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/types.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
  64static unsigned int xennet_max_queues;
  65module_param_named(max_queues, xennet_max_queues, uint, 0644);
  66MODULE_PARM_DESC(max_queues,
  67		 "Maximum number of queues per virtual interface");
  68
  69static bool __read_mostly xennet_trusted = true;
  70module_param_named(trusted, xennet_trusted, bool, 0644);
  71MODULE_PARM_DESC(trusted, "Is the backend trusted");
  72
  73#define XENNET_TIMEOUT  (5 * HZ)
  74
  75static const struct ethtool_ops xennet_ethtool_ops;
  76
  77struct 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 NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
  86#define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
  87
  88/* Minimum number of Rx slots (includes slot for GSO metadata). */
  89#define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
  90
  91/* Queue name is interface name with "-qNNN" appended */
  92#define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
  93
  94/* IRQ name is queue name with "-tx" or "-rx" appended */
  95#define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
  96
  97static DECLARE_WAIT_QUEUE_HEAD(module_wq);
  98
  99struct netfront_stats {
 100	u64			packets;
 101	u64			bytes;
 102	struct u64_stats_sync	syncp;
 103};
 104
 105struct netfront_info;
 106
 107struct netfront_queue {
 108	unsigned int id; /* Queue ID, 0-based */
 109	char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
 110	struct netfront_info *info;
 111
 112	struct bpf_prog __rcu *xdp_prog;
 113
 114	struct napi_struct napi;
 115
 116	/* Split event channels support, tx_* == rx_* when using
 117	 * single event channel.
 118	 */
 119	unsigned int tx_evtchn, rx_evtchn;
 120	unsigned int tx_irq, rx_irq;
 121	/* Only used when split event channels support is enabled */
 122	char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
 123	char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
 124
 125	spinlock_t   tx_lock;
 126	struct xen_netif_tx_front_ring tx;
 127	int tx_ring_ref;
 128
 129	/*
 130	 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
 131	 * are linked from tx_skb_freelist through tx_link.
 132	 */
 133	struct sk_buff *tx_skbs[NET_TX_RING_SIZE];
 134	unsigned short tx_link[NET_TX_RING_SIZE];
 135#define TX_LINK_NONE 0xffff
 136#define TX_PENDING   0xfffe
 137	grant_ref_t gref_tx_head;
 138	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
 139	struct page *grant_tx_page[NET_TX_RING_SIZE];
 140	unsigned tx_skb_freelist;
 141	unsigned int tx_pend_queue;
 142
 143	spinlock_t   rx_lock ____cacheline_aligned_in_smp;
 144	struct xen_netif_rx_front_ring rx;
 145	int rx_ring_ref;
 146
 147	struct timer_list rx_refill_timer;
 148
 149	struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
 150	grant_ref_t gref_rx_head;
 151	grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
 152
 153	unsigned int rx_rsp_unconsumed;
 154	spinlock_t rx_cons_lock;
 155
 156	struct page_pool *page_pool;
 157	struct xdp_rxq_info xdp_rxq;
 158};
 159
 160struct netfront_info {
 161	struct list_head list;
 162	struct net_device *netdev;
 163
 164	struct xenbus_device *xbdev;
 165
 166	/* Multi-queue support */
 167	struct netfront_queue *queues;
 168
 169	/* Statistics */
 170	struct netfront_stats __percpu *rx_stats;
 171	struct netfront_stats __percpu *tx_stats;
 172
 173	/* XDP state */
 174	bool netback_has_xdp_headroom;
 175	bool netfront_xdp_enabled;
 176
 177	/* Is device behaving sane? */
 178	bool broken;
 179
 180	/* Should skbs be bounced into a zeroed buffer? */
 181	bool bounce;
 182
 183	atomic_t rx_gso_checksum_fixup;
 184};
 185
 186struct netfront_rx_info {
 187	struct xen_netif_rx_response rx;
 188	struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
 189};
 190
 191/*
 192 * Access macros for acquiring freeing slots in tx_skbs[].
 193 */
 194
 195static void add_id_to_list(unsigned *head, unsigned short *list,
 196			   unsigned short id)
 197{
 198	list[id] = *head;
 199	*head = id;
 200}
 201
 202static unsigned short get_id_from_list(unsigned *head, unsigned short *list)
 203{
 204	unsigned int id = *head;
 205
 206	if (id != TX_LINK_NONE) {
 207		*head = list[id];
 208		list[id] = TX_LINK_NONE;
 209	}
 210	return id;
 211}
 212
 213static int xennet_rxidx(RING_IDX idx)
 214{
 215	return idx & (NET_RX_RING_SIZE - 1);
 216}
 217
 218static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
 219					 RING_IDX ri)
 220{
 221	int i = xennet_rxidx(ri);
 222	struct sk_buff *skb = queue->rx_skbs[i];
 223	queue->rx_skbs[i] = NULL;
 224	return skb;
 225}
 226
 227static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
 228					    RING_IDX ri)
 229{
 230	int i = xennet_rxidx(ri);
 231	grant_ref_t ref = queue->grant_rx_ref[i];
 232	queue->grant_rx_ref[i] = INVALID_GRANT_REF;
 233	return ref;
 234}
 235
 236#ifdef CONFIG_SYSFS
 237static const struct attribute_group xennet_dev_group;
 238#endif
 239
 240static bool xennet_can_sg(struct net_device *dev)
 241{
 242	return dev->features & NETIF_F_SG;
 243}
 244
 245
 246static void rx_refill_timeout(struct timer_list *t)
 247{
 248	struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer);
 249	napi_schedule(&queue->napi);
 250}
 251
 252static int netfront_tx_slot_available(struct netfront_queue *queue)
 253{
 254	return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
 255		(NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
 256}
 257
 258static void xennet_maybe_wake_tx(struct netfront_queue *queue)
 259{
 260	struct net_device *dev = queue->info->netdev;
 261	struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
 262
 263	if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
 264	    netfront_tx_slot_available(queue) &&
 265	    likely(netif_running(dev)))
 266		netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
 267}
 268
 269
 270static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
 271{
 272	struct sk_buff *skb;
 273	struct page *page;
 274
 275	skb = __netdev_alloc_skb(queue->info->netdev,
 276				 RX_COPY_THRESHOLD + NET_IP_ALIGN,
 277				 GFP_ATOMIC | __GFP_NOWARN);
 278	if (unlikely(!skb))
 279		return NULL;
 280
 281	page = page_pool_alloc_pages(queue->page_pool,
 282				     GFP_ATOMIC | __GFP_NOWARN | __GFP_ZERO);
 283	if (unlikely(!page)) {
 284		kfree_skb(skb);
 285		return NULL;
 286	}
 287	skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
 288
 289	/* Align ip header to a 16 bytes boundary */
 290	skb_reserve(skb, NET_IP_ALIGN);
 291	skb->dev = queue->info->netdev;
 292
 293	return skb;
 294}
 295
 296
 297static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
 298{
 299	RING_IDX req_prod = queue->rx.req_prod_pvt;
 300	int notify;
 301	int err = 0;
 302
 303	if (unlikely(!netif_carrier_ok(queue->info->netdev)))
 304		return;
 305
 306	for (req_prod = queue->rx.req_prod_pvt;
 307	     req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
 308	     req_prod++) {
 309		struct sk_buff *skb;
 310		unsigned short id;
 311		grant_ref_t ref;
 312		struct page *page;
 313		struct xen_netif_rx_request *req;
 314
 315		skb = xennet_alloc_one_rx_buffer(queue);
 316		if (!skb) {
 317			err = -ENOMEM;
 318			break;
 319		}
 320
 321		id = xennet_rxidx(req_prod);
 322
 323		BUG_ON(queue->rx_skbs[id]);
 324		queue->rx_skbs[id] = skb;
 325
 326		ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
 327		WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
 328		queue->grant_rx_ref[id] = ref;
 329
 330		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
 331
 332		req = RING_GET_REQUEST(&queue->rx, req_prod);
 333		gnttab_page_grant_foreign_access_ref_one(ref,
 334							 queue->info->xbdev->otherend_id,
 335							 page,
 336							 0);
 337		req->id = id;
 338		req->gref = ref;
 339	}
 340
 341	queue->rx.req_prod_pvt = req_prod;
 342
 343	/* Try again later if there are not enough requests or skb allocation
 344	 * failed.
 345	 * Enough requests is quantified as the sum of newly created slots and
 346	 * the unconsumed slots at the backend.
