Linux Audio

Check our new training course

Loading...
v3.1
   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#include <linux/module.h>
  33#include <linux/kernel.h>
  34#include <linux/netdevice.h>
  35#include <linux/etherdevice.h>
  36#include <linux/skbuff.h>
  37#include <linux/ethtool.h>
  38#include <linux/if_ether.h>
  39#include <linux/tcp.h>
  40#include <linux/udp.h>
  41#include <linux/moduleparam.h>
  42#include <linux/mm.h>
  43#include <linux/slab.h>
  44#include <net/ip.h>
 
 
 
  45
  46#include <xen/xen.h>
  47#include <xen/xenbus.h>
  48#include <xen/events.h>
  49#include <xen/page.h>
 
  50#include <xen/grant_table.h>
  51
  52#include <xen/interface/io/netif.h>
  53#include <xen/interface/memory.h>
  54#include <xen/interface/grant_table.h>
  55
 
 
 
 
 
 
 
 
 
 
 
 
 
  56static const struct ethtool_ops xennet_ethtool_ops;
  57
  58struct netfront_cb {
  59	struct page *page;
  60	unsigned offset;
  61};
  62
  63#define NETFRONT_SKB_CB(skb)	((struct netfront_cb *)((skb)->cb))
  64
  65#define RX_COPY_THRESHOLD 256
  66
  67#define GRANT_INVALID_REF	0
 
 
 
 
 
 
 
  68
  69#define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE)
  70#define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE)
  71#define TX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
 
  72
  73struct netfront_stats {
  74	u64			rx_packets;
  75	u64			tx_packets;
  76	u64			rx_bytes;
  77	u64			tx_bytes;
  78	struct u64_stats_sync	syncp;
  79};
  80
  81struct netfront_info {
  82	struct list_head list;
  83	struct net_device *netdev;
 
 
 
 
 
  84
  85	struct napi_struct napi;
  86
  87	unsigned int evtchn;
  88	struct xenbus_device *xbdev;
 
 
 
 
 
 
  89
  90	spinlock_t   tx_lock;
  91	struct xen_netif_tx_front_ring tx;
  92	int tx_ring_ref;
  93
  94	/*
  95	 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
  96	 * are linked from tx_skb_freelist through skb_entry.link.
  97	 *
  98	 *  NB. Freelist index entries are always going to be less than
  99	 *  PAGE_OFFSET, whereas pointers to skbs will always be equal or
 100	 *  greater than PAGE_OFFSET: we use this property to distinguish
 101	 *  them.
 102	 */
 103	union skb_entry {
 104		struct sk_buff *skb;
 105		unsigned long link;
 106	} tx_skbs[NET_TX_RING_SIZE];
 107	grant_ref_t gref_tx_head;
 108	grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
 
 109	unsigned tx_skb_freelist;
 
 110
 111	spinlock_t   rx_lock ____cacheline_aligned_in_smp;
 112	struct xen_netif_rx_front_ring rx;
 113	int rx_ring_ref;
 114
 115	/* Receive-ring batched refills. */
 116#define RX_MIN_TARGET 8
 117#define RX_DFL_MIN_TARGET 64
 118#define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256)
 119	unsigned rx_min_target, rx_max_target, rx_target;
 120	struct sk_buff_head rx_batch;
 121
 122	struct timer_list rx_refill_timer;
 123
 124	struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
 125	grant_ref_t gref_rx_head;
 126	grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
 127
 128	unsigned long rx_pfn_array[NET_RX_RING_SIZE];
 129	struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1];
 130	struct mmu_update rx_mmu[NET_RX_RING_SIZE];
 
 
 
 
 
 
 
 
 
 
 
 
 131
 132	/* Statistics */
 133	struct netfront_stats __percpu *stats;
 
 
 
 
 
 
 
 
 134
 135	unsigned long rx_gso_checksum_fixup;
 
 
 
 136};
 137
 138struct netfront_rx_info {
 139	struct xen_netif_rx_response rx;
 140	struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
 141};
 142
 143static void skb_entry_set_link(union skb_entry *list, unsigned short id)
 144{
 145	list->link = id;
 146}
 147
 148static int skb_entry_is_link(const union skb_entry *list)
 149{
 150	BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link));
 151	return (unsigned long)list->skb < PAGE_OFFSET;
 152}
 153
 154/*
 155 * Access macros for acquiring freeing slots in tx_skbs[].
 156 */
 157
 158static void add_id_to_freelist(unsigned *head, union skb_entry *list,
 159			       unsigned short id)
 160{
 161	skb_entry_set_link(&list[id], *head);
 162	*head = id;
 163}
 164
 165static unsigned short get_id_from_freelist(unsigned *head,
 166					   union skb_entry *list)
 167{
 168	unsigned int id = *head;
 169	*head = list[id].link;
 
 
 
 
 170	return id;
 171}
 172
 173static int xennet_rxidx(RING_IDX idx)
 174{
 175	return idx & (NET_RX_RING_SIZE - 1);
 176}
 177
 178static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np,
 179					 RING_IDX ri)
 180{
 181	int i = xennet_rxidx(ri);
 182	struct sk_buff *skb = np->rx_skbs[i];
 183	np->rx_skbs[i] = NULL;
 184	return skb;
 185}
 186
 187static grant_ref_t xennet_get_rx_ref(struct netfront_info *np,
 188					    RING_IDX ri)
 189{
 190	int i = xennet_rxidx(ri);
 191	grant_ref_t ref = np->grant_rx_ref[i];
 192	np->grant_rx_ref[i] = GRANT_INVALID_REF;
 193	return ref;
 194}
 195
 196#ifdef CONFIG_SYSFS
 197static int xennet_sysfs_addif(struct net_device *netdev);
 198static void xennet_sysfs_delif(struct net_device *netdev);
 199#else /* !CONFIG_SYSFS */
 200#define xennet_sysfs_addif(dev) (0)
 201#define xennet_sysfs_delif(dev) do { } while (0)
 202#endif
 203
 204static int xennet_can_sg(struct net_device *dev)
 205{
 206	return dev->features & NETIF_F_SG;
 207}
 208
 209
 210static void rx_refill_timeout(unsigned long data)
 211{
 212	struct net_device *dev = (struct net_device *)data;
 213	struct netfront_info *np = netdev_priv(dev);
 214	napi_schedule(&np->napi);
 215}
 216
 217static int netfront_tx_slot_available(struct netfront_info *np)
 218{
 219	return (np->tx.req_prod_pvt - np->tx.rsp_cons) <
 220		(TX_MAX_TARGET - MAX_SKB_FRAGS - 2);
 221}
 222
 223static void xennet_maybe_wake_tx(struct net_device *dev)
 224{
 225	struct netfront_info *np = netdev_priv(dev);
 
 226
 227	if (unlikely(netif_queue_stopped(dev)) &&
 228	    netfront_tx_slot_available(np) &&
 229	    likely(netif_running(dev)))
 230		netif_wake_queue(dev);
 231}
 232
 233static void xennet_alloc_rx_buffers(struct net_device *dev)
 
 234{
 235	unsigned short id;
 236	struct netfront_info *np = netdev_priv(dev);
 237	struct sk_buff *skb;
 238	struct page *page;
 239	int i, batch_target, notify;
 240	RING_IDX req_prod = np->rx.req_prod_pvt;
 241	grant_ref_t ref;
 242	unsigned long pfn;
 243	void *vaddr;
 244	struct xen_netif_rx_request *req;
 245
 246	if (unlikely(!netif_carrier_ok(dev)))
 247		return;
 248
 249	/*
 250	 * Allocate skbuffs greedily, even though we batch updates to the
 251	 * receive ring. This creates a less bursty demand on the memory
 252	 * allocator, so should reduce the chance of failed allocation requests
 253	 * both for ourself and for other kernel subsystems.
 254	 */
 255	batch_target = np->rx_target - (req_prod - np->rx.rsp_cons);
 256	for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) {
 257		skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN,
 258					 GFP_ATOMIC | __GFP_NOWARN);
 259		if (unlikely(!skb))
 260			goto no_skb;
 
 
 
 
 
 
 261
 262		/* Align ip header to a 16 bytes boundary */
 263		skb_reserve(skb, NET_IP_ALIGN);
 264
 265		page = alloc_page(GFP_ATOMIC | __GFP_NOWARN);
 266		if (!page) {
 267			kfree_skb(skb);
 268no_skb:
 269			/* Any skbuffs queued for refill? Force them out. */
 270			if (i != 0)
 271				goto refill;
 272			/* Could not allocate any skbuffs. Try again later. */
 273			mod_timer(&np->rx_refill_timer,
 274				  jiffies + (HZ/10));
 275			break;
 276		}
 277
 278		skb_shinfo(skb)->frags[0].page = page;
 279		skb_shinfo(skb)->nr_frags = 1;
 280		__skb_queue_tail(&np->rx_batch, skb);
 281	}
 
 282
 283	/* Is the batch large enough to be worthwhile? */
 284	if (i < (np->rx_target/2)) {
 285		if (req_prod > np->rx.sring->req_prod)
 286			goto push;
 287		return;
 288	}
 289
 290	/* Adjust our fill target if we risked running out of buffers. */
 291	if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) &&
 292	    ((np->rx_target *= 2) > np->rx_max_target))
 293		np->rx_target = np->rx_max_target;
 294
 295 refill:
 296	for (i = 0; ; i++) {
 297		skb = __skb_dequeue(&np->rx_batch);
 298		if (skb == NULL)
 299			break;
 300
 301		skb->dev = dev;
 302
 303		id = xennet_rxidx(req_prod + i);
 304
 305		BUG_ON(np->rx_skbs[id]);
 306		np->rx_skbs[id] = skb;
 
 
 307
 308		ref = gnttab_claim_grant_reference(&np->gref_rx_head);
 309		BUG_ON((signed short)ref < 0);
 310		np->grant_rx_ref[id] = ref;
 
 
 311
 312		pfn = page_to_pfn(skb_shinfo(skb)->frags[0].page);
 313		vaddr = page_address(skb_shinfo(skb)->frags[0].page);
 314
 315		req = RING_GET_REQUEST(&np->rx, req_prod + i);
 316		gnttab_grant_foreign_access_ref(ref,
 317						np->xbdev->otherend_id,
 318						pfn_to_mfn(pfn),
 319						0);
 320
 
 
 
 
 
 
 
 
 
 
 
 321		req->id = id;
 322		req->gref = ref;
 323	}
 324
 325	wmb();		/* barrier so backend seens requests */
 326
 327	/* Above is a suitable barrier to ensure backend will see requests. */
 328	np->rx.req_prod_pvt = req_prod + i;
 329 push:
 330	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify);
 
 
 
 
 
 
 
 
 331	if (notify)
 332		notify_remote_via_irq(np->netdev->irq);
 333}
 334
 335static int xennet_open(struct net_device *dev)
 336{
 337	struct netfront_info *np = netdev_priv(dev);
 
 
 
 338
 339	napi_enable(&np->napi);
 
 340
 341	spin_lock_bh(&np->rx_lock);
 342	if (netif_carrier_ok(dev)) {
 343		xennet_alloc_rx_buffers(dev);
 344		np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
 345		if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
 346			napi_schedule(&np->napi);
 
 
 
 
 
 
 347	}
 348	spin_unlock_bh(&np->rx_lock);
 349
 350	netif_start_queue(dev);
 351
 352	return 0;
 353}
 354
 355static void xennet_tx_buf_gc(struct net_device *dev)
 356{
 357	RING_IDX cons, prod;
 358	unsigned short id;
 359	struct netfront_info *np = netdev_priv(dev);
 360	struct sk_buff *skb;
 
 
 
 361
 362	BUG_ON(!netif_carrier_ok(dev));
 363
 364	do {
 365		prod = np->tx.sring->rsp_prod;
 
 
 
 
 
 366		rmb(); /* Ensure we see responses up to 'rp'. */
 367
 368		for (cons = np->tx.rsp_cons; cons != prod; cons++) {
 369			struct xen_netif_tx_response *txrsp;
 370
 371			txrsp = RING_GET_RESPONSE(&np->tx, cons);
 372			if (txrsp->status == XEN_NETIF_RSP_NULL)
 
 
 373				continue;
 374
 375			id  = txrsp->id;
 376			skb = np->tx_skbs[id].skb;
 377			if (unlikely(gnttab_query_foreign_access(
 378				np->grant_tx_ref[id]) != 0)) {
 379				printk(KERN_ALERT "xennet_tx_buf_gc: warning "
 380				       "-- grant still in use by backend "
 381				       "domain.\n");
 382				BUG();
 
 
 
 
 
 
 
 
 
 
 
 
 
 383			}
 384			gnttab_end_foreign_access_ref(
 385				np->grant_tx_ref[id], GNTMAP_readonly);
 386			gnttab_release_grant_reference(
 387				&np->gref_tx_head, np->grant_tx_ref[id]);
 388			np->grant_tx_ref[id] = GRANT_INVALID_REF;
 389			add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id);
 
 390			dev_kfree_skb_irq(skb);
 391		}
 392
 393		np->tx.rsp_cons = prod;
 394
 395		/*
 396		 * Set a new event, then check for race with update of tx_cons.
 397		 * Note that it is essential to schedule a callback, no matter
 398		 * how few buffers are pending. Even if there is space in the
 399		 * transmit ring, higher layers may be blocked because too much
 400		 * data is outstanding: in such cases notification from Xen is
 401		 * likely to be the only kick that we'll get.
 402		 */
 403		np->tx.sring->rsp_event =
 404			prod + ((np->tx.sring->req_prod - prod) >> 1) + 1;
 405		mb();		/* update shared area */
 406	} while ((cons == prod) && (prod != np->tx.sring->rsp_prod));
 407
 408	xennet_maybe_wake_tx(dev);
 
 
 
 
 
 
 409}
 410
 411static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev,
 412			      struct xen_netif_tx_request *tx)
 
 
 
 
 
 
 
 
 
 413{
 414	struct netfront_info *np = netdev_priv(dev);
 415	char *data = skb->data;
 416	unsigned long mfn;
 417	RING_IDX prod = np->tx.req_prod_pvt;
 418	int frags = skb_shinfo(skb)->nr_frags;
 419	unsigned int offset = offset_in_page(data);
 420	unsigned int len = skb_headlen(skb);
 421	unsigned int id;
 
 422	grant_ref_t ref;
 423	int i;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 424
 425	/* While the header overlaps a page boundary (including being
 426	   larger than a page), split it it into page-sized chunks. */
 427	while (len > PAGE_SIZE - offset) {
 428		tx->size = PAGE_SIZE - offset;
 429		tx->flags |= XEN_NETTXF_more_data;
 430		len -= tx->size;
 431		data += tx->size;
 432		offset = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 433
 434		id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
 435		np->tx_skbs[id].skb = skb_get(skb);
 436		tx = RING_GET_REQUEST(&np->tx, prod++);
 437		tx->id = id;
 438		ref = gnttab_claim_grant_reference(&np->gref_tx_head);
 439		BUG_ON((signed short)ref < 0);
 440
 441		mfn = virt_to_mfn(data);
 442		gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
 443						mfn, GNTMAP_readonly);
 444
 445		tx->gref = np->grant_tx_ref[id] = ref;
 446		tx->offset = offset;
 447		tx->size = len;
 448		tx->flags = 0;
 449	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 450
 451	/* Grant backend access to each skb fragment page. */
 452	for (i = 0; i < frags; i++) {
 453		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 
 
 454
 455		tx->flags |= XEN_NETTXF_more_data;
 
