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