queue->xdp_rxq);
2229 err_free_pp:
2230 	page_pool_destroy(queue->page_pool);
2231 	queue->page_pool = NULL;
2232 	return err;
2233 }
2234 
xennet_create_queues(struct netfront_info * info,unsigned int * num_queues)2235 static int xennet_create_queues(struct netfront_info *info,
2236 				unsigned int *num_queues)
2237 {
2238 	unsigned int i;
2239 	int ret;
2240 
2241 	info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
2242 			       GFP_KERNEL);
2243 	if (!info->queues)
2244 		return -ENOMEM;
2245 
2246 	for (i = 0; i < *num_queues; i++) {
2247 		struct netfront_queue *queue = &info->queues[i];
2248 
2249 		queue->id = i;
2250 		queue->info = info;
2251 
2252 		ret = xennet_init_queue(queue);
2253 		if (ret < 0) {
2254 			dev_warn(&info->xbdev->dev,
2255 				 "only created %d queues\n", i);
2256 			*num_queues = i;
2257 			break;
2258 		}
2259 
2260 		/* use page pool recycling instead of buddy allocator */
2261 		ret = xennet_create_page_pool(queue);
2262 		if (ret < 0) {
2263 			dev_err(&info->xbdev->dev, "can't allocate page pool\n");
2264 			*num_queues = i;
2265 			return ret;
2266 		}
2267 
2268 		netif_napi_add(queue->info->netdev, &queue->napi,
2269 			       xennet_poll, 64);
2270 		if (netif_running(info->netdev))
2271 			napi_enable(&queue->napi);
2272 	}
2273 
2274 	netif_set_real_num_tx_queues(info->netdev, *num_queues);
2275 
2276 	if (*num_queues == 0) {
2277 		dev_err(&info->xbdev->dev, "no queues\n");
2278 		return -EINVAL;
2279 	}
2280 	return 0;
2281 }
2282 
2283 /* Common code used when first setting up, and when resuming. */
talk_to_netback(struct xenbus_device * dev,struct netfront_info * info)2284 static int talk_to_netback(struct xenbus_device *dev,
2285 			   struct netfront_info *info)
2286 {
2287 	const char *message;
2288 	struct xenbus_transaction xbt;
2289 	int err;
2290 	unsigned int feature_split_evtchn;
2291 	unsigned int i = 0;
2292 	unsigned int max_queues = 0;
2293 	struct netfront_queue *queue = NULL;
2294 	unsigned int num_queues = 1;
2295 
2296 	info->netdev->irq = 0;
2297 
2298 	/* Check if backend is trusted. */
2299 	info->bounce = !xennet_trusted ||
2300 		       !xenbus_read_unsigned(dev->nodename, "trusted", 1);
2301 
2302 	/* Check if backend supports multiple queues */
2303 	max_queues = xenbus_read_unsigned(info->xbdev->otherend,
2304 					  "multi-queue-max-queues", 1);
2305 	num_queues = min(max_queues, xennet_max_queues);
2306 
2307 	/* Check feature-split-event-channels */
2308 	feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
2309 					"feature-split-event-channels", 0);
2310 
2311 	/* Read mac addr. */
2312 	err = xen_net_read_mac(dev, info->netdev->dev_addr);
2313 	if (err) {
2314 		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
2315 		goto out_unlocked;
2316 	}
2317 
2318 	info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend,
2319 							      "feature-xdp-headroom", 0);
2320 	if (info->netback_has_xdp_headroom) {
2321 		/* set the current xen-netfront xdp state */
2322 		err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ?