 456
 457		id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
 458		np->tx_skbs[id].skb = skb_get(skb);
 459		tx = RING_GET_REQUEST(&np->tx, prod++);
 460		tx->id = id;
 461		ref = gnttab_claim_grant_reference(&np->gref_tx_head);
 462		BUG_ON((signed short)ref < 0);
 463
 464		mfn = pfn_to_mfn(page_to_pfn(frag->page));
 465		gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id,
 466						mfn, GNTMAP_readonly);
 
 
 
 467
 468		tx->gref = np->grant_tx_ref[id] = ref;
 469		tx->offset = frag->page_offset;
 470		tx->size = frag->size;
 471		tx->flags = 0;
 
 
 472	}
 473
 474	np->tx.req_prod_pvt = prod;
 475}
 476
 477static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
 
 
 
 
 
 
 
 
 
 
 
 478{
 479	unsigned short id;
 480	struct netfront_info *np = netdev_priv(dev);
 481	struct netfront_stats *stats = this_cpu_ptr(np->stats);
 482	struct xen_netif_tx_request *tx;
 483	struct xen_netif_extra_info *extra;
 484	char *data = skb->data;
 485	RING_IDX i;
 486	grant_ref_t ref;
 487	unsigned long mfn;
 488	int notify;
 489	int frags = skb_shinfo(skb)->nr_frags;
 490	unsigned int offset = offset_in_page(data);
 491	unsigned int len = skb_headlen(skb);
 492
 493	frags += DIV_ROUND_UP(offset + len, PAGE_SIZE);
 494	if (unlikely(frags > MAX_SKB_FRAGS + 1)) {
 495		printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n",
 496		       frags);
 497		dump_stack();
 498		goto drop;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 499	}
 
 500
 501	spin_lock_irq(&np->tx_lock);
 
 502
 503	if (unlikely(!netif_carrier_ok(dev) ||
 504		     (frags > 1 && !xennet_can_sg(dev)) ||
 505		     netif_needs_gso(skb, netif_skb_features(skb)))) {
 506		spin_unlock_irq(&np->tx_lock);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 507		goto drop;
 508	}
 509
 510	i = np->tx.req_prod_pvt;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 511
 512	id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs);
 513	np->tx_skbs[id].skb = skb;
 514
 515	tx = RING_GET_REQUEST(&np->tx, i);
 516
 517	tx->id   = id;
 518	ref = gnttab_claim_grant_reference(&np->gref_tx_head);
 519	BUG_ON((signed short)ref < 0);
 520	mfn = virt_to_mfn(data);
 521	gnttab_grant_foreign_access_ref(
 522		ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly);
 523	tx->gref = np->grant_tx_ref[id] = ref;
 524	tx->offset = offset;
 525	tx->size = len;
 526	extra = NULL;
 
 
 
 
 
 
 
 
 527
 528	tx->flags = 0;
 529	if (skb->ip_summed == CHECKSUM_PARTIAL)
 530		/* local packet? */
 531		tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated;
 
 532	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
 533		/* remote but checksummed. */
 534		tx->flags |= XEN_NETTXF_data_validated;
 535
 
 536	if (skb_shinfo(skb)->gso_size) {
 537		struct xen_netif_extra_info *gso;
 538
 539		gso = (struct xen_netif_extra_info *)
 540			RING_GET_REQUEST(&np->tx, ++i);
 541
 542		if (extra)
 543			extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE;
 544		else
 545			tx->flags |= XEN_NETTXF_extra_info;
 546
 547		gso->u.gso.size = skb_shinfo(skb)->gso_size;
 548		gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4;
 
 
 549		gso->u.gso.pad = 0;
 550		gso->u.gso.features = 0;
 551
 552		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
 553		gso->flags = 0;
 554		extra = gso;
 555	}
 556
 557	np->tx.req_prod_pvt = i + 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 558
 559	xennet_make_frags(skb, dev, tx);
 560	tx->size = skb->len;
 561
 562	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify);
 563	if (notify)
 564		notify_remote_via_irq(np->netdev->irq);
 565
 566	u64_stats_update_begin(&stats->syncp);
 567	stats->tx_bytes += skb->len;
 568	stats->tx_packets++;
 569	u64_stats_update_end(&stats->syncp);
 570
 571	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
 572	xennet_tx_buf_gc(dev);
 573
 574	if (!netfront_tx_slot_available(np))
 575		netif_stop_queue(dev);
 576
 577	spin_unlock_irq(&np->tx_lock);
 578
 579	return NETDEV_TX_OK;
 580
 581 drop:
 582	dev->stats.tx_dropped++;
 583	dev_kfree_skb(skb);
 584	return NETDEV_TX_OK;
 585}
 586
 587static int xennet_close(struct net_device *dev)
 588{
 589	struct netfront_info *np = netdev_priv(dev);
 590	netif_stop_queue(np->netdev);
 591	napi_disable(&np->napi);
 
 
 
 
 
 
 592	return 0;
 593}
 594
 595static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 596				grant_ref_t ref)
 597{
 598	int new = xennet_rxidx(np->rx.req_prod_pvt);
 599
 600	BUG_ON(np->rx_skbs[new]);
 601	np->rx_skbs[new] = skb;
 602	np->grant_rx_ref[new] = ref;
 603	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new;
 604	RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref;
 605	np->rx.req_prod_pvt++;
 606}
 607
 608static int xennet_get_extras(struct netfront_info *np,
 609			     struct xen_netif_extra_info *extras,
 610			     RING_IDX rp)
 611
 612{
 613	struct xen_netif_extra_info *extra;
 614	struct device *dev = &np->netdev->dev;
 615	RING_IDX cons = np->rx.rsp_cons;
 616	int err = 0;
 617
 618	do {
 619		struct sk_buff *skb;
 620		grant_ref_t ref;
 621
 622		if (unlikely(cons + 1 == rp)) {
 623			if (net_ratelimit())
 624				dev_warn(dev, "Missing extra info\n");
 625			err = -EBADR;
 626			break;
 627		}
 628
 629		extra = (struct xen_netif_extra_info *)
 630			RING_GET_RESPONSE(&np->rx, ++cons);
 631
 632		if (unlikely(!extra->type ||
 633			     extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
 634			if (net_ratelimit())
 635				dev_warn(dev, "Invalid extra type: %d\n",
 636					extra->type);
 637			err = -EINVAL;
 638		} else {
 639			memcpy(&extras[extra->type - 1], extra,
 640			       sizeof(*extra));
 641		}
 642
 643		skb = xennet_get_rx_skb(np, cons);
 644		ref = xennet_get_rx_ref(np, cons);
 645		xennet_move_rx_slot(np, skb, ref);
 646	} while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE);
 647
 648	np->rx.rsp_cons = cons;
 649	return err;
 650}
 651
 652static int xennet_get_responses(struct netfront_info *np,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 653				struct netfront_rx_info *rinfo, RING_IDX rp,
 654				struct sk_buff_head *list)
 
 655{
 656	struct xen_netif_rx_response *rx = &rinfo->rx;
 
 
 
 657	struct xen_netif_extra_info *extras = rinfo->extras;
 658	struct device *dev = &np->netdev->dev;
 659	RING_IDX cons = np->rx.rsp_cons;
 660	struct sk_buff *skb = xennet_get_rx_skb(np, cons);
 661	grant_ref_t ref = xennet_get_rx_ref(np, cons);
 662	int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD);
 663	int frags = 1;
 664	int err = 0;
 665	unsigned long ret;
 666
 667	if (rx->flags & XEN_NETRXF_extra_info) {
 668		err = xennet_get_extras(np, extras, rp);
 669		cons = np->rx.rsp_cons;
 670	}
 
 671
 672	for (;;) {
 673		if (unlikely(rx->status < 0 ||
 674			     rx->offset + rx->status > PAGE_SIZE)) {
 675			if (net_ratelimit())
 676				dev_warn(dev, "rx->offset: %x, size: %u\n",
 677					 rx->offset, rx->status);
 678			xennet_move_rx_slot(np, skb, ref);
 679			err = -EINVAL;
 680			goto next;
 681		}
 
 
 682
 
 683		/*
 684		 * This definitely indicates a bug, either in this driver or in
 685		 * the backend driver. In future this should flag the bad
 686		 * situation to the system controller to reboot the backed.
 687		 */
 688		if (ref == GRANT_INVALID_REF) {
 689			if (net_ratelimit())
 690				dev_warn(dev, "Bad rx response id %d.\n",
 691					 rx->id);
 692			err = -EINVAL;
 693			goto next;
 694		}
 695
 696		ret = gnttab_end_foreign_access_ref(ref, 0);
 697		BUG_ON(!ret);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 698
 699		gnttab_release_grant_reference(&np->gref_rx_head, ref);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 700
 701		__skb_queue_tail(list, skb);
 702
 703next:
 704		if (!(rx->flags & XEN_NETRXF_more_data))
 705			break;
 706
 707		if (cons + frags == rp) {
 708			if (net_ratelimit())
 709				dev_warn(dev, "Need more frags\n");
 710			err = -ENOENT;
 711			break;
 712		}
 713
 714		rx = RING_GET_RESPONSE(&np->rx, cons + frags);
 715		skb = xennet_get_rx_skb(np, cons + frags);
 716		ref = xennet_get_rx_ref(np, cons + frags);
 717		frags++;
 
 718	}
 719
 720	if (unlikely(frags > max)) {
 721		if (net_ratelimit())
 722			dev_warn(dev, "Too many frags\n");
 723		err = -E2BIG;
 724	}
 725
 726	if (unlikely(err))
 727		np->rx.rsp_cons = cons + frags;
 728
 729	return err;
 730}
 731
 732static int xennet_set_skb_gso(struct sk_buff *skb,
 733			      struct xen_netif_extra_info *gso)
 734{
 735	if (!gso->u.gso.size) {
 736		if (net_ratelimit())
 737			printk(KERN_WARNING "GSO size must not be zero.\n");
 738		return -EINVAL;
 739	}
 740
 741	/* Currently only TCPv4 S.O. is supported. */
 742	if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) {
 743		if (net_ratelimit())
 744			printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type);
 745		return -EINVAL;
 746	}
 747
 748	skb_shinfo(skb)->gso_size = gso->u.gso.size;
 749	skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
 
 
 
 750
 751	/* Header must be checked, and gso_segs computed. */
 752	skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 753	skb_shinfo(skb)->gso_segs = 0;
 754
 755	return 0;
 756}
 757
 758static RING_IDX xennet_fill_frags(struct netfront_info *np,
 759				  struct sk_buff *skb,
 760				  struct sk_buff_head *list)
 761{
 762	struct skb_shared_info *shinfo = skb_shinfo(skb);
 763	int nr_frags = shinfo->nr_frags;
 764	RING_IDX cons = np->rx.rsp_cons;
 765	skb_frag_t *frag = shinfo->frags + nr_frags;
 766	struct sk_buff *nskb;
 767
 768	while ((nskb = __skb_dequeue(list))) {
 769		struct xen_netif_rx_response *rx =
 770			RING_GET_RESPONSE(&np->rx, ++cons);
 771
 772		frag->page = skb_shinfo(nskb)->frags[0].page;
 773		frag->page_offset = rx->offset;
 774		frag->size = rx->status;
 775
 776		skb->data_len += rx->status;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 777
 778		skb_shinfo(nskb)->nr_frags = 0;
 779		kfree_skb(nskb);
 780
 781		frag++;
 782		nr_frags++;
 783	}
 784
 785	shinfo->nr_frags = nr_frags;
 786	return cons;
 
 787}
 788
 789static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
 790{
 791	struct iphdr *iph;
 792	unsigned char *th;
 793	int err = -EPROTO;
 794	int recalculate_partial_csum = 0;
 795
 796	/*
 797	 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
 798	 * peers can fail to set NETRXF_csum_blank when sending a GSO
 799	 * frame. In this case force the SKB to CHECKSUM_PARTIAL and
 800	 * recalculate the partial checksum.
 801	 */
 802	if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
 803		struct netfront_info *np = netdev_priv(dev);
 804		np->rx_gso_checksum_fixup++;
 805		skb->ip_summed = CHECKSUM_PARTIAL;
 806		recalculate_partial_csum = 1;
 807	}
 808
 809	/* A non-CHECKSUM_PARTIAL SKB does not require setup. */
 810	if (skb->ip_summed != CHECKSUM_PARTIAL)
 811		return 0;
 812
 813	if (skb->protocol != htons(ETH_P_IP))
 814		goto out;
 815
 816	iph = (void *)skb->data;
 817	th = skb->data + 4 * iph->ihl;
 818	if (th >= skb_tail_pointer(skb))
 819		goto out;
 820
 821	skb->csum_start = th - skb->head;
 822	switch (iph->protocol) {
 823	case IPPROTO_TCP:
 824		skb->csum_offset = offsetof(struct tcphdr, check);
 825
 826		if (recalculate_partial_csum) {
 827			struct tcphdr *tcph = (struct tcphdr *)th;
 828			tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
 829							 skb->len - iph->ihl*4,
 830							 IPPROTO_TCP, 0);
 831		}
 832		break;
 833	case IPPROTO_UDP:
 834		skb->csum_offset = offsetof(struct udphdr, check);
 835
 836		if (recalculate_partial_csum) {
 837			struct udphdr *udph = (struct udphdr *)th;
 838			udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr,
 839							 skb->len - iph->ihl*4,
 840							 IPPROTO_UDP, 0);
 841		}
 842		break;
 843	default:
 844		if (net_ratelimit())
 845			printk(KERN_ERR "Attempting to checksum a non-"
 846			       "TCP/UDP packet, dropping a protocol"
 847			       " %d packet", iph->protocol);
 848		goto out;
 849	}
 850
 851	if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb))
 852		goto out;
 853
 854	err = 0;
 855
 856out:
 857	return err;
 858}
 859
 860static int handle_incoming_queue(struct net_device *dev,
 861				 struct sk_buff_head *rxq)
 862{
 863	struct netfront_info *np = netdev_priv(dev);
 864	struct netfront_stats *stats = this_cpu_ptr(np->stats);
 865	int packets_dropped = 0;
 866	struct sk_buff *skb;
 867
 868	while ((skb = __skb_dequeue(rxq)) != NULL) {
 869		struct page *page = NETFRONT_SKB_CB(skb)->page;
 870		void *vaddr = page_address(page);
 871		unsigned offset = NETFRONT_SKB_CB(skb)->offset;
 872
 873		memcpy(skb->data, vaddr + offset,
 874		       skb_headlen(skb));
 875
 876		if (page != skb_shinfo(skb)->frags[0].page)
 877			__free_page(page);
 878
 879		/* Ethernet work: Delayed to here as it peeks the header. */
 880		skb->protocol = eth_type_trans(skb, dev);
 
 881
 882		if (checksum_setup(dev, skb)) {
 883			kfree_skb(skb);
 884			packets_dropped++;
 885			dev->stats.rx_errors++;
 886			continue;
 887		}
 888
 889		u64_stats_update_begin(&stats->syncp);
 890		stats->rx_packets++;
 891		stats->rx_bytes += skb->len;
 892		u64_stats_update_end(&stats->syncp);
 893
 894		/* Pass it up. */
 895		netif_receive_skb(skb);
 896	}
 897
 898	return packets_dropped;
 899}
 900
 901static int xennet_poll(struct napi_struct *napi, int budget)
 902{
 903	struct netfront_info *np = container_of(napi, struct netfront_info, napi);
 904	struct net_device *dev = np->netdev;
 905	struct sk_buff *skb;
 906	struct netfront_rx_info rinfo;
 907	struct xen_netif_rx_response *rx = &rinfo.rx;
 908	struct xen_netif_extra_info *extras = rinfo.extras;
 909	RING_IDX i, rp;
 910	int work_done;
 911	struct sk_buff_head rxq;
 912	struct sk_buff_head errq;
 913	struct sk_buff_head tmpq;
 914	unsigned long flags;
 915	unsigned int len;
 916	int err;
 
