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/types.h>
  49 #include <linux/bpf_trace.h>
  50
  51 #include <xen/xen.h>
  52 #include <xen/xenbus.h>
  53 #include <xen/events.h>
  54 #include <xen/page.h>
  55 #include <xen/platform_pci.h>
  56 #include <xen/grant_table.h>
  57
  58 #include <xen/interface/io/netif.h>
  59 #include <xen/interface/memory.h>
  60 #include <xen/interface/grant_table.h>
  61
  62 /* Module parameters */
  63 #define MAX_QUEUES_DEFAULT 8
  64 static unsigned int xennet_max_queues;
  65 module_param_named(max_queues, xennet_max_queues, uint, 0644);
  66 MODULE_PARM_DESC(max_queues,
  67                  "Maximum number of queues per virtual interface");
  68
  69 static bool __read_mostly xennet_trusted = true;
  70 module_param_named(trusted, xennet_trusted, bool, 0644);
  71 MODULE_PARM_DESC(trusted, "Is the backend trusted");
  72
  73 #define XENNET_TIMEOUT  (5 * HZ)
  74
  75 static const struct ethtool_ops xennet_ethtool_ops;
  76
  77 struct netfront_cb {
  78         int pull_to;
  79 };
  80
  81 #define NETFRONT_SKB_CB(skb)    ((struct netfront_cb *)((skb)->cb))
  82
  83 #define RX_COPY_THRESHOLD 256
  84
  85 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE)
  86 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE)
  87
  88 /* Minimum number of Rx slots (includes slot for GSO metadata). */
  89 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1)
  90
  91 /* Queue name is interface name with "-qNNN" appended */
  92 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6)
  93
  94 /* IRQ name is queue name with "-tx" or "-rx" appended */
  95 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3)
  96
  97 static DECLARE_WAIT_QUEUE_HEAD(module_wq);
  98
  99 struct netfront_stats {
 100         u64                     packets;
 101         u64                     bytes;
 102         struct u64_stats_sync   syncp;
 103 };
 104
 105 struct netfront_info;
 106
 107 struct netfront_queue {
 108         unsigned int id; /* Queue ID, 0-based */
 109         char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */
 110         struct netfront_info *info;
 111
 112         struct bpf_prog __rcu *xdp_prog;
 113
 114         struct napi_struct napi;
 115
 116         /* Split event channels support, tx_* == rx_* when using
 117          * single event channel.
 118          */
 119         unsigned int tx_evtchn, rx_evtchn;
 120         unsigned int tx_irq, rx_irq;
 121         /* Only used when split event channels support is enabled */
 122         char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */
 123         char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */
 124
 125         spinlock_t   tx_lock;
 126         struct xen_netif_tx_front_ring tx;
 127         int tx_ring_ref;
 128
 129         /*
 130          * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries
 131          * are linked from tx_skb_freelist through tx_link.
 132          */
 133         struct sk_buff *tx_skbs[NET_TX_RING_SIZE];
 134         unsigned short tx_link[NET_TX_RING_SIZE];
 135 #define TX_LINK_NONE 0xffff
 136 #define TX_PENDING   0xfffe
 137         grant_ref_t gref_tx_head;
 138         grant_ref_t grant_tx_ref[NET_TX_RING_SIZE];
 139         struct page *grant_tx_page[NET_TX_RING_SIZE];
 140         unsigned tx_skb_freelist;
 141         unsigned int tx_pend_queue;
 142
 143         spinlock_t   rx_lock ____cacheline_aligned_in_smp;
 144         struct xen_netif_rx_front_ring rx;
 145         int rx_ring_ref;
 146
 147         struct timer_list rx_refill_timer;
 148
 149         struct sk_buff *rx_skbs[NET_RX_RING_SIZE];
 150         grant_ref_t gref_rx_head;
 151         grant_ref_t grant_rx_ref[NET_RX_RING_SIZE];
 152
 153         unsigned int rx_rsp_unconsumed;
 154         spinlock_t rx_cons_lock;
 155
 156         struct page_pool *page_pool;
 157         struct xdp_rxq_info xdp_rxq;
 158 };
 159
 160 struct netfront_info {
 161         struct list_head list;
 162         struct net_device *netdev;
 163
 164         struct xenbus_device *xbdev;
 165
 166         /* Multi-queue support */
 167         struct netfront_queue *queues;
 168
 169         /* Statistics */
 170         struct netfront_stats __percpu *rx_stats;
 171         struct netfront_stats __percpu *tx_stats;
 172
 173         /* XDP state */
 174         bool netback_has_xdp_headroom;
 175         bool netfront_xdp_enabled;
 176
 177         /* Is device behaving sane? */
 178         bool broken;
 179
 180         /* Should skbs be bounced into a zeroed buffer? */
 181         bool bounce;
 182
 183         atomic_t rx_gso_checksum_fixup;
 184 };
 185
 186 struct netfront_rx_info {
 187         struct xen_netif_rx_response rx;
 188         struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1];
 189 };
 190
 191 /*
 192  * Access macros for acquiring freeing slots in tx_skbs[].
 193  */
 194
 195 static void add_id_to_list(unsigned *head, unsigned short *list,
 196                            unsigned short id)
 197 {
 198         list[id] = *head;
 199         *head = id;
 200 }
 201
 202 static unsigned short get_id_from_list(unsigned *head, unsigned short *list)
 203 {
 204         unsigned int id = *head;
 205
 206         if (id != TX_LINK_NONE) {
 207                 *head = list[id];
 208                 list[id] = TX_LINK_NONE;
 209         }
 210         return id;
 211 }
 212
 213 static int xennet_rxidx(RING_IDX idx)
 214 {
 215         return idx & (NET_RX_RING_SIZE - 1);
 216 }
 217
 218 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue,
 219                                          RING_IDX ri)
 220 {
 221         int i = xennet_rxidx(ri);
 222         struct sk_buff *skb = queue->rx_skbs[i];
 223         queue->rx_skbs[i] = NULL;
 224         return skb;
 225 }
 226
 227 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue,
 228                                             RING_IDX ri)
 229 {
 230         int i = xennet_rxidx(ri);
 231         grant_ref_t ref = queue->grant_rx_ref[i];
 232         queue->grant_rx_ref[i] = INVALID_GRANT_REF;
 233         return ref;
 234 }
 235
 236 #ifdef CONFIG_SYSFS
 237 static const struct attribute_group xennet_dev_group;
 238 #endif
 239
 240 static bool xennet_can_sg(struct net_device *dev)
 241 {
 242         return dev->features & NETIF_F_SG;
 243 }
 244
 245
 246 static void rx_refill_timeout(struct timer_list *t)
 247 {
 248         struct netfront_queue *queue = timer_container_of(queue, t,
 249                                                           rx_refill_timer);
 250         napi_schedule(&queue->napi);
 251 }
 252
 253 static int netfront_tx_slot_available(struct netfront_queue *queue)
 254 {
 255         return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) <
 256                 (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1);
 257 }
 258
 259 static void xennet_maybe_wake_tx(struct netfront_queue *queue)
 260 {
 261         struct net_device *dev = queue->info->netdev;
 262         struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id);
 263
 264         if (unlikely(netif_tx_queue_stopped(dev_queue)) &&
 265             netfront_tx_slot_available(queue) &&
 266             likely(netif_running(dev)))
 267                 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id));
 268 }
 269
 270
 271 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue)
 272 {
 273         struct sk_buff *skb;
 274         struct page *page;
 275
 276         skb = __netdev_alloc_skb(queue->info->netdev,
 277                                  RX_COPY_THRESHOLD + NET_IP_ALIGN,
 278                                  GFP_ATOMIC | __GFP_NOWARN);
 279         if (unlikely(!skb))
 280                 return NULL;
 281
 282         page = page_pool_alloc_pages(queue->page_pool,
 283                                      GFP_ATOMIC | __GFP_NOWARN | __GFP_ZERO);
 284         if (unlikely(!page)) {
 285                 kfree_skb(skb);
 286                 return NULL;
 287         }
 288         skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE);
 289         skb_mark_for_recycle(skb);
 290
 291         /* Align ip header to a 16 bytes boundary */
 292         skb_reserve(skb, NET_IP_ALIGN);
 293         skb->dev = queue->info->netdev;
 294
 295         return skb;
 296 }
 297
 298
 299 static void xennet_alloc_rx_buffers(struct netfront_queue *queue)
 300 {
 301         RING_IDX req_prod = queue->rx.req_prod_pvt;
 302         int notify;
 303         int err = 0;
 304
 305         if (unlikely(!netif_carrier_ok(queue->info->netdev)))
 306                 return;
 307
 308         for (req_prod = queue->rx.req_prod_pvt;
 309              req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE;
 310              req_prod++) {
 311                 struct sk_buff *skb;
 312                 unsigned short id;
 313                 grant_ref_t ref;
 314                 struct page *page;
 315                 struct xen_netif_rx_request *req;
 316
 317                 skb = xennet_alloc_one_rx_buffer(queue);
 318                 if (!skb) {
 319                         err = -ENOMEM;
 320                         break;
 321                 }
 322
 323                 id = xennet_rxidx(req_prod);
 324
 325                 BUG_ON(queue->rx_skbs[id]);
 326                 queue->rx_skbs[id] = skb;
 327
 328                 ref = gnttab_claim_grant_reference(&queue->gref_rx_head);
 329                 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
 330                 queue->grant_rx_ref[id] = ref;
 331
 332                 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
 333
 334                 req = RING_GET_REQUEST(&queue->rx, req_prod);
 335                 gnttab_page_grant_foreign_access_ref_one(ref,
 336                                                          queue->info->xbdev->otherend_id,
 337                                                          page,
 338                                                          0);
 339                 req->id = id;
 340                 req->gref = ref;
 341         }
 342
 343         queue->rx.req_prod_pvt = req_prod;
 344
 345         /* Try again later if there are not enough requests or skb allocation
 346          * failed.
 347          * Enough requests is quantified as the sum of newly created slots and
 348          * the unconsumed slots at the backend.