 347	 */
 348	if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
 349	    unlikely(err)) {
 350		mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
 351		return;
 352	}
 353
 354	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
 355	if (notify)
 356		notify_remote_via_irq(queue->rx_irq);
 357}
 358
 359static int xennet_open(struct net_device *dev)
 360{
 361	struct netfront_info *np = netdev_priv(dev);
 362	unsigned int num_queues = dev->real_num_tx_queues;
 363	unsigned int i = 0;
 364	struct netfront_queue *queue = NULL;
 365
 366	if (!np->queues || np->broken)
 367		return -ENODEV;
 368
 369	for (i = 0; i < num_queues; ++i) {
 370		queue = &np->queues[i];
 371		napi_enable(&queue->napi);
 372
 373		spin_lock_bh(&queue->rx_lock);
 374		if (netif_carrier_ok(dev)) {
 375			xennet_alloc_rx_buffers(queue);
 376			queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
 377			if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
 378				napi_schedule(&queue->napi);
 379		}
 380		spin_unlock_bh(&queue->rx_lock);
 381	}
 382
 383	netif_tx_start_all_queues(dev);
 384
 385	return 0;
 386}
 387
 388static bool xennet_tx_buf_gc(struct netfront_queue *queue)
 389{
 390	RING_IDX cons, prod;
 391	unsigned short id;
 392	struct sk_buff *skb;
 393	bool more_to_do;
 394	bool work_done = false;
 395	const struct device *dev = &queue->info->netdev->dev;
 396
 397	BUG_ON(!netif_carrier_ok(queue->info->netdev));
 398
 399	do {
 400		prod = queue->tx.sring->rsp_prod;
 401		if (RING_RESPONSE_PROD_OVERFLOW(&queue->tx, prod)) {
 402			dev_alert(dev, "Illegal number of responses %u\n",
 403				  prod - queue->tx.rsp_cons);
 404			goto err;
 405		}
 406		rmb(); /* Ensure we see responses up to 'rp'. */
 407
 408		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
 409			struct xen_netif_tx_response txrsp;
 410
 411			work_done = true;
 412
 413			RING_COPY_RESPONSE(&queue->tx, cons, &txrsp);
 414			if (txrsp.status == XEN_NETIF_RSP_NULL)
 415				continue;
 416
 417			id = txrsp.id;
 418			if (id >= RING_SIZE(&queue->tx)) {
 419				dev_alert(dev,
 420					  "Response has incorrect id (%u)\n",
 421					  id);
 422				goto err;
 423			}
 424			if (queue->tx_link[id] != TX_PENDING) {
 425				dev_alert(dev,
 426					  "Response for inactive request\n");
 427				goto err;
 428			}
 429
 430			queue->tx_link[id] = TX_LINK_NONE;
 431			skb = queue->tx_skbs[id];
 432			queue->tx_skbs[id] = NULL;
 433			if (unlikely(!gnttab_end_foreign_access_ref(
 434				queue->grant_tx_ref[id]))) {
 435				dev_alert(dev,
 436					  "Grant still in use by backend domain\n");
 437				goto err;
 438			}
 439			gnttab_release_grant_reference(
 440				&queue->gref_tx_head, queue->grant_tx_ref[id]);
 441			queue->grant_tx_ref[id] = INVALID_GRANT_REF;
 442			queue->grant_tx_page[id] = NULL;
 443			add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, id);
 444			dev_kfree_skb_irq(skb);
 445		}
 446
 447		queue->tx.rsp_cons = prod;
 448
 449		RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
 450	} while (more_to_do);
 451
 452	xennet_maybe_wake_tx(queue);
 453
 454	return work_done;
 455
 456 err:
 457	queue->info->broken = true;
 458	dev_alert(dev, "Disabled for further use\n");
 459
 460	return work_done;
 461}
 462
 463struct xennet_gnttab_make_txreq {
 464	struct netfront_queue *queue;
 465	struct sk_buff *skb;
 466	struct page *page;
 467	struct xen_netif_tx_request *tx;      /* Last request on ring page */
 468	struct xen_netif_tx_request tx_local; /* Last request local copy*/
 469	unsigned int size;
 470};
 471
 472static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
 473				  unsigned int len, void *data)
 474{
 475	struct xennet_gnttab_make_txreq *info = data;
 476	unsigned int id;
 477	struct xen_netif_tx_request *tx;
 478	grant_ref_t ref;
 479	/* convenient aliases */
 480	struct page *page = info->page;
 481	struct netfront_queue *queue = info->queue;
 482	struct sk_buff *skb = info->skb;
 483
 484	id = get_id_from_list(&queue->tx_skb_freelist, queue->tx_link);
 485	tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
 486	ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
 487	WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
 488
 489	gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
 490					gfn, GNTMAP_readonly);
 491
 492	queue->tx_skbs[id] = skb;
 493	queue->grant_tx_page[id] = page;
 494	queue->grant_tx_ref[id] = ref;
 495
 496	info->tx_local.id = id;
 497	info->tx_local.gref = ref;
 498	info->tx_local.offset = offset;
 499	info->tx_local.size = len;
 500	info->tx_local.flags = 0;
 501
 502	*tx = info->tx_local;
 503
 504	/*
 505	 * Put the request in the pending queue, it will be set to be pending
 506	 * when the producer index is about to be raised.
 507	 */
 508	add_id_to_list(&queue->tx_pend_queue, queue->tx_link, id);
 509
 510	info->tx = tx;
 511	info->size += info->tx_local.size;
 512}
 513
 514static struct xen_netif_tx_request *xennet_make_first_txreq(
 515	struct xennet_gnttab_make_txreq *info,
 516	unsigned int offset, unsigned int len)
 517{
 518	info->size = 0;
 519
 520	gnttab_for_one_grant(info->page, offset, len, xennet_tx_setup_grant, info);
 521
 522	return info->tx;
 523}
 524
 525static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
 526				  unsigned int len, void *data)
 527{
 528	struct xennet_gnttab_make_txreq *info = data;
 529
 530	info->tx->flags |= XEN_NETTXF_more_data;
 531	skb_get(info->skb);
 532	xennet_tx_setup_grant(gfn, offset, len, data);
 533}
 534
 535static void xennet_make_txreqs(
 536	struct xennet_gnttab_make_txreq *info,
 537	struct page *page,
 538	unsigned int offset, unsigned int len)
 539{
 540	/* Skip unused frames from start of page */
 541	page += offset >> PAGE_SHIFT;
 542	offset &= ~PAGE_MASK;
 543
 544	while (len) {
 545		info->page = page;
 546		info->size = 0;
 547
 548		gnttab_foreach_grant_in_range(page, offset, len,
 549					      xennet_make_one_txreq,
 550					      info);
 551
 552		page++;
 553		offset = 0;
 554		len -= info->size;
 555	}
 556}
 557
 558/*
 559 * Count how many ring slots are required to send this skb. Each frag
 560 * might be a compound page.
 561 */
 562static int xennet_count_skb_slots(struct sk_buff *skb)
 563{
 564	int i, frags = skb_shinfo(skb)->nr_frags;
 565	int slots;
 566
 567	slots = gnttab_count_grant(offset_in_page(skb->data),
 568				   skb_headlen(skb));
 569
 570	for (i = 0; i < frags; i++) {
 571		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 572		unsigned long size = skb_frag_size(frag);
 573		unsigned long offset = skb_frag_off(frag);
 574
 575		/* Skip unused frames from start of page */
 576		offset &= ~PAGE_MASK;
 577
 578		slots += gnttab_count_grant(offset, size);
 579	}
 580
 581	return slots;
 582}
 583
 584static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
 585			       struct net_device *sb_dev)
 586{
 587	unsigned int num_queues = dev->real_num_tx_queues;
 588	u32 hash;
 589	u16 queue_idx;
 590
 591	/* First, check if there is only one queue */
 592	if (num_queues == 1) {
 593		queue_idx = 0;
 594	} else {
 595		hash = skb_get_hash(skb);
 596		queue_idx = hash % num_queues;
 597	}
 598
 599	return queue_idx;
 600}
 601
 602static void xennet_mark_tx_pending(struct netfront_queue *queue)
 603{
 604	unsigned int i;
 605
 606	while ((i = get_id_from_list(&queue->tx_pend_queue, queue->tx_link)) !=
 607	       TX_LINK_NONE)
 608		queue->tx_link[i] = TX_PENDING;
 609}
 610
 611static int xennet_xdp_xmit_one(struct net_device *dev,
 612			       struct netfront_queue *queue,
 613			       struct xdp_frame *xdpf)
 614{
 615	struct netfront_info *np = netdev_priv(dev);
 616	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
 617	struct xennet_gnttab_make_txreq info = {
 618		.queue = queue,
 619		.skb = NULL,
 620		.page = virt_to_page(xdpf->data),
 621	};
 622	int notify;
 623
 624	xennet_make_first_txreq(&info,
 625				offset_in_page(xdpf->data),
 626				xdpf->len);
 627
 628	xennet_mark_tx_pending(queue);
 629
 630	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
 631	if (notify)
 632		notify_remote_via_irq(queue->tx_irq);
 633
 634	u64_stats_update_begin(&tx_stats->syncp);
 635	tx_stats->bytes += xdpf->len;
 636	tx_stats->packets++;
 637	u64_stats_update_end(&tx_stats->syncp);
 638
 639	xennet_tx_buf_gc(queue);
 640
 641	return 0;
 642}
 643
 644static int xennet_xdp_xmit(struct net_device *dev, int n,
 645			   struct xdp_frame **frames, u32 flags)
 646{
 647	unsigned int num_queues = dev->real_num_tx_queues;
 648	struct netfront_info *np = netdev_priv(dev);
 649	struct netfront_queue *queue = NULL;
 650	unsigned long irq_flags;
 651	int nxmit = 0;
 652	int i;
 653
 654	if (unlikely(np->broken))
 655		return -ENODEV;
 656	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
 657		return -EINVAL;
 658
 659	queue = &np->queues[smp_processor_id() % num_queues];
 660
 661	spin_lock_irqsave(&queue->tx_lock, irq_flags);
 662	for (i = 0; i < n; i++) {
 663		struct xdp_frame *xdpf = frames[i];
 664
 665		if (!xdpf)
 666			continue;
 667		if (xennet_xdp_xmit_one(dev, queue, xdpf))
 668			break;
 669		nxmit++;
 670	}
 671	spin_unlock_irqrestore(&queue->tx_lock, irq_flags);
 672
 673	return nxmit;
 674}
 675
 676static struct sk_buff *bounce_skb(const struct sk_buff *skb)
 677{
 678	unsigned int headerlen = skb_headroom(skb);
 679	/* Align size to allocate full pages and avoid contiguous data leaks */
 680	unsigned int size = ALIGN(skb_end_offset(skb) + skb->data_len,
 681				  XEN_PAGE_SIZE);
 682	struct sk_buff *n = alloc_skb(size, GFP_ATOMIC | __GFP_ZERO);
 683
 684	if (!n)
 685		return NULL;
 686
 687	if (!IS_ALIGNED((uintptr_t)n->head, XEN_PAGE_SIZE)) {
 688		WARN_ONCE(1, "misaligned skb allocated\n");
 689		kfree_skb(n);
 690		return NULL;
 691	}
 692
 693	/* Set the data pointer */
 694	skb_reserve(n, headerlen);
 695	/* Set the tail pointer and length */
 696	skb_put(n, skb->len);
 697
 698	BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
 699
 700	skb_copy_header(n, skb);
 701	return n;
 702}
 703
 704#define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
 705
 706static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
 707{
 708	struct netfront_info *np = netdev_priv(dev);
 709	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
 710	struct xen_netif_tx_request *first_tx;
 711	unsigned int i;
 712	int notify;
 713	int slots;
 714	struct page *page;
 715	unsigned int offset;
 716	unsigned int len;
 717	unsigned long flags;
 718	struct netfront_queue *queue = NULL;
 719	struct xennet_gnttab_make_txreq info = { };
 720	unsigned int num_queues = dev->real_num_tx_queues;
 721	u16 queue_index;
 722	struct sk_buff *nskb;
 723
 724	/* Drop the packet if no queues are set up */
 725	if (num_queues < 1)
 726		goto drop;
 727	if (unlikely(np->broken))
 728		goto drop;
 729	/* Determine which queue to transmit this SKB on */
 730	queue_index = skb_get_queue_mapping(skb);
 731	queue = &np->queues[queue_index];
 732
 733	/* If skb->len is too big for wire format, drop skb and alert
 734	 * user about misconfiguration.