 917
 918	spin_lock(&np->rx_lock);
 919
 920	skb_queue_head_init(&rxq);
 921	skb_queue_head_init(&errq);
 922	skb_queue_head_init(&tmpq);
 923
 924	rp = np->rx.sring->rsp_prod;
 
 
 
 
 
 
 
 925	rmb(); /* Ensure we see queued responses up to 'rp'. */
 926
 927	i = np->rx.rsp_cons;
 928	work_done = 0;
 929	while ((i != rp) && (work_done < budget)) {
 930		memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx));
 931		memset(extras, 0, sizeof(rinfo.extras));
 932
 933		err = xennet_get_responses(np, &rinfo, rp, &tmpq);
 
 934
 935		if (unlikely(err)) {
 
 
 
 
 936err:
 937			while ((skb = __skb_dequeue(&tmpq)))
 938				__skb_queue_tail(&errq, skb);
 939			dev->stats.rx_errors++;
 940			i = np->rx.rsp_cons;
 941			continue;
 942		}
 943
 944		skb = __skb_dequeue(&tmpq);
 945
 946		if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
 947			struct xen_netif_extra_info *gso;
 948			gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
 949
 950			if (unlikely(xennet_set_skb_gso(skb, gso))) {
 951				__skb_queue_head(&tmpq, skb);
 952				np->rx.rsp_cons += skb_queue_len(&tmpq);
 
 
 953				goto err;
 954			}
 955		}
 956
 957		NETFRONT_SKB_CB(skb)->page = skb_shinfo(skb)->frags[0].page;
 958		NETFRONT_SKB_CB(skb)->offset = rx->offset;
 
 
 
 
 
 
 959
 960		len = rx->status;
 961		if (len > RX_COPY_THRESHOLD)
 962			len = RX_COPY_THRESHOLD;
 963		skb_put(skb, len);
 964
 965		if (rx->status > len) {
 966			skb_shinfo(skb)->frags[0].page_offset =
 967				rx->offset + len;
 968			skb_shinfo(skb)->frags[0].size = rx->status - len;
 969			skb->data_len = rx->status - len;
 970		} else {
 971			skb_shinfo(skb)->frags[0].page = NULL;
 972			skb_shinfo(skb)->nr_frags = 0;
 973		}
 974
 975		i = xennet_fill_frags(np, skb, &tmpq);
 976
 977		/*
 978		 * Truesize approximates the size of true data plus
 979		 * any supervisor overheads. Adding hypervisor
 980		 * overheads has been shown to significantly reduce
 981		 * achievable bandwidth with the default receive
 982		 * buffer size. It is therefore not wise to account
 983		 * for it here.
 984		 *
 985		 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set
 986		 * to RX_COPY_THRESHOLD + the supervisor
 987		 * overheads. Here, we add the size of the data pulled
 988		 * in xennet_fill_frags().
 989		 *
 990		 * We also adjust for any unused space in the main
 991		 * data area by subtracting (RX_COPY_THRESHOLD -
 992		 * len). This is especially important with drivers
 993		 * which split incoming packets into header and data,
 994		 * using only 66 bytes of the main data area (see the
 995		 * e1000 driver for example.)  On such systems,
 996		 * without this last adjustement, our achievable
 997		 * receive throughout using the standard receive
 998		 * buffer size was cut by 25%(!!!).
 999		 */
1000		skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len);
1001		skb->len += skb->data_len;
1002
1003		if (rx->flags & XEN_NETRXF_csum_blank)
1004			skb->ip_summed = CHECKSUM_PARTIAL;
1005		else if (rx->flags & XEN_NETRXF_data_validated)
1006			skb->ip_summed = CHECKSUM_UNNECESSARY;
1007
1008		__skb_queue_tail(&rxq, skb);
1009
1010		np->rx.rsp_cons = ++i;
 
1011		work_done++;
1012	}
 
 
1013
1014	__skb_queue_purge(&errq);
1015
1016	work_done -= handle_incoming_queue(dev, &rxq);
1017
1018	/* If we get a callback with very few responses, reduce fill target. */
1019	/* NB. Note exponential increase, linear decrease. */
1020	if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) >
1021	     ((3*np->rx_target) / 4)) &&
1022	    (--np->rx_target < np->rx_min_target))
1023		np->rx_target = np->rx_min_target;
1024
1025	xennet_alloc_rx_buffers(dev);
1026
1027	if (work_done < budget) {
1028		int more_to_do = 0;
1029
1030		local_irq_save(flags);
1031
1032		RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do);
1033		if (!more_to_do)
1034			__napi_complete(napi);
1035
1036		local_irq_restore(flags);
 
 
1037	}
1038
1039	spin_unlock(&np->rx_lock);
1040
1041	return work_done;
1042}
1043
1044static int xennet_change_mtu(struct net_device *dev, int mtu)
1045{
1046	int max = xennet_can_sg(dev) ? 65535 - ETH_HLEN : ETH_DATA_LEN;
1047
1048	if (mtu > max)
1049		return -EINVAL;
1050	dev->mtu = mtu;
1051	return 0;
1052}
1053
1054static struct rtnl_link_stats64 *xennet_get_stats64(struct net_device *dev,
1055						    struct rtnl_link_stats64 *tot)
1056{
1057	struct netfront_info *np = netdev_priv(dev);
1058	int cpu;
1059
1060	for_each_possible_cpu(cpu) {
1061		struct netfront_stats *stats = per_cpu_ptr(np->stats, cpu);
 
1062		u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1063		unsigned int start;
1064
1065		do {
1066			start = u64_stats_fetch_begin_bh(&stats->syncp);
 
 
 
1067
1068			rx_packets = stats->rx_packets;
1069			tx_packets = stats->tx_packets;
1070			rx_bytes = stats->rx_bytes;
1071			tx_bytes = stats->tx_bytes;
1072		} while (u64_stats_fetch_retry_bh(&stats->syncp, start));
1073
1074		tot->rx_packets += rx_packets;
1075		tot->tx_packets += tx_packets;
1076		tot->rx_bytes   += rx_bytes;
1077		tot->tx_bytes   += tx_bytes;
1078	}
1079
1080	tot->rx_errors  = dev->stats.rx_errors;
1081	tot->tx_dropped = dev->stats.tx_dropped;
1082
1083	return tot;
1084}
1085
1086static void xennet_release_tx_bufs(struct netfront_info *np)
1087{
1088	struct sk_buff *skb;
1089	int i;
1090
1091	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1092		/* Skip over entries which are actually freelist references */
1093		if (skb_entry_is_link(&np->tx_skbs[i]))
1094			continue;
1095
1096		skb = np->tx_skbs[i].skb;
1097		gnttab_end_foreign_access_ref(np->grant_tx_ref[i],
1098					      GNTMAP_readonly);
1099		gnttab_release_grant_reference(&np->gref_tx_head,
1100					       np->grant_tx_ref[i]);
1101		np->grant_tx_ref[i] = GRANT_INVALID_REF;
1102		add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i);
 
1103		dev_kfree_skb_irq(skb);
1104	}
1105}
1106
1107static void xennet_release_rx_bufs(struct netfront_info *np)
1108{
1109	struct mmu_update      *mmu = np->rx_mmu;
1110	struct multicall_entry *mcl = np->rx_mcl;
1111	struct sk_buff_head free_list;
1112	struct sk_buff *skb;
1113	unsigned long mfn;
1114	int xfer = 0, noxfer = 0, unused = 0;
1115	int id, ref;
1116
1117	dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n",
1118			 __func__);
1119	return;
1120
1121	skb_queue_head_init(&free_list);
1122
1123	spin_lock_bh(&np->rx_lock);
1124
1125	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1126		ref = np->grant_rx_ref[id];
1127		if (ref == GRANT_INVALID_REF) {
1128			unused++;
 
 
1129			continue;
1130		}
1131
1132		skb = np->rx_skbs[id];
1133		mfn = gnttab_end_foreign_transfer_ref(ref);
1134		gnttab_release_grant_reference(&np->gref_rx_head, ref);
1135		np->grant_rx_ref[id] = GRANT_INVALID_REF;
1136
1137		if (0 == mfn) {
1138			skb_shinfo(skb)->nr_frags = 0;
1139			dev_kfree_skb(skb);
1140			noxfer++;
1141			continue;
1142		}
1143
1144		if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1145			/* Remap the page. */
1146			struct page *page = skb_shinfo(skb)->frags[0].page;
1147			unsigned long pfn = page_to_pfn(page);
1148			void *vaddr = page_address(page);
1149
1150			MULTI_update_va_mapping(mcl, (unsigned long)vaddr,
1151						mfn_pte(mfn, PAGE_KERNEL),
1152						0);
1153			mcl++;
1154			mmu->ptr = ((u64)mfn << PAGE_SHIFT)
1155				| MMU_MACHPHYS_UPDATE;
1156			mmu->val = pfn;
1157			mmu++;
1158
1159			set_phys_to_machine(pfn, mfn);
1160		}
1161		__skb_queue_tail(&free_list, skb);
1162		xfer++;
1163	}
1164
1165	dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n",
1166		 __func__, xfer, noxfer, unused);
1167
1168	if (xfer) {
1169		if (!xen_feature(XENFEAT_auto_translated_physmap)) {
1170			/* Do all the remapping work and M2P updates. */
1171			MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu,
1172					 NULL, DOMID_SELF);
1173			mcl++;
1174			HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl);
1175		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1176	}
1177
1178	__skb_queue_purge(&free_list);
 
 
 
 
 
 
 
 
 
 
 
 
 
1179
1180	spin_unlock_bh(&np->rx_lock);
1181}
1182
1183static void xennet_uninit(struct net_device *dev)
1184{
1185	struct netfront_info *np = netdev_priv(dev);
1186	xennet_release_tx_bufs(np);
1187	xennet_release_rx_bufs(np);
1188	gnttab_free_grant_references(np->gref_tx_head);
1189	gnttab_free_grant_references(np->gref_rx_head);
 
1190}
1191
1192static u32 xennet_fix_features(struct net_device *dev, u32 features)
1193{
1194	struct netfront_info *np = netdev_priv(dev);
1195	int val;
 
 
 
1196
1197	if (features & NETIF_F_SG) {
1198		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg",
1199				 "%d", &val) < 0)
1200			val = 0;
 
 
 
1201
1202		if (!val)
1203			features &= ~NETIF_F_SG;
 
 
 
1204	}
 
1205
1206	if (features & NETIF_F_TSO) {
1207		if (xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1208				 "feature-gso-tcpv4", "%d", &val) < 0)
1209			val = 0;
1210
1211		if (!val)
1212			features &= ~NETIF_F_TSO;
1213	}
1214
1215	return features;
 
 
 
 
 
 
 
1216}
1217
1218static int xennet_set_features(struct net_device *dev, u32 features)
1219{
1220	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1221		netdev_info(dev, "Reducing MTU because no SG offload");
1222		dev->mtu = ETH_DATA_LEN;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1223	}
1224
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1225	return 0;
1226}
1227
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1228static const struct net_device_ops xennet_netdev_ops = {
1229	.ndo_open            = xennet_open,
1230	.ndo_uninit          = xennet_uninit,
 
1231	.ndo_stop            = xennet_close,
1232	.ndo_start_xmit      = xennet_start_xmit,
1233	.ndo_change_mtu	     = xennet_change_mtu,
1234	.ndo_get_stats64     = xennet_get_stats64,
1235	.ndo_set_mac_address = eth_mac_addr,
1236	.ndo_validate_addr   = eth_validate_addr,
1237	.ndo_fix_features    = xennet_fix_features,
1238	.ndo_set_features    = xennet_set_features,
 
 
 
 
 
 
1239};
1240
1241static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev)
1242{
1243	int i, err;
 
 
 
 
 
 
 
 
 
1244	struct net_device *netdev;
1245	struct netfront_info *np;
1246
1247	netdev = alloc_etherdev(sizeof(struct netfront_info));
1248	if (!netdev) {
1249		printk(KERN_WARNING "%s> alloc_etherdev failed.\n",
1250		       __func__);
1251		return ERR_PTR(-ENOMEM);
1252	}
1253
1254	np                   = netdev_priv(netdev);
1255	np->xbdev            = dev;
1256
1257	spin_lock_init(&np->tx_lock);
1258	spin_lock_init(&np->rx_lock);
1259
1260	skb_queue_head_init(&np->rx_batch);
1261	np->rx_target     = RX_DFL_MIN_TARGET;
1262	np->rx_min_target = RX_DFL_MIN_TARGET;
1263	np->rx_max_target = RX_MAX_TARGET;
1264
1265	init_timer(&np->rx_refill_timer);
1266	np->rx_refill_timer.data = (unsigned long)netdev;
1267	np->rx_refill_timer.function = rx_refill_timeout;
1268
1269	err = -ENOMEM;
1270	np->stats = alloc_percpu(struct netfront_stats);
1271	if (np->stats == NULL)
 
 
 
1272		goto exit;
1273
1274	/* Initialise tx_skbs as a free chain containing every entry. */
1275	np->tx_skb_freelist = 0;
1276	for (i = 0; i < NET_TX_RING_SIZE; i++) {
1277		skb_entry_set_link(&np->tx_skbs[i], i+1);
1278		np->grant_tx_ref[i] = GRANT_INVALID_REF;
1279	}
1280
1281	/* Clear out rx_skbs */
1282	for (i = 0; i < NET_RX_RING_SIZE; i++) {
1283		np->rx_skbs[i] = NULL;
1284		np->grant_rx_ref[i] = GRANT_INVALID_REF;
1285	}
1286
1287	/* A grant for every tx ring slot */
1288	if (gnttab_alloc_grant_references(TX_MAX_TARGET,
1289					  &np->gref_tx_head) < 0) {
1290		printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n");
1291		err = -ENOMEM;
1292		goto exit_free_stats;
1293	}
1294	/* A grant for every rx ring slot */
1295	if (gnttab_alloc_grant_references(RX_MAX_TARGET,
1296					  &np->gref_rx_head) < 0) {
1297		printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n");
1298		err = -ENOMEM;
1299		goto exit_free_tx;
1300	}
1301
1302	netdev->netdev_ops	= &xennet_netdev_ops;
1303
1304	netif_napi_add(netdev, &np->napi, xennet_poll, 64);
1305	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1306				  NETIF_F_GSO_ROBUST;
1307	netdev->hw_features	= NETIF_F_IP_CSUM | NETIF_F_SG | NETIF_F_TSO;
 
 
1308
1309	/*
1310         * Assume that all hw features are available for now. This set
1311         * will be adjusted by the call to netdev_update_features() in
1312         * xennet_connect() which is the earliest point where we can
1313         * negotiate with the backend regarding supported features.
1314         */
1315	netdev->features |= netdev->hw_features;
 
 
1316
1317	SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops);
 
 
1318	SET_NETDEV_DEV(netdev, &dev->dev);
1319
1320	np->netdev = netdev;
 
1321
1322	netif_carrier_off(netdev);
1323
 
 
 
 
 
 
 
 
 