 349          */
 350         if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN ||
 351             unlikely(err)) {
 352                 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10));
 353                 return;
 354         }
 355
 356         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify);
 357         if (notify)
 358                 notify_remote_via_irq(queue->rx_irq);
 359 }
 360
 361 static int xennet_open(struct net_device *dev)
 362 {
 363         struct netfront_info *np = netdev_priv(dev);
 364         unsigned int num_queues = dev->real_num_tx_queues;
 365         unsigned int i = 0;
 366         struct netfront_queue *queue = NULL;
 367
 368         if (!np->queues || np->broken)
 369                 return -ENODEV;
 370
 371         for (i = 0; i < num_queues; ++i) {
 372                 queue = &np->queues[i];
 373                 napi_enable(&queue->napi);
 374
 375                 spin_lock_bh(&queue->rx_lock);
 376                 if (netif_carrier_ok(dev)) {
 377                         xennet_alloc_rx_buffers(queue);
 378                         queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1;
 379                         if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))
 380                                 napi_schedule(&queue->napi);
 381                 }
 382                 spin_unlock_bh(&queue->rx_lock);
 383         }
 384
 385         netif_tx_start_all_queues(dev);
 386
 387         return 0;
 388 }
 389
 390 static bool xennet_tx_buf_gc(struct netfront_queue *queue)
 391 {
 392         RING_IDX cons, prod;
 393         unsigned short id;
 394         struct sk_buff *skb;
 395         bool more_to_do;
 396         bool work_done = false;
 397         const struct device *dev = &queue->info->netdev->dev;
 398
 399         BUG_ON(!netif_carrier_ok(queue->info->netdev));
 400
 401         do {
 402                 prod = queue->tx.sring->rsp_prod;
 403                 if (RING_RESPONSE_PROD_OVERFLOW(&queue->tx, prod)) {
 404                         dev_alert(dev, "Illegal number of responses %u\n",
 405                                   prod - queue->tx.rsp_cons);
 406                         goto err;
 407                 }
 408                 rmb(); /* Ensure we see responses up to 'rp'. */
 409
 410                 for (cons = queue->tx.rsp_cons; cons != prod; cons++) {
 411                         struct xen_netif_tx_response txrsp;
 412
 413                         work_done = true;
 414
 415                         RING_COPY_RESPONSE(&queue->tx, cons, &txrsp);
 416                         if (txrsp.status == XEN_NETIF_RSP_NULL)
 417                                 continue;
 418
 419                         id = txrsp.id;
 420                         if (id >= RING_SIZE(&queue->tx)) {
 421                                 dev_alert(dev,
 422                                           "Response has incorrect id (%u)\n",
 423                                           id);
 424                                 goto err;
 425                         }
 426                         if (queue->tx_link[id] != TX_PENDING) {
 427                                 dev_alert(dev,
 428                                           "Response for inactive request\n");
 429                                 goto err;
 430                         }
 431
 432                         queue->tx_link[id] = TX_LINK_NONE;
 433                         skb = queue->tx_skbs[id];
 434                         queue->tx_skbs[id] = NULL;
 435                         if (unlikely(!gnttab_end_foreign_access_ref(
 436                                 queue->grant_tx_ref[id]))) {
 437                                 dev_alert(dev,
 438                                           "Grant still in use by backend domain\n");
 439                                 goto err;
 440                         }
 441                         gnttab_release_grant_reference(
 442                                 &queue->gref_tx_head, queue->grant_tx_ref[id]);
 443                         queue->grant_tx_ref[id] = INVALID_GRANT_REF;
 444                         queue->grant_tx_page[id] = NULL;
 445                         add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, id);
 446                         dev_kfree_skb_irq(skb);
 447                 }
 448
 449                 queue->tx.rsp_cons = prod;
 450
 451                 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do);
 452         } while (more_to_do);
 453
 454         xennet_maybe_wake_tx(queue);
 455
 456         return work_done;
 457
 458  err:
 459         queue->info->broken = true;
 460         dev_alert(dev, "Disabled for further use\n");
 461
 462         return work_done;
 463 }
 464
 465 struct xennet_gnttab_make_txreq {
 466         struct netfront_queue *queue;
 467         struct sk_buff *skb;
 468         struct page *page;
 469         struct xen_netif_tx_request *tx;      /* Last request on ring page */
 470         struct xen_netif_tx_request tx_local; /* Last request local copy*/
 471         unsigned int size;
 472 };
 473
 474 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset,
 475                                   unsigned int len, void *data)
 476 {
 477         struct xennet_gnttab_make_txreq *info = data;
 478         unsigned int id;
 479         struct xen_netif_tx_request *tx;
 480         grant_ref_t ref;
 481         /* convenient aliases */
 482         struct page *page = info->page;
 483         struct netfront_queue *queue = info->queue;
 484         struct sk_buff *skb = info->skb;
 485
 486         id = get_id_from_list(&queue->tx_skb_freelist, queue->tx_link);
 487         tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
 488         ref = gnttab_claim_grant_reference(&queue->gref_tx_head);
 489         WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref));
 490
 491         gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id,
 492                                         gfn, GNTMAP_readonly);
 493
 494         queue->tx_skbs[id] = skb;
 495         queue->grant_tx_page[id] = page;
 496         queue->grant_tx_ref[id] = ref;
 497
 498         info->tx_local.id = id;
 499         info->tx_local.gref = ref;
 500         info->tx_local.offset = offset;
 501         info->tx_local.size = len;
 502         info->tx_local.flags = 0;
 503
 504         *tx = info->tx_local;
 505
 506         /*
 507          * Put the request in the pending queue, it will be set to be pending
 508          * when the producer index is about to be raised.
 509          */
 510         add_id_to_list(&queue->tx_pend_queue, queue->tx_link, id);
 511
 512         info->tx = tx;
 513         info->size += info->tx_local.size;
 514 }
 515
 516 static struct xen_netif_tx_request *xennet_make_first_txreq(
 517         struct xennet_gnttab_make_txreq *info,
 518         unsigned int offset, unsigned int len)
 519 {
 520         info->size = 0;
 521
 522         gnttab_for_one_grant(info->page, offset, len, xennet_tx_setup_grant, info);
 523
 524         return info->tx;
 525 }
 526
 527 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset,
 528                                   unsigned int len, void *data)
 529 {
 530         struct xennet_gnttab_make_txreq *info = data;
 531
 532         info->tx->flags |= XEN_NETTXF_more_data;
 533         skb_get(info->skb);
 534         xennet_tx_setup_grant(gfn, offset, len, data);
 535 }
 536
 537 static void xennet_make_txreqs(
 538         struct xennet_gnttab_make_txreq *info,
 539         struct page *page,
 540         unsigned int offset, unsigned int len)
 541 {
 542         /* Skip unused frames from start of page */
 543         page += offset >> PAGE_SHIFT;
 544         offset &= ~PAGE_MASK;
 545
 546         while (len) {
 547                 info->page = page;
 548                 info->size = 0;
 549
 550                 gnttab_foreach_grant_in_range(page, offset, len,
 551                                               xennet_make_one_txreq,
 552                                               info);
 553
 554                 page++;
 555                 offset = 0;
 556                 len -= info->size;
 557         }
 558 }
 559
 560 /*
 561  * Count how many ring slots are required to send this skb. Each frag
 562  * might be a compound page.
 563  */
 564 static int xennet_count_skb_slots(struct sk_buff *skb)
 565 {
 566         int i, frags = skb_shinfo(skb)->nr_frags;
 567         int slots;
 568
 569         slots = gnttab_count_grant(offset_in_page(skb->data),
 570                                    skb_headlen(skb));
 571
 572         for (i = 0; i < frags; i++) {
 573                 skb_frag_t *frag = skb_shinfo(skb)->frags + i;
 574                 unsigned long size = skb_frag_size(frag);
 575                 unsigned long offset = skb_frag_off(frag);
 576
 577                 /* Skip unused frames from start of page */
 578                 offset &= ~PAGE_MASK;
 579
 580                 slots += gnttab_count_grant(offset, size);
 581         }
 582
 583         return slots;
 584 }
 585
 586 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb,
 587                                struct net_device *sb_dev)
 588 {
 589         unsigned int num_queues = dev->real_num_tx_queues;
 590         u32 hash;
 591         u16 queue_idx;
 592
 593         /* First, check if there is only one queue */
 594         if (num_queues == 1) {
 595                 queue_idx = 0;
 596         } else {
 597                 hash = skb_get_hash(skb);
 598                 queue_idx = hash % num_queues;
 599         }
 600
 601         return queue_idx;
 602 }
 603
 604 static void xennet_mark_tx_pending(struct netfront_queue *queue)
 605 {
 606         unsigned int i;
 607
 608         while ((i = get_id_from_list(&queue->tx_pend_queue, queue->tx_link)) !=
 609                TX_LINK_NONE)
 610                 queue->tx_link[i] = TX_PENDING;
 611 }
 612
 613 static int xennet_xdp_xmit_one(struct net_device *dev,
 614                                struct netfront_queue *queue,
 615                                struct xdp_frame *xdpf)
 616 {
 617         struct netfront_info *np = netdev_priv(dev);
 618         struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
 619         struct xennet_gnttab_make_txreq info = {
 620                 .queue = queue,
 621                 .skb = NULL,
 622                 .page = virt_to_page(xdpf->data),
 623         };
 624         int notify;
 625
 626         xennet_make_first_txreq(&info,
 627                                 offset_in_page(xdpf->data),
 628                                 xdpf->len);
 629
 630         xennet_mark_tx_pending(queue);
 631
 632         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
 633         if (notify)
 634                 notify_remote_via_irq(queue->tx_irq);
 635
 636         u64_stats_update_begin(&tx_stats->syncp);
 637         tx_stats->bytes += xdpf->len;
 638         tx_stats->packets++;
 639         u64_stats_update_end(&tx_stats->syncp);
 640
 641         xennet_tx_buf_gc(queue);
 642
 643         return 0;
 644 }
 645
 646 static int xennet_xdp_xmit(struct net_device *dev, int n,
 647                            struct xdp_frame **frames, u32 flags)
 648 {
 649         unsigned int num_queues = dev->real_num_tx_queues;
 650         struct netfront_info *np = netdev_priv(dev);
 651         struct netfront_queue *queue = NULL;
 652         unsigned long irq_flags;
 653         int nxmit = 0;
 654         int i;
 655
 656         if (unlikely(np->broken))
 657                 return -ENODEV;
 658         if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK))
 659                 return -EINVAL;
 660
 661         queue = &np->queues[smp_processor_id() % num_queues];
 662
 663         spin_lock_irqsave(&queue->tx_lock, irq_flags);
 664         for (i = 0; i < n; i++) {
 665                 struct xdp_frame *xdpf = frames[i];
 666
 667                 if (!xdpf)
 668                         continue;
 669                 if (xennet_xdp_xmit_one(dev, queue, xdpf))
 670                         break;
 671                 nxmit++;
 672         }
 673         spin_unlock_irqrestore(&queue->tx_lock, irq_flags);
 674
 675         return nxmit;
 676 }
 677
 678 static struct sk_buff *bounce_skb(const struct sk_buff *skb)
 679 {
 680         unsigned int headerlen = skb_headroom(skb);
 681         /* Align size to allocate full pages and avoid contiguous data leaks */
 682         unsigned int size = ALIGN(skb_end_offset(skb) + skb->data_len,
 683                                   XEN_PAGE_SIZE);
 684         struct sk_buff *n = alloc_skb(size, GFP_ATOMIC | __GFP_ZERO);
 685
 686         if (!n)
 687                 return NULL;
 688
 689         if (!IS_ALIGNED((uintptr_t)n->head, XEN_PAGE_SIZE)) {
 690                 WARN_ONCE(1, "misaligned skb allocated\n");
 691                 kfree_skb(n);
 692                 return NULL;
 693         }
 694
 695         /* Set the data pointer */
 696         skb_reserve(n, headerlen);
 697         /* Set the tail pointer and length */
 698         skb_put(n, skb->len);
 699
 700         BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
 701
 702         skb_copy_header(n, skb);
 703         return n;
 704 }
 705
 706 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1)
 707
 708 static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev)
 709 {
 710         struct netfront_info *np = netdev_priv(dev);
 711         struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats);
 712         struct xen_netif_tx_request *first_tx;
 713         unsigned int i;
 714         int notify;
 715         int slots;
 716         struct page *page;
 717         unsigned int offset;
 718         unsigned int len;
 719         unsigned long flags;
 720         struct netfront_queue *queue = NULL;
 721         struct xennet_gnttab_make_txreq info = { };
 722         unsigned int num_queues = dev->real_num_tx_queues;
 723         u16 queue_index;
 724         struct sk_buff *nskb;
 725
 726         /* Drop the packet if no queues are set up */
 727         if (num_queues < 1)
 728                 goto drop;
 729         if (unlikely(np->broken))
 730                 goto drop;
 731         /* Determine which queue to transmit this SKB on */
 732         queue_index = skb_get_queue_mapping(skb);
 733         queue = &np->queues[queue_index];
 734
 735         /* If skb->len is too big for wire format, drop skb and alert
 736          * user about misconfiguration.