 735	 */
 736	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
 737		net_alert_ratelimited(
 738			"xennet: skb->len = %u, too big for wire format\n",
 739			skb->len);
 740		goto drop;
 741	}
 742
 743	slots = xennet_count_skb_slots(skb);
 744	if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
 745		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
 746				    slots, skb->len);
 747		if (skb_linearize(skb))
 748			goto drop;
 749	}
 750
 751	page = virt_to_page(skb->data);
 752	offset = offset_in_page(skb->data);
 753
 754	/* The first req should be at least ETH_HLEN size or the packet will be
 755	 * dropped by netback.
 756	 *
 757	 * If the backend is not trusted bounce all data to zeroed pages to
 758	 * avoid exposing contiguous data on the granted page not belonging to
 759	 * the skb.
 760	 */
 761	if (np->bounce || unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
 762		nskb = bounce_skb(skb);
 763		if (!nskb)
 764			goto drop;
 765		dev_consume_skb_any(skb);
 766		skb = nskb;
 767		page = virt_to_page(skb->data);
 768		offset = offset_in_page(skb->data);
 769	}
 770
 771	len = skb_headlen(skb);
 772
 773	spin_lock_irqsave(&queue->tx_lock, flags);
 774
 775	if (unlikely(!netif_carrier_ok(dev) ||
 776		     (slots > 1 && !xennet_can_sg(dev)) ||
 777		     netif_needs_gso(skb, netif_skb_features(skb)))) {
 778		spin_unlock_irqrestore(&queue->tx_lock, flags);
 779		goto drop;
 780	}
 781
 782	/* First request for the linear area. */
 783	info.queue = queue;
 784	info.skb = skb;
 785	info.page = page;
 786	first_tx = xennet_make_first_txreq(&info, offset, len);
 787	offset += info.tx_local.size;
 788	if (offset == PAGE_SIZE) {
 789		page++;
 790		offset = 0;
 791	}
 792	len -= info.tx_local.size;
 793
 794	if (skb->ip_summed == CHECKSUM_PARTIAL)
 795		/* local packet? */
 796		first_tx->flags |= XEN_NETTXF_csum_blank |
 797				   XEN_NETTXF_data_validated;
 798	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
 799		/* remote but checksummed. */
 800		first_tx->flags |= XEN_NETTXF_data_validated;
 801
 802	/* Optional extra info after the first request. */
 803	if (skb_shinfo(skb)->gso_size) {
 804		struct xen_netif_extra_info *gso;
 805
 806		gso = (struct xen_netif_extra_info *)
 807			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
 808
 809		first_tx->flags |= XEN_NETTXF_extra_info;
 810
 811		gso->u.gso.size = skb_shinfo(skb)->gso_size;
 812		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
 813			XEN_NETIF_GSO_TYPE_TCPV6 :
 814			XEN_NETIF_GSO_TYPE_TCPV4;
 815		gso->u.gso.pad = 0;
 816		gso->u.gso.features = 0;
 817
 818		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
 819		gso->flags = 0;
 820	}
 821
 822	/* Requests for the rest of the linear area. */
 823	xennet_make_txreqs(&info, page, offset, len);
 824
 825	/* Requests for all the frags. */
 826	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 827		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 828		xennet_make_txreqs(&info, skb_frag_page(frag),
 829					skb_frag_off(frag),
 830					skb_frag_size(frag));
 831	}
 832
 833	/* First request has the packet length. */
 834	first_tx->size = skb->len;
 835
 836	/* timestamp packet in software */
 837	skb_tx_timestamp(skb);
 838
 839	xennet_mark_tx_pending(queue);
 840
 841	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
 842	if (notify)
 843		notify_remote_via_irq(queue->tx_irq);
 844
 845	u64_stats_update_begin(&tx_stats->syncp);
 846	tx_stats->bytes += skb->len;
 847	tx_stats->packets++;
 848	u64_stats_update_end(&tx_stats->syncp);
 849
 850	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
 851	xennet_tx_buf_gc(queue);
 852
 853	if (!netfront_tx_slot_available(queue))
 854		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
 855
 856	spin_unlock_irqrestore(&queue->tx_lock, flags);
 857
 858	return NETDEV_TX_OK;
 859
 860 drop:
 861	dev->stats.tx_dropped++;
 862	dev_kfree_skb_any(skb);
 863	return NETDEV_TX_OK;
 864}
 865
 866static int xennet_close(struct net_device *dev)
 867{
 868	struct netfront_info *np = netdev_priv(dev);
 869	unsigned int num_queues = dev->real_num_tx_queues;
 870	unsigned int i;
 871	struct netfront_queue *queue;
 872	netif_tx_stop_all_queues(np->netdev);
 873	for (i = 0; i < num_queues; ++i) {
 874		queue = &np->queues[i];
 875		napi_disable(&queue->napi);
 876	}
 877	return 0;
 878}
 879
 880static void xennet_destroy_queues(struct netfront_info *info)
 881{
 882	unsigned int i;
 883
 884	for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
 885		struct netfront_queue *queue = &info->queues[i];
 886
 887		if (netif_running(info->netdev))
 888			napi_disable(&queue->napi);
 889		netif_napi_del(&queue->napi);
 890	}
 891
 892	kfree(info->queues);
 893	info->queues = NULL;
 894}
 895
 896static void xennet_uninit(struct net_device *dev)
 897{
 898	struct netfront_info *np = netdev_priv(dev);
 899	xennet_destroy_queues(np);
 900}
 901
 902static void xennet_set_rx_rsp_cons(struct netfront_queue *queue, RING_IDX val)
 903{
 904	unsigned long flags;
 905
 906	spin_lock_irqsave(&queue->rx_cons_lock, flags);
 907	queue->rx.rsp_cons = val;
 908	queue->rx_rsp_unconsumed = XEN_RING_NR_UNCONSUMED_RESPONSES(&queue->rx);
 909	spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
 910}
 911
 912static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
 913				grant_ref_t ref)
 914{
 915	int new = xennet_rxidx(queue->rx.req_prod_pvt);
 916
 917	BUG_ON(queue->rx_skbs[new]);
 918	queue->rx_skbs[new] = skb;
 919	queue->grant_rx_ref[new] = ref;
 920	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
 921	RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
 922	queue->rx.req_prod_pvt++;
 923}
 924
 925static int xennet_get_extras(struct netfront_queue *queue,
 926			     struct xen_netif_extra_info *extras,
 927			     RING_IDX rp)
 928
 929{
 930	struct xen_netif_extra_info extra;
 931	struct device *dev = &queue->info->netdev->dev;
 932	RING_IDX cons = queue->rx.rsp_cons;
 933	int err = 0;
 934
 935	do {
 936		struct sk_buff *skb;
 937		grant_ref_t ref;
 938
 939		if (unlikely(cons + 1 == rp)) {
 940			if (net_ratelimit())
 941				dev_warn(dev, "Missing extra info\n");
 942			err = -EBADR;
 943			break;
 944		}
 945
 946		RING_COPY_RESPONSE(&queue->rx, ++cons, &extra);
 947
 948		if (unlikely(!extra.type ||
 949			     extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
 950			if (net_ratelimit())
 951				dev_warn(dev, "Invalid extra type: %d\n",
 952					 extra.type);
 953			err = -EINVAL;
 954		} else {
 955			extras[extra.type - 1] = extra;
 956		}
 957
 958		skb = xennet_get_rx_skb(queue, cons);
 959		ref = xennet_get_rx_ref(queue, cons);
 960		xennet_move_rx_slot(queue, skb, ref);
 961	} while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
 962
 963	xennet_set_rx_rsp_cons(queue, cons);
 964	return err;
 965}
 966
 967static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
 968		   struct xen_netif_rx_response *rx, struct bpf_prog *prog,
 969		   struct xdp_buff *xdp, bool *need_xdp_flush)
 970{
 971	struct xdp_frame *xdpf;
 972	u32 len = rx->status;
 973	u32 act;
 974	int err;
 975
 976	xdp_init_buff(xdp, XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
 977		      &queue->xdp_rxq);
 978	xdp_prepare_buff(xdp, page_address(pdata), XDP_PACKET_HEADROOM,
 979			 len, false);
 980
 981	act = bpf_prog_run_xdp(prog, xdp);
 982	switch (act) {
 983	case XDP_TX:
 984		get_page(pdata);
 985		xdpf = xdp_convert_buff_to_frame(xdp);
 986		err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0);
 987		if (unlikely(!err))
 988			xdp_return_frame_rx_napi(xdpf);
 989		else if (unlikely(err < 0))
 990			trace_xdp_exception(queue->info->netdev, prog, act);
 991		break;
 992	case XDP_REDIRECT:
 993		get_page(pdata);
 994		err = xdp_do_redirect(queue->info->netdev, xdp, prog);
 995		*need_xdp_flush = true;
 996		if (unlikely(err))
 997			trace_xdp_exception(queue->info->netdev, prog, act);
 998		break;
 999	case XDP_PASS:
1000	case XDP_DROP:
1001		break;
1002
1003	case XDP_ABORTED:
1004		trace_xdp_exception(queue->info->netdev, prog, act);
1005		break;
1006
1007	default:
1008		bpf_warn_invalid_xdp_action(queue->info->netdev, prog, act);
1009	}
1010
1011	return act;
1012}
1013
1014static int xennet_get_responses(struct netfront_queue *queue,
1015				struct netfront_rx_info *rinfo, RING_IDX rp,
1016				struct sk_buff_head *list,
1017				bool *need_xdp_flush)
1018{
1019	struct xen_netif_rx_response *rx = &rinfo->rx, rx_local;
1020	int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
1021	RING_IDX cons = queue->rx.rsp_cons;
1022	struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
1023	struct xen_netif_extra_info *extras = rinfo->extras;
1024	grant_ref_t ref = xennet_get_rx_ref(queue, cons);
1025	struct device *dev = &queue->info->netdev->dev;
1026	struct bpf_prog *xdp_prog;
1027	struct xdp_buff xdp;
1028	int slots = 1;
1029	int err = 0;
1030	u32 verdict;
1031
1032	if (rx->flags & XEN_NETRXF_extra_info) {
1033		err = xennet_get_extras(queue, extras, rp);
1034		if (!err) {
1035			if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) {
1036				struct xen_netif_extra_info *xdp;
1037
1038				xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1];
1039				rx->offset = xdp->u.xdp.headroom;
1040			}
1041		}
1042		cons = queue->rx.rsp_cons;
1043	}
1044
1045	for (;;) {
1046		/*
1047		 * This definitely indicates a bug, either in this driver or in
1048		 * the backend driver. In future this should flag the bad
1049		 * situation to the system controller to reboot the backend.