1324	return netdev;
1325
1326 exit_free_tx:
1327	gnttab_free_grant_references(np->gref_tx_head);
1328 exit_free_stats:
1329	free_percpu(np->stats);
1330 exit:
1331	free_netdev(netdev);
1332	return ERR_PTR(err);
1333}
1334
1335/**
1336 * Entry point to this code when a new device is created.  Allocate the basic
1337 * structures and the ring buffers for communication with the backend, and
1338 * inform the backend of the appropriate details for those.
1339 */
1340static int __devinit netfront_probe(struct xenbus_device *dev,
1341				    const struct xenbus_device_id *id)
1342{
1343	int err;
1344	struct net_device *netdev;
1345	struct netfront_info *info;
1346
1347	netdev = xennet_create_dev(dev);
1348	if (IS_ERR(netdev)) {
1349		err = PTR_ERR(netdev);
1350		xenbus_dev_fatal(dev, err, "creating netdev");
1351		return err;
1352	}
1353
1354	info = netdev_priv(netdev);
1355	dev_set_drvdata(&dev->dev, info);
1356
1357	err = register_netdev(info->netdev);
1358	if (err) {
1359		printk(KERN_WARNING "%s: register_netdev err=%d\n",
1360		       __func__, err);
1361		goto fail;
1362	}
1363
1364	err = xennet_sysfs_addif(info->netdev);
1365	if (err) {
1366		unregister_netdev(info->netdev);
1367		printk(KERN_WARNING "%s: add sysfs failed err=%d\n",
1368		       __func__, err);
1369		goto fail;
1370	}
1371
1372	return 0;
1373
1374 fail:
1375	free_netdev(netdev);
1376	dev_set_drvdata(&dev->dev, NULL);
1377	return err;
1378}
1379
1380static void xennet_end_access(int ref, void *page)
1381{
1382	/* This frees the page as a side-effect */
1383	if (ref != GRANT_INVALID_REF)
1384		gnttab_end_foreign_access(ref, 0, (unsigned long)page);
1385}
1386
1387static void xennet_disconnect_backend(struct netfront_info *info)
1388{
1389	/* Stop old i/f to prevent errors whilst we rebuild the state. */
1390	spin_lock_bh(&info->rx_lock);
1391	spin_lock_irq(&info->tx_lock);
1392	netif_carrier_off(info->netdev);
1393	spin_unlock_irq(&info->tx_lock);
1394	spin_unlock_bh(&info->rx_lock);
1395
1396	if (info->netdev->irq)
1397		unbind_from_irqhandler(info->netdev->irq, info->netdev);
1398	info->evtchn = info->netdev->irq = 0;
 
1399
1400	/* End access and free the pages */
1401	xennet_end_access(info->tx_ring_ref, info->tx.sring);
1402	xennet_end_access(info->rx_ring_ref, info->rx.sring);
 
 
 
 
 
1403
1404	info->tx_ring_ref = GRANT_INVALID_REF;
1405	info->rx_ring_ref = GRANT_INVALID_REF;
1406	info->tx.sring = NULL;
1407	info->rx.sring = NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1408}
1409
1410/**
1411 * We are reconnecting to the backend, due to a suspend/resume, or a backend
1412 * driver restart.  We tear down our netif structure and recreate it, but
1413 * leave the device-layer structures intact so that this is transparent to the
1414 * rest of the kernel.
1415 */
1416static int netfront_resume(struct xenbus_device *dev)
1417{
1418	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1419
1420	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1421
 
 
 
 
1422	xennet_disconnect_backend(info);
 
 
 
 
 
 
1423	return 0;
1424}
1425
1426static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1427{
1428	char *s, *e, *macstr;
1429	int i;
1430
1431	macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1432	if (IS_ERR(macstr))
1433		return PTR_ERR(macstr);
1434
1435	for (i = 0; i < ETH_ALEN; i++) {
1436		mac[i] = simple_strtoul(s, &e, 16);
1437		if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1438			kfree(macstr);
1439			return -ENOENT;
1440		}
1441		s = e+1;
1442	}
1443
1444	kfree(macstr);
1445	return 0;
1446}
1447
1448static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1449{
1450	struct net_device *dev = dev_id;
1451	struct netfront_info *np = netdev_priv(dev);
1452	unsigned long flags;
1453
1454	spin_lock_irqsave(&np->tx_lock, flags);
 
 
1455
1456	if (likely(netif_carrier_ok(dev))) {
1457		xennet_tx_buf_gc(dev);
1458		/* Under tx_lock: protects access to rx shared-ring indexes. */
1459		if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
1460			napi_schedule(&np->napi);
1461	}
 
 
1462
1463	spin_unlock_irqrestore(&np->tx_lock, flags);
1464
1465	return IRQ_HANDLED;
 
 
 
 
1466}
1467
1468static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1469{
1470	struct xen_netif_tx_sring *txs;
1471	struct xen_netif_rx_sring *rxs;
1472	int err;
1473	struct net_device *netdev = info->netdev;
1474
1475	info->tx_ring_ref = GRANT_INVALID_REF;
1476	info->rx_ring_ref = GRANT_INVALID_REF;
1477	info->rx.sring = NULL;
1478	info->tx.sring = NULL;
1479	netdev->irq = 0;
1480
1481	err = xen_net_read_mac(dev, netdev->dev_addr);
1482	if (err) {
1483		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
1484		goto fail;
1485	}
1486
1487	txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1488	if (!txs) {
1489		err = -ENOMEM;
1490		xenbus_dev_fatal(dev, err, "allocating tx ring page");
 
1491		goto fail;
1492	}
1493	SHARED_RING_INIT(txs);
1494	FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE);
1495
1496	err = xenbus_grant_ring(dev, virt_to_mfn(txs));
1497	if (err < 0) {
1498		free_page((unsigned long)txs);
 
 
 
 
 
 
 
 
 
1499		goto fail;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1500	}
 
1501
1502	info->tx_ring_ref = err;
1503	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
1504	if (!rxs) {
 
 
 
 
 
 
 
1505		err = -ENOMEM;
1506		xenbus_dev_fatal(dev, err, "allocating rx ring page");
1507		goto fail;
1508	}
1509	SHARED_RING_INIT(rxs);
1510	FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE);
1511
1512	err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
1513	if (err < 0) {
1514		free_page((unsigned long)rxs);
1515		goto fail;
 
 
1516	}
1517	info->rx_ring_ref = err;
1518
1519	err = xenbus_alloc_evtchn(dev, &info->evtchn);
1520	if (err)
1521		goto fail;
1522
1523	err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt,
1524					0, netdev->name, netdev);
1525	if (err < 0)
1526		goto fail;
1527	netdev->irq = err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1528	return 0;
1529
1530 fail:
 
 
 
1531	return err;
1532}
1533
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1534/* Common code used when first setting up, and when resuming. */
1535static int talk_to_netback(struct xenbus_device *dev,
1536			   struct netfront_info *info)
1537{
1538	const char *message;
1539	struct xenbus_transaction xbt;
1540	int err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1541
1542	/* Create shared ring, alloc event channel. */
1543	err = setup_netfront(dev, info);
1544	if (err)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1545		goto out;
 
 
 
 
 
 
 
 
 
 
1546
1547again:
1548	err = xenbus_transaction_start(&xbt);
1549	if (err) {
1550		xenbus_dev_fatal(dev, err, "starting transaction");
1551		goto destroy_ring;
1552	}
1553
1554	err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u",
1555			    info->tx_ring_ref);
1556	if (err) {
1557		message = "writing tx ring-ref";
1558		goto abort_transaction;
1559	}
1560	err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u",
1561			    info->rx_ring_ref);
1562	if (err) {
1563		message = "writing rx ring-ref";
1564		goto abort_transaction;
1565	}
1566	err = xenbus_printf(xbt, dev->nodename,
1567			    "event-channel", "%u", info->evtchn);
1568	if (err) {
1569		message = "writing event-channel";
1570		goto abort_transaction;
 
 
 
 
 
 
 
 
1571	}
1572
 
1573	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
1574			    1);
1575	if (err) {
1576		message = "writing request-rx-copy";
1577		goto abort_transaction;
1578	}
1579
1580	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
1581	if (err) {
1582		message = "writing feature-rx-notify";
1583		goto abort_transaction;
1584	}
1585
1586	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
1587	if (err) {
1588		message = "writing feature-sg";
1589		goto abort_transaction;
1590	}
1591
1592	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
1593	if (err) {
1594		message = "writing feature-gso-tcpv4";
1595		goto abort_transaction;
1596	}
1597
 
 
 
 
 
 
 
 
 
 
 
 
 
1598	err = xenbus_transaction_end(xbt, 0);
1599	if (err) {
1600		if (err == -EAGAIN)
1601			goto again;
1602		xenbus_dev_fatal(dev, err, "completing transaction");
1603		goto destroy_ring;
1604	}
1605
1606	return 0;
1607
1608 abort_transaction:
1609	xenbus_transaction_end(xbt, 1);
1610	xenbus_dev_fatal(dev, err, "%s", message);
 
 
1611 destroy_ring:
1612	xennet_disconnect_backend(info);
 
 
1613 out:
 
 
 
1614	return err;
1615}
1616
1617static int xennet_connect(struct net_device *dev)
1618{
1619	struct netfront_info *np = netdev_priv(dev);
1620	int i, requeue_idx, err;
1621	struct sk_buff *skb;
1622	grant_ref_t ref;
1623	struct xen_netif_rx_request *req;
1624	unsigned int feature_rx_copy;
1625
1626	err = xenbus_scanf(XBT_NIL, np->xbdev->otherend,
1627			   "feature-rx-copy", "%u", &feature_rx_copy);
1628	if (err != 1)
1629		feature_rx_copy = 0;
1630
1631	if (!feature_rx_copy) {
1632		dev_info(&dev->dev,
1633			 "backend does not support copying receive path\n");
1634		return -ENODEV;
1635	}
1636
1637	err = talk_to_netback(np->xbdev, np);
1638	if (err)
1639		return err;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1640
1641	rtnl_lock();
1642	netdev_update_features(dev);
1643	rtnl_unlock();
1644
1645	spin_lock_bh(&np->rx_lock);
1646	spin_lock_irq(&np->tx_lock);
1647
1648	/* Step 1: Discard all pending TX packet fragments. */
1649	xennet_release_tx_bufs(np);
1650
1651	/* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */
1652	for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) {
1653		if (!np->rx_skbs[i])
1654			continue;
1655
1656		skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i);
1657		ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i);
1658		req = RING_GET_REQUEST(&np->rx, requeue_idx);
1659
1660		gnttab_grant_foreign_access_ref(
1661			ref, np->xbdev->otherend_id,
1662			pfn_to_mfn(page_to_pfn(skb_shinfo(skb)->
1663					       frags->page)),
1664			0);
1665		req->gref = ref;
1666		req->id   = requeue_idx;
1667
1668		requeue_idx++;
1669	}
1670
1671	np->rx.req_prod_pvt = requeue_idx;
1672
1673	/*
1674	 * Step 3: All public and private state should now be sane.  Get
1675	 * ready to start sending and receiving packets and give the driver
1676	 * domain a kick because we've probably just requeued some
1677	 * packets.
1678	 */
 
 
 
 
1679	netif_carrier_on(np->netdev);
1680	notify_remote_via_irq(np->netdev->irq);
1681	xennet_tx_buf_gc(dev);
1682	xennet_alloc_rx_buffers(dev);
1683
1684	spin_unlock_irq(&np->tx_lock);
1685	spin_unlock_bh(&np->rx_lock);
 
 
 
 
 
 
1686
1687	return 0;
1688}
1689
1690/**
1691 * Callback received when the backend's state changes.
1692 */
1693static void netback_changed(struct xenbus_device *dev,
1694			    enum xenbus_state backend_state)
1695{
1696	struct netfront_info *np = dev_get_drvdata(&dev->dev);
1697	struct net_device *netdev = np->netdev;
1698
1699	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
1700
 
 
1701	switch (backend_state) {
1702	case XenbusStateInitialising:
1703	case XenbusStateInitialised:
1704	case XenbusStateReconfiguring:
1705	case XenbusStateReconfigured:
1706	case XenbusStateConnected:
1707	case XenbusStateUnknown:
1708	case XenbusStateClosed:
1709		break;
1710
1711	case XenbusStateInitWait:
1712		if (dev->state != XenbusStateInitialising)
1713			break;
1714		if (xennet_connect(netdev) != 0)
1715			break;
1716		xenbus_switch_state(dev, XenbusStateConnected);
1717		netif_notify_peers(netdev);
1718		break;
1719
 
 
 
 
 
 
 
 
1720	case XenbusStateClosing:
1721		xenbus_frontend_closed(dev);
1722		break;
1723	}
1724}
1725
1726static const struct xennet_stat {
1727	char name[ETH_GSTRING_LEN];
1728	u16 offset;
1729} xennet_stats[] = {
1730	{
1731		"rx_gso_checksum_fixup",
1732		offsetof(struct netfront_info, rx_gso_checksum_fixup)
1733	},
1734};
1735
1736static int xennet_get_sset_count(struct net_device *dev, int string_set)
1737{
1738	switch (string_set) {
1739	case ETH_SS_STATS:
1740		return ARRAY_SIZE(xennet_stats);
1741	default:
1742		return -EINVAL;
1743	}
1744}
1745
1746static void xennet_get_ethtool_stats(struct net_device *dev,
1747				     struct ethtool_stats *stats, u64 * data)
1748{
1749	void *np = netdev_priv(dev);
1750	int i;
1751
1752	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1753		data[i] = *(unsigned long *)(np + xennet_stats[i].offset);
1754}
1755
1756static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
1757{
1758	int i;
1759
1760	switch (stringset) {
1761	case ETH_SS_STATS:
1762		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
1763			memcpy(data + i * ETH_GSTRING_LEN,
1764			       xennet_stats[i].name, ETH_GSTRING_LEN);
1765		break;
1766	}
1767}
1768
1769static const struct ethtool_ops xennet_ethtool_ops =
1770{
1771	.get_link = ethtool_op_get_link,
1772
1773	.get_sset_count = xennet_get_sset_count,
1774	.get_ethtool_stats = xennet_get_ethtool_stats,
1775	.get_strings = xennet_get_strings,
 
1776};
1777
1778#ifdef CONFIG_SYSFS
1779static ssize_t show_rxbuf_min(struct device *dev,
1780			      struct device_attribute *attr, char *buf)
1781{
1782	struct net_device *netdev = to_net_dev(dev);
1783	struct netfront_info *info = netdev_priv(netdev);
1784
1785	return sprintf(buf, "%u\n", info->rx_min_target);
1786}
1787
1788static ssize_t store_rxbuf_min(struct device *dev,
1789			       struct device_attribute *attr,
1790			       const char *buf, size_t len)
1791{
1792	struct net_device *netdev = to_net_dev(dev);
1793	struct netfront_info *np = netdev_priv(netdev);
1794	char *endp;
1795	unsigned long target;
1796
1797	if (!capable(CAP_NET_ADMIN))
1798		return -EPERM;
1799
1800	target = simple_strtoul(buf, &endp, 0);
1801	if (endp == buf)
1802		return -EBADMSG;
1803
1804	if (target < RX_MIN_TARGET)
1805		target = RX_MIN_TARGET;
1806	if (target > RX_MAX_TARGET)
1807		target = RX_MAX_TARGET;
1808
1809	spin_lock_bh(&np->rx_lock);
1810	if (target > np->rx_max_target)
1811		np->rx_max_target = target;
1812	np->rx_min_target = target;
1813	if (target > np->rx_target)
1814		np->rx_target = target;
1815
1816	xennet_alloc_rx_buffers(netdev);
1817
1818	spin_unlock_bh(&np->rx_lock);
1819	return len;
1820}
1821
1822static ssize_t show_rxbuf_max(struct device *dev,
1823			      struct device_attribute *attr, char *buf)
1824{
1825	struct net_device *netdev = to_net_dev(dev);
1826	struct netfront_info *info = netdev_priv(netdev);
 