 737          */
 738         if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) {
 739                 net_alert_ratelimited(
 740                         "xennet: skb->len = %u, too big for wire format\n",
 741                         skb->len);
 742                 goto drop;
 743         }
 744
 745         slots = xennet_count_skb_slots(skb);
 746         if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) {
 747                 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n",
 748                                     slots, skb->len);
 749                 if (skb_linearize(skb))
 750                         goto drop;
 751         }
 752
 753         page = virt_to_page(skb->data);
 754         offset = offset_in_page(skb->data);
 755
 756         /* The first req should be at least ETH_HLEN size or the packet will be
 757          * dropped by netback.
 758          *
 759          * If the backend is not trusted bounce all data to zeroed pages to
 760          * avoid exposing contiguous data on the granted page not belonging to
 761          * the skb.
 762          */
 763         if (np->bounce || unlikely(PAGE_SIZE - offset < ETH_HLEN)) {
 764                 nskb = bounce_skb(skb);
 765                 if (!nskb)
 766                         goto drop;
 767                 dev_consume_skb_any(skb);
 768                 skb = nskb;
 769                 page = virt_to_page(skb->data);
 770                 offset = offset_in_page(skb->data);
 771         }
 772
 773         len = skb_headlen(skb);
 774
 775         spin_lock_irqsave(&queue->tx_lock, flags);
 776
 777         if (unlikely(!netif_carrier_ok(dev) ||
 778                      (slots > 1 && !xennet_can_sg(dev)) ||
 779                      netif_needs_gso(skb, netif_skb_features(skb)))) {
 780                 spin_unlock_irqrestore(&queue->tx_lock, flags);
 781                 goto drop;
 782         }
 783
 784         /* First request for the linear area. */
 785         info.queue = queue;
 786         info.skb = skb;
 787         info.page = page;
 788         first_tx = xennet_make_first_txreq(&info, offset, len);
 789         offset += info.tx_local.size;
 790         if (offset == PAGE_SIZE) {
 791                 page++;
 792                 offset = 0;
 793         }
 794         len -= info.tx_local.size;
 795
 796         if (skb->ip_summed == CHECKSUM_PARTIAL)
 797                 /* local packet? */
 798                 first_tx->flags |= XEN_NETTXF_csum_blank |
 799                                    XEN_NETTXF_data_validated;
 800         else if (skb->ip_summed == CHECKSUM_UNNECESSARY)
 801                 /* remote but checksummed. */
 802                 first_tx->flags |= XEN_NETTXF_data_validated;
 803
 804         /* Optional extra info after the first request. */
 805         if (skb_shinfo(skb)->gso_size) {
 806                 struct xen_netif_extra_info *gso;
 807
 808                 gso = (struct xen_netif_extra_info *)
 809                         RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++);
 810
 811                 first_tx->flags |= XEN_NETTXF_extra_info;
 812
 813                 gso->u.gso.size = skb_shinfo(skb)->gso_size;
 814                 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ?
 815                         XEN_NETIF_GSO_TYPE_TCPV6 :
 816                         XEN_NETIF_GSO_TYPE_TCPV4;
 817                 gso->u.gso.pad = 0;
 818                 gso->u.gso.features = 0;
 819
 820                 gso->type = XEN_NETIF_EXTRA_TYPE_GSO;
 821                 gso->flags = 0;
 822         }
 823
 824         /* Requests for the rest of the linear area. */
 825         xennet_make_txreqs(&info, page, offset, len);
 826
 827         /* Requests for all the frags. */
 828         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 829                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 830                 xennet_make_txreqs(&info, skb_frag_page(frag),
 831                                         skb_frag_off(frag),
 832                                         skb_frag_size(frag));
 833         }
 834
 835         /* First request has the packet length. */
 836         first_tx->size = skb->len;
 837
 838         /* timestamp packet in software */
 839         skb_tx_timestamp(skb);
 840
 841         xennet_mark_tx_pending(queue);
 842
 843         RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify);
 844         if (notify)
 845                 notify_remote_via_irq(queue->tx_irq);
 846
 847         u64_stats_update_begin(&tx_stats->syncp);
 848         tx_stats->bytes += skb->len;
 849         tx_stats->packets++;
 850         u64_stats_update_end(&tx_stats->syncp);
 851
 852         /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */
 853         xennet_tx_buf_gc(queue);
 854
 855         if (!netfront_tx_slot_available(queue))
 856                 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id));
 857
 858         spin_unlock_irqrestore(&queue->tx_lock, flags);
 859
 860         return NETDEV_TX_OK;
 861
 862  drop:
 863         dev->stats.tx_dropped++;
 864         dev_kfree_skb_any(skb);
 865         return NETDEV_TX_OK;
 866 }
 867
 868 static int xennet_close(struct net_device *dev)
 869 {
 870         struct netfront_info *np = netdev_priv(dev);
 871         unsigned int num_queues = np->queues ? dev->real_num_tx_queues : 0;
 872         unsigned int i;
 873         struct netfront_queue *queue;
 874         netif_tx_stop_all_queues(np->netdev);
 875         for (i = 0; i < num_queues; ++i) {
 876                 queue = &np->queues[i];
 877                 napi_disable(&queue->napi);
 878         }
 879         return 0;
 880 }
 881
 882 static void xennet_destroy_queues(struct netfront_info *info)
 883 {
 884         unsigned int i;
 885
 886         if (!info->queues)
 887                 return;
 888
 889         for (i = 0; i < info->netdev->real_num_tx_queues; i++) {
 890                 struct netfront_queue *queue = &info->queues[i];
 891
 892                 if (netif_running(info->netdev))
 893                         napi_disable(&queue->napi);
 894                 netif_napi_del(&queue->napi);
 895         }
 896
 897         kfree(info->queues);
 898         info->queues = NULL;
 899 }
 900
 901 static void xennet_uninit(struct net_device *dev)
 902 {
 903         struct netfront_info *np = netdev_priv(dev);
 904         xennet_destroy_queues(np);
 905 }
 906
 907 static void xennet_set_rx_rsp_cons(struct netfront_queue *queue, RING_IDX val)
 908 {
 909         unsigned long flags;
 910
 911         spin_lock_irqsave(&queue->rx_cons_lock, flags);
 912         queue->rx.rsp_cons = val;
 913         queue->rx_rsp_unconsumed = XEN_RING_NR_UNCONSUMED_RESPONSES(&queue->rx);
 914         spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
 915 }
 916
 917 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb,
 918                                 grant_ref_t ref)
 919 {
 920         int new = xennet_rxidx(queue->rx.req_prod_pvt);
 921
 922         BUG_ON(queue->rx_skbs[new]);
 923         queue->rx_skbs[new] = skb;
 924         queue->grant_rx_ref[new] = ref;
 925         RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new;
 926         RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref;
 927         queue->rx.req_prod_pvt++;
 928 }
 929
 930 static int xennet_get_extras(struct netfront_queue *queue,
 931                              struct xen_netif_extra_info *extras,
 932                              RING_IDX rp)
 933
 934 {
 935         struct xen_netif_extra_info extra;
 936         struct device *dev = &queue->info->netdev->dev;
 937         RING_IDX cons = queue->rx.rsp_cons;
 938         int err = 0;
 939
 940         do {
 941                 struct sk_buff *skb;
 942                 grant_ref_t ref;
 943
 944                 if (unlikely(cons + 1 == rp)) {
 945                         if (net_ratelimit())
 946                                 dev_warn(dev, "Missing extra info\n");
 947                         err = -EBADR;
 948                         break;
 949                 }
 950
 951                 RING_COPY_RESPONSE(&queue->rx, ++cons, &extra);
 952
 953                 if (unlikely(!extra.type ||
 954                              extra.type >= XEN_NETIF_EXTRA_TYPE_MAX)) {
 955                         if (net_ratelimit())
 956                                 dev_warn(dev, "Invalid extra type: %d\n",
 957                                          extra.type);
 958                         err = -EINVAL;
 959                 } else {
 960                         extras[extra.type - 1] = extra;
 961                 }
 962
 963                 skb = xennet_get_rx_skb(queue, cons);
 964                 ref = xennet_get_rx_ref(queue, cons);
 965                 xennet_move_rx_slot(queue, skb, ref);
 966         } while (extra.flags & XEN_NETIF_EXTRA_FLAG_MORE);
 967
 968         xennet_set_rx_rsp_cons(queue, cons);
 969         return err;
 970 }
 971
 972 static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata,
 973                    struct xen_netif_rx_response *rx, struct bpf_prog *prog,
 974                    struct xdp_buff *xdp, bool *need_xdp_flush)
 975 {
 976         struct xdp_frame *xdpf;
 977         u32 len = rx->status;
 978         u32 act;
 979         int err;
 980
 981         xdp_init_buff(xdp, XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
 982                       &queue->xdp_rxq);
 983         xdp_prepare_buff(xdp, page_address(pdata), XDP_PACKET_HEADROOM,
 984                          len, false);
 985
 986         act = bpf_prog_run_xdp(prog, xdp);
 987         switch (act) {
 988         case XDP_TX:
 989                 xdpf = xdp_convert_buff_to_frame(xdp);
 990                 if (unlikely(!xdpf)) {
 991                         trace_xdp_exception(queue->info->netdev, prog, act);
 992                         break;
 993                 }
 994                 get_page(pdata);
 995                 err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0);
 996                 if (unlikely(err <= 0)) {
 997                         if (err < 0)
 998                                 trace_xdp_exception(queue->info->netdev, prog, act);
 999                         xdp_return_frame_rx_napi(xdpf);
1000                 }
1001                 break;
1002         case XDP_REDIRECT:
1003                 get_page(pdata);
1004                 err = xdp_do_redirect(queue->info->netdev, xdp, prog);
1005                 *need_xdp_flush = true;
1006                 if (unlikely(err)) {
1007                         trace_xdp_exception(queue->info->netdev, prog, act);
1008                         xdp_return_buff(xdp);
1009                 }
1010                 break;
1011         case XDP_PASS:
1012         case XDP_DROP:
1013                 break;
1014
1015         case XDP_ABORTED:
1016                 trace_xdp_exception(queue->info->netdev, prog, act);
1017                 break;
1018
1019         default:
1020                 bpf_warn_invalid_xdp_action(queue->info->netdev, prog, act);
1021         }
1022
1023         return act;
1024 }
1025
1026 static int xennet_get_responses(struct netfront_queue *queue,
1027                                 struct netfront_rx_info *rinfo, RING_IDX rp,
1028                                 struct sk_buff_head *list,
1029                                 bool *need_xdp_flush)
1030 {
1031         struct xen_netif_rx_response *rx = &rinfo->rx, rx_local;
1032         int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD);
1033         RING_IDX cons = queue->rx.rsp_cons;
1034         struct sk_buff *skb = xennet_get_rx_skb(queue, cons);
1035         struct xen_netif_extra_info *extras = rinfo->extras;
1036         grant_ref_t ref = xennet_get_rx_ref(queue, cons);
1037         struct device *dev = &queue->info->netdev->dev;
1038         struct bpf_prog *xdp_prog;
1039         struct xdp_buff xdp;
1040         int slots = 1;
1041         int err = 0;
1042         u32 verdict;
1043
1044         if (rx->flags & XEN_NETRXF_extra_info) {
1045                 err = xennet_get_extras(queue, extras, rp);
1046                 if (!err) {
1047                         if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) {
1048                                 struct xen_netif_extra_info *xdp;
1049
1050                                 xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1];
1051                                 rx->offset = xdp->u.xdp.headroom;
1052                         }
1053                 }
1054                 cons = queue->rx.rsp_cons;
1055         }
1056
1057         for (;;) {
1058                 /*
1059                  * This definitely indicates a bug, either in this driver or in
1060                  * the backend driver. In future this should flag the bad
1061                  * situation to the system controller to reboot the backend.