2323 					  NETBACK_XDP_HEADROOM_ENABLE :
2324 					  NETBACK_XDP_HEADROOM_DISABLE);
2325 		if (err)
2326 			goto out_unlocked;
2327 	}
2328 
2329 	rtnl_lock();
2330 	if (info->queues)
2331 		xennet_destroy_queues(info);
2332 
2333 	/* For the case of a reconnect reset the "broken" indicator. */
2334 	info->broken = false;
2335 
2336 	err = xennet_create_queues(info, &num_queues);
2337 	if (err < 0) {
2338 		xenbus_dev_fatal(dev, err, "creating queues");
2339 		kfree(info->queues);
2340 		info->queues = NULL;
2341 		goto out;
2342 	}
2343 	rtnl_unlock();
2344 
2345 	/* Create shared ring, alloc event channel -- for each queue */
2346 	for (i = 0; i < num_queues; ++i) {
2347 		queue = &info->queues[i];
2348 		err = setup_netfront(dev, queue, feature_split_evtchn);
2349 		if (err)
2350 			goto destroy_ring;
2351 	}
2352 
2353 again:
2354 	err = xenbus_transaction_start(&xbt);
2355 	if (err) {
2356 		xenbus_dev_fatal(dev, err, "starting transaction");
2357 		goto destroy_ring;
2358 	}
2359 
2360 	if (xenbus_exists(XBT_NIL,
2361 			  info->xbdev->otherend, "multi-queue-max-queues")) {
2362 		/* Write the number of queues */
2363 		err = xenbus_printf(xbt, dev->nodename,
2364 				    "multi-queue-num-queues", "%u", num_queues);
2365 		if (err) {
2366 			message = "writing multi-queue-num-queues";
2367 			goto abort_transaction_no_dev_fatal;
2368 		}
2369 	}
2370 
2371 	if (num_queues == 1) {
2372 		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
2373 		if (err)
2374 			goto abort_transaction_no_dev_fatal;
2375 	} else {
2376 		/* Write the keys for each queue */
2377 		for (i = 0; i < num_queues; ++i) {
2378 			queue = &info->queues[i];
2379 			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
2380 			if (err)
2381 				goto abort_transaction_no_dev_fatal;
2382 		}
2383 	}
2384 
2385 	/* The remaining keys are not queue-specific */
2386 	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
2387 			    1);
2388 	if (err) {
2389 		message = "writing request-rx-copy";
2390 		goto abort_transaction;
2391 	}
2392 
2393 	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
2394 	if (err) {
2395 		message = "writing feature-rx-notify";
2396 		goto abort_transaction;
2397 	}
2398 
2399 	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
2400 	if (err) {
2401 		message = "writing feature-sg";
2402 		goto abort_transaction;
2403 	}
2404 
2405 	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
2406 	if (err) {
2407 		message = "writing feature-gso-tcpv4";
2408 		goto abort_transaction;
2409 	}
2410 
2411 	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
2412 	if (err) {
2413 		message = "writing feature-gso-tcpv6";
2414 		goto abort_transaction;
2415 	}
2416 
2417 	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
2418 			   "1");
2419 	if (err) {
2420 		message = "writing feature-ipv6-csum-offload";
2421 		goto abort_transaction;
2422 	}
2423 
2424 	err = xenbus_transaction_end(xbt, 0);
2425 	if (err) {
2426 		if (err == -EAGAIN)
2427 			goto again;
2428 		xenbus_dev_fatal(dev, err, "completing transaction");
2429 		goto destroy_ring;
2430 	}
2431 
2432 	return 0;
2433 
2434  abort_transaction:
2435 	xenbus_dev_fatal(dev, err, "%s", message);
2436 abort_transaction_no_dev_fatal:
2437 	xenbus_transaction_end(xbt, 1);
2438  destroy_ring:
2439 	xennet_disconnect_backend(info);
2440 	rtnl_lock();
2441 	xennet_destroy_queues(info);
2442  out:
2443 	rtnl_unlock();
2444 out_unlocked:
2445 	device_unregister(&dev->dev);
2446 	return err;
2447 }
2448 
xennet_connect(struct net_device * dev)2449 static int xennet_connect(struct net_device *dev)
2450 {
2451 	struct netfront_info *np = netdev_priv(dev);
2452 	unsigned int num_queues = 0;
2453 	int err;
2454 	unsigned int j = 0;
2455 	struct netfront_queue *queue = NULL;
2456 
2457 	if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
2458 		dev_info(&dev->dev,
2459 			 "backend does not support copying receive path\n");
2460 		return -ENODEV;
2461 	}
2462 
2463 	err = talk_to_netback(np->xbdev, np);
2464 	if (err)
2465 		return err;
2466 	if (np->netback_has_xdp_headroom)
2467 		pr_info("backend supports XDP headroom\n");
2468 	if (np->bounce)
2469 		dev_info(&np->xbdev->dev,
2470 			 "bouncing transmitted data to zeroed pages\n");
2471 
2472 	/* talk_to_netback() sets the correct number of queues */
2473 	num_queues = dev->real_num_tx_queues;
2474 
2475 	if (dev->reg_state == NETREG_UNINITIALIZED) {
2476 		err = register_netdev(dev);
2477 		if (err) {
2478 			pr_warn("%s: register_netdev err=%d\n", __func__, err);
2479 			device_unregister(&np->xbdev->dev);
2480 			return err;
2481 		}
2482 	}
2483 
2484 	rtnl_lock();
2485 	netdev_update_features(dev);
2486 	rtnl_unlock();
2487 
2488 	/*
2489 	 * All public and private state should now be sane.  Get
2490 	 * ready to start sending and receiving packets and give the driver
2491 	 * domain a kick because we've probably just requeued some
2492 	 * packets.