 
 
 
 
1827
1828	return sprintf(buf, "%u\n", info->rx_max_target);
1829}
 
 
1830
1831static ssize_t store_rxbuf_max(struct device *dev,
1832			       struct device_attribute *attr,
1833			       const char *buf, size_t len)
1834{
1835	struct net_device *netdev = to_net_dev(dev);
1836	struct netfront_info *np = netdev_priv(netdev);
1837	char *endp;
1838	unsigned long target;
1839
1840	if (!capable(CAP_NET_ADMIN))
1841		return -EPERM;
1842
1843	target = simple_strtoul(buf, &endp, 0);
1844	if (endp == buf)
1845		return -EBADMSG;
1846
1847	if (target < RX_MIN_TARGET)
1848		target = RX_MIN_TARGET;
1849	if (target > RX_MAX_TARGET)
1850		target = RX_MAX_TARGET;
1851
1852	spin_lock_bh(&np->rx_lock);
1853	if (target < np->rx_min_target)
1854		np->rx_min_target = target;
1855	np->rx_max_target = target;
1856	if (target < np->rx_target)
1857		np->rx_target = target;
1858
1859	xennet_alloc_rx_buffers(netdev);
 
1860
1861	spin_unlock_bh(&np->rx_lock);
1862	return len;
 
 
 
 
 
 
 
1863}
1864
1865static ssize_t show_rxbuf_cur(struct device *dev,
1866			      struct device_attribute *attr, char *buf)
1867{
1868	struct net_device *netdev = to_net_dev(dev);
1869	struct netfront_info *info = netdev_priv(netdev);
1870
1871	return sprintf(buf, "%u\n", info->rx_target);
1872}
1873
1874static struct device_attribute xennet_attrs[] = {
1875	__ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min),
1876	__ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max),
1877	__ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL),
1878};
1879
1880static int xennet_sysfs_addif(struct net_device *netdev)
1881{
1882	int i;
1883	int err;
1884
1885	for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) {
1886		err = device_create_file(&netdev->dev,
1887					   &xennet_attrs[i]);
1888		if (err)
1889			goto fail;
1890	}
1891	return 0;
1892
1893 fail:
1894	while (--i >= 0)
1895		device_remove_file(&netdev->dev, &xennet_attrs[i]);
1896	return err;
1897}
1898
1899static void xennet_sysfs_delif(struct net_device *netdev)
1900{
1901	int i;
1902
1903	for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++)
1904		device_remove_file(&netdev->dev, &xennet_attrs[i]);
1905}
1906
1907#endif /* CONFIG_SYSFS */
1908
1909static struct xenbus_device_id netfront_ids[] = {
1910	{ "vif" },
1911	{ "" }
1912};
1913
1914
1915static int __devexit xennet_remove(struct xenbus_device *dev)
1916{
1917	struct netfront_info *info = dev_get_drvdata(&dev->dev);
1918
1919	dev_dbg(&dev->dev, "%s\n", dev->nodename);
1920
1921	unregister_netdev(info->netdev);
1922
1923	xennet_disconnect_backend(info);
1924
1925	del_timer_sync(&info->rx_refill_timer);
1926
1927	xennet_sysfs_delif(info->netdev);
1928
1929	free_percpu(info->stats);
1930
1931	free_netdev(info->netdev);
1932
1933	return 0;
1934}
1935
1936static struct xenbus_driver netfront_driver = {
1937	.name = "vif",
1938	.owner = THIS_MODULE,
1939	.ids = netfront_ids,
1940	.probe = netfront_probe,
1941	.remove = __devexit_p(xennet_remove),
1942	.resume = netfront_resume,
1943	.otherend_changed = netback_changed,
1944};
1945
1946static int __init netif_init(void)
1947{
1948	if (!xen_domain())
1949		return -ENODEV;
1950
1951	if (xen_initial_domain())
1952		return 0;
 
 
1953
1954	printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n");
 
 
 
 
 
1955
1956	return xenbus_register_frontend(&netfront_driver);
1957}
1958module_init(netif_init);
1959
1960
1961static void __exit netif_exit(void)
1962{
1963	if (xen_initial_domain())
1964		return;
1965
1966	xenbus_unregister_driver(&netfront_driver);
1967}
1968module_exit(netif_exit);
1969
1970MODULE_DESCRIPTION("Xen virtual network device frontend");
1971MODULE_LICENSE("GPL");
1972MODULE_ALIAS("xen:vif");
1973MODULE_ALIAS("xennet");
v6.13.7
   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	skb_mark_for_recycle(skb);
 289
 290	/* Align ip header to a 16 bytes boundary */
 291	skb_reserve(skb, NET_IP_ALIGN);
 292	skb->dev = queue->info->netdev;
 293
 294	return skb;
 295}
 296
 
 
 
 
 
 
 
 
 
 
 
 
 297
 298static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
 299{
 300	RING_IDX req_prod = queue->rx.req_prod_pvt;
 301	int notify;
 302	int err = 0;
 303
 304	if (unlikely(!netif_carrier_ok(queue->info->netdev)))
 
 
 
 305		return;
 
 
 
 
 
 
 
 
 
 
 
 
 306
 307	for (req_prod = queue->rx.req_prod_pvt;
 308	     req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
 309	     req_prod++) {
 310		struct sk_buff *skb;
 311		unsigned short id;
 312		grant_ref_t ref;
 313		struct page *page;
 314		struct xen_netif_rx_request *req;
 315
 316		skb = xennet_alloc_one_rx_buffer(queue);
 317		if (!skb) {
 318			err = -ENOMEM;
 319			break;
 320		}
 321
 322		id = xennet_rxidx(req_prod);
 
 323
 324		BUG_ON(queue->rx_skbs[id]);
 325		queue->rx_skbs[id] = skb;
 
 
 
 326
 327		ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
 328		WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
 329		queue->grant_rx_ref[id] = ref;
 330
 331		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
 332
 333		req = RING_GET_REQUEST(&queue->rx, req_prod);
 334		gnttab_page_grant_foreign_access_ref_one(ref,
 335							 queue->info->xbdev->otherend_id,
 336							 page,
 337							 0);
 338		req->id = id;
 339		req->gref = ref;
 340	}
 341
 342	queue->rx.req_prod_pvt = req_prod;
 343
 344	/* Try again later if there are not enough requests or skb allocation
 345	 * failed.
 346	 * Enough requests is quantified as the sum of newly created slots and
 347	 * the unconsumed slots at the backend.
 348	 */
 349	if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
 350	    unlikely(err)) {
 351		mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
 352		return;
 353	}
 354
 355	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
 356	if (notify)
 357		notify_remote_via_irq(queue->rx_irq);
 358}
 359
 360static int xennet_open(struct net_device *dev)
 361{
 362	struct netfront_info *np = netdev_priv(dev);
 363	unsigned int num_queues = dev->real_num_tx_queues;
 364	unsigned int i = 0;
 365	struct netfront_queue *queue = NULL;
 366
 367	if (!np->queues || np->broken)
 368		return -ENODEV;
 369
 370	for (i = 0; i < num_queues; ++i) {
 371		queue = &np->queues[i];
 372		napi_enable(&queue->napi);
 373
 374		spin_lock_bh(&queue->rx_lock);
 375		if (netif_carrier_ok(dev)) {
 376			xennet_alloc_rx_buffers(queue);
 377			queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
 378			if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
 379				napi_schedule(&queue->napi);
 380		}
 381		spin_unlock_bh(&queue->rx_lock);
 382	}
 
 383
 384	netif_tx_start_all_queues(dev);
 385
 386	return 0;
 387}
 388
 389static bool xennet_tx_buf_gc(struct netfront_queue *queue)
 390{
 391	RING_IDX cons, prod;
 392	unsigned short id;
 
 393	struct sk_buff *skb;
 394	bool more_to_do;
 395	bool work_done = false;
 396	const struct device *dev = &queue->info->netdev->dev;
 397
 398	BUG_ON(!netif_carrier_ok(queue->info->netdev));
 399
 400	do {
 401		prod = queue->tx.sring->rsp_prod;
 402		if (RING_RESPONSE_PROD_OVERFLOW(&queue->tx, prod)) {
 403			dev_alert(dev, "Illegal number of responses %u\n",
 404				  prod - queue->tx.rsp_cons);
 405			goto err;
 406		}
 407		rmb(); /* Ensure we see responses up to 'rp'. */
 408
 409		for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
 410			struct xen_netif_tx_response txrsp;
 411
 412			work_done = true;
 413
 414			RING_COPY_RESPONSE(&queue->tx, cons, &txrsp);
 415			if (txrsp.status == XEN_NETIF_RSP_NULL)
 416				continue;
 417
 418			id = txrsp.id;
 419			if (id >= RING_SIZE(&queue->tx)) {
 420				dev_alert(dev,
 421					  "Response has incorrect id (%u)\n",
 422					  id);
 423				goto err;
 424			}
 425			if (queue->tx_link[id] != TX_PENDING) {
 426				dev_alert(dev,
 427					  "Response for inactive request\n");
 428				goto err;
 429			}
 430
 431			queue->tx_link[id] = TX_LINK_NONE;
 432			skb = queue->tx_skbs[id];
 433			queue->tx_skbs[id] = NULL;
 434			if (unlikely(!gnttab_end_foreign_access_ref(
 435				queue->grant_tx_ref[id]))) {
 436				dev_alert(dev,
 437					  "Grant still in use by backend domain\n");
 438				goto err;
 439			}
 
 
 440			gnttab_release_grant_reference(
 441				&queue->gref_tx_head, queue->grant_tx_ref[id]);
 442			queue->grant_tx_ref[id] = INVALID_GRANT_REF;
 443			queue->grant_tx_page[id] = NULL;
 444			add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, id);
 445			dev_kfree_skb_irq(skb);
 446		}
 447
 448		queue->tx.rsp_cons = prod;
 449
 450		RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
 451	} while (more_to_do);
 452
 453	xennet_maybe_wake_tx(queue);
 
 
 
 
 
 
 
 
 454
 455	return work_done;
 456
 457 err:
 458	queue->info->broken = true;
 459	dev_alert(dev, "Disabled for further use\n");
 460
 461	return work_done;
 462}
 463
 464struct xennet_gnttab_make_txreq {
 465	struct netfront_queue *queue;
 466	struct sk_buff *skb;
 467	struct page *page;
 468	struct xen_netif_tx_request *tx;      /* Last request on ring page */
 469	struct xen_netif_tx_request tx_local; /* Last request local copy*/
 470	unsigned int size;
 471};
 472
 473static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
 474				  unsigned int len, void *data)
 475{
 476	struct xennet_gnttab_make_txreq *info = data;
 
 
 
 
 
 
 477	unsigned int id;
 478	struct xen_netif_tx_request *tx;
 479	grant_ref_t ref;
 480	/* convenient aliases */
 481	struct page *page = info->page;
 482	struct netfront_queue *queue = info->queue;
 483	struct sk_buff *skb = info->skb;
 484
 485	id = get_id_from_list(&queue->tx_skb_freelist, queue->tx_link);
 486	tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
 487	ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
 488	WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
 489
 490	gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
 491					gfn, GNTMAP_readonly);
 492
 493	queue->tx_skbs[id] = skb;
 494	queue->grant_tx_page[id] = page;
 495	queue->grant_tx_ref[id] = ref;
 496
 497	info->tx_local.id = id;
 498	info->tx_local.gref = ref;
 499	info->tx_local.offset = offset;
 500	info->tx_local.size = len;
 501	info->tx_local.flags = 0;
 502
 503	*tx = info->tx_local;
 504
 505	/*
 506	 * Put the request in the pending queue, it will be set to be pending
 507	 * when the producer index is about to be raised.
 508	 */
 509	add_id_to_list(&queue->tx_pend_queue, queue->tx_link, id);
 510
 511	info->tx = tx;
 512	info->size += info->tx_local.size;
 513}
 514
 515static struct xen_netif_tx_request *xennet_make_first_txreq(
 516	struct xennet_gnttab_make_txreq *info,
 517	unsigned int offset, unsigned int len)
 518{
 519	info->size = 0;
 520
 521	gnttab_for_one_grant(info->page, offset, len, xennet_tx_setup_grant, info);
 522
 523	return info->tx;
 524}
 525
 526static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
 527				  unsigned int len, void *data)
 528{
 529	struct xennet_gnttab_make_txreq *info = data;
 530
 531	info->tx->flags |= XEN_NETTXF_more_data;
 532	skb_get(info->skb);
 533	xennet_tx_setup_grant(gfn, offset, len, data);
 534}
 535
 536static void xennet_make_txreqs(
 537	struct xennet_gnttab_make_txreq *info,
 538	struct page *page,
 539	unsigned int offset, unsigned int len)
 540{
 541	/* Skip unused frames from start of page */
 542	page += offset >> PAGE_SHIFT;
 543	offset &= ~PAGE_MASK;
 544
 545	while (len) {
 546		info->page = page;
 547		info->size = 0;
 548
 549		gnttab_foreach_grant_in_range(page, offset, len,
 550					      xennet_make_one_txreq,
 551					      info);
 552
 553		page++;
 554		offset = 0;
 555		len -= info->size;
 
 
 
 
 
 
 
 
 556	}
 557}
 558
 559/*
 560 * Count how many ring slots are required to send this skb. Each frag
 561 * might be a compound page.
 562 */
 563static int xennet_count_skb_slots(struct sk_buff *skb)
 564{
 565	int i, frags = skb_shinfo(skb)->nr_frags;
 566	int slots;
 567
 568	slots = gnttab_count_grant(offset_in_page(skb->data),
 569				   skb_headlen(skb));
 570
 
 571	for (i = 0; i < frags; i++) {
 572		skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 573		unsigned long size = skb_frag_size(frag);
 574		unsigned long offset = skb_frag_off(frag);
 575
 576		/* Skip unused frames from start of page */
 577		offset &= ~PAGE_MASK;
 578
 579		slots += gnttab_count_grant(offset, size);
 580	}
 581
 582	return slots;
 583}
 
 584
 585static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
 586			       struct net_device *sb_dev)
 587{
 588	unsigned int num_queues = dev->real_num_tx_queues;
 589	u32 hash;
 590	u16 queue_idx;
 591
 592	/* First, check if there is only one queue */
 593	if (num_queues == 1) {
 594		queue_idx = 0;
 595	} else {
 596		hash = skb_get_hash(skb);
 597		queue_idx = hash % num_queues;
 598	}
 599
 600	return queue_idx;
 601}
 602
 603static void xennet_mark_tx_pending(struct netfront_queue *queue)
 604{
 605	unsigned int i;
 606
 607	while ((i = get_id_from_list(&queue->tx_pend_queue, queue->tx_link)) !=
 608	       TX_LINK_NONE)
 609		queue->tx_link[i] = TX_PENDING;
 610}
 611
 612static int xennet_xdp_xmit_one(struct net_device *dev,
 613			       struct netfront_queue *queue,
 614			       struct xdp_frame *xdpf)
 615{
 
 616	struct netfront_info *np = netdev_priv(dev);
 617	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
 618	struct xennet_gnttab_make_txreq info = {
 619		.queue = queue,
 620		.skb = NULL,
 621		.page = virt_to_page(xdpf->data),
 622	};
 