1062                  */
1063                 if (ref == INVALID_GRANT_REF) {
1064                         if (net_ratelimit())
1065                                 dev_warn(dev, "Bad rx response id %d.\n",
1066                                          rx->id);
1067                         err = -EINVAL;
1068                         goto next;
1069                 }
1070
1071                 if (unlikely(rx->status < 0 ||
1072                              rx->offset + rx->status > XEN_PAGE_SIZE)) {
1073                         if (net_ratelimit())
1074                                 dev_warn(dev, "rx->offset: %u, size: %d\n",
1075                                          rx->offset, rx->status);
1076                         xennet_move_rx_slot(queue, skb, ref);
1077                         err = -EINVAL;
1078                         goto next;
1079                 }
1080
1081                 if (!gnttab_end_foreign_access_ref(ref)) {
1082                         dev_alert(dev,
1083                                   "Grant still in use by backend domain\n");
1084                         queue->info->broken = true;
1085                         dev_alert(dev, "Disabled for further use\n");
1086                         return -EINVAL;
1087                 }
1088
1089                 gnttab_release_grant_reference(&queue->gref_rx_head, ref);
1090
1091                 rcu_read_lock();
1092                 xdp_prog = rcu_dereference(queue->xdp_prog);
1093                 if (xdp_prog) {
1094                         if (!(rx->flags & XEN_NETRXF_more_data)) {
1095                                 /* currently only a single page contains data */
1096                                 verdict = xennet_run_xdp(queue,
1097                                                          skb_frag_page(&skb_shinfo(skb)->frags[0]),
1098                                                          rx, xdp_prog, &xdp, need_xdp_flush);
1099                                 if (verdict != XDP_PASS)
1100                                         err = -EINVAL;
1101                         } else {
1102                                 /* drop the frame */
1103                                 err = -EINVAL;
1104                         }
1105                 }
1106                 rcu_read_unlock();
1107
1108                 __skb_queue_tail(list, skb);
1109
1110 next:
1111                 if (!(rx->flags & XEN_NETRXF_more_data))
1112                         break;
1113
1114                 if (cons + slots == rp) {
1115                         if (net_ratelimit())
1116                                 dev_warn(dev, "Need more slots\n");
1117                         err = -ENOENT;
1118                         break;
1119                 }
1120
1121                 RING_COPY_RESPONSE(&queue->rx, cons + slots, &rx_local);
1122                 rx = &rx_local;
1123                 skb = xennet_get_rx_skb(queue, cons + slots);
1124                 ref = xennet_get_rx_ref(queue, cons + slots);
1125                 slots++;
1126         }
1127
1128         if (unlikely(slots > max)) {
1129                 if (net_ratelimit())
1130                         dev_warn(dev, "Too many slots\n");
1131                 err = -E2BIG;
1132         }
1133
1134         if (unlikely(err))
1135                 xennet_set_rx_rsp_cons(queue, cons + slots);
1136
1137         return err;
1138 }
1139
1140 static int xennet_set_skb_gso(struct sk_buff *skb,
1141                               struct xen_netif_extra_info *gso)
1142 {
1143         if (!gso->u.gso.size) {
1144                 if (net_ratelimit())
1145                         pr_warn("GSO size must not be zero\n");
1146                 return -EINVAL;
1147         }
1148
1149         if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 &&
1150             gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) {
1151                 if (net_ratelimit())
1152                         pr_warn("Bad GSO type %d\n", gso->u.gso.type);
1153                 return -EINVAL;
1154         }
1155
1156         skb_shinfo(skb)->gso_size = gso->u.gso.size;
1157         skb_shinfo(skb)->gso_type =
1158                 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ?
1159                 SKB_GSO_TCPV4 :
1160                 SKB_GSO_TCPV6;
1161
1162         /* Header must be checked, and gso_segs computed. */
1163         skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
1164         skb_shinfo(skb)->gso_segs = 0;
1165
1166         return 0;
1167 }
1168
1169 static int xennet_fill_frags(struct netfront_queue *queue,
1170                              struct sk_buff *skb,
1171                              struct sk_buff_head *list)
1172 {
1173         RING_IDX cons = queue->rx.rsp_cons;
1174         struct sk_buff *nskb;
1175
1176         while ((nskb = __skb_dequeue(list))) {
1177                 struct xen_netif_rx_response rx;
1178                 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0];
1179
1180                 RING_COPY_RESPONSE(&queue->rx, ++cons, &rx);
1181
1182                 if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) {
1183                         unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1184
1185                         BUG_ON(pull_to < skb_headlen(skb));
1186                         __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1187                 }
1188                 if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) {
1189                         xennet_set_rx_rsp_cons(queue,
1190                                                ++cons + skb_queue_len(list));
1191                         kfree_skb(nskb);
1192                         return -ENOENT;
1193                 }
1194
1195                 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags,
1196                                 skb_frag_page(nfrag),
1197                                 rx.offset, rx.status, PAGE_SIZE);
1198
1199                 skb_shinfo(nskb)->nr_frags = 0;
1200                 kfree_skb(nskb);
1201         }
1202
1203         xennet_set_rx_rsp_cons(queue, cons);
1204
1205         return 0;
1206 }
1207
1208 static int checksum_setup(struct net_device *dev, struct sk_buff *skb)
1209 {
1210         bool recalculate_partial_csum = false;
1211
1212         /*
1213          * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy
1214          * peers can fail to set NETRXF_csum_blank when sending a GSO
1215          * frame. In this case force the SKB to CHECKSUM_PARTIAL and
1216          * recalculate the partial checksum.
1217          */
1218         if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) {
1219                 struct netfront_info *np = netdev_priv(dev);
1220                 atomic_inc(&np->rx_gso_checksum_fixup);
1221                 skb->ip_summed = CHECKSUM_PARTIAL;
1222                 recalculate_partial_csum = true;
1223         }
1224
1225         /* A non-CHECKSUM_PARTIAL SKB does not require setup. */
1226         if (skb->ip_summed != CHECKSUM_PARTIAL)
1227                 return 0;
1228
1229         return skb_checksum_setup(skb, recalculate_partial_csum);
1230 }
1231
1232 static int handle_incoming_queue(struct netfront_queue *queue,
1233                                  struct sk_buff_head *rxq)
1234 {
1235         struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats);
1236         int packets_dropped = 0;
1237         struct sk_buff *skb;
1238
1239         while ((skb = __skb_dequeue(rxq)) != NULL) {
1240                 int pull_to = NETFRONT_SKB_CB(skb)->pull_to;
1241
1242                 if (pull_to > skb_headlen(skb))
1243                         __pskb_pull_tail(skb, pull_to - skb_headlen(skb));
1244
1245                 /* Ethernet work: Delayed to here as it peeks the header. */
1246                 skb->protocol = eth_type_trans(skb, queue->info->netdev);
1247                 skb_reset_network_header(skb);
1248
1249                 if (checksum_setup(queue->info->netdev, skb)) {
1250                         kfree_skb(skb);
1251                         packets_dropped++;
1252                         queue->info->netdev->stats.rx_errors++;
1253                         continue;
1254                 }
1255
1256                 u64_stats_update_begin(&rx_stats->syncp);
1257                 rx_stats->packets++;
1258                 rx_stats->bytes += skb->len;
1259                 u64_stats_update_end(&rx_stats->syncp);
1260
1261                 /* Pass it up. */
1262                 napi_gro_receive(&queue->napi, skb);
1263         }
1264
1265         return packets_dropped;
1266 }
1267
1268 static int xennet_poll(struct napi_struct *napi, int budget)
1269 {
1270         struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi);
1271         struct net_device *dev = queue->info->netdev;
1272         struct sk_buff *skb;
1273         struct netfront_rx_info rinfo;
1274         struct xen_netif_rx_response *rx = &rinfo.rx;
1275         struct xen_netif_extra_info *extras = rinfo.extras;
1276         RING_IDX i, rp;
1277         int work_done;
1278         struct sk_buff_head rxq;
1279         struct sk_buff_head errq;
1280         struct sk_buff_head tmpq;
1281         int err;
1282         bool need_xdp_flush = false;
1283
1284         spin_lock(&queue->rx_lock);
1285
1286         skb_queue_head_init(&rxq);
1287         skb_queue_head_init(&errq);
1288         skb_queue_head_init(&tmpq);
1289
1290         rp = queue->rx.sring->rsp_prod;
1291         if (RING_RESPONSE_PROD_OVERFLOW(&queue->rx, rp)) {
1292                 dev_alert(&dev->dev, "Illegal number of responses %u\n",
1293                           rp - queue->rx.rsp_cons);
1294                 queue->info->broken = true;
1295                 spin_unlock(&queue->rx_lock);
1296                 return 0;
1297         }
1298         rmb(); /* Ensure we see queued responses up to 'rp'. */
1299
1300         i = queue->rx.rsp_cons;
1301         work_done = 0;
1302         while ((i != rp) && (work_done < budget)) {
1303                 RING_COPY_RESPONSE(&queue->rx, i, rx);
1304                 memset(extras, 0, sizeof(rinfo.extras));
1305
1306                 err = xennet_get_responses(queue, &rinfo, rp, &tmpq,
1307                                            &need_xdp_flush);
1308
1309                 if (unlikely(err)) {
1310                         if (queue->info->broken) {
1311                                 spin_unlock(&queue->rx_lock);
1312                                 return 0;
1313                         }
1314 err:
1315                         while ((skb = __skb_dequeue(&tmpq)))
1316                                 __skb_queue_tail(&errq, skb);
1317                         dev->stats.rx_errors++;
1318                         i = queue->rx.rsp_cons;
1319                         continue;
1320                 }
1321
1322                 skb = __skb_dequeue(&tmpq);
1323
1324                 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) {
1325                         struct xen_netif_extra_info *gso;
1326                         gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1];
1327
1328                         if (unlikely(xennet_set_skb_gso(skb, gso))) {
1329                                 __skb_queue_head(&tmpq, skb);
1330                                 xennet_set_rx_rsp_cons(queue,
1331                                                        queue->rx.rsp_cons +
1332                                                        skb_queue_len(&tmpq));
1333                                 goto err;
1334                         }
1335                 }
1336
1337                 NETFRONT_SKB_CB(skb)->pull_to = rx->status;
1338                 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD)
1339                         NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD;
1340
1341                 skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset);
1342                 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status);
1343                 skb->data_len = rx->status;
1344                 skb->len += rx->status;
1345
1346                 if (unlikely(xennet_fill_frags(queue, skb, &tmpq)))
1347                         goto err;
1348
1349                 if (rx->flags & XEN_NETRXF_csum_blank)
1350                         skb->ip_summed = CHECKSUM_PARTIAL;
1351                 else if (rx->flags & XEN_NETRXF_data_validated)
1352                         skb->ip_summed = CHECKSUM_UNNECESSARY;
1353
1354                 __skb_queue_tail(&rxq, skb);
1355
1356                 i = queue->rx.rsp_cons + 1;
1357                 xennet_set_rx_rsp_cons(queue, i);
1358                 work_done++;
1359         }
1360         if (need_xdp_flush)
1361                 xdp_do_flush();
1362
1363         __skb_queue_purge(&errq);
1364
1365         work_done -= handle_incoming_queue(queue, &rxq);
1366
1367         xennet_alloc_rx_buffers(queue);
1368
1369         if (work_done < budget) {
1370                 int more_to_do = 0;
1371
1372                 napi_complete_done(napi, work_done);
1373
1374                 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do);
1375                 if (more_to_do)
1376                         napi_schedule(napi);
1377         }
1378
1379         spin_unlock(&queue->rx_lock);
1380
1381         return work_done;
1382 }
1383
1384 static int xennet_change_mtu(struct net_device *dev, int mtu)
1385 {
1386         int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN;