1050		 */
1051		if (ref == INVALID_GRANT_REF) {
1052			if (net_ratelimit())
1053				dev_warn(dev, "Bad rx response id %d.\n",
1054					 rx->id);
1055			err = -EINVAL;
1056			goto next;
1057		}
1058
1059		if (unlikely(rx->status < 0 ||
1060			     rx->offset + rx->status > XEN_PAGE_SIZE)) {
1061			if (net_ratelimit())
1062				dev_warn(dev, "rx->offset: %u, size: %d\n",
1063					 rx->offset, rx->status);
1064			xennet_move_rx_slot(queue, skb, ref);
1065			err = -EINVAL;
1066			goto next;
1067		}
1068
1069		if (!gnttab_end_foreign_access_ref(ref)) {
1070			dev_alert(dev,
1071				  "Grant still in use by backend domain\n");
1072			queue->info->broken = true;
1073			dev_alert(dev, "Disabled for further use\n");
1074			return -EINVAL;
1075		}
1076
1077		gnttab_release_grant_reference(&queue->gref_rx_head, ref);
1078
1079		rcu_read_lock();
1080		xdp_prog = rcu_dereference(queue->xdp_prog);
1081		if (xdp_prog) {
1082			if (!(rx->flags & XEN_NETRXF_more_data)) {
1083				/* currently only a single page contains data */
1084				verdict = xennet_run_xdp(queue,
1085							 skb_frag_page(&skb_shinfo(skb)->frags[0]),
1086							 rx, xdp_prog, &xdp, need_xdp_flush);
1087				if (verdict != XDP_PASS)
1088					err = -EINVAL;
1089			} else {
1090				/* drop the frame */
1091				err = -EINVAL;
1092			}
1093		}
1094		rcu_read_unlock();
1095
1096		__skb_queue_tail(list, skb);
1097
1098next:
1099		if (!(rx->flags & XEN_NETRXF_more_data))
1100			break;
1101
1102		if (cons + slots == rp) {
1103			if (net_ratelimit())
1104				dev_warn(dev, "Need more slots\n");
1105			err = -ENOENT;
1106			break;
1107		}
1108
1109		RING_COPY_RESPONSE(&queue->rx, cons + slots, &rx_local);
1110		rx = &rx_local;
1111		skb = xennet_get_rx_skb(queue, cons + slots);
1112		ref = xennet_get_rx_ref(queue, cons + slots);
1113		slots++;
1114	}
1115
1116	if (unlikely(slots > max)) {
1117		if (net_ratelimit())
1118			dev_warn(dev, "Too many slots\n");
1119		err = -E2BIG;
1120	}
1121
1122	if (unlikely(err))
1123		xennet_set_rx_rsp_cons(queue, cons + slots);
1124
1125	return err;
1126}
1127
1128static int xennet_set_skb_gso(struct sk_buff *skb,
1129			      struct xen_netif_extra_info *gso)
1130{
1131	if (!gso->u.gso.size) {
1132		if (net_ratelimit())
1133			pr_warn("GSO size must not be zero\n");
1134		return -EINVAL;
1135	}
1136
1137	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
1138	    gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
1139		if (net_ratelimit())
1140			pr_warn("Bad GSO type %d\n", gso->u.gso.type);
1141		return -EINVAL;
1142	}
1143
1144	skb_shinfo(skb)->gso_size = gso->u.gso.size;
1145	skb_shinfo(skb)->gso_type =
1146		(gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
1147		SKB_GSO_TCPV4 :
1148		SKB_GSO_TCPV6;
1149
1150	/* Header must be checked, and gso_segs computed. */
1151	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1152	skb_shinfo(skb)->gso_segs = 0;
1153
1154	return 0;
1155}
1156
1157static int xennet_fill_frags(struct netfront_queue *queue,
1158			     struct sk_buff *skb,
1159			     struct sk_buff_head *list)
1160{
1161	RING_IDX cons = queue->rx.rsp_cons;
1162	struct sk_buff *nskb;
1163
1164	while ((nskb = __skb_dequeue(list))) {
1165		struct xen_netif_rx_response rx;
1166		skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
1167
1168		RING_COPY_RESPONSE(&queue->rx, ++cons, &rx);
1169
1170		if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
1171			unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1172
1173			BUG_ON(pull_to < skb_headlen(skb));
1174			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1175		}
1176		if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
1177			xennet_set_rx_rsp_cons(queue,
1178					       ++cons + skb_queue_len(list));
1179			kfree_skb(nskb);
1180			return -ENOENT;
1181		}
1182
1183		skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
1184				skb_frag_page(nfrag),
1185				rx.offset, rx.status, PAGE_SIZE);
1186
1187		skb_shinfo(nskb)->nr_frags = 0;
1188		kfree_skb(nskb);
1189	}
1190
1191	xennet_set_rx_rsp_cons(queue, cons);
1192
1193	return 0;
1194}
1195
1196static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
1197{
1198	bool recalculate_partial_csum = false;
1199
1200	/*
1201	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1202	 * peers can fail to set NETRXF_csum_blank when sending a GSO
1203	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1204	 * recalculate the partial checksum.
1205	 */
1206	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1207		struct netfront_info *np = netdev_priv(dev);
1208		atomic_inc(&np->rx_gso_checksum_fixup);
1209		skb->ip_summed = CHECKSUM_PARTIAL;
1210		recalculate_partial_csum = true;
1211	}
1212
1213	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1214	if (skb->ip_summed != CHECKSUM_PARTIAL)
1215		return 0;
1216
1217	return skb_checksum_setup(skb, recalculate_partial_csum);
1218}
1219
1220static int handle_incoming_queue(struct netfront_queue *queue,
1221				 struct sk_buff_head *rxq)
1222{
1223	struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
1224	int packets_dropped = 0;
1225	struct sk_buff *skb;
1226
1227	while ((skb = __skb_dequeue(rxq)) != NULL) {
1228		int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1229
1230		if (pull_to > skb_headlen(skb))
1231			__pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1232
1233		/* Ethernet work: Delayed to here as it peeks the header. */
1234		skb->protocol = eth_type_trans(skb, queue->info->netdev);
1235		skb_reset_network_header(skb);
1236
1237		if (checksum_setup(queue->info->netdev, skb)) {
1238			kfree_skb(skb);
1239			packets_dropped++;
1240			queue->info->netdev->stats.rx_errors++;
1241			continue;
1242		}
1243
1244		u64_stats_update_begin(&rx_stats->syncp);
1245		rx_stats->packets++;
1246		rx_stats->bytes += skb->len;
1247		u64_stats_update_end(&rx_stats->syncp);
1248
1249		/* Pass it up. */
1250		napi_gro_receive(&queue->napi, skb);
1251	}
1252
1253	return packets_dropped;
1254}
1255
1256static int xennet_poll(struct napi_struct *napi, int budget)
1257{
1258	struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
1259	struct net_device *dev = queue->info->netdev;
1260	struct sk_buff *skb;
1261	struct netfront_rx_info rinfo;
1262	struct xen_netif_rx_response *rx = &rinfo.rx;
1263	struct xen_netif_extra_info *extras = rinfo.extras;
1264	RING_IDX i, rp;
1265	int work_done;
1266	struct sk_buff_head rxq;
1267	struct sk_buff_head errq;
1268	struct sk_buff_head tmpq;
1269	int err;
1270	bool need_xdp_flush = false;
1271
1272	spin_lock(&queue->rx_lock);
1273
1274	skb_queue_head_init(&rxq);
1275	skb_queue_head_init(&errq);
1276	skb_queue_head_init(&tmpq);
1277
1278	rp = queue->rx.sring->rsp_prod;
1279	if (RING_RESPONSE_PROD_OVERFLOW(&queue->rx, rp)) {
1280		dev_alert(&dev->dev, "Illegal number of responses %u\n",
1281			  rp - queue->rx.rsp_cons);
1282		queue->info->broken = true;
1283		spin_unlock(&queue->rx_lock);
1284		return 0;
1285	}
1286	rmb(); /* Ensure we see queued responses up to 'rp'. */
1287
1288	i = queue->rx.rsp_cons;
1289	work_done = 0;
1290	while ((i != rp) && (work_done < budget)) {
1291		RING_COPY_RESPONSE(&queue->rx, i, rx);
1292		memset(extras, 0, sizeof(rinfo.extras));
1293
1294		err = xennet_get_responses(queue, &rinfo, rp, &tmpq,
1295					   &need_xdp_flush);
1296
1297		if (unlikely(err)) {
1298			if (queue->info->broken) {
1299				spin_unlock(&queue->rx_lock);
1300				return 0;
1301			}
1302err:
1303			while ((skb = __skb_dequeue(&tmpq)))
1304				__skb_queue_tail(&errq, skb);
1305			dev->stats.rx_errors++;
1306			i = queue->rx.rsp_cons;
1307			continue;
1308		}
1309
1310		skb = __skb_dequeue(&tmpq);
1311
1312		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1313			struct xen_netif_extra_info *gso;
1314			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1315
1316			if (unlikely(xennet_set_skb_gso(skb, gso))) {
1317				__skb_queue_head(&tmpq, skb);
1318				xennet_set_rx_rsp_cons(queue,
1319						       queue->rx.rsp_cons +
1320						       skb_queue_len(&tmpq));
1321				goto err;
1322			}
1323		}
1324
1325		NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1326		if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1327			NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1328
1329		skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset);
1330		skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1331		skb->data_len = rx->status;
1332		skb->len += rx->status;
1333
1334		if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
1335			goto err;
1336
1337		if (rx->flags & XEN_NETRXF_csum_blank)
1338			skb->ip_summed = CHECKSUM_PARTIAL;
1339		else if (rx->flags & XEN_NETRXF_data_validated)
1340			skb->ip_summed = CHECKSUM_UNNECESSARY;
1341
1342		__skb_queue_tail(&rxq, skb);
1343
1344		i = queue->rx.rsp_cons + 1;
1345		xennet_set_rx_rsp_cons(queue, i);
1346		work_done++;
1347	}
1348	if (need_xdp_flush)
1349		xdp_do_flush();
1350
1351	__skb_queue_purge(&errq);
1352
1353	work_done -= handle_incoming_queue(queue, &rxq);
1354
1355	xennet_alloc_rx_buffers(queue);
1356
1357	if (work_done < budget) {
1358		int more_to_do = 0;
1359
1360		napi_complete_done(napi, work_done);
1361
1362		RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1363		if (more_to_do)
1364			napi_schedule(napi);
1365	}
1366
1367	spin_unlock(&queue->rx_lock);
1368
1369	return work_done;
1370}
1371
1372static int xennet_change_mtu(struct net_device *dev, int mtu)
1373{
1374	int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1375
1376	if (mtu > max)
1377		return -EINVAL;
1378	dev->mtu = mtu;
1379	return 0;
1380}
1381
1382static void xennet_get_stats64(struct net_device *dev,
1383			       struct rtnl_link_stats64 *tot)
1384{
1385	struct netfront_info *np = netdev_priv(dev);
1386	int cpu;
1387
1388	for_each_possible_cpu(cpu) {
1389		struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1390		struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1391		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1392		unsigned int start;
1393
1394		do {
1395			start = u64_stats_fetch_begin(&tx_stats->syncp);
1396			tx_packets = tx_stats->packets;
1397			tx_bytes = tx_stats->bytes;
1398		} while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1399
1400		do {
1401			start = u64_stats_fetch_begin(&rx_stats->syncp);
1402			rx_packets = rx_stats->packets;
1403			rx_bytes = rx_stats->bytes;
1404		} while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1405
1406		tot->rx_packets += rx_packets;
1407		tot->tx_packets += tx_packets;
1408		tot->rx_bytes   += rx_bytes;
1409		tot->tx_bytes   += tx_bytes;
1410	}
1411
1412	tot->rx_errors  = dev->stats.rx_errors;
1413	tot->tx_dropped = dev->stats.tx_dropped;
1414}
1415
1416static void xennet_release_tx_bufs(struct netfront_queue *queue)
1417{
1418	struct sk_buff *skb;
1419	int i;
1420
1421	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1422		/* Skip over entries which are actually freelist references */
1423		if (!queue->tx_skbs[i])
1424			continue;
1425
1426		skb = queue->tx_skbs[i];
1427		queue->tx_skbs[i] = NULL;
1428		get_page(queue->grant_tx_page[i]);
1429		gnttab_end_foreign_access(queue->grant_tx_ref[i],
1430					  queue->grant_tx_page[i]);
1431		queue->grant_tx_page[i] = NULL;
1432		queue->grant_tx_ref[i] = INVALID_GRANT_REF;
1433		add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, i);
1434		dev_kfree_skb_irq(skb);
1435	}
1436}
1437
1438static void xennet_release_rx_bufs(struct netfront_queue *queue)
1439{
1440	int id, ref;
1441
1442	spin_lock_bh(&queue->rx_lock);
1443
1444	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1445		struct sk_buff *skb;
1446		struct page *page;
1447
1448		skb = queue->rx_skbs[id];
1449		if (!skb)
1450			continue;
1451
1452		ref = queue->grant_rx_ref[id];
1453		if (ref == INVALID_GRANT_REF)
1454			continue;
1455
1456		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1457
1458		/* gnttab_end_foreign_access() needs a page ref until
1459		 * foreign access is ended (which may be deferred).