 623	int notify;
 624
 625	xennet_make_first_txreq(&info,
 626				offset_in_page(xdpf->data),
 627				xdpf->len);
 628
 629	xennet_mark_tx_pending(queue);
 630
 631	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
 632	if (notify)
 633		notify_remote_via_irq(queue->tx_irq);
 634
 635	u64_stats_update_begin(&tx_stats->syncp);
 636	tx_stats->bytes += xdpf->len;
 637	tx_stats->packets++;
 638	u64_stats_update_end(&tx_stats->syncp);
 639
 640	xennet_tx_buf_gc(queue);
 641
 642	return 0;
 643}
 644
 645static int xennet_xdp_xmit(struct net_device *dev, int n,
 646			   struct xdp_frame **frames, u32 flags)
 647{
 648	unsigned int num_queues = dev->real_num_tx_queues;
 649	struct netfront_info *np = netdev_priv(dev);
 650	struct netfront_queue *queue = NULL;
 651	unsigned long irq_flags;
 652	int nxmit = 0;
 653	int i;
 654
 655	if (unlikely(np->broken))
 656		return -ENODEV;
 657	if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
 658		return -EINVAL;
 659
 660	queue = &np->queues[smp_processor_id() % num_queues];
 661
 662	spin_lock_irqsave(&queue->tx_lock, irq_flags);
 663	for (i = 0; i < n; i++) {
 664		struct xdp_frame *xdpf = frames[i];
 665
 666		if (!xdpf)
 667			continue;
 668		if (xennet_xdp_xmit_one(dev, queue, xdpf))
 669			break;
 670		nxmit++;
 671	}
 672	spin_unlock_irqrestore(&queue->tx_lock, irq_flags);
 673
 674	return nxmit;
 675}
 676
 677static struct sk_buff *bounce_skb(const struct sk_buff *skb)
 678{
 679	unsigned int headerlen = skb_headroom(skb);
 680	/* Align size to allocate full pages and avoid contiguous data leaks */
 681	unsigned int size = ALIGN(skb_end_offset(skb) + skb->data_len,
 682				  XEN_PAGE_SIZE);
 683	struct sk_buff *n = alloc_skb(size, GFP_ATOMIC | __GFP_ZERO);
 684
 685	if (!n)
 686		return NULL;
 687
 688	if (!IS_ALIGNED((uintptr_t)n->head, XEN_PAGE_SIZE)) {
 689		WARN_ONCE(1, "misaligned skb allocated\n");
 690		kfree_skb(n);
 691		return NULL;
 692	}
 693
 694	/* Set the data pointer */
 695	skb_reserve(n, headerlen);
 696	/* Set the tail pointer and length */
 697	skb_put(n, skb->len);
 698
 699	BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
 700
 701	skb_copy_header(n, skb);
 702	return n;
 703}
 704
 705#define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
 706
 707static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
 708{
 709	struct netfront_info *np = netdev_priv(dev);
 710	struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
 711	struct xen_netif_tx_request *first_tx;
 712	unsigned int i;
 713	int notify;
 714	int slots;
 715	struct page *page;
 716	unsigned int offset;
 717	unsigned int len;
 718	unsigned long flags;
 719	struct netfront_queue *queue = NULL;
 720	struct xennet_gnttab_make_txreq info = { };
 721	unsigned int num_queues = dev->real_num_tx_queues;
 722	u16 queue_index;
 723	struct sk_buff *nskb;
 724
 725	/* Drop the packet if no queues are set up */
 726	if (num_queues < 1)
 727		goto drop;
 728	if (unlikely(np->broken))
 729		goto drop;
 730	/* Determine which queue to transmit this SKB on */
 731	queue_index = skb_get_queue_mapping(skb);
 732	queue = &np->queues[queue_index];
 733
 734	/* If skb->len is too big for wire format, drop skb and alert
 735	 * user about misconfiguration.
 736	 */
 737	if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
 738		net_alert_ratelimited(
 739			"xennet: skb->len = %u, too big for wire format\n",
 740			skb->len);
 741		goto drop;
 742	}
 743
 744	slots = xennet_count_skb_slots(skb);
 745	if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
 746		net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
 747				    slots, skb->len);
 748		if (skb_linearize(skb))
 749			goto drop;
 750	}
 751
 752	page = virt_to_page(skb->data);
 753	offset = offset_in_page(skb->data);
 754
 755	/* The first req should be at least ETH_HLEN size or the packet will be
 756	 * dropped by netback.
 757	 *
 758	 * If the backend is not trusted bounce all data to zeroed pages to
 759	 * avoid exposing contiguous data on the granted page not belonging to
 760	 * the skb.
 761	 */
 762	if (np->bounce || unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
 763		nskb = bounce_skb(skb);
 764		if (!nskb)
 765			goto drop;
 766		dev_consume_skb_any(skb);
 767		skb = nskb;
 768		page = virt_to_page(skb->data);
 769		offset = offset_in_page(skb->data);
 770	}
 771
 772	len = skb_headlen(skb);
 
 773
 774	spin_lock_irqsave(&queue->tx_lock, flags);
 775
 776	if (unlikely(!netif_carrier_ok(dev) ||
 777		     (slots > 1 && !xennet_can_sg(dev)) ||
 778		     netif_needs_gso(skb, netif_skb_features(skb)))) {
 779		spin_unlock_irqrestore(&queue->tx_lock, flags);
 780		goto drop;
 781	}
 782
 783	/* First request for the linear area. */
 784	info.queue = queue;
 785	info.skb = skb;
 786	info.page = page;
 787	first_tx = xennet_make_first_txreq(&info, offset, len);
 788	offset += info.tx_local.size;
 789	if (offset == PAGE_SIZE) {
 790		page++;
 791		offset = 0;
 792	}
 793	len -= info.tx_local.size;
 794
 
 795	if (skb->ip_summed == CHECKSUM_PARTIAL)
 796		/* local packet? */
 797		first_tx->flags |= XEN_NETTXF_csum_blank |
 798				   XEN_NETTXF_data_validated;
 799	else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
 800		/* remote but checksummed. */
 801		first_tx->flags |= XEN_NETTXF_data_validated;
 802
 803	/* Optional extra info after the first request. */
 804	if (skb_shinfo(skb)->gso_size) {
 805		struct xen_netif_extra_info *gso;
 806
 807		gso = (struct xen_netif_extra_info *)
 808			RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
 809
 810		first_tx->flags |= XEN_NETTXF_extra_info;
 
 
 
 811
 812		gso->u.gso.size = skb_shinfo(skb)->gso_size;
 813		gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
 814			XEN_NETIF_GSO_TYPE_TCPV6 :
 815			XEN_NETIF_GSO_TYPE_TCPV4;
 816		gso->u.gso.pad = 0;
 817		gso->u.gso.features = 0;
 818
 819		gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
 820		gso->flags = 0;
 
 821	}
 822
 823	/* Requests for the rest of the linear area. */
 824	xennet_make_txreqs(&info, page, offset, len);
 825
 826	/* Requests for all the frags. */
 827	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 828		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 829		xennet_make_txreqs(&info, skb_frag_page(frag),
 830					skb_frag_off(frag),
 831					skb_frag_size(frag));
 832	}
 833
 834	/* First request has the packet length. */
 835	first_tx->size = skb->len;
 836
 837	/* timestamp packet in software */
 838	skb_tx_timestamp(skb);
 839
 840	xennet_mark_tx_pending(queue);
 
 841
 842	RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
 843	if (notify)
 844		notify_remote_via_irq(queue->tx_irq);
 845
 846	u64_stats_update_begin(&tx_stats->syncp);
 847	tx_stats->bytes += skb->len;
 848	tx_stats->packets++;
 849	u64_stats_update_end(&tx_stats->syncp);
 850
 851	/* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
 852	xennet_tx_buf_gc(queue);
 853
 854	if (!netfront_tx_slot_available(queue))
 855		netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
 856
 857	spin_unlock_irqrestore(&queue->tx_lock, flags);
 858
 859	return NETDEV_TX_OK;
 860
 861 drop:
 862	dev->stats.tx_dropped++;
 863	dev_kfree_skb_any(skb);
 864	return NETDEV_TX_OK;
 865}
 866
 867static int xennet_close(struct net_device *dev)
 868{
 869	struct netfront_info *np = netdev_priv(dev);
 870	unsigned int num_queues = np->queues ? dev->real_num_tx_queues : 0;
 871	unsigned int i;
 872	struct netfront_queue *queue;
 873	netif_tx_stop_all_queues(np->netdev);
 874	for (i = 0; i < num_queues; ++i) {
 875		queue = &np->queues[i];
 876		napi_disable(&queue->napi);
 877	}
 878	return 0;
 879}
 880
 881static void xennet_destroy_queues(struct netfront_info *info)
 882{
 883	unsigned int i;
 884
 885	if (!info->queues)
 886		return;
 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
 900static void xennet_uninit(struct net_device *dev)
 901{
 902	struct netfront_info *np = netdev_priv(dev);
 903	xennet_destroy_queues(np);
 904}
 905
 906static 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 = XEN_RING_NR_UNCONSUMED_RESPONSES(&queue->rx);
 913	spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
 914}
 915
 916static 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
 929static 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
 971static 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_init_buff(xdp, XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
 981		      &queue->xdp_rxq);
 982	xdp_prepare_buff(xdp, page_address(pdata), XDP_PACKET_HEADROOM,
 983			 len, false);
 984
 985	act = bpf_prog_run_xdp(prog, xdp);
 986	switch (act) {
 987	case XDP_TX:
 988		get_page(pdata);
 989		xdpf = xdp_convert_buff_to_frame(xdp);
 990		err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0);
 991		if (unlikely(!err))
 992			xdp_return_frame_rx_napi(xdpf);
 993		else 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(queue->info->netdev, prog, act);
1013	}
1014
1015	return act;
1016}
1017
1018static 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		/*
1051		 * This definitely indicates a bug, either in this driver or in
1052		 * the backend driver. In future this should flag the bad
1053		 * situation to the system controller to reboot the backend.
1054		 */
1055		if (ref == INVALID_GRANT_REF) {
1056			if (net_ratelimit())
1057				dev_warn(dev, "Bad rx response id %d.\n",
1058					 rx->id);
1059			err = -EINVAL;
1060			goto next;
1061		}
1062
1063		if (unlikely(rx->status < 0 ||
1064			     rx->offset + rx->status > XEN_PAGE_SIZE)) {
1065			if (net_ratelimit())
1066				dev_warn(dev, "rx->offset: %u, size: %d\n",
1067					 rx->offset, rx->status);
1068			xennet_move_rx_slot(queue, skb, ref);
1069			err = -EINVAL;
1070			goto next;
1071		}
1072
1073		if (!gnttab_end_foreign_access_ref(ref)) {
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
1102next:
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
1132static 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
1161static 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
1200static 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
1224static 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
1260static 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			}
1306err:
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
1376static 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	WRITE_ONCE(dev->mtu, mtu);
1383	return 0;
1384}
1385
1386static 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(&tx_stats->syncp);
1400			tx_packets = tx_stats->packets;
1401			tx_bytes = tx_stats->bytes;
1402		} while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1403
1404		do {
1405			start = u64_stats_fetch_begin(&rx_stats->syncp);
1406			rx_packets = rx_stats->packets;
1407			rx_bytes = rx_stats->bytes;
1408		} while (u64_stats_fetch_retry(&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
1420static 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					  queue->grant_tx_page[i]);
1435		queue->grant_tx_page[i] = NULL;
1436		queue->grant_tx_ref[i] = INVALID_GRANT_REF;
1437		add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, i);
1438		dev_kfree_skb_irq(skb);
1439	}
1440}
1441
1442static void xennet_release_rx_bufs(struct netfront_queue *queue)
1443{
 
 
 
 
 
 
1444	int id, ref;
1445
1446	spin_lock_bh(&queue->rx_lock);
 
 
 
 
 
 
1447
1448	for (id = 0; id < NET_RX_RING_SIZE; id++) {
1449		struct sk_buff *skb;
1450		struct page *page;
1451
1452		skb = queue->rx_skbs[id];
1453		if (!skb)
1454			continue;
 
1455
1456		ref = queue->grant_rx_ref[id];
1457		if (ref == INVALID_GRANT_REF)
 
 
 
 
 
 
 
1458			continue;
 
1459
1460		page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
 
 
 
 
1461
1462		/* gnttab_end_foreign_access() needs a page ref until
1463		 * foreign access is ended (which may be deferred).
1464		 */
1465		get_page(page);
1466		gnttab_end_foreign_access(ref, page);
1467		queue->grant_rx_ref[id] = INVALID_GRANT_REF;
 
 
1468
1469		kfree_skb(skb);
 
 
 
1470	}
1471
1472	spin_unlock_bh(&queue->rx_lock);
1473}
1474
1475static netdev_features_t xennet_fix_features(struct net_device *dev,
1476	netdev_features_t features)
1477{
1478	struct netfront_info *np = netdev_priv(dev);
1479
1480	if (features & NETIF_F_SG &&
1481	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1482		features &= ~NETIF_F_SG;
1483
1484	if (features & NETIF_F_IPV6_CSUM &&
1485	    !xenbus_read_unsigned(np->xbdev->otherend,
1486				  "feature-ipv6-csum-offload", 0))
1487		features &= ~NETIF_F_IPV6_CSUM;
1488
1489	if (features & NETIF_F_TSO &&
1490	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1491		features &= ~NETIF_F_TSO;
1492
1493	if (features & NETIF_F_TSO6 &&
1494	    !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1495		features &= ~NETIF_F_TSO6;
1496
1497	return features;
1498}
1499
1500static int xennet_set_features(struct net_device *dev,
1501	netdev_features_t features)
1502{
1503	if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1504		netdev_info(dev, "Reducing MTU because no SG offload");
1505		dev->mtu = ETH_DATA_LEN;
1506	}
1507
1508	return 0;
1509}
1510
1511static bool xennet_handle_tx(struct netfront_queue *queue, unsigned int *eoi)
1512{
1513	unsigned long flags;
1514
1515	if (unlikely(queue->info->broken))
1516		return false;
1517
1518	spin_lock_irqsave(&queue->tx_lock, flags);
1519	if (xennet_tx_buf_gc(queue))
1520		*eoi = 0;
1521	spin_unlock_irqrestore(&queue->tx_lock, flags);
1522
1523	return true;
1524}
1525
1526static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1527{
1528	unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1529
1530	if (likely(xennet_handle_tx(dev_id, &eoiflag)))
1531		xen_irq_lateeoi(irq, eoiflag);
1532
1533	return IRQ_HANDLED;
1534}
1535
1536static bool xennet_handle_rx(struct netfront_queue *queue, unsigned int *eoi)
1537{
1538	unsigned int work_queued;
1539	unsigned long flags;
1540
1541	if (unlikely(queue->info->broken))
1542		return false;
1543
1544	spin_lock_irqsave(&queue->rx_cons_lock, flags);
1545	work_queued = XEN_RING_NR_UNCONSUMED_RESPONSES(&queue->rx);
1546	if (work_queued > queue->rx_rsp_unconsumed) {
1547		queue->rx_rsp_unconsumed = work_queued;
1548		*eoi = 0;
1549	} else if (unlikely(work_queued < queue->rx_rsp_unconsumed)) {
1550		const struct device *dev = &queue->info->netdev->dev;
1551
1552		spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
1553		dev_alert(dev, "RX producer index going backwards\n");
1554		dev_alert(dev, "Disabled for further use\n");
1555		queue->info->broken = true;
1556		return false;
1557	}
1558	spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
1559
1560	if (likely(netif_carrier_ok(queue->info->netdev) && work_queued))
1561		napi_schedule(&queue->napi);
 