2493 	 */
2494 	netif_tx_lock_bh(np->netdev);
2495 	netif_device_attach(np->netdev);
2496 	netif_tx_unlock_bh(np->netdev);
2497 
2498 	netif_carrier_on(np->netdev);
2499 	for (j = 0; j < num_queues; ++j) {
2500 		queue = &np->queues[j];
2501 
2502 		notify_remote_via_irq(queue->tx_irq);
2503 		if (queue->tx_irq != queue->rx_irq)
2504 			notify_remote_via_irq(queue->rx_irq);
2505 
2506 		spin_lock_irq(&queue->tx_lock);
2507 		xennet_tx_buf_gc(queue);
2508 		spin_unlock_irq(&queue->tx_lock);
2509 
2510 		spin_lock_bh(&queue->rx_lock);
2511 		xennet_alloc_rx_buffers(queue);
2512 		spin_unlock_bh(&queue->rx_lock);
2513 	}
2514 
2515 	return 0;
2516 }
2517 
2518 /**
2519  * Callback received when the backend's state changes.
2520  */
netback_changed(struct xenbus_device * dev,enum xenbus_state backend_state)2521 static void netback_changed(struct xenbus_device *dev,
2522 			    enum xenbus_state backend_state)
2523 {
2524 	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2525 	struct net_device *netdev = np->netdev;
2526 
2527 	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2528 
2529 	wake_up_all(&module_wq);
2530 
2531 	switch (backend_state) {
2532 	case XenbusStateInitialising:
2533 	case XenbusStateInitialised:
2534 	case XenbusStateReconfiguring:
2535 	case XenbusStateReconfigured:
2536 	case XenbusStateUnknown:
2537 		break;
2538 
2539 	case XenbusStateInitWait:
2540 		if (dev->state != XenbusStateInitialising)
2541 			break;
2542 		if (xennet_connect(netdev) != 0)
2543 			break;
2544 		xenbus_switch_state(dev, XenbusStateConnected);
2545 		break;
2546 
2547 	case XenbusStateConnected:
2548 		netdev_notify_peers(netdev);
2549 		break;
2550 
2551 	case XenbusStateClosed:
2552 		if (dev->state == XenbusStateClosed)
2553 			break;
2554 		fallthrough;	/* Missed the backend's CLOSING state */
2555 	case XenbusStateClosing:
2556 		xenbus_frontend_closed(dev);
2557 		break;
2558 	}
2559 }
2560 
2561 static const struct xennet_stat {
2562 	char name[ETH_GSTRING_LEN];
2563 	u16 offset;
2564 } xennet_stats[] = {
2565 	{
2566 		"rx_gso_checksum_fixup",
2567 		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2568 	},
2569 };
2570 
xennet_get_sset_count(struct net_device * dev,int string_set)2571 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2572 {
2573 	switch (string_set) {
2574 	case ETH_SS_STATS:
2575 		return ARRAY_SIZE(xennet_stats);
2576 	default:
2577 		return -EINVAL;
2578 	}
2579 }
2580 
xennet_get_ethtool_stats(struct net_device * dev,struct ethtool_stats * stats,u64 * data)2581 static void xennet_get_ethtool_stats(struct net_device *dev,
2582 				     struct ethtool_stats *stats, u64 * data)
2583 {
2584 	void *np = netdev_priv(dev);
2585 	int i;
2586 
2587 	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2588 		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2589 }
2590 
xennet_get_strings(struct net_device * dev,u32 stringset,u8 * data)2591 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2592 {
2593 	int i;
2594 
2595 	switch (stringset) {
2596 	case ETH_SS_STATS:
2597 		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2598 			memcpy(data + i * ETH_GSTRING_LEN,
2599 			       xennet_stats[i].name, ETH_GSTRING_LEN);
2600 		break;
2601 	}
2602 }
2603 
2604 static const struct ethtool_ops xennet_ethtool_ops =
2605 {
2606 	.get_link = ethtool_op_get_link,
2607 
2608 	.get_sset_count = xennet_get_sset_count,
2609 	.get_ethtool_stats = xennet_get_ethtool_stats,
2610 	.get_strings = xennet_get_strings,
2611 	.get_ts_info = ethtool_op_get_ts_info,
2612 };
2613 
2614 #ifdef CONFIG_SYSFS
show_rxbuf(struct device * dev,struct device_attribute * attr,char * buf)2615 static ssize_t show_rxbuf(struct device *dev,
2616 			  struct device_attribute *attr, char *buf)
2617 {
2618 	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2619 }
2620 