1460		 */
1461		get_page(page);
1462		gnttab_end_foreign_access(ref, page);
1463		queue->grant_rx_ref[id] = INVALID_GRANT_REF;
1464
1465		kfree_skb(skb);
1466	}
1467
1468	spin_unlock_bh(&queue->rx_lock);
1469}
1470
1471static netdev_features_t xennet_fix_features(struct net_device *dev,
1472	netdev_features_t features)
1473{
1474	struct netfront_info *np = netdev_priv(dev);
1475
1476	if (features & NETIF_F_SG &&
1477	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1478		features &= ~NETIF_F_SG;
1479
1480	if (features & NETIF_F_IPV6_CSUM &&
1481	    !xenbus_read_unsigned(np->xbdev->otherend,
1482				  "feature-ipv6-csum-offload", 0))
1483		features &= ~NETIF_F_IPV6_CSUM;
1484
1485	if (features & NETIF_F_TSO &&
1486	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1487		features &= ~NETIF_F_TSO;
1488
1489	if (features & NETIF_F_TSO6 &&
1490	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1491		features &= ~NETIF_F_TSO6;
1492
1493	return features;
1494}
1495
1496static int xennet_set_features(struct net_device *dev,
1497	netdev_features_t features)
1498{
1499	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1500		netdev_info(dev, "Reducing MTU because no SG offload");
1501		dev->mtu = ETH_DATA_LEN;
1502	}
1503
1504	return 0;
1505}
1506
1507static bool xennet_handle_tx(struct netfront_queue *queue, unsigned int *eoi)
1508{
1509	unsigned long flags;
1510
1511	if (unlikely(queue->info->broken))
1512		return false;
1513
1514	spin_lock_irqsave(&queue->tx_lock, flags);
1515	if (xennet_tx_buf_gc(queue))
1516		*eoi = 0;
1517	spin_unlock_irqrestore(&queue->tx_lock, flags);
1518
1519	return true;
1520}
1521
1522static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1523{
1524	unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1525
1526	if (likely(xennet_handle_tx(dev_id, &eoiflag)))
1527		xen_irq_lateeoi(irq, eoiflag);
1528
1529	return IRQ_HANDLED;
1530}
1531
1532static bool xennet_handle_rx(struct netfront_queue *queue, unsigned int *eoi)
1533{
1534	unsigned int work_queued;
1535	unsigned long flags;
1536
1537	if (unlikely(queue->info->broken))
1538		return false;
1539
1540	spin_lock_irqsave(&queue->rx_cons_lock, flags);
1541	work_queued = XEN_RING_NR_UNCONSUMED_RESPONSES(&queue->rx);
1542	if (work_queued > queue->rx_rsp_unconsumed) {
1543		queue->rx_rsp_unconsumed = work_queued;
1544		*eoi = 0;
1545	} else if (unlikely(work_queued < queue->rx_rsp_unconsumed)) {
1546		const struct device *dev = &queue->info->netdev->dev;
1547
1548		spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
1549		dev_alert(dev, "RX producer index going backwards\n");
1550		dev_alert(dev, "Disabled for further use\n");
1551		queue->info->broken = true;
1552		return false;
1553	}
1554	spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
1555
1556	if (likely(netif_carrier_ok(queue->info->netdev) && work_queued))
1557		napi_schedule(&queue->napi);
1558
1559	return true;
1560}
1561
1562static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1563{
1564	unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1565
1566	if (likely(xennet_handle_rx(dev_id, &eoiflag)))
1567		xen_irq_lateeoi(irq, eoiflag);
1568
1569	return IRQ_HANDLED;
1570}
1571
1572static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1573{
1574	unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1575
1576	if (xennet_handle_tx(dev_id, &eoiflag) &&
1577	    xennet_handle_rx(dev_id, &eoiflag))
1578		xen_irq_lateeoi(irq, eoiflag);
1579
1580	return IRQ_HANDLED;
1581}
1582
1583#ifdef CONFIG_NET_POLL_CONTROLLER
1584static void xennet_poll_controller(struct net_device *dev)
1585{
1586	/* Poll each queue */
1587	struct netfront_info *info = netdev_priv(dev);
1588	unsigned int num_queues = dev->real_num_tx_queues;
1589	unsigned int i;
1590
1591	if (info->broken)
1592		return;
1593
1594	for (i = 0; i < num_queues; ++i)
1595		xennet_interrupt(0, &info->queues[i]);
1596}
1597#endif
1598
1599#define NETBACK_XDP_HEADROOM_DISABLE	0
1600#define NETBACK_XDP_HEADROOM_ENABLE	1
1601
1602static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
1603{
1604	int err;
1605	unsigned short headroom;
1606
1607	headroom = xdp ? XDP_PACKET_HEADROOM : 0;
1608	err = xenbus_printf(XBT_NIL, np->xbdev->nodename,
1609			    "xdp-headroom", "%hu",
1610			    headroom);
1611	if (err)
1612		pr_warn("Error writing xdp-headroom\n");
1613
1614	return err;
1615}
1616
1617static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1618			  struct netlink_ext_ack *extack)
1619{
1620	unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
1621	struct netfront_info *np = netdev_priv(dev);
1622	struct bpf_prog *old_prog;
1623	unsigned int i, err;
1624
1625	if (dev->mtu > max_mtu) {
1626		netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu);
1627		return -EINVAL;
1628	}
1629
1630	if (!np->netback_has_xdp_headroom)
1631		return 0;
1632
1633	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);
1634
1635	err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE :
1636				  NETBACK_XDP_HEADROOM_DISABLE);
1637	if (err)
1638		return err;
1639
1640	/* avoid the race with XDP headroom adjustment */
1641	wait_event(module_wq,
1642		   xenbus_read_driver_state(np->xbdev->otherend) ==
1643		   XenbusStateReconfigured);
1644	np->netfront_xdp_enabled = true;
1645
1646	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
1647
1648	if (prog)
1649		bpf_prog_add(prog, dev->real_num_tx_queues);
1650
1651	for (i = 0; i < dev->real_num_tx_queues; ++i)
1652		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
1653
1654	if (old_prog)
1655		for (i = 0; i < dev->real_num_tx_queues; ++i)
1656			bpf_prog_put(old_prog);
1657
1658	xenbus_switch_state(np->xbdev, XenbusStateConnected);
1659
1660	return 0;
1661}
1662
1663static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1664{
1665	struct netfront_info *np = netdev_priv(dev);
1666
1667	if (np->broken)
1668		return -ENODEV;
1669
1670	switch (xdp->command) {
1671	case XDP_SETUP_PROG:
1672		return xennet_xdp_set(dev, xdp->prog, xdp->extack);
1673	default:
1674		return -EINVAL;
1675	}
1676}
1677
1678static const struct net_device_ops xennet_netdev_ops = {
1679	.ndo_uninit          = xennet_uninit,
1680	.ndo_open            = xennet_open,
1681	.ndo_stop            = xennet_close,
1682	.ndo_start_xmit      = xennet_start_xmit,
1683	.ndo_change_mtu	     = xennet_change_mtu,
1684	.ndo_get_stats64     = xennet_get_stats64,
1685	.ndo_set_mac_address = eth_mac_addr,
1686	.ndo_validate_addr   = eth_validate_addr,
1687	.ndo_fix_features    = xennet_fix_features,
1688	.ndo_set_features    = xennet_set_features,
1689	.ndo_select_queue    = xennet_select_queue,
1690	.ndo_bpf            = xennet_xdp,
1691	.ndo_xdp_xmit	    = xennet_xdp_xmit,
1692#ifdef CONFIG_NET_POLL_CONTROLLER
1693	.ndo_poll_controller = xennet_poll_controller,
1694#endif
1695};
1696
1697static void xennet_free_netdev(struct net_device *netdev)
1698{
1699	struct netfront_info *np = netdev_priv(netdev);
1700
1701	free_percpu(np->rx_stats);
1702	free_percpu(np->tx_stats);
1703	free_netdev(netdev);
1704}
1705
1706static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1707{
1708	int err;
1709	struct net_device *netdev;
1710	struct netfront_info *np;
1711
1712	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1713	if (!netdev)
1714		return ERR_PTR(-ENOMEM);
1715
1716	np                   = netdev_priv(netdev);
1717	np->xbdev            = dev;
1718
1719	np->queues = NULL;
1720
1721	err = -ENOMEM;
1722	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1723	if (np->rx_stats == NULL)
1724		goto exit;
1725	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1726	if (np->tx_stats == NULL)
1727		goto exit;
1728
1729	netdev->netdev_ops	= &xennet_netdev_ops;
1730
1731	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1732				  NETIF_F_GSO_ROBUST;
1733	netdev->hw_features	= NETIF_F_SG |
1734				  NETIF_F_IPV6_CSUM |
1735				  NETIF_F_TSO | NETIF_F_TSO6;
1736
1737	/*
1738         * Assume that all hw features are available for now. This set
1739         * will be adjusted by the call to netdev_update_features() in
1740         * xennet_connect() which is the earliest point where we can
1741         * negotiate with the backend regarding supported features.