 
1562
1563	return true;
1564}
 
1565
1566static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1567{
1568	unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1569
1570	if (likely(xennet_handle_rx(dev_id, &eoiflag)))
1571		xen_irq_lateeoi(irq, eoiflag);
1572
1573	return IRQ_HANDLED;
1574}
1575
1576static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1577{
1578	unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1579
1580	if (xennet_handle_tx(dev_id, &eoiflag) &&
1581	    xennet_handle_rx(dev_id, &eoiflag))
1582		xen_irq_lateeoi(irq, eoiflag);
1583
1584	return IRQ_HANDLED;
1585}
1586
1587#ifdef CONFIG_NET_POLL_CONTROLLER
1588static void xennet_poll_controller(struct net_device *dev)
1589{
1590	/* Poll each queue */
1591	struct netfront_info *info = netdev_priv(dev);
1592	unsigned int num_queues = dev->real_num_tx_queues;
1593	unsigned int i;
1594
1595	if (info->broken)
1596		return;
1597
1598	for (i = 0; i < num_queues; ++i)
1599		xennet_interrupt(0, &info->queues[i]);
1600}
1601#endif
1602
1603#define NETBACK_XDP_HEADROOM_DISABLE	0
1604#define NETBACK_XDP_HEADROOM_ENABLE	1
1605
1606static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
1607{
1608	int err;
1609	unsigned short headroom;
1610
1611	headroom = xdp ? XDP_PACKET_HEADROOM : 0;
1612	err = xenbus_printf(XBT_NIL, np->xbdev->nodename,
1613			    "xdp-headroom", "%hu",
1614			    headroom);
1615	if (err)
1616		pr_warn("Error writing xdp-headroom\n");
1617
1618	return err;
1619}
1620
1621static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1622			  struct netlink_ext_ack *extack)
1623{
1624	unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
1625	struct netfront_info *np = netdev_priv(dev);
1626	struct bpf_prog *old_prog;
1627	unsigned int i, err;
1628
1629	if (dev->mtu > max_mtu) {
1630		netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu);
1631		return -EINVAL;
1632	}
1633
1634	if (!np->netback_has_xdp_headroom)
1635		return 0;
1636
1637	xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);
1638
1639	err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE :
1640				  NETBACK_XDP_HEADROOM_DISABLE);
1641	if (err)
1642		return err;
1643
1644	/* avoid the race with XDP headroom adjustment */
1645	wait_event(module_wq,
1646		   xenbus_read_driver_state(np->xbdev->otherend) ==
1647		   XenbusStateReconfigured);
1648	np->netfront_xdp_enabled = true;
1649
1650	old_prog = rtnl_dereference(np->queues[0].xdp_prog);
1651
1652	if (prog)
1653		bpf_prog_add(prog, dev->real_num_tx_queues);
1654
1655	for (i = 0; i < dev->real_num_tx_queues; ++i)
1656		rcu_assign_pointer(np->queues[i].xdp_prog, prog);
1657
1658	if (old_prog)
1659		for (i = 0; i < dev->real_num_tx_queues; ++i)
1660			bpf_prog_put(old_prog);
1661
1662	xenbus_switch_state(np->xbdev, XenbusStateConnected);
1663
1664	return 0;
1665}
1666
1667static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1668{
1669	struct netfront_info *np = netdev_priv(dev);
1670
1671	if (np->broken)
1672		return -ENODEV;
1673
1674	switch (xdp->command) {
1675	case XDP_SETUP_PROG:
1676		return xennet_xdp_set(dev, xdp->prog, xdp->extack);
1677	default:
1678		return -EINVAL;
1679	}
1680}
1681
1682static const struct net_device_ops xennet_netdev_ops = {
 
1683	.ndo_uninit          = xennet_uninit,
1684	.ndo_open            = xennet_open,
1685	.ndo_stop            = xennet_close,
1686	.ndo_start_xmit      = xennet_start_xmit,
1687	.ndo_change_mtu	     = xennet_change_mtu,
1688	.ndo_get_stats64     = xennet_get_stats64,
1689	.ndo_set_mac_address = eth_mac_addr,
1690	.ndo_validate_addr   = eth_validate_addr,
1691	.ndo_fix_features    = xennet_fix_features,
1692	.ndo_set_features    = xennet_set_features,
1693	.ndo_select_queue    = xennet_select_queue,
1694	.ndo_bpf            = xennet_xdp,
1695	.ndo_xdp_xmit	    = xennet_xdp_xmit,
1696#ifdef CONFIG_NET_POLL_CONTROLLER
1697	.ndo_poll_controller = xennet_poll_controller,
1698#endif
1699};
1700
1701static void xennet_free_netdev(struct net_device *netdev)
1702{
1703	struct netfront_info *np = netdev_priv(netdev);
1704
1705	free_percpu(np->rx_stats);
1706	free_percpu(np->tx_stats);
1707	free_netdev(netdev);
1708}
1709
1710static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1711{
1712	int err;
1713	struct net_device *netdev;
1714	struct netfront_info *np;
1715
1716	netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1717	if (!netdev)
 
 
1718		return ERR_PTR(-ENOMEM);
 
1719
1720	np                   = netdev_priv(netdev);
1721	np->xbdev            = dev;
1722
1723	np->queues = NULL;
 
 
 
 
 
 
 
 
 
 
1724
1725	err = -ENOMEM;
1726	np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1727	if (np->rx_stats == NULL)
1728		goto exit;
1729	np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1730	if (np->tx_stats == NULL)
1731		goto exit;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1732
1733	netdev->netdev_ops	= &xennet_netdev_ops;
1734
 
1735	netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1736				  NETIF_F_GSO_ROBUST;
1737	netdev->hw_features	= NETIF_F_SG |
1738				  NETIF_F_IPV6_CSUM |
1739				  NETIF_F_TSO | NETIF_F_TSO6;
1740
1741	/*
1742         * Assume that all hw features are available for now. This set
1743         * will be adjusted by the call to netdev_update_features() in
1744         * xennet_connect() which is the earliest point where we can
1745         * negotiate with the backend regarding supported features.
1746         */
1747	netdev->features |= netdev->hw_features;
1748	netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
1749			       NETDEV_XDP_ACT_NDO_XMIT;
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 */
1782static 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
1805static void xennet_end_access(int ref, void *page)
1806{
1807	/* This frees the page as a side-effect */
1808	if (ref != INVALID_GRANT_REF)
1809		gnttab_end_foreign_access(ref, virt_to_page(page));
1810}
1811
1812static 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 = INVALID_GRANT_REF;
1846		queue->rx_ring_ref = INVALID_GRANT_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 */
1860static 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
1880static 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
1902static 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
1921bind_fail:
1922	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1923	queue->tx_evtchn = 0;
1924fail:
1925	return err;
1926}
1927
1928static 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
1959bind_rx_fail:
1960	unbind_from_irqhandler(queue->tx_irq, queue);
1961	queue->tx_irq = 0;
1962bind_tx_fail:
1963	xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1964	queue->rx_evtchn = 0;
1965alloc_rx_evtchn_fail:
1966	xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1967	queue->tx_evtchn = 0;
1968fail:
1969	return err;
1970}
1971
1972static 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;
1977	int err;
 
1978
1979	queue->tx_ring_ref = INVALID_GRANT_REF;
1980	queue->rx_ring_ref = INVALID_GRANT_REF;
1981	queue->rx.sring = NULL;
1982	queue->tx.sring = NULL;
 
1983
1984	err = xenbus_setup_ring(dev, GFP_NOIO | __GFP_HIGH, (void **)&txs,
1985				1, &queue->tx_ring_ref);
1986	if (err)
1987		goto fail;
 
1988
1989	XEN_FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1990
1991	err = xenbus_setup_ring(dev, GFP_NOIO | __GFP_HIGH, (void **)&rxs,
1992				1, &queue->rx_ring_ref);
1993	if (err)
1994		goto fail;
 
 
 
1995
1996	XEN_FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
1997
1998	if (feature_split_evtchn)
1999		err = setup_netfront_split(queue);
2000	/* setup single event channel if
2001	 *  a) feature-split-event-channels == 0
2002	 *  b) feature-split-event-channels == 1 but failed to setup
2003	 */
2004	if (!feature_split_evtchn || err)
2005		err = setup_netfront_single(queue);
2006
2007	if (err)
2008		goto fail;
2009
2010	return 0;
2011
2012 fail:
2013	xenbus_teardown_ring((void **)&queue->rx.sring, 1, &queue->rx_ring_ref);
2014	xenbus_teardown_ring((void **)&queue->tx.sring, 1, &queue->tx_ring_ref);
2015
2016	return err;
2017}
2018
2019/* Queue-specific initialisation
2020 * This used to be done in xennet_create_dev() but must now
2021 * be run per-queue.
2022 */
2023static int xennet_init_queue(struct netfront_queue *queue)
2024{
2025	unsigned short i;
2026	int err = 0;
2027	char *devid;
2028
2029	spin_lock_init(&queue->tx_lock);
2030	spin_lock_init(&queue->rx_lock);
2031	spin_lock_init(&queue->rx_cons_lock);
2032
2033	timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
2034
2035	devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
2036	snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
2037		 devid, queue->id);
2038
2039	/* Initialise tx_skb_freelist as a free chain containing every entry. */
2040	queue->tx_skb_freelist = 0;
2041	queue->tx_pend_queue = TX_LINK_NONE;
2042	for (i = 0; i < NET_TX_RING_SIZE; i++) {
2043		queue->tx_link[i] = i + 1;
2044		queue->grant_tx_ref[i] = INVALID_GRANT_REF;
2045		queue->grant_tx_page[i] = NULL;
2046	}
2047	queue->tx_link[NET_TX_RING_SIZE - 1] = TX_LINK_NONE;
2048
2049	/* Clear out rx_skbs */
2050	for (i = 0; i < NET_RX_RING_SIZE; i++) {
2051		queue->rx_skbs[i] = NULL;
2052		queue->grant_rx_ref[i] = INVALID_GRANT_REF;
2053	}
2054
2055	/* A grant for every tx ring slot */
2056	if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2057					  &queue->gref_tx_head) < 0) {
2058		pr_alert("can't alloc tx grant refs\n");
2059		err = -ENOMEM;
2060		goto exit;
 
2061	}
 
 
2062
2063	/* A grant for every rx ring slot */
2064	if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
2065					  &queue->gref_rx_head) < 0) {
2066		pr_alert("can't alloc rx grant refs\n");
2067		err = -ENOMEM;
2068		goto exit_free_tx;
2069	}
 
2070
2071	return 0;
 
 
2072
2073 exit_free_tx:
2074	gnttab_free_grant_references(queue->gref_tx_head);
2075 exit:
2076	return err;
2077}
2078
2079static int write_queue_xenstore_keys(struct netfront_queue *queue,
2080			   struct xenbus_transaction *xbt, int write_hierarchical)
2081{
2082	/* Write the queue-specific keys into XenStore in the traditional
2083	 * way for a single queue, or in a queue subkeys for multiple
2084	 * queues.
2085	 */
2086	struct xenbus_device *dev = queue->info->xbdev;
2087	int err;
2088	const char *message;
2089	char *path;
2090	size_t pathsize;
2091
2092	/* Choose the correct place to write the keys */
2093	if (write_hierarchical) {
2094		pathsize = strlen(dev->nodename) + 10;
2095		path = kzalloc(pathsize, GFP_KERNEL);
2096		if (!path) {
2097			err = -ENOMEM;
2098			message = "out of memory while writing ring references";
2099			goto error;
2100		}
2101		snprintf(path, pathsize, "%s/queue-%u",
2102				dev->nodename, queue->id);
2103	} else {
2104		path = (char *)dev->nodename;
2105	}
2106
2107	/* Write ring references */
2108	err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
2109			queue->tx_ring_ref);
2110	if (err) {
2111		message = "writing tx-ring-ref";
2112		goto error;
2113	}
2114
2115	err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
2116			queue->rx_ring_ref);
2117	if (err) {
2118		message = "writing rx-ring-ref";
2119		goto error;
2120	}
2121
2122	/* Write event channels; taking into account both shared
2123	 * and split event channel scenarios.
2124	 */
2125	if (queue->tx_evtchn == queue->rx_evtchn) {
2126		/* Shared event channel */
2127		err = xenbus_printf(*xbt, path,
2128				"event-channel", "%u", queue->tx_evtchn);
2129		if (err) {
2130			message = "writing event-channel";
2131			goto error;
2132		}
2133	} else {
2134		/* Split event channels */
2135		err = xenbus_printf(*xbt, path,
2136				"event-channel-tx", "%u", queue->tx_evtchn);
2137		if (err) {
2138			message = "writing event-channel-tx";
2139			goto error;
2140		}
2141
2142		err = xenbus_printf(*xbt, path,
2143				"event-channel-rx", "%u", queue->rx_evtchn);
2144		if (err) {
2145			message = "writing event-channel-rx";
2146			goto error;
2147		}
2148	}
2149
2150	if (write_hierarchical)
2151		kfree(path);
2152	return 0;
2153
2154error:
2155	if (write_hierarchical)
2156		kfree(path);
2157	xenbus_dev_fatal(dev, err, "%s", message);
2158	return err;
2159}
2160
2161
2162
2163static int xennet_create_page_pool(struct netfront_queue *queue)
2164{
2165	int err;
2166	struct page_pool_params pp_params = {
2167		.order = 0,
2168		.flags = 0,
2169		.pool_size = NET_RX_RING_SIZE,
2170		.nid = NUMA_NO_NODE,
2171		.dev = &queue->info->netdev->dev,
2172		.offset = XDP_PACKET_HEADROOM,
2173		.max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
2174	};
2175
2176	queue->page_pool = page_pool_create(&pp_params);
2177	if (IS_ERR(queue->page_pool)) {
2178		err = PTR_ERR(queue->page_pool);
2179		queue->page_pool = NULL;
2180		return err;
2181	}
2182
2183	err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev,
2184			       queue->id, 0);
2185	if (err) {
2186		netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n");
2187		goto err_free_pp;
2188	}
2189
2190	err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq,
2191					 MEM_TYPE_PAGE_POOL, queue->page_pool);
2192	if (err) {
2193		netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n");
2194		goto err_unregister_rxq;
2195	}
2196	return 0;
2197
2198err_unregister_rxq:
2199	xdp_rxq_info_unreg(&queue->xdp_rxq);
2200err_free_pp:
2201	page_pool_destroy(queue->page_pool);
2202	queue->page_pool = NULL;
2203	return err;
2204}
2205
2206static int xennet_create_queues(struct netfront_info *info,
2207				unsigned int *num_queues)
2208{
2209	unsigned int i;
2210	int ret;
2211
2212	info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
2213			       GFP_KERNEL);
2214	if (!info->queues)
2215		return -ENOMEM;
2216
2217	for (i = 0; i < *num_queues; i++) {
2218		struct netfront_queue *queue = &info->queues[i];
2219
2220		queue->id = i;
2221		queue->info = info;
2222
2223		ret = xennet_init_queue(queue);
2224		if (ret < 0) {
2225			dev_warn(&info->xbdev->dev,
2226				 "only created %d queues\n", i);
2227			*num_queues = i;
2228			break;
2229		}
2230
2231		/* use page pool recycling instead of buddy allocator */
2232		ret = xennet_create_page_pool(queue);
2233		if (ret < 0) {
2234			dev_err(&info->xbdev->dev, "can't allocate page pool\n");
2235			*num_queues = i;
2236			return ret;
2237		}
2238
2239		netif_napi_add(queue->info->netdev, &queue->napi, xennet_poll);
2240		if (netif_running(info->netdev))
2241			napi_enable(&queue->napi);
2242	}
2243
2244	netif_set_real_num_tx_queues(info->netdev, *num_queues);
2245
2246	if (*num_queues == 0) {
2247		dev_err(&info->xbdev->dev, "no queues\n");
2248		return -EINVAL;
2249	}
2250	return 0;
2251}
2252
2253/* Common code used when first setting up, and when resuming. */
2254static int talk_to_netback(struct xenbus_device *dev,
2255			   struct netfront_info *info)
2256{
2257	const char *message;
2258	struct xenbus_transaction xbt;
2259	int err;
2260	unsigned int feature_split_evtchn;
2261	unsigned int i = 0;
2262	unsigned int max_queues = 0;
2263	struct netfront_queue *queue = NULL;
2264	unsigned int num_queues = 1;
2265	u8 addr[ETH_ALEN];
2266
2267	info->netdev->irq = 0;
2268
2269	/* Check if backend is trusted. */
2270	info->bounce = !xennet_trusted ||
2271		       !xenbus_read_unsigned(dev->nodename, "trusted", 1);
2272
2273	/* Check if backend supports multiple queues */
2274	max_queues = xenbus_read_unsigned(info->xbdev->otherend,
2275					  "multi-queue-max-queues", 1);
2276	num_queues = min(max_queues, xennet_max_queues);
2277
2278	/* Check feature-split-event-channels */
2279	feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
2280					"feature-split-event-channels", 0);
2281
2282	/* Read mac addr. */
2283	err = xen_net_read_mac(dev, addr);
2284	if (err) {
2285		xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
2286		goto out_unlocked;
2287	}
2288	eth_hw_addr_set(info->netdev, addr);
2289
2290	info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend,
2291							      "feature-xdp-headroom", 0);
2292	if (info->netback_has_xdp_headroom) {
2293		/* set the current xen-netfront xdp state */
2294		err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ?
2295					  NETBACK_XDP_HEADROOM_ENABLE :
2296					  NETBACK_XDP_HEADROOM_DISABLE);
2297		if (err)
2298			goto out_unlocked;
2299	}
2300
2301	rtnl_lock();
2302	if (info->queues)
2303		xennet_destroy_queues(info);
2304
2305	/* For the case of a reconnect reset the "broken" indicator. */
2306	info->broken = false;
2307
2308	err = xennet_create_queues(info, &num_queues);
2309	if (err < 0) {
2310		xenbus_dev_fatal(dev, err, "creating queues");
2311		kfree(info->queues);
2312		info->queues = NULL;
2313		goto out;
2314	}
2315	rtnl_unlock();
2316
2317	/* Create shared ring, alloc event channel -- for each queue */
2318	for (i = 0; i < num_queues; ++i) {
2319		queue = &info->queues[i];
2320		err = setup_netfront(dev, queue, feature_split_evtchn);
2321		if (err)
2322			goto destroy_ring;
2323	}
2324
2325again:
2326	err = xenbus_transaction_start(&xbt);
2327	if (err) {
2328		xenbus_dev_fatal(dev, err, "starting transaction");
2329		goto destroy_ring;
2330	}
2331
2332	if (xenbus_exists(XBT_NIL,
2333			  info->xbdev->otherend, "multi-queue-max-queues")) {
2334		/* Write the number of queues */
2335		err = xenbus_printf(xbt, dev->nodename,
2336				    "multi-queue-num-queues", "%u", num_queues);
2337		if (err) {
2338			message = "writing multi-queue-num-queues";
2339			goto abort_transaction_no_dev_fatal;
2340		}
 
