drivers/net/xen-netfront.c

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