1742         */
1743	netdev->features |= netdev->hw_features;
1744	netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
1745			       NETDEV_XDP_ACT_NDO_XMIT;
1746
1747	netdev->ethtool_ops = &xennet_ethtool_ops;
1748	netdev->min_mtu = ETH_MIN_MTU;
1749	netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1750	SET_NETDEV_DEV(netdev, &dev->dev);
1751
1752	np->netdev = netdev;
1753	np->netfront_xdp_enabled = false;
1754
1755	netif_carrier_off(netdev);
1756
1757	do {
1758		xenbus_switch_state(dev, XenbusStateInitialising);
1759		err = wait_event_timeout(module_wq,
1760				 xenbus_read_driver_state(dev->otherend) !=
1761				 XenbusStateClosed &&
1762				 xenbus_read_driver_state(dev->otherend) !=
1763				 XenbusStateUnknown, XENNET_TIMEOUT);
1764	} while (!err);
1765
1766	return netdev;
1767
1768 exit:
1769	xennet_free_netdev(netdev);
1770	return ERR_PTR(err);
1771}
1772
1773/*
1774 * Entry point to this code when a new device is created.  Allocate the basic
1775 * structures and the ring buffers for communication with the backend, and
1776 * inform the backend of the appropriate details for those.
1777 */
1778static int netfront_probe(struct xenbus_device *dev,
1779			  const struct xenbus_device_id *id)
1780{
1781	int err;
1782	struct net_device *netdev;
1783	struct netfront_info *info;
1784
1785	netdev = xennet_create_dev(dev);
1786	if (IS_ERR(netdev)) {
1787		err = PTR_ERR(netdev);
1788		xenbus_dev_fatal(dev, err, "creating netdev");
1789		return err;
1790	}
1791
1792	info = netdev_priv(netdev);
1793	dev_set_drvdata(&dev->dev, info);
1794#ifdef CONFIG_SYSFS
1795	info->netdev->sysfs_groups[0] = &xennet_dev_group;
1796#endif
1797
1798	return 0;
1799}
1800
1801static void xennet_end_access(int ref, void *page)
1802{
1803	/* This frees the page as a side-effect */
1804	if (ref != INVALID_GRANT_REF)
1805		gnttab_end_foreign_access(ref, virt_to_page(page));
1806}
1807
1808static void xennet_disconnect_backend(struct netfront_info *info)
1809{
1810	unsigned int i = 0;
1811	unsigned int num_queues = info->netdev->real_num_tx_queues;
1812
1813	netif_carrier_off(info->netdev);
1814
1815	for (i = 0; i < num_queues && info->queues; ++i) {
1816		struct netfront_queue *queue = &info->queues[i];
1817
1818		del_timer_sync(&queue->rx_refill_timer);
1819
1820		if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1821			unbind_from_irqhandler(queue->tx_irq, queue);
1822		if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1823			unbind_from_irqhandler(queue->tx_irq, queue);
1824			unbind_from_irqhandler(queue->rx_irq, queue);
1825		}
1826		queue->tx_evtchn = queue->rx_evtchn = 0;
1827		queue->tx_irq = queue->rx_irq = 0;
1828
1829		if (netif_running(info->netdev))
1830			napi_synchronize(&queue->napi);
1831
1832		xennet_release_tx_bufs(queue);
1833		xennet_release_rx_bufs(queue);
1834		gnttab_free_grant_references(queue->gref_tx_head);
1835		gnttab_free_grant_references(queue->gref_rx_head);
1836
1837		/* End access and free the pages */
1838		xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1839		xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1840
1841		queue->tx_ring_ref = INVALID_GRANT_REF;
1842		queue->rx_ring_ref = INVALID_GRANT_REF;
1843		queue->tx.sring = NULL;
1844		queue->rx.sring = NULL;
1845
1846		page_pool_destroy(queue->page_pool);
1847	}
1848}
1849
1850/*
1851 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1852 * driver restart.  We tear down our netif structure and recreate it, but
1853 * leave the device-layer structures intact so that this is transparent to the
1854 * rest of the kernel.
1855 */
1856static int netfront_resume(struct xenbus_device *dev)
1857{
1858	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1859
1860	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1861
1862	netif_tx_lock_bh(info->netdev);
1863	netif_device_detach(info->netdev);
1864	netif_tx_unlock_bh(info->netdev);
1865
1866	xennet_disconnect_backend(info);
1867
1868	rtnl_lock();
1869	if (info->queues)
1870		xennet_destroy_queues(info);
1871	rtnl_unlock();
1872
1873	return 0;
1874}
1875
1876static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1877{
1878	char *s, *e, *macstr;
1879	int i;
1880
1881	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1882	if (IS_ERR(macstr))
1883		return PTR_ERR(macstr);
1884
1885	for (i = 0; i < ETH_ALEN; i++) {
1886		mac[i] = simple_strtoul(s, &e, 16);
1887		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1888			kfree(macstr);
1889			return -ENOENT;
1890		}
1891		s = e+1;
1892	}
1893
1894	kfree(macstr);
1895	return 0;
1896}
1897
1898static int setup_netfront_single(struct netfront_queue *queue)
1899{
1900	int err;
1901
1902	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1903	if (err < 0)
1904		goto fail;
1905
1906	err = bind_evtchn_to_irqhandler_lateeoi(queue->tx_evtchn,
1907						xennet_interrupt, 0,
1908						queue->info->netdev->name,
1909						queue);
1910	if (err < 0)
1911		goto bind_fail;
1912	queue->rx_evtchn = queue->tx_evtchn;
1913	queue->rx_irq = queue->tx_irq = err;
1914
1915	return 0;
1916
1917bind_fail:
1918	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1919	queue->tx_evtchn = 0;
1920fail:
1921	return err;
1922}
1923
1924static int setup_netfront_split(struct netfront_queue *queue)
1925{
1926	int err;
1927
1928	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1929	if (err < 0)
1930		goto fail;
1931	err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1932	if (err < 0)
1933		goto alloc_rx_evtchn_fail;
1934
1935	snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1936		 "%s-tx", queue->name);
1937	err = bind_evtchn_to_irqhandler_lateeoi(queue->tx_evtchn,
1938						xennet_tx_interrupt, 0,
1939						queue->tx_irq_name, queue);
1940	if (err < 0)
1941		goto bind_tx_fail;
1942	queue->tx_irq = err;
1943
1944	snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1945		 "%s-rx", queue->name);
1946	err = bind_evtchn_to_irqhandler_lateeoi(queue->rx_evtchn,
1947						xennet_rx_interrupt, 0,
1948						queue->rx_irq_name, queue);
1949	if (err < 0)
1950		goto bind_rx_fail;
1951	queue->rx_irq = err;
1952
1953	return 0;
1954
1955bind_rx_fail:
1956	unbind_from_irqhandler(queue->tx_irq, queue);
1957	queue->tx_irq = 0;
1958bind_tx_fail:
1959	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1960	queue->rx_evtchn = 0;
1961alloc_rx_evtchn_fail:
1962	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1963	queue->tx_evtchn = 0;
1964fail:
1965	return err;
1966}
1967
1968static int setup_netfront(struct xenbus_device *dev,
1969			struct netfront_queue *queue, unsigned int feature_split_evtchn)
1970{
1971	struct xen_netif_tx_sring *txs;
1972	struct xen_netif_rx_sring *rxs;
1973	int err;
1974
1975	queue->tx_ring_ref = INVALID_GRANT_REF;
1976	queue->rx_ring_ref = INVALID_GRANT_REF;
1977	queue->rx.sring = NULL;
1978	queue->tx.sring = NULL;
1979
1980	err = xenbus_setup_ring(dev, GFP_NOIO | __GFP_HIGH, (void **)&txs,
1981				1, &queue->tx_ring_ref);
1982	if (err)
1983		goto fail;
1984
1985	XEN_FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1986
1987	err = xenbus_setup_ring(dev, GFP_NOIO | __GFP_HIGH, (void **)&rxs,
1988				1, &queue->rx_ring_ref);
1989	if (err)
1990		goto fail;
1991
1992	XEN_FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1993
1994	if (feature_split_evtchn)
1995		err = setup_netfront_split(queue);
1996	/* setup single event channel if
1997	 *  a) feature-split-event-channels == 0
1998	 *  b) feature-split-event-channels == 1 but failed to setup
1999	 */
2000	if (!feature_split_evtchn || err)
2001		err = setup_netfront_single(queue);
2002
2003	if (err)
2004		goto fail;
2005
2006	return 0;
2007
2008 fail:
2009	xenbus_teardown_ring((void **)&queue->rx.sring, 1, &queue->rx_ring_ref);
2010	xenbus_teardown_ring((void **)&queue->tx.sring, 1, &queue->tx_ring_ref);
2011
2012	return err;
2013}
2014
2015/* Queue-specific initialisation
2016 * This used to be done in xennet_create_dev() but must now
2017 * be run per-queue.
