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