1387
1388         if (mtu > max)
1389                 return -EINVAL;
1390         WRITE_ONCE(dev->mtu, mtu);
1391         return 0;
1392 }
1393
1394 static void xennet_get_stats64(struct net_device *dev,
1395                                struct rtnl_link_stats64 *tot)
1396 {
1397         struct netfront_info *np = netdev_priv(dev);
1398         int cpu;
1399
1400         for_each_possible_cpu(cpu) {
1401                 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu);
1402                 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu);
1403                 u64 rx_packets, rx_bytes, tx_packets, tx_bytes;
1404                 unsigned int start;
1405
1406                 do {
1407                         start = u64_stats_fetch_begin(&tx_stats->syncp);
1408                         tx_packets = tx_stats->packets;
1409                         tx_bytes = tx_stats->bytes;
1410                 } while (u64_stats_fetch_retry(&tx_stats->syncp, start));
1411
1412                 do {
1413                         start = u64_stats_fetch_begin(&rx_stats->syncp);
1414                         rx_packets = rx_stats->packets;
1415                         rx_bytes = rx_stats->bytes;
1416                 } while (u64_stats_fetch_retry(&rx_stats->syncp, start));
1417
1418                 tot->rx_packets += rx_packets;
1419                 tot->tx_packets += tx_packets;
1420                 tot->rx_bytes   += rx_bytes;
1421                 tot->tx_bytes   += tx_bytes;
1422         }
1423
1424         tot->rx_errors  = dev->stats.rx_errors;
1425         tot->tx_dropped = dev->stats.tx_dropped;
1426 }
1427
1428 static void xennet_release_tx_bufs(struct netfront_queue *queue)
1429 {
1430         struct sk_buff *skb;
1431         int i;
1432
1433         for (i = 0; i < NET_TX_RING_SIZE; i++) {
1434                 /* Skip over entries which are actually freelist references */
1435                 if (!queue->tx_skbs[i])
1436                         continue;
1437
1438                 skb = queue->tx_skbs[i];
1439                 queue->tx_skbs[i] = NULL;
1440                 get_page(queue->grant_tx_page[i]);
1441                 gnttab_end_foreign_access(queue->grant_tx_ref[i],
1442                                           queue->grant_tx_page[i]);
1443                 queue->grant_tx_page[i] = NULL;
1444                 queue->grant_tx_ref[i] = INVALID_GRANT_REF;
1445                 add_id_to_list(&queue->tx_skb_freelist, queue->tx_link, i);
1446                 dev_kfree_skb_irq(skb);
1447         }
1448 }
1449
1450 static void xennet_release_rx_bufs(struct netfront_queue *queue)
1451 {
1452         int id, ref;
1453
1454         spin_lock_bh(&queue->rx_lock);
1455
1456         for (id = 0; id < NET_RX_RING_SIZE; id++) {
1457                 struct sk_buff *skb;
1458                 struct page *page;
1459
1460                 skb = queue->rx_skbs[id];
1461                 if (!skb)
1462                         continue;
1463
1464                 ref = queue->grant_rx_ref[id];
1465                 if (ref == INVALID_GRANT_REF)
1466                         continue;
1467
1468                 page = skb_frag_page(&skb_shinfo(skb)->frags[0]);
1469
1470                 /* gnttab_end_foreign_access() needs a page ref until
1471                  * foreign access is ended (which may be deferred).
1472                  */
1473                 get_page(page);
1474                 gnttab_end_foreign_access(ref, page);
1475                 queue->grant_rx_ref[id] = INVALID_GRANT_REF;
1476
1477                 kfree_skb(skb);
1478         }
1479
1480         spin_unlock_bh(&queue->rx_lock);
1481 }
1482
1483 static netdev_features_t xennet_fix_features(struct net_device *dev,
1484         netdev_features_t features)
1485 {
1486         struct netfront_info *np = netdev_priv(dev);
1487
1488         if (features & NETIF_F_SG &&
1489             !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0))
1490                 features &= ~NETIF_F_SG;
1491
1492         if (features & NETIF_F_IPV6_CSUM &&
1493             !xenbus_read_unsigned(np->xbdev->otherend,
1494                                   "feature-ipv6-csum-offload", 0))
1495                 features &= ~NETIF_F_IPV6_CSUM;
1496
1497         if (features & NETIF_F_TSO &&
1498             !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0))
1499                 features &= ~NETIF_F_TSO;
1500
1501         if (features & NETIF_F_TSO6 &&
1502             !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0))
1503                 features &= ~NETIF_F_TSO6;
1504
1505         return features;
1506 }
1507
1508 static int xennet_set_features(struct net_device *dev,
1509         netdev_features_t features)
1510 {
1511         if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) {
1512                 netdev_info(dev, "Reducing MTU because no SG offload");
1513                 dev->mtu = ETH_DATA_LEN;
1514         }
1515
1516         return 0;
1517 }
1518
1519 static bool xennet_handle_tx(struct netfront_queue *queue, unsigned int *eoi)
1520 {
1521         unsigned long flags;
1522
1523         if (unlikely(queue->info->broken))
1524                 return false;
1525
1526         spin_lock_irqsave(&queue->tx_lock, flags);
1527         if (xennet_tx_buf_gc(queue))
1528                 *eoi = 0;
1529         spin_unlock_irqrestore(&queue->tx_lock, flags);
1530
1531         return true;
1532 }
1533
1534 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id)
1535 {
1536         unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1537
1538         if (likely(xennet_handle_tx(dev_id, &eoiflag)))
1539                 xen_irq_lateeoi(irq, eoiflag);
1540
1541         return IRQ_HANDLED;
1542 }
1543
1544 static bool xennet_handle_rx(struct netfront_queue *queue, unsigned int *eoi)
1545 {
1546         unsigned int work_queued;
1547         unsigned long flags;
1548
1549         if (unlikely(queue->info->broken))
1550                 return false;
1551
1552         spin_lock_irqsave(&queue->rx_cons_lock, flags);
1553         work_queued = XEN_RING_NR_UNCONSUMED_RESPONSES(&queue->rx);
1554         if (work_queued > queue->rx_rsp_unconsumed) {
1555                 queue->rx_rsp_unconsumed = work_queued;
1556                 *eoi = 0;
1557         } else if (unlikely(work_queued < queue->rx_rsp_unconsumed)) {
1558                 const struct device *dev = &queue->info->netdev->dev;
1559
1560                 spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
1561                 dev_alert(dev, "RX producer index going backwards\n");
1562                 dev_alert(dev, "Disabled for further use\n");
1563                 queue->info->broken = true;
1564                 return false;
1565         }
1566         spin_unlock_irqrestore(&queue->rx_cons_lock, flags);
1567
1568         if (likely(netif_carrier_ok(queue->info->netdev) && work_queued))
1569                 napi_schedule(&queue->napi);
1570
1571         return true;
1572 }
1573
1574 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id)
1575 {
1576         unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1577
1578         if (likely(xennet_handle_rx(dev_id, &eoiflag)))
1579                 xen_irq_lateeoi(irq, eoiflag);
1580
1581         return IRQ_HANDLED;
1582 }
1583
1584 static irqreturn_t xennet_interrupt(int irq, void *dev_id)
1585 {
1586         unsigned int eoiflag = XEN_EOI_FLAG_SPURIOUS;
1587
1588         if (xennet_handle_tx(dev_id, &eoiflag) &&
1589             xennet_handle_rx(dev_id, &eoiflag))
1590                 xen_irq_lateeoi(irq, eoiflag);
1591
1592         return IRQ_HANDLED;
1593 }
1594
1595 #ifdef CONFIG_NET_POLL_CONTROLLER
1596 static void xennet_poll_controller(struct net_device *dev)
1597 {
1598         /* Poll each queue */
1599         struct netfront_info *info = netdev_priv(dev);
1600         unsigned int num_queues = dev->real_num_tx_queues;
1601         unsigned int i;
1602
1603         if (info->broken)
1604                 return;
1605
1606         for (i = 0; i < num_queues; ++i)
1607                 xennet_interrupt(0, &info->queues[i]);
1608 }
1609 #endif
1610
1611 #define NETBACK_XDP_HEADROOM_DISABLE    0
1612 #define NETBACK_XDP_HEADROOM_ENABLE     1
1613
1614 static int talk_to_netback_xdp(struct netfront_info *np, int xdp)
1615 {
1616         int err;
1617         unsigned short headroom;
1618
1619         headroom = xdp ? XDP_PACKET_HEADROOM : 0;
1620         err = xenbus_printf(XBT_NIL, np->xbdev->nodename,
1621                             "xdp-headroom", "%hu",
1622                             headroom);
1623         if (err)
1624                 pr_warn("Error writing xdp-headroom\n");
1625
1626         return err;
1627 }
1628
1629 static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog,
1630                           struct netlink_ext_ack *extack)
1631 {
1632         unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM;
1633         struct netfront_info *np = netdev_priv(dev);
1634         struct bpf_prog *old_prog;
1635         unsigned int i, err;
1636
1637         if (dev->mtu > max_mtu) {
1638                 netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu);
1639                 return -EINVAL;
1640         }
1641
1642         if (!np->netback_has_xdp_headroom)
1643                 return 0;
1644
1645         xenbus_switch_state(np->xbdev, XenbusStateReconfiguring);
1646
1647         err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE :
1648                                   NETBACK_XDP_HEADROOM_DISABLE);
1649         if (err)
1650                 return err;
1651
1652         /* avoid the race with XDP headroom adjustment */
1653         wait_event(module_wq,
1654                    xenbus_read_driver_state(np->xbdev->otherend) ==
1655                    XenbusStateReconfigured);
1656         np->netfront_xdp_enabled = true;
1657
1658         old_prog = rtnl_dereference(np->queues[0].xdp_prog);
1659
1660         if (prog)
1661                 bpf_prog_add(prog, dev->real_num_tx_queues);
1662
1663         for (i = 0; i < dev->real_num_tx_queues; ++i)
1664                 rcu_assign_pointer(np->queues[i].xdp_prog, prog);
1665
1666         if (old_prog)
1667                 for (i = 0; i < dev->real_num_tx_queues; ++i)
1668                         bpf_prog_put(old_prog);
1669
1670         xenbus_switch_state(np->xbdev, XenbusStateConnected);
1671
1672         return 0;
1673 }
1674
1675 static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp)
1676 {
1677         struct netfront_info *np = netdev_priv(dev);
1678
1679         if (np->broken)
1680                 return -ENODEV;
1681
1682         switch (xdp->command) {
1683         case XDP_SETUP_PROG:
1684                 return xennet_xdp_set(dev, xdp->prog, xdp->extack);
1685         default:
1686                 return -EINVAL;
1687         }
1688 }
1689
1690 static const struct net_device_ops xennet_netdev_ops = {
1691         .ndo_uninit          = xennet_uninit,
1692         .ndo_open            = xennet_open,
1693         .ndo_stop            = xennet_close,
1694         .ndo_start_xmit      = xennet_start_xmit,
1695         .ndo_change_mtu      = xennet_change_mtu,
1696         .ndo_get_stats64     = xennet_get_stats64,
1697         .ndo_set_mac_address = eth_mac_addr,
1698         .ndo_validate_addr   = eth_validate_addr,
1699         .ndo_fix_features    = xennet_fix_features,
1700         .ndo_set_features    = xennet_set_features,
1701         .ndo_select_queue    = xennet_select_queue,
1702         .ndo_bpf            = xennet_xdp,
1703         .ndo_xdp_xmit       = xennet_xdp_xmit,
1704 #ifdef CONFIG_NET_POLL_CONTROLLER
1705         .ndo_poll_controller = xennet_poll_controller,
1706 #endif
1707 };
1708
1709 static void xennet_free_netdev(struct net_device *netdev)
1710 {
1711         struct netfront_info *np = netdev_priv(netdev);
1712
1713         free_percpu(np->rx_stats);
1714         free_percpu(np->tx_stats);
1715         free_netdev(netdev);
1716 }
1717
1718 static struct net_device *xennet_create_dev(struct xenbus_device *dev)
1719 {
1720         int err;
1721         struct net_device *netdev;
1722         struct netfront_info *np;
1723
1724         netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues);
1725         if (!netdev)
1726                 return ERR_PTR(-ENOMEM);
1727
1728         np                   = netdev_priv(netdev);
1729         np->xbdev            = dev;
1730
1731         np->queues = NULL;
1732
1733         err = -ENOMEM;
1734         np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1735         if (np->rx_stats == NULL)
1736                 goto exit;
1737         np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats);
1738         if (np->tx_stats == NULL)
1739                 goto exit;
1740
1741         netdev->netdev_ops      = &xennet_netdev_ops;
1742
1743         netdev->features        = NETIF_F_IP_CSUM | NETIF_F_RXCSUM |
1744                                   NETIF_F_GSO_ROBUST;
1745         netdev->hw_features     = NETIF_F_SG |
1746                                   NETIF_F_IPV6_CSUM |
1747                                   NETIF_F_TSO | NETIF_F_TSO6;
1748
1749         /*
1750          * Assume that all hw features are available for now. This set
1751          * will be adjusted by the call to netdev_update_features() in
1752          * xennet_connect() which is the earliest point where we can
1753          * negotiate with the backend regarding supported features.