2018 */
2019static int xennet_init_queue(struct netfront_queue *queue)
2020{
2021	unsigned short i;
2022	int err = 0;
2023	char *devid;
2024
2025	spin_lock_init(&queue->tx_lock);
2026	spin_lock_init(&queue->rx_lock);
2027	spin_lock_init(&queue->rx_cons_lock);
2028
2029	timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
2030
2031	devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
2032	snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
2033		 devid, queue->id);
2034
2035	/* Initialise tx_skb_freelist as a free chain containing every entry. */
2036	queue->tx_skb_freelist = 0;
2037	queue->tx_pend_queue = TX_LINK_NONE;
2038	for (i = 0; i < NET_TX_RING_SIZE; i++) {
2039		queue->tx_link[i] = i + 1;
2040		queue->grant_tx_ref[i] = INVALID_GRANT_REF;
2041		queue->grant_tx_page[i] = NULL;
2042	}
2043	queue->tx_link[NET_TX_RING_SIZE - 1] = TX_LINK_NONE;
2044
2045	/* Clear out rx_skbs */
2046	for (i = 0; i < NET_RX_RING_SIZE; i++) {
2047		queue->rx_skbs[i] = NULL;
2048		queue->grant_rx_ref[i] = INVALID_GRANT_REF;
2049	}
2050
2051	/* A grant for every tx ring slot */
2052	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2053					  &queue->gref_tx_head) < 0) {
2054		pr_alert("can't alloc tx grant refs\n");
2055		err = -ENOMEM;
2056		goto exit;
2057	}
2058
2059	/* A grant for every rx ring slot */
2060	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
2061					  &queue->gref_rx_head) < 0) {
2062		pr_alert("can't alloc rx grant refs\n");
2063		err = -ENOMEM;
2064		goto exit_free_tx;
2065	}
2066
2067	return 0;
2068
2069 exit_free_tx:
2070	gnttab_free_grant_references(queue->gref_tx_head);
2071 exit:
2072	return err;
2073}
2074
2075static int write_queue_xenstore_keys(struct netfront_queue *queue,
2076			   struct xenbus_transaction *xbt, int write_hierarchical)
2077{
2078	/* Write the queue-specific keys into XenStore in the traditional
2079	 * way for a single queue, or in a queue subkeys for multiple
2080	 * queues.
2081	 */
2082	struct xenbus_device *dev = queue->info->xbdev;
2083	int err;
2084	const char *message;
2085	char *path;
2086	size_t pathsize;
2087
2088	/* Choose the correct place to write the keys */
2089	if (write_hierarchical) {
2090		pathsize = strlen(dev->nodename) + 10;
2091		path = kzalloc(pathsize, GFP_KERNEL);
2092		if (!path) {
2093			err = -ENOMEM;
2094			message = "out of memory while writing ring references";
2095			goto error;
2096		}
2097		snprintf(path, pathsize, "%s/queue-%u",
2098				dev->nodename, queue->id);
2099	} else {
2100		path = (char *)dev->nodename;
2101	}
2102
2103	/* Write ring references */
2104	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
2105			queue->tx_ring_ref);
2106	if (err) {
2107		message = "writing tx-ring-ref";
2108		goto error;
2109	}
2110
2111	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
2112			queue->rx_ring_ref);
2113	if (err) {
2114		message = "writing rx-ring-ref";
2115		goto error;
2116	}
2117
2118	/* Write event channels; taking into account both shared
2119	 * and split event channel scenarios.
2120	 */
2121	if (queue->tx_evtchn == queue->rx_evtchn) {
2122		/* Shared event channel */
2123		err = xenbus_printf(*xbt, path,
2124				"event-channel", "%u", queue->tx_evtchn);
2125		if (err) {
2126			message = "writing event-channel";
2127			goto error;
2128		}
2129	} else {
2130		/* Split event channels */
2131		err = xenbus_printf(*xbt, path,
2132				"event-channel-tx", "%u", queue->tx_evtchn);
2133		if (err) {
2134			message = "writing event-channel-tx";
2135			goto error;
2136		}
2137
2138		err = xenbus_printf(*xbt, path,
2139				"event-channel-rx", "%u", queue->rx_evtchn);
2140		if (err) {
2141			message = "writing event-channel-rx";
2142			goto error;
2143		}
2144	}
2145
2146	if (write_hierarchical)
2147		kfree(path);
2148	return 0;
2149
2150error:
2151	if (write_hierarchical)
2152		kfree(path);
2153	xenbus_dev_fatal(dev, err, "%s", message);
2154	return err;
2155}
2156
2157
2158
2159static int xennet_create_page_pool(struct netfront_queue *queue)
2160{
2161	int err;
2162	struct page_pool_params pp_params = {
2163		.order = 0,
2164		.flags = 0,
2165		.pool_size = NET_RX_RING_SIZE,
2166		.nid = NUMA_NO_NODE,
2167		.dev = &queue->info->netdev->dev,
2168		.offset = XDP_PACKET_HEADROOM,
2169		.max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
2170	};
2171
2172	queue->page_pool = page_pool_create(&pp_params);
2173	if (IS_ERR(queue->page_pool)) {
2174		err = PTR_ERR(queue->page_pool);
2175		queue->page_pool = NULL;
2176		return err;
2177	}
2178
2179	err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev,
2180			       queue->id, 0);
2181	if (err) {
2182		netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n");
2183		goto err_free_pp;
2184	}
2185
2186	err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq,
2187					 MEM_TYPE_PAGE_POOL, queue->page_pool);
2188	if (err) {
2189		netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n");
2190		goto err_unregister_rxq;
2191	}
2192	return 0;
2193
2194err_unregister_rxq:
2195	xdp_rxq_info_unreg(&queue->xdp_rxq);
2196err_free_pp:
2197	page_pool_destroy(queue->page_pool);
2198	queue->page_pool = NULL;
2199	return err;
2200}
2201
2202static int xennet_create_queues(struct netfront_info *info,
2203				unsigned int *num_queues)
2204{
2205	unsigned int i;
2206	int ret;
2207
2208	info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
2209			       GFP_KERNEL);
2210	if (!info->queues)
2211		return -ENOMEM;
2212
2213	for (i = 0; i < *num_queues; i++) {
2214		struct netfront_queue *queue = &info->queues[i];
2215
2216		queue->id = i;
2217		queue->info = info;
2218
2219		ret = xennet_init_queue(queue);
2220		if (ret < 0) {
2221			dev_warn(&info->xbdev->dev,
2222				 "only created %d queues\n", i);
2223			*num_queues = i;
2224			break;
2225		}
2226
2227		/* use page pool recycling instead of buddy allocator */
2228		ret = xennet_create_page_pool(queue);
2229		if (ret < 0) {
2230			dev_err(&info->xbdev->dev, "can't allocate page pool\n");
2231			*num_queues = i;
2232			return ret;
2233		}
2234
2235		netif_napi_add(queue->info->netdev, &queue->napi, xennet_poll);
2236		if (netif_running(info->netdev))
2237			napi_enable(&queue->napi);
2238	}
2239
2240	netif_set_real_num_tx_queues(info->netdev, *num_queues);
2241
2242	if (*num_queues == 0) {
2243		dev_err(&info->xbdev->dev, "no queues\n");
2244		return -EINVAL;
2245	}
2246	return 0;
2247}
2248
2249/* Common code used when first setting up, and when resuming. */
2250static int talk_to_netback(struct xenbus_device *dev,
2251			   struct netfront_info *info)
2252{
2253	const char *message;
2254	struct xenbus_transaction xbt;
2255	int err;
2256	unsigned int feature_split_evtchn;
2257	unsigned int i = 0;
2258	unsigned int max_queues = 0;
2259	struct netfront_queue *queue = NULL;
2260	unsigned int num_queues = 1;
2261	u8 addr[ETH_ALEN];
2262
2263	info->netdev->irq = 0;
2264
2265	/* Check if backend is trusted. */
2266	info->bounce = !xennet_trusted ||
2267		       !xenbus_read_unsigned(dev->nodename, "trusted", 1);
2268
2269	/* Check if backend supports multiple queues */
2270	max_queues = xenbus_read_unsigned(info->xbdev->otherend,
2271					  "multi-queue-max-queues", 1);
2272	num_queues = min(max_queues, xennet_max_queues);
2273
2274	/* Check feature-split-event-channels */
2275	feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
2276					"feature-split-event-channels", 0);
2277
2278	/* Read mac addr. */
2279	err = xen_net_read_mac(dev, addr);
2280	if (err) {
2281		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
2282		goto out_unlocked;
2283	}
2284	eth_hw_addr_set(info->netdev, addr);
2285
2286	info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend,
2287							      "feature-xdp-headroom", 0);
2288	if (info->netback_has_xdp_headroom) {
2289		/* set the current xen-netfront xdp state */
2290		err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ?
