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"); 2544