1754          */
1755         netdev->features |= netdev->hw_features;
1756         netdev->xdp_features = NETDEV_XDP_ACT_BASIC | NETDEV_XDP_ACT_REDIRECT |
1757                                NETDEV_XDP_ACT_NDO_XMIT;
1758
1759         netdev->ethtool_ops = &xennet_ethtool_ops;
1760         netdev->min_mtu = ETH_MIN_MTU;
1761         netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE;
1762         SET_NETDEV_DEV(netdev, &dev->dev);
1763
1764         np->netdev = netdev;
1765         np->netfront_xdp_enabled = false;
1766
1767         netif_carrier_off(netdev);
1768
1769         do {
1770                 xenbus_switch_state(dev, XenbusStateInitialising);
1771                 err = wait_event_timeout(module_wq,
1772                                  xenbus_read_driver_state(dev->otherend) !=
1773                                  XenbusStateClosed &&
1774                                  xenbus_read_driver_state(dev->otherend) !=
1775                                  XenbusStateUnknown, XENNET_TIMEOUT);
1776         } while (!err);
1777
1778         return netdev;
1779
1780  exit:
1781         xennet_free_netdev(netdev);
1782         return ERR_PTR(err);
1783 }
1784
1785 /*
1786  * Entry point to this code when a new device is created.  Allocate the basic
1787  * structures and the ring buffers for communication with the backend, and
1788  * inform the backend of the appropriate details for those.
1789  */
1790 static int netfront_probe(struct xenbus_device *dev,
1791                           const struct xenbus_device_id *id)
1792 {
1793         int err;
1794         struct net_device *netdev;
1795         struct netfront_info *info;
1796
1797         netdev = xennet_create_dev(dev);
1798         if (IS_ERR(netdev)) {
1799                 err = PTR_ERR(netdev);
1800                 xenbus_dev_fatal(dev, err, "creating netdev");
1801                 return err;
1802         }
1803
1804         info = netdev_priv(netdev);
1805         dev_set_drvdata(&dev->dev, info);
1806 #ifdef CONFIG_SYSFS
1807         info->netdev->sysfs_groups[0] = &xennet_dev_group;
1808 #endif
1809
1810         return 0;
1811 }
1812
1813 static void xennet_end_access(int ref, void *page)
1814 {
1815         /* This frees the page as a side-effect */
1816         if (ref != INVALID_GRANT_REF)
1817                 gnttab_end_foreign_access(ref, virt_to_page(page));
1818 }
1819
1820 static void xennet_disconnect_backend(struct netfront_info *info)
1821 {
1822         unsigned int i = 0;
1823         unsigned int num_queues = info->netdev->real_num_tx_queues;
1824
1825         netif_carrier_off(info->netdev);
1826
1827         for (i = 0; i < num_queues && info->queues; ++i) {
1828                 struct netfront_queue *queue = &info->queues[i];
1829
1830                 timer_delete_sync(&queue->rx_refill_timer);
1831
1832                 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq))
1833                         unbind_from_irqhandler(queue->tx_irq, queue);
1834                 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) {
1835                         unbind_from_irqhandler(queue->tx_irq, queue);
1836                         unbind_from_irqhandler(queue->rx_irq, queue);
1837                 }
1838                 queue->tx_evtchn = queue->rx_evtchn = 0;
1839                 queue->tx_irq = queue->rx_irq = 0;
1840
1841                 if (netif_running(info->netdev))
1842                         napi_synchronize(&queue->napi);
1843
1844                 xennet_release_tx_bufs(queue);
1845                 xennet_release_rx_bufs(queue);
1846                 gnttab_free_grant_references(queue->gref_tx_head);
1847                 gnttab_free_grant_references(queue->gref_rx_head);
1848
1849                 /* End access and free the pages */
1850                 xennet_end_access(queue->tx_ring_ref, queue->tx.sring);
1851                 xennet_end_access(queue->rx_ring_ref, queue->rx.sring);
1852
1853                 queue->tx_ring_ref = INVALID_GRANT_REF;
1854                 queue->rx_ring_ref = INVALID_GRANT_REF;
1855                 queue->tx.sring = NULL;
1856                 queue->rx.sring = NULL;
1857
1858                 page_pool_destroy(queue->page_pool);
1859         }
1860 }
1861
1862 /*
1863  * We are reconnecting to the backend, due to a suspend/resume, or a backend
1864  * driver restart.  We tear down our netif structure and recreate it, but
1865  * leave the device-layer structures intact so that this is transparent to the
1866  * rest of the kernel.
1867  */
1868 static int netfront_resume(struct xenbus_device *dev)
1869 {
1870         struct netfront_info *info = dev_get_drvdata(&dev->dev);
1871
1872         dev_dbg(&dev->dev, "%s\n", dev->nodename);
1873
1874         netif_tx_lock_bh(info->netdev);
1875         netif_device_detach(info->netdev);
1876         netif_tx_unlock_bh(info->netdev);
1877
1878         xennet_disconnect_backend(info);
1879
1880         rtnl_lock();
1881         if (info->queues)
1882                 xennet_destroy_queues(info);
1883         rtnl_unlock();
1884
1885         return 0;
1886 }
1887
1888 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
1889 {
1890         char *s, *e, *macstr;
1891         int i;
1892
1893         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
1894         if (IS_ERR(macstr))
1895                 return PTR_ERR(macstr);
1896
1897         for (i = 0; i < ETH_ALEN; i++) {
1898                 mac[i] = simple_strtoul(s, &e, 16);
1899                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
1900                         kfree(macstr);
1901                         return -ENOENT;
1902                 }
1903                 s = e+1;
1904         }
1905
1906         kfree(macstr);
1907         return 0;
1908 }
1909
1910 static int setup_netfront_single(struct netfront_queue *queue)
1911 {
1912         int err;
1913
1914         err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1915         if (err < 0)
1916                 goto fail;
1917
1918         err = bind_evtchn_to_irqhandler_lateeoi(queue->tx_evtchn,
1919                                                 xennet_interrupt, 0,
1920                                                 queue->info->netdev->name,
1921                                                 queue);
1922         if (err < 0)
1923                 goto bind_fail;
1924         queue->rx_evtchn = queue->tx_evtchn;
1925         queue->rx_irq = queue->tx_irq = err;
1926
1927         return 0;
1928
1929 bind_fail:
1930         xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1931         queue->tx_evtchn = 0;
1932 fail:
1933         return err;
1934 }
1935
1936 static int setup_netfront_split(struct netfront_queue *queue)
1937 {
1938         int err;
1939
1940         err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn);
1941         if (err < 0)
1942                 goto fail;
1943         err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn);
1944         if (err < 0)
1945                 goto alloc_rx_evtchn_fail;
1946
1947         snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name),
1948                  "%s-tx", queue->name);
1949         err = bind_evtchn_to_irqhandler_lateeoi(queue->tx_evtchn,
1950                                                 xennet_tx_interrupt, 0,
1951                                                 queue->tx_irq_name, queue);
1952         if (err < 0)
1953                 goto bind_tx_fail;
1954         queue->tx_irq = err;
1955
1956         snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name),
1957                  "%s-rx", queue->name);
1958         err = bind_evtchn_to_irqhandler_lateeoi(queue->rx_evtchn,
1959                                                 xennet_rx_interrupt, 0,
1960                                                 queue->rx_irq_name, queue);
1961         if (err < 0)
1962                 goto bind_rx_fail;
1963         queue->rx_irq = err;
1964
1965         return 0;
1966
1967 bind_rx_fail:
1968         unbind_from_irqhandler(queue->tx_irq, queue);
1969         queue->tx_irq = 0;
1970 bind_tx_fail:
1971         xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn);
1972         queue->rx_evtchn = 0;
1973 alloc_rx_evtchn_fail:
1974         xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn);
1975         queue->tx_evtchn = 0;
1976 fail:
1977         return err;
1978 }
1979
1980 static int setup_netfront(struct xenbus_device *dev,
1981                         struct netfront_queue *queue, unsigned int feature_split_evtchn)
1982 {
1983         struct xen_netif_tx_sring *txs;
1984         struct xen_netif_rx_sring *rxs;
1985         int err;
1986
1987         queue->tx_ring_ref = INVALID_GRANT_REF;
1988         queue->rx_ring_ref = INVALID_GRANT_REF;
1989         queue->rx.sring = NULL;
1990         queue->tx.sring = NULL;
1991
1992         err = xenbus_setup_ring(dev, GFP_NOIO | __GFP_HIGH, (void **)&txs,
1993                                 1, &queue->tx_ring_ref);
1994         if (err)
1995                 goto fail;
1996
1997         XEN_FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE);
1998
1999         err = xenbus_setup_ring(dev, GFP_NOIO | __GFP_HIGH, (void **)&rxs,
2000                                 1, &queue->rx_ring_ref);
2001         if (err)
2002                 goto fail;
2003
2004         XEN_FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE);
2005
2006         if (feature_split_evtchn)
2007                 err = setup_netfront_split(queue);
2008         /* setup single event channel if
2009          *  a) feature-split-event-channels == 0
2010          *  b) feature-split-event-channels == 1 but failed to setup
2011          */
2012         if (!feature_split_evtchn || err)
2013                 err = setup_netfront_single(queue);
2014
2015         if (err)
2016                 goto fail;
2017
2018         return 0;
2019
2020  fail:
2021         xenbus_teardown_ring((void **)&queue->rx.sring, 1, &queue->rx_ring_ref);
2022         xenbus_teardown_ring((void **)&queue->tx.sring, 1, &queue->tx_ring_ref);
2023
2024         return err;
2025 }
2026
2027 /* Queue-specific initialisation
2028  * This used to be done in xennet_create_dev() but must now
2029  * be run per-queue.