2291					  NETBACK_XDP_HEADROOM_ENABLE :
2292					  NETBACK_XDP_HEADROOM_DISABLE);
2293		if (err)
2294			goto out_unlocked;
2295	}
2296
2297	rtnl_lock();
2298	if (info->queues)
2299		xennet_destroy_queues(info);
2300
2301	/* For the case of a reconnect reset the "broken" indicator. */
2302	info->broken = false;
2303
2304	err = xennet_create_queues(info, &num_queues);
2305	if (err < 0) {
2306		xenbus_dev_fatal(dev, err, "creating queues");
2307		kfree(info->queues);
2308		info->queues = NULL;
2309		goto out;
2310	}
2311	rtnl_unlock();
2312
2313	/* Create shared ring, alloc event channel -- for each queue */
2314	for (i = 0; i < num_queues; ++i) {
2315		queue = &info->queues[i];
2316		err = setup_netfront(dev, queue, feature_split_evtchn);
2317		if (err)
2318			goto destroy_ring;
2319	}
2320
2321again:
2322	err = xenbus_transaction_start(&xbt);
2323	if (err) {
2324		xenbus_dev_fatal(dev, err, "starting transaction");
2325		goto destroy_ring;
2326	}
2327
2328	if (xenbus_exists(XBT_NIL,
2329			  info->xbdev->otherend, "multi-queue-max-queues")) {
2330		/* Write the number of queues */
2331		err = xenbus_printf(xbt, dev->nodename,
2332				    "multi-queue-num-queues", "%u", num_queues);
2333		if (err) {
2334			message = "writing multi-queue-num-queues";
2335			goto abort_transaction_no_dev_fatal;
2336		}
2337	}
2338
2339	if (num_queues == 1) {
2340		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
2341		if (err)
2342			goto abort_transaction_no_dev_fatal;
2343	} else {
2344		/* Write the keys for each queue */
2345		for (i = 0; i < num_queues; ++i) {
2346			queue = &info->queues[i];
2347			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
2348			if (err)
2349				goto abort_transaction_no_dev_fatal;
2350		}
2351	}
2352
2353	/* The remaining keys are not queue-specific */
2354	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
2355			    1);
2356	if (err) {
2357		message = "writing request-rx-copy";
2358		goto abort_transaction;
2359	}
2360
2361	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
2362	if (err) {
2363		message = "writing feature-rx-notify";
2364		goto abort_transaction;
2365	}
2366
2367	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
2368	if (err) {
2369		message = "writing feature-sg";
2370		goto abort_transaction;
2371	}
2372
2373	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
2374	if (err) {
2375		message = "writing feature-gso-tcpv4";
2376		goto abort_transaction;
2377	}
2378
2379	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
2380	if (err) {
2381		message = "writing feature-gso-tcpv6";
2382		goto abort_transaction;
2383	}
2384
2385	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
2386			   "1");
2387	if (err) {
2388		message = "writing feature-ipv6-csum-offload";
2389		goto abort_transaction;
2390	}
2391
2392	err = xenbus_transaction_end(xbt, 0);
2393	if (err) {
2394		if (err == -EAGAIN)
2395			goto again;
2396		xenbus_dev_fatal(dev, err, "completing transaction");
2397		goto destroy_ring;
2398	}
2399
2400	return 0;
2401
2402 abort_transaction:
2403	xenbus_dev_fatal(dev, err, "%s", message);
2404abort_transaction_no_dev_fatal:
2405	xenbus_transaction_end(xbt, 1);
2406 destroy_ring:
2407	xennet_disconnect_backend(info);
2408	rtnl_lock();
2409	xennet_destroy_queues(info);
2410 out:
2411	rtnl_unlock();
2412out_unlocked:
2413	device_unregister(&dev->dev);
2414	return err;
2415}
2416
2417static int xennet_connect(struct net_device *dev)
2418{
2419	struct netfront_info *np = netdev_priv(dev);
2420	unsigned int num_queues = 0;
2421	int err;
2422	unsigned int j = 0;
2423	struct netfront_queue *queue = NULL;
2424
2425	if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
2426		dev_info(&dev->dev,
2427			 "backend does not support copying receive path\n");
2428		return -ENODEV;
2429	}
2430
2431	err = talk_to_netback(np->xbdev, np);
2432	if (err)
2433		return err;
2434	if (np->netback_has_xdp_headroom)
2435		pr_info("backend supports XDP headroom\n");
2436	if (np->bounce)
2437		dev_info(&np->xbdev->dev,
2438			 "bouncing transmitted data to zeroed pages\n");
2439
2440	/* talk_to_netback() sets the correct number of queues */
2441	num_queues = dev->real_num_tx_queues;
2442
2443	if (dev->reg_state == NETREG_UNINITIALIZED) {
2444		err = register_netdev(dev);
2445		if (err) {
2446			pr_warn("%s: register_netdev err=%d\n", __func__, err);
2447			device_unregister(&np->xbdev->dev);
2448			return err;
2449		}
2450	}
2451
2452	rtnl_lock();
2453	netdev_update_features(dev);
2454	rtnl_unlock();
2455
2456	/*
2457	 * All public and private state should now be sane.  Get
2458	 * ready to start sending and receiving packets and give the driver
2459	 * domain a kick because we've probably just requeued some
2460	 * packets.
2461	 */
2462	netif_tx_lock_bh(np->netdev);
2463	netif_device_attach(np->netdev);
2464	netif_tx_unlock_bh(np->netdev);
2465
2466	netif_carrier_on(np->netdev);
2467	for (j = 0; j < num_queues; ++j) {
2468		queue = &np->queues[j];
2469
2470		notify_remote_via_irq(queue->tx_irq);
2471		if (queue->tx_irq != queue->rx_irq)
2472			notify_remote_via_irq(queue->rx_irq);
2473
2474		spin_lock_bh(&queue->rx_lock);
2475		xennet_alloc_rx_buffers(queue);
2476		spin_unlock_bh(&queue->rx_lock);
2477	}
2478
2479	return 0;
2480}
2481
2482/*
2483 * Callback received when the backend's state changes.
2484 */
2485static void netback_changed(struct xenbus_device *dev,
2486			    enum xenbus_state backend_state)
2487{
2488	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2489	struct net_device *netdev = np->netdev;
2490
2491	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2492
2493	wake_up_all(&module_wq);
2494
2495	switch (backend_state) {
2496	case XenbusStateInitialising:
2497	case XenbusStateInitialised:
2498	case XenbusStateReconfiguring:
2499	case XenbusStateReconfigured:
2500	case XenbusStateUnknown:
2501		break;
2502
2503	case XenbusStateInitWait:
2504		if (dev->state != XenbusStateInitialising)
2505			break;
2506		if (xennet_connect(netdev) != 0)
2507			break;
2508		xenbus_switch_state(dev, XenbusStateConnected);
2509		break;
2510
2511	case XenbusStateConnected:
2512		netdev_notify_peers(netdev);
2513		break;
2514
2515	case XenbusStateClosed:
2516		if (dev->state == XenbusStateClosed)
2517			break;
2518		fallthrough;	/* Missed the backend's CLOSING state */
2519	case XenbusStateClosing:
2520		xenbus_frontend_closed(dev);
2521		break;
2522	}
2523}
2524
2525static const struct xennet_stat {
2526	char name[ETH_GSTRING_LEN];
2527	u16 offset;
2528} xennet_stats[] = {
2529	{
2530		"rx_gso_checksum_fixup",
2531		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2532	},
2533};
2534
2535static int xennet_get_sset_count(struct net_device *dev, int string_set)
2536{
2537	switch (string_set) {
2538	case ETH_SS_STATS:
2539		return ARRAY_SIZE(xennet_stats);
2540	default:
2541		return -EINVAL;
2542	}
2543}
2544
2545static void xennet_get_ethtool_stats(struct net_device *dev,
2546				     struct ethtool_stats *stats, u64 * data)
2547{
2548	void *np = netdev_priv(dev);
2549	int i;
2550
2551	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2552		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2553}
2554
2555static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2556{
2557	int i;
2558
2559	switch (stringset) {
2560	case ETH_SS_STATS:
2561		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2562			memcpy(data + i * ETH_GSTRING_LEN,
2563			       xennet_stats[i].name, ETH_GSTRING_LEN);
2564		break;
2565	}
2566}
2567
2568static const struct ethtool_ops xennet_ethtool_ops =
2569{
2570	.get_link = ethtool_op_get_link,
2571
2572	.get_sset_count = xennet_get_sset_count,
2573	.get_ethtool_stats = xennet_get_ethtool_stats,
2574	.get_strings = xennet_get_strings,
2575	.get_ts_info = ethtool_op_get_ts_info,
2576};
2577
2578#ifdef CONFIG_SYSFS
2579static ssize_t show_rxbuf(struct device *dev,
2580			  struct device_attribute *attr, char *buf)
2581{
2582	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2583}
2584
2585static ssize_t store_rxbuf(struct device *dev,
2586			   struct device_attribute *attr,
2587			   const char *buf, size_t len)
2588{
2589	char *endp;
2590
2591	if (!capable(CAP_NET_ADMIN))
2592		return -EPERM;
2593
2594	simple_strtoul(buf, &endp, 0);
2595	if (endp == buf)
2596		return -EBADMSG;
2597
2598	/* rxbuf_min and rxbuf_max are no longer configurable. */
2599
2600	return len;
2601}
2602
2603static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2604static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2605static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2606
2607static struct attribute *xennet_dev_attrs[] = {
2608	&dev_attr_rxbuf_min.attr,
2609	&dev_attr_rxbuf_max.attr,
2610	&dev_attr_rxbuf_cur.attr,
2611	NULL
2612};
2613
2614static const struct attribute_group xennet_dev_group = {
2615	.attrs = xennet_dev_attrs
2616};
2617#endif /* CONFIG_SYSFS */
2618
2619static void xennet_bus_close(struct xenbus_device *dev)
2620{
2621	int ret;
2622
2623	if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2624		return;
2625	do {
2626		xenbus_switch_state(dev, XenbusStateClosing);
2627		ret = wait_event_timeout(module_wq,
2628				   xenbus_read_driver_state(dev->otherend) ==
2629				   XenbusStateClosing ||
2630				   xenbus_read_driver_state(dev->otherend) ==
2631				   XenbusStateClosed ||
2632				   xenbus_read_driver_state(dev->otherend) ==
2633				   XenbusStateUnknown,
2634				   XENNET_TIMEOUT);
2635	} while (!ret);
2636
2637	if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2638		return;
2639
2640	do {
2641		xenbus_switch_state(dev, XenbusStateClosed);
2642		ret = wait_event_timeout(module_wq,
2643				   xenbus_read_driver_state(dev->otherend) ==
2644				   XenbusStateClosed ||
2645				   xenbus_read_driver_state(dev->otherend) ==
2646				   XenbusStateUnknown,
2647				   XENNET_TIMEOUT);
2648	} while (!ret);
2649}
2650
2651static void xennet_remove(struct xenbus_device *dev)
2652{
2653	struct netfront_info *info = dev_get_drvdata(&dev->dev);
2654
2655	xennet_bus_close(dev);
2656	xennet_disconnect_backend(info);
2657
2658	if (info->netdev->reg_state == NETREG_REGISTERED)
2659		unregister_netdev(info->netdev);
2660
2661	if (info->queues) {
2662		rtnl_lock();
2663		xennet_destroy_queues(info);
2664		rtnl_unlock();
2665	}
2666	xennet_free_netdev(info->netdev);
2667}
2668
2669static const struct xenbus_device_id netfront_ids[] = {
2670	{ "vif" },
2671	{ "" }
2672};
2673
2674static struct xenbus_driver netfront_driver = {
2675	.ids = netfront_ids,
2676	.probe = netfront_probe,
2677	.remove = xennet_remove,
2678	.resume = netfront_resume,
2679	.otherend_changed = netback_changed,
2680};
2681
2682static int __init netif_init(void)
2683{
2684	if (!xen_domain())
2685		return -ENODEV;
2686
2687	if (!xen_has_pv_nic_devices())
2688		return -ENODEV;
2689
2690	pr_info("Initialising Xen virtual ethernet driver\n");
2691
2692	/* Allow as many queues as there are CPUs inut max. 8 if user has not
2693	 * specified a value.
2694	 */
2695	if (xennet_max_queues == 0)
2696		xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2697					  num_online_cpus());
2698
2699	return xenbus_register_frontend(&netfront_driver);
2700}
2701module_init(netif_init);
2702
2703
2704static void __exit netif_exit(void)
2705{
2706	xenbus_unregister_driver(&netfront_driver);
2707}
2708module_exit(netif_exit);
2709
2710MODULE_DESCRIPTION("Xen virtual network device frontend");
2711MODULE_LICENSE("GPL");
2712MODULE_ALIAS("xen:vif");
2713MODULE_ALIAS("xennet");