drivers/net/xen-netfront.c

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