2030  */
2031 static int xennet_init_queue(struct netfront_queue *queue)
2032 {
2033         unsigned short i;
2034         int err = 0;
2035         char *devid;
2036
2037         spin_lock_init(&queue->tx_lock);
2038         spin_lock_init(&queue->rx_lock);
2039         spin_lock_init(&queue->rx_cons_lock);
2040
2041         timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0);
2042
2043         devid = strrchr(queue->info->xbdev->nodename, '/') + 1;
2044         snprintf(queue->name, sizeof(queue->name), "vif%s-q%u",
2045                  devid, queue->id);
2046
2047         /* Initialise tx_skb_freelist as a free chain containing every entry. */
2048         queue->tx_skb_freelist = 0;
2049         queue->tx_pend_queue = TX_LINK_NONE;
2050         for (i = 0; i < NET_TX_RING_SIZE; i++) {
2051                 queue->tx_link[i] = i + 1;
2052                 queue->grant_tx_ref[i] = INVALID_GRANT_REF;
2053                 queue->grant_tx_page[i] = NULL;
2054         }
2055         queue->tx_link[NET_TX_RING_SIZE - 1] = TX_LINK_NONE;
2056
2057         /* Clear out rx_skbs */
2058         for (i = 0; i < NET_RX_RING_SIZE; i++) {
2059                 queue->rx_skbs[i] = NULL;
2060                 queue->grant_rx_ref[i] = INVALID_GRANT_REF;
2061         }
2062
2063         /* A grant for every tx ring slot */
2064         if (gnttab_alloc_grant_references(NET_TX_RING_SIZE,
2065                                           &queue->gref_tx_head) < 0) {
2066                 pr_alert("can't alloc tx grant refs\n");
2067                 err = -ENOMEM;
2068                 goto exit;
2069         }
2070
2071         /* A grant for every rx ring slot */
2072         if (gnttab_alloc_grant_references(NET_RX_RING_SIZE,
2073                                           &queue->gref_rx_head) < 0) {
2074                 pr_alert("can't alloc rx grant refs\n");
2075                 err = -ENOMEM;
2076                 goto exit_free_tx;
2077         }
2078
2079         return 0;
2080
2081  exit_free_tx:
2082         gnttab_free_grant_references(queue->gref_tx_head);
2083  exit:
2084         return err;
2085 }
2086
2087 static int write_queue_xenstore_keys(struct netfront_queue *queue,
2088                            struct xenbus_transaction *xbt, int write_hierarchical)
2089 {
2090         /* Write the queue-specific keys into XenStore in the traditional
2091          * way for a single queue, or in a queue subkeys for multiple
2092          * queues.
2093          */
2094         struct xenbus_device *dev = queue->info->xbdev;
2095         int err;
2096         const char *message;
2097         char *path;
2098         size_t pathsize;
2099
2100         /* Choose the correct place to write the keys */
2101         if (write_hierarchical) {
2102                 pathsize = strlen(dev->nodename) + 10;
2103                 path = kzalloc(pathsize, GFP_KERNEL);
2104                 if (!path) {
2105                         err = -ENOMEM;
2106                         message = "out of memory while writing ring references";
2107                         goto error;
2108                 }
2109                 snprintf(path, pathsize, "%s/queue-%u",
2110                                 dev->nodename, queue->id);
2111         } else {
2112                 path = (char *)dev->nodename;
2113         }
2114
2115         /* Write ring references */
2116         err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u",
2117                         queue->tx_ring_ref);
2118         if (err) {
2119                 message = "writing tx-ring-ref";
2120                 goto error;
2121         }
2122
2123         err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u",
2124                         queue->rx_ring_ref);
2125         if (err) {
2126                 message = "writing rx-ring-ref";
2127                 goto error;
2128         }
2129
2130         /* Write event channels; taking into account both shared
2131          * and split event channel scenarios.
2132          */
2133         if (queue->tx_evtchn == queue->rx_evtchn) {
2134                 /* Shared event channel */
2135                 err = xenbus_printf(*xbt, path,
2136                                 "event-channel", "%u", queue->tx_evtchn);
2137                 if (err) {
2138                         message = "writing event-channel";
2139                         goto error;
2140                 }
2141         } else {
2142                 /* Split event channels */
2143                 err = xenbus_printf(*xbt, path,
2144                                 "event-channel-tx", "%u", queue->tx_evtchn);
2145                 if (err) {
2146                         message = "writing event-channel-tx";
2147                         goto error;
2148                 }
2149
2150                 err = xenbus_printf(*xbt, path,
2151                                 "event-channel-rx", "%u", queue->rx_evtchn);
2152                 if (err) {
2153                         message = "writing event-channel-rx";
2154                         goto error;
2155                 }
2156         }
2157
2158         if (write_hierarchical)
2159                 kfree(path);
2160         return 0;
2161
2162 error:
2163         if (write_hierarchical)
2164                 kfree(path);
2165         xenbus_dev_fatal(dev, err, "%s", message);
2166         return err;
2167 }
2168
2169
2170
2171 static int xennet_create_page_pool(struct netfront_queue *queue)
2172 {
2173         int err;
2174         struct page_pool_params pp_params = {
2175                 .order = 0,
2176                 .flags = 0,
2177                 .pool_size = NET_RX_RING_SIZE,
2178                 .nid = NUMA_NO_NODE,
2179                 .dev = &queue->info->netdev->dev,
2180                 .offset = XDP_PACKET_HEADROOM,
2181                 .max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM,
2182         };
2183
2184         queue->page_pool = page_pool_create(&pp_params);
2185         if (IS_ERR(queue->page_pool)) {
2186                 err = PTR_ERR(queue->page_pool);
2187                 queue->page_pool = NULL;
2188                 return err;
2189         }
2190
2191         err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev,
2192                                queue->id, 0);
2193         if (err) {
2194                 netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n");
2195                 goto err_free_pp;
2196         }
2197
2198         err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq,
2199                                          MEM_TYPE_PAGE_POOL, queue->page_pool);
2200         if (err) {
2201                 netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n");
2202                 goto err_unregister_rxq;
2203         }
2204         return 0;
2205
2206 err_unregister_rxq:
2207         xdp_rxq_info_unreg(&queue->xdp_rxq);
2208 err_free_pp:
2209         page_pool_destroy(queue->page_pool);
2210         queue->page_pool = NULL;
2211         return err;
2212 }
2213
2214 static int xennet_create_queues(struct netfront_info *info,
2215                                 unsigned int *num_queues)
2216 {
2217         unsigned int i;
2218         int ret;
2219
2220         info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue),
2221                                GFP_KERNEL);
2222         if (!info->queues)
2223                 return -ENOMEM;
2224
2225         for (i = 0; i < *num_queues; i++) {
2226                 struct netfront_queue *queue = &info->queues[i];
2227
2228                 queue->id = i;
2229                 queue->info = info;
2230
2231                 ret = xennet_init_queue(queue);
2232                 if (ret < 0) {
2233                         dev_warn(&info->xbdev->dev,
2234                                  "only created %d queues\n", i);
2235                         *num_queues = i;
2236                         break;
2237                 }
2238
2239                 /* use page pool recycling instead of buddy allocator */
2240                 ret = xennet_create_page_pool(queue);
2241                 if (ret < 0) {
2242                         dev_err(&info->xbdev->dev, "can't allocate page pool\n");
2243                         *num_queues = i;
2244                         return ret;
2245                 }
2246
2247                 netif_napi_add(queue->info->netdev, &queue->napi, xennet_poll);
2248                 if (netif_running(info->netdev))
2249                         napi_enable(&queue->napi);
2250         }
2251
2252         netif_set_real_num_tx_queues(info->netdev, *num_queues);
2253
2254         if (*num_queues == 0) {
2255                 dev_err(&info->xbdev->dev, "no queues\n");
2256                 return -EINVAL;
2257         }
2258         return 0;
2259 }
2260
2261 /* Common code used when first setting up, and when resuming. */
2262 static int talk_to_netback(struct xenbus_device *dev,
2263                            struct netfront_info *info)
2264 {
2265         const char *message;
2266         struct xenbus_transaction xbt;
2267         int err;
2268         unsigned int feature_split_evtchn;
2269         unsigned int i = 0;
2270         unsigned int max_queues = 0;
2271         struct netfront_queue *queue = NULL;
2272         unsigned int num_queues = 1;
2273         u8 addr[ETH_ALEN];
2274
2275         info->netdev->irq = 0;
2276
2277         /* Check if backend is trusted. */
2278         info->bounce = !xennet_trusted ||
2279                        !xenbus_read_unsigned(dev->nodename, "trusted", 1);
2280
2281         /* Check if backend supports multiple queues */
2282         max_queues = xenbus_read_unsigned(info->xbdev->otherend,
2283                                           "multi-queue-max-queues", 1);
2284         num_queues = min(max_queues, xennet_max_queues);
2285
2286         /* Check feature-split-event-channels */
2287         feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend,
2288                                         "feature-split-event-channels", 0);
2289
2290         /* Read mac addr. */
2291         err = xen_net_read_mac(dev, addr);
2292         if (err) {
2293                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
2294                 goto out_unlocked;
2295         }
2296         eth_hw_addr_set(info->netdev, addr);
2297
2298         info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend,
2299                                                               "feature-xdp-headroom", 0);
2300         if (info->netback_has_xdp_headroom) {
2301                 /* set the current xen-netfront xdp state */
2302                 err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ?