store_rxbuf(struct device * dev,struct device_attribute * attr,const char * buf,size_t len)2621 static ssize_t store_rxbuf(struct device *dev,
2622 			   struct device_attribute *attr,
2623 			   const char *buf, size_t len)
2624 {
2625 	char *endp;
2626 	unsigned long target;
2627 
2628 	if (!capable(CAP_NET_ADMIN))
2629 		return -EPERM;
2630 
2631 	target = simple_strtoul(buf, &endp, 0);
2632 	if (endp == buf)
2633 		return -EBADMSG;
2634 
2635 	/* rxbuf_min and rxbuf_max are no longer configurable. */
2636 
2637 	return len;
2638 }
2639 
2640 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2641 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2642 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2643 
2644 static struct attribute *xennet_dev_attrs[] = {
2645 	&dev_attr_rxbuf_min.attr,
2646 	&dev_attr_rxbuf_max.attr,
2647 	&dev_attr_rxbuf_cur.attr,
2648 	NULL
2649 };
2650 
2651 static const struct attribute_group xennet_dev_group = {
2652 	.attrs = xennet_dev_attrs
2653 };
2654 #endif /* CONFIG_SYSFS */
2655 
xennet_bus_close(struct xenbus_device * dev)2656 static void xennet_bus_close(struct xenbus_device *dev)
2657 {
2658 	int ret;
2659 
2660 	if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2661 		return;
2662 	do {
2663 		xenbus_switch_state(dev, XenbusStateClosing);
2664 		ret = wait_event_timeout(module_wq,
2665 				   xenbus_read_driver_state(dev->otherend) ==
2666 				   XenbusStateClosing ||
2667 				   xenbus_read_driver_state(dev->otherend) ==
2668 				   XenbusStateClosed ||
2669 				   xenbus_read_driver_state(dev->otherend) ==
2670 				   XenbusStateUnknown,
2671 				   XENNET_TIMEOUT);
2672 	} while (!ret);
2673 
2674 	if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2675 		return;
2676 
2677 	do {
2678 		xenbus_switch_state(dev, XenbusStateClosed);
2679 		ret = wait_event_timeout(module_wq,
2680 				   xenbus_read_driver_state(dev->otherend) ==
2681 				   XenbusStateClosed ||
2682 				   xenbus_read_driver_state(dev->otherend) ==
2683 				   XenbusStateUnknown,
2684 				   XENNET_TIMEOUT);
2685 	} while (!ret);
2686 }
2687 
xennet_remove(struct xenbus_device * dev)2688 static int xennet_remove(struct xenbus_device *dev)
2689 {
2690 	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2691 
2692 	xennet_bus_close(dev);
2693 	xennet_disconnect_backend(info);
2694 
2695 	if (info->netdev->reg_state == NETREG_REGISTERED)
2696 		unregister_netdev(info->netdev);
2697 
2698 	if (info->queues) {
2699 		rtnl_lock();
2700 		xennet_destroy_queues(info);
2701 		rtnl_unlock();
2702 	}
2703 	xennet_free_netdev(info->netdev);
2704 
2705 	return 0;
2706 }
2707 
2708 static const struct xenbus_device_id netfront_ids[] = {
2709 	{ "vif" },
2710 	{ "" }
2711 };
2712 
2713 static struct xenbus_driver netfront_driver = {
2714 	.ids = netfront_ids,
2715 	.probe = netfront_probe,
2716 	.remove = xennet_remove,
2717 	.resume = netfront_resume,
2718 	.otherend_changed = netback_changed,
2719 };
2720 
netif_init(void)2721 static int __init netif_init(void)
2722 {
2723 	if (!xen_domain())
2724 		return -ENODEV;
2725 
2726 	if (!xen_has_pv_nic_devices())
2727 		return -ENODEV;
2728 
2729 	pr_info("Initialising Xen virtual ethernet driver\n");
2730 
2731 	/* Allow as many queues as there are CPUs inut max. 8 if user has not
2732 	 * specified a value.
2733 	 */
2734 	if (xennet_max_queues == 0)
2735 		xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2736 					  num_online_cpus());
2737 
2738 	return xenbus_register_frontend(&netfront_driver);
2739 }
2740 module_init(netif_init);
2741 
2742 
netif_exit(void)2743 static void __exit netif_exit(void)
2744 {
2745 	xenbus_unregister_driver(&netfront_driver);
2746 }
2747 module_exit(netif_exit);
2748 
2749 MODULE_DESCRIPTION("Xen virtual network device frontend");
2750 MODULE_LICENSE("GPL");
2751 MODULE_ALIAS("xen:vif");
2752 MODULE_ALIAS("xennet");
2753