 
2341	}
2342
2343	if (num_queues == 1) {
2344		err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
2345		if (err)
2346			goto abort_transaction_no_dev_fatal;
2347	} else {
2348		/* Write the keys for each queue */
2349		for (i = 0; i < num_queues; ++i) {
2350			queue = &info->queues[i];
2351			err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
2352			if (err)
2353				goto abort_transaction_no_dev_fatal;
2354		}
2355	}
2356
2357	/* The remaining keys are not queue-specific */
2358	err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
2359			    1);
2360	if (err) {
2361		message = "writing request-rx-copy";
2362		goto abort_transaction;
2363	}
2364
2365	err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
2366	if (err) {
2367		message = "writing feature-rx-notify";
2368		goto abort_transaction;
2369	}
2370
2371	err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
2372	if (err) {
2373		message = "writing feature-sg";
2374		goto abort_transaction;
2375	}
2376
2377	err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
2378	if (err) {
2379		message = "writing feature-gso-tcpv4";
2380		goto abort_transaction;
2381	}
2382
2383	err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
2384	if (err) {
2385		message = "writing feature-gso-tcpv6";
2386		goto abort_transaction;
2387	}
2388
2389	err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
2390			   "1");
2391	if (err) {
2392		message = "writing feature-ipv6-csum-offload";
2393		goto abort_transaction;
2394	}
2395
2396	err = xenbus_transaction_end(xbt, 0);
2397	if (err) {
2398		if (err == -EAGAIN)
2399			goto again;
2400		xenbus_dev_fatal(dev, err, "completing transaction");
2401		goto destroy_ring;
2402	}
2403
2404	return 0;
2405
2406 abort_transaction:
 
2407	xenbus_dev_fatal(dev, err, "%s", message);
2408abort_transaction_no_dev_fatal:
2409	xenbus_transaction_end(xbt, 1);
2410 destroy_ring:
2411	xennet_disconnect_backend(info);
2412	rtnl_lock();
2413	xennet_destroy_queues(info);
2414 out:
2415	rtnl_unlock();
2416out_unlocked:
2417	device_unregister(&dev->dev);
2418	return err;
2419}
2420
2421static int xennet_connect(struct net_device *dev)
2422{
2423	struct netfront_info *np = netdev_priv(dev);
2424	unsigned int num_queues = 0;
2425	int err;
2426	unsigned int j = 0;
2427	struct netfront_queue *queue = NULL;
 
 
 
 
 
 
2428
2429	if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
2430		dev_info(&dev->dev,
2431			 "backend does not support copying receive path\n");
2432		return -ENODEV;
2433	}
2434
2435	err = talk_to_netback(np->xbdev, np);
2436	if (err)
2437		return err;
2438	if (np->netback_has_xdp_headroom)
2439		pr_info("backend supports XDP headroom\n");
2440	if (np->bounce)
2441		dev_info(&np->xbdev->dev,
2442			 "bouncing transmitted data to zeroed pages\n");
2443
2444	/* talk_to_netback() sets the correct number of queues */
2445	num_queues = dev->real_num_tx_queues;
2446
2447	if (dev->reg_state == NETREG_UNINITIALIZED) {
2448		err = register_netdev(dev);
2449		if (err) {
2450			pr_warn("%s: register_netdev err=%d\n", __func__, err);
2451			device_unregister(&np->xbdev->dev);
2452			return err;
2453		}
2454	}
2455
2456	rtnl_lock();
2457	netdev_update_features(dev);
2458	rtnl_unlock();
2459
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2460	/*
2461	 * All public and private state should now be sane.  Get
2462	 * ready to start sending and receiving packets and give the driver
2463	 * domain a kick because we've probably just requeued some
2464	 * packets.
2465	 */
2466	netif_tx_lock_bh(np->netdev);
2467	netif_device_attach(np->netdev);
2468	netif_tx_unlock_bh(np->netdev);
2469
2470	netif_carrier_on(np->netdev);
2471	for (j = 0; j < num_queues; ++j) {
2472		queue = &np->queues[j];
 
2473
2474		notify_remote_via_irq(queue->tx_irq);
2475		if (queue->tx_irq != queue->rx_irq)
2476			notify_remote_via_irq(queue->rx_irq);
2477
2478		spin_lock_bh(&queue->rx_lock);
2479		xennet_alloc_rx_buffers(queue);
2480		spin_unlock_bh(&queue->rx_lock);
2481	}
2482
2483	return 0;
2484}
2485
2486/*
2487 * Callback received when the backend's state changes.
2488 */
2489static void netback_changed(struct xenbus_device *dev,
2490			    enum xenbus_state backend_state)
2491{
2492	struct netfront_info *np = dev_get_drvdata(&dev->dev);
2493	struct net_device *netdev = np->netdev;
2494
2495	dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2496
2497	wake_up_all(&module_wq);
2498
2499	switch (backend_state) {
2500	case XenbusStateInitialising:
2501	case XenbusStateInitialised:
2502	case XenbusStateReconfiguring:
2503	case XenbusStateReconfigured:
 
2504	case XenbusStateUnknown:
 
2505		break;
2506
2507	case XenbusStateInitWait:
2508		if (dev->state != XenbusStateInitialising)
2509			break;
2510		if (xennet_connect(netdev) != 0)
2511			break;
2512		xenbus_switch_state(dev, XenbusStateConnected);
 
2513		break;
2514
2515	case XenbusStateConnected:
2516		netdev_notify_peers(netdev);
2517		break;
2518
2519	case XenbusStateClosed:
2520		if (dev->state == XenbusStateClosed)
2521			break;
2522		fallthrough;	/* Missed the backend's CLOSING state */
2523	case XenbusStateClosing:
2524		xenbus_frontend_closed(dev);
2525		break;
2526	}
2527}
2528
2529static const struct xennet_stat {
2530	char name[ETH_GSTRING_LEN];
2531	u16 offset;
2532} xennet_stats[] = {
2533	{
2534		"rx_gso_checksum_fixup",
2535		offsetof(struct netfront_info, rx_gso_checksum_fixup)
2536	},
2537};
2538
2539static int xennet_get_sset_count(struct net_device *dev, int string_set)
2540{
2541	switch (string_set) {
2542	case ETH_SS_STATS:
2543		return ARRAY_SIZE(xennet_stats);
2544	default:
2545		return -EINVAL;
2546	}
2547}
2548
2549static void xennet_get_ethtool_stats(struct net_device *dev,
2550				     struct ethtool_stats *stats, u64 * data)
2551{
2552	void *np = netdev_priv(dev);
2553	int i;
2554
2555	for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2556		data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2557}
2558
2559static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2560{
2561	int i;
2562
2563	switch (stringset) {
2564	case ETH_SS_STATS:
2565		for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2566			memcpy(data + i * ETH_GSTRING_LEN,
2567			       xennet_stats[i].name, ETH_GSTRING_LEN);
2568		break;
2569	}
2570}
2571
2572static const struct ethtool_ops xennet_ethtool_ops =
2573{
2574	.get_link = ethtool_op_get_link,
2575
2576	.get_sset_count = xennet_get_sset_count,
2577	.get_ethtool_stats = xennet_get_ethtool_stats,
2578	.get_strings = xennet_get_strings,
2579	.get_ts_info = ethtool_op_get_ts_info,
2580};
2581
2582#ifdef CONFIG_SYSFS
2583static ssize_t show_rxbuf(struct device *dev,
2584			  struct device_attribute *attr, char *buf)
2585{
2586	return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
 
 
 
2587}
2588
2589static ssize_t store_rxbuf(struct device *dev,
2590			   struct device_attribute *attr,
2591			   const char *buf, size_t len)
2592{
 
 
2593	char *endp;
 
2594
2595	if (!capable(CAP_NET_ADMIN))
2596		return -EPERM;
2597
2598	simple_strtoul(buf, &endp, 0);
2599	if (endp == buf)
2600		return -EBADMSG;
2601
2602	/* rxbuf_min and rxbuf_max are no longer configurable. */
 
 
 
 
 
 
 
 
 
 
 
 
2603
 
2604	return len;
2605}
2606
2607static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2608static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2609static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2610
2611static struct attribute *xennet_dev_attrs[] = {
2612	&dev_attr_rxbuf_min.attr,
2613	&dev_attr_rxbuf_max.attr,
2614	&dev_attr_rxbuf_cur.attr,
2615	NULL
2616};
2617
2618static const struct attribute_group xennet_dev_group = {
2619	.attrs = xennet_dev_attrs
2620};
2621#endif /* CONFIG_SYSFS */
2622
2623static void xennet_bus_close(struct xenbus_device *dev)
 
 
2624{
2625	int ret;
 
 
 
 
 
 
2626
2627	if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2628		return;
2629	do {
2630		xenbus_switch_state(dev, XenbusStateClosing);
2631		ret = wait_event_timeout(module_wq,
2632				   xenbus_read_driver_state(dev->otherend) ==
2633				   XenbusStateClosing ||
2634				   xenbus_read_driver_state(dev->otherend) ==
2635				   XenbusStateClosed ||
2636				   xenbus_read_driver_state(dev->otherend) ==
2637				   XenbusStateUnknown,
2638				   XENNET_TIMEOUT);
2639	} while (!ret);
 
 
2640
2641	if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2642		return;
2643
2644	do {
2645		xenbus_switch_state(dev, XenbusStateClosed);
2646		ret = wait_event_timeout(module_wq,
2647				   xenbus_read_driver_state(dev->otherend) ==
2648				   XenbusStateClosed ||
2649				   xenbus_read_driver_state(dev->otherend) ==
2650				   XenbusStateUnknown,
2651				   XENNET_TIMEOUT);
2652	} while (!ret);
2653}
2654
2655static void xennet_remove(struct xenbus_device *dev)
 
2656{
2657	struct netfront_info *info = dev_get_drvdata(&dev->dev);
 
 
 
 
2658
2659	xennet_bus_close(dev);
2660	xennet_disconnect_backend(info);
 
 
 
2661
2662	if (info->netdev->reg_state == NETREG_REGISTERED)
2663		unregister_netdev(info->netdev);
 
 
2664
2665	if (info->queues) {
2666		rtnl_lock();
2667		xennet_destroy_queues(info);
2668		rtnl_unlock();
 
2669	}
2670	xennet_free_netdev(info->netdev);
 
 
 
 
 
2671}
2672
2673static const struct xenbus_device_id netfront_ids[] = {
 
 
 
 
 
 
 
 
 
 
2674	{ "vif" },
2675	{ "" }
2676};
2677
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2678static struct xenbus_driver netfront_driver = {
 
 
2679	.ids = netfront_ids,
2680	.probe = netfront_probe,
2681	.remove = xennet_remove,
2682	.resume = netfront_resume,
2683	.otherend_changed = netback_changed,
2684};
2685
2686static int __init netif_init(void)
2687{
2688	if (!xen_domain())
2689		return -ENODEV;
2690
2691	if (!xen_has_pv_nic_devices())
2692		return -ENODEV;
2693
2694	pr_info("Initialising Xen virtual ethernet driver\n");
2695
2696	/* Allow as many queues as there are CPUs inut max. 8 if user has not
2697	 * specified a value.
2698	 */
2699	if (xennet_max_queues == 0)
2700		xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2701					  num_online_cpus());
2702
2703	return xenbus_register_frontend(&netfront_driver);
2704}
2705module_init(netif_init);
2706
2707
2708static void __exit netif_exit(void)
2709{
 
 
 
2710	xenbus_unregister_driver(&netfront_driver);
2711}
2712module_exit(netif_exit);
2713
2714MODULE_DESCRIPTION("Xen virtual network device frontend");
2715MODULE_LICENSE("GPL");
2716MODULE_ALIAS("xen:vif");
2717MODULE_ALIAS("xennet");