2303                                           NETBACK_XDP_HEADROOM_ENABLE :
2304                                           NETBACK_XDP_HEADROOM_DISABLE);
2305                 if (err)
2306                         goto out_unlocked;
2307         }
2308
2309         rtnl_lock();
2310         if (info->queues)
2311                 xennet_destroy_queues(info);
2312
2313         /* For the case of a reconnect reset the "broken" indicator. */
2314         info->broken = false;
2315
2316         err = xennet_create_queues(info, &num_queues);
2317         if (err < 0) {
2318                 xenbus_dev_fatal(dev, err, "creating queues");
2319                 kfree(info->queues);
2320                 info->queues = NULL;
2321                 goto out;
2322         }
2323         rtnl_unlock();
2324
2325         /* Create shared ring, alloc event channel -- for each queue */
2326         for (i = 0; i < num_queues; ++i) {
2327                 queue = &info->queues[i];
2328                 err = setup_netfront(dev, queue, feature_split_evtchn);
2329                 if (err)
2330                         goto destroy_ring;
2331         }
2332
2333 again:
2334         err = xenbus_transaction_start(&xbt);
2335         if (err) {
2336                 xenbus_dev_fatal(dev, err, "starting transaction");
2337                 goto destroy_ring;
2338         }
2339
2340         if (xenbus_exists(XBT_NIL,
2341                           info->xbdev->otherend, "multi-queue-max-queues")) {
2342                 /* Write the number of queues */
2343                 err = xenbus_printf(xbt, dev->nodename,
2344                                     "multi-queue-num-queues", "%u", num_queues);
2345                 if (err) {
2346                         message = "writing multi-queue-num-queues";
2347                         goto abort_transaction_no_dev_fatal;
2348                 }
2349         }
2350
2351         if (num_queues == 1) {
2352                 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */
2353                 if (err)
2354                         goto abort_transaction_no_dev_fatal;
2355         } else {
2356                 /* Write the keys for each queue */
2357                 for (i = 0; i < num_queues; ++i) {
2358                         queue = &info->queues[i];
2359                         err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */
2360                         if (err)
2361                                 goto abort_transaction_no_dev_fatal;
2362                 }
2363         }
2364
2365         /* The remaining keys are not queue-specific */
2366         err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u",
2367                             1);
2368         if (err) {
2369                 message = "writing request-rx-copy";
2370                 goto abort_transaction;
2371         }
2372
2373         err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1);
2374         if (err) {
2375                 message = "writing feature-rx-notify";
2376                 goto abort_transaction;
2377         }
2378
2379         err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1);
2380         if (err) {
2381                 message = "writing feature-sg";
2382                 goto abort_transaction;
2383         }
2384
2385         err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1);
2386         if (err) {
2387                 message = "writing feature-gso-tcpv4";
2388                 goto abort_transaction;
2389         }
2390
2391         err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1");
2392         if (err) {
2393                 message = "writing feature-gso-tcpv6";
2394                 goto abort_transaction;
2395         }
2396
2397         err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload",
2398                            "1");
2399         if (err) {
2400                 message = "writing feature-ipv6-csum-offload";
2401                 goto abort_transaction;
2402         }
2403
2404         err = xenbus_transaction_end(xbt, 0);
2405         if (err) {
2406                 if (err == -EAGAIN)
2407                         goto again;
2408                 xenbus_dev_fatal(dev, err, "completing transaction");
2409                 goto destroy_ring;
2410         }
2411
2412         return 0;
2413
2414  abort_transaction:
2415         xenbus_dev_fatal(dev, err, "%s", message);
2416 abort_transaction_no_dev_fatal:
2417         xenbus_transaction_end(xbt, 1);
2418  destroy_ring:
2419         xennet_disconnect_backend(info);
2420         rtnl_lock();
2421         xennet_destroy_queues(info);
2422  out:
2423         rtnl_unlock();
2424 out_unlocked:
2425         device_unregister(&dev->dev);
2426         return err;
2427 }
2428
2429 static int xennet_connect(struct net_device *dev)
2430 {
2431         struct netfront_info *np = netdev_priv(dev);
2432         unsigned int num_queues = 0;
2433         int err;
2434         unsigned int j = 0;
2435         struct netfront_queue *queue = NULL;
2436
2437         if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) {
2438                 dev_info(&dev->dev,
2439                          "backend does not support copying receive path\n");
2440                 return -ENODEV;
2441         }
2442
2443         err = talk_to_netback(np->xbdev, np);
2444         if (err)
2445                 return err;
2446         if (np->netback_has_xdp_headroom)
2447                 pr_info("backend supports XDP headroom\n");
2448         if (np->bounce)
2449                 dev_info(&np->xbdev->dev,
2450                          "bouncing transmitted data to zeroed pages\n");
2451
2452         /* talk_to_netback() sets the correct number of queues */
2453         num_queues = dev->real_num_tx_queues;
2454
2455         if (dev->reg_state == NETREG_UNINITIALIZED) {
2456                 err = register_netdev(dev);
2457                 if (err) {
2458                         pr_warn("%s: register_netdev err=%d\n", __func__, err);
2459                         device_unregister(&np->xbdev->dev);
2460                         return err;
2461                 }
2462         }
2463
2464         rtnl_lock();
2465         netdev_update_features(dev);
2466         rtnl_unlock();
2467
2468         /*
2469          * All public and private state should now be sane.  Get
2470          * ready to start sending and receiving packets and give the driver
2471          * domain a kick because we've probably just requeued some
2472          * packets.
2473          */
2474         netif_tx_lock_bh(np->netdev);
2475         netif_device_attach(np->netdev);
2476         netif_tx_unlock_bh(np->netdev);
2477
2478         netif_carrier_on(np->netdev);
2479         for (j = 0; j < num_queues; ++j) {
2480                 queue = &np->queues[j];
2481
2482                 notify_remote_via_irq(queue->tx_irq);
2483                 if (queue->tx_irq != queue->rx_irq)
2484                         notify_remote_via_irq(queue->rx_irq);
2485
2486                 spin_lock_bh(&queue->rx_lock);
2487                 xennet_alloc_rx_buffers(queue);
2488                 spin_unlock_bh(&queue->rx_lock);
2489         }
2490
2491         return 0;
2492 }
2493
2494 /*
2495  * Callback received when the backend's state changes.
2496  */
2497 static void netback_changed(struct xenbus_device *dev,
2498                             enum xenbus_state backend_state)
2499 {
2500         struct netfront_info *np = dev_get_drvdata(&dev->dev);
2501         struct net_device *netdev = np->netdev;
2502
2503         dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state));
2504
2505         wake_up_all(&module_wq);
2506
2507         switch (backend_state) {
2508         case XenbusStateInitialising:
2509         case XenbusStateInitialised:
2510         case XenbusStateReconfiguring:
2511         case XenbusStateReconfigured:
2512         case XenbusStateUnknown:
2513                 break;
2514
2515         case XenbusStateInitWait:
2516                 if (dev->state != XenbusStateInitialising)
2517                         break;
2518                 if (xennet_connect(netdev) != 0)
2519                         break;
2520                 xenbus_switch_state(dev, XenbusStateConnected);
2521                 break;
2522
2523         case XenbusStateConnected:
2524                 netdev_notify_peers(netdev);
2525                 break;
2526
2527         case XenbusStateClosed:
2528                 if (dev->state == XenbusStateClosed)
2529                         break;
2530                 fallthrough;    /* Missed the backend's CLOSING state */
2531         case XenbusStateClosing:
2532                 xenbus_frontend_closed(dev);
2533                 break;
2534         }
2535 }
2536
2537 static const struct xennet_stat {
2538         char name[ETH_GSTRING_LEN];
2539         u16 offset;
2540 } xennet_stats[] = {
2541         {
2542                 "rx_gso_checksum_fixup",
2543                 offsetof(struct netfront_info, rx_gso_checksum_fixup)
2544         },
2545 };
2546
2547 static int xennet_get_sset_count(struct net_device *dev, int string_set)
2548 {
2549         switch (string_set) {
2550         case ETH_SS_STATS:
2551                 return ARRAY_SIZE(xennet_stats);
2552         default:
2553                 return -EINVAL;
2554         }
2555 }
2556
2557 static void xennet_get_ethtool_stats(struct net_device *dev,
2558                                      struct ethtool_stats *stats, u64 * data)
2559 {
2560         void *np = netdev_priv(dev);
2561         int i;
2562
2563         for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2564                 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset));
2565 }
2566
2567 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data)
2568 {
2569         int i;
2570
2571         switch (stringset) {
2572         case ETH_SS_STATS:
2573                 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++)
2574                         memcpy(data + i * ETH_GSTRING_LEN,
2575                                xennet_stats[i].name, ETH_GSTRING_LEN);
2576                 break;
2577         }
2578 }
2579
2580 static const struct ethtool_ops xennet_ethtool_ops =
2581 {
2582         .get_link = ethtool_op_get_link,
2583
2584         .get_sset_count = xennet_get_sset_count,
2585         .get_ethtool_stats = xennet_get_ethtool_stats,
2586         .get_strings = xennet_get_strings,
2587         .get_ts_info = ethtool_op_get_ts_info,
2588 };
2589
2590 #ifdef CONFIG_SYSFS
2591 static ssize_t show_rxbuf(struct device *dev,
2592                           struct device_attribute *attr, char *buf)
2593 {
2594         return sprintf(buf, "%lu\n", NET_RX_RING_SIZE);
2595 }
2596
2597 static ssize_t store_rxbuf(struct device *dev,
2598                            struct device_attribute *attr,
2599                            const char *buf, size_t len)
2600 {
2601         char *endp;
2602
2603         if (!capable(CAP_NET_ADMIN))
2604                 return -EPERM;
2605
2606         simple_strtoul(buf, &endp, 0);
2607         if (endp == buf)
2608                 return -EBADMSG;
2609
2610         /* rxbuf_min and rxbuf_max are no longer configurable. */
2611
2612         return len;
2613 }
2614
2615 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf);
2616 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf);
2617 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL);
2618
2619 static struct attribute *xennet_dev_attrs[] = {
2620         &dev_attr_rxbuf_min.attr,
2621         &dev_attr_rxbuf_max.attr,
2622         &dev_attr_rxbuf_cur.attr,
2623         NULL
2624 };
2625
2626 static const struct attribute_group xennet_dev_group = {
2627         .attrs = xennet_dev_attrs
2628 };
2629 #endif /* CONFIG_SYSFS */
2630
2631 static void xennet_bus_close(struct xenbus_device *dev)
2632 {
2633         int ret;
2634
2635         if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2636                 return;
2637         do {
2638                 xenbus_switch_state(dev, XenbusStateClosing);
2639                 ret = wait_event_timeout(module_wq,
2640                                    xenbus_read_driver_state(dev->otherend) ==
2641                                    XenbusStateClosing ||
2642                                    xenbus_read_driver_state(dev->otherend) ==
2643                                    XenbusStateClosed ||
2644                                    xenbus_read_driver_state(dev->otherend) ==
2645                                    XenbusStateUnknown,
2646                                    XENNET_TIMEOUT);
2647         } while (!ret);
2648
2649         if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed)
2650                 return;
2651
2652         do {
2653                 xenbus_switch_state(dev, XenbusStateClosed);
2654                 ret = wait_event_timeout(module_wq,
2655                                    xenbus_read_driver_state(dev->otherend) ==
2656                                    XenbusStateClosed ||
2657                                    xenbus_read_driver_state(dev->otherend) ==
2658                                    XenbusStateUnknown,
2659                                    XENNET_TIMEOUT);
2660         } while (!ret);
2661 }
2662
2663 static void xennet_remove(struct xenbus_device *dev)
2664 {
2665         struct netfront_info *info = dev_get_drvdata(&dev->dev);
2666
2667         xennet_bus_close(dev);
2668         xennet_disconnect_backend(info);
2669
2670         if (info->netdev->reg_state == NETREG_REGISTERED)
2671                 unregister_netdev(info->netdev);
2672
2673         if (info->queues) {
2674                 rtnl_lock();
2675                 xennet_destroy_queues(info);
2676                 rtnl_unlock();
2677         }
2678         xennet_free_netdev(info->netdev);
2679 }
2680
2681 static const struct xenbus_device_id netfront_ids[] = {
2682         { "vif" },
2683         { "" }
2684 };
2685
2686 static struct xenbus_driver netfront_driver = {
2687         .ids = netfront_ids,
2688         .probe = netfront_probe,
2689         .remove = xennet_remove,
2690         .resume = netfront_resume,
2691         .otherend_changed = netback_changed,
2692 };
2693
2694 static int __init netif_init(void)
2695 {
2696         if (!xen_domain())
2697                 return -ENODEV;
2698
2699         if (!xen_has_pv_nic_devices())
2700                 return -ENODEV;
2701
2702         pr_info("Initialising Xen virtual ethernet driver\n");
2703
2704         /* Allow as many queues as there are CPUs inut max. 8 if user has not
2705          * specified a value.
2706          */
2707         if (xennet_max_queues == 0)
2708                 xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT,
2709                                           num_online_cpus());
2710
2711         return xenbus_register_frontend(&netfront_driver);
2712 }
2713 module_init(netif_init);
2714
2715
2716 static void __exit netif_exit(void)
2717 {
2718         xenbus_unregister_driver(&netfront_driver);
2719 }
2720 module_exit(netif_exit);
2721
2722 MODULE_DESCRIPTION("Xen virtual network device frontend");
2723 MODULE_LICENSE("GPL");
2724 MODULE_ALIAS("xen:vif");
2725 MODULE_ALIAS("xennet");