1 /* 2 * Virtual network driver for conversing with remote driver backends. 3 * 4 * Copyright (c) 2002-2005, K A Fraser 5 * Copyright (c) 2005, XenSource Ltd 6 * 7 * This program is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU General Public License version 2 9 * as published by the Free Software Foundation; or, when distributed 10 * separately from the Linux kernel or incorporated into other 11 * software packages, subject to the following license: 12 * 13 * Permission is hereby granted, free of charge, to any person obtaining a copy 14 * of this source file (the "Software"), to deal in the Software without 15 * restriction, including without limitation the rights to use, copy, modify, 16 * merge, publish, distribute, sublicense, and/or sell copies of the Software, 17 * and to permit persons to whom the Software is furnished to do so, subject to 18 * the following conditions: 19 * 20 * The above copyright notice and this permission notice shall be included in 21 * all copies or substantial portions of the Software. 22 * 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 24 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 25 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 26 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 27 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 28 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 29 * IN THE SOFTWARE. 30 */ 31 32 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt 33 34 #include <linux/module.h> 35 #include <linux/kernel.h> 36 #include <linux/netdevice.h> 37 #include <linux/etherdevice.h> 38 #include <linux/skbuff.h> 39 #include <linux/ethtool.h> 40 #include <linux/if_ether.h> 41 #include <net/tcp.h> 42 #include <linux/udp.h> 43 #include <linux/moduleparam.h> 44 #include <linux/mm.h> 45 #include <linux/slab.h> 46 #include <net/ip.h> 47 #include <linux/bpf.h> 48 #include <net/page_pool.h> 49 #include <linux/bpf_trace.h> 50 51 #include <xen/xen.h> 52 #include <xen/xenbus.h> 53 #include <xen/events.h> 54 #include <xen/page.h> 55 #include <xen/platform_pci.h> 56 #include <xen/grant_table.h> 57 58 #include <xen/interface/io/netif.h> 59 #include <xen/interface/memory.h> 60 #include <xen/interface/grant_table.h> 61 62 /* Module parameters */ 63 #define MAX_QUEUES_DEFAULT 8 64 static unsigned int xennet_max_queues; 65 module_param_named(max_queues, xennet_max_queues, uint, 0644); 66 MODULE_PARM_DESC(max_queues, 67 "Maximum number of queues per virtual interface"); 68 69 #define XENNET_TIMEOUT (5 * HZ) 70 71 static const struct ethtool_ops xennet_ethtool_ops; 72 73 struct netfront_cb { 74 int pull_to; 75 }; 76 77 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb)) 78 79 #define RX_COPY_THRESHOLD 256 80 81 #define GRANT_INVALID_REF 0 82 83 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE) 84 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE) 85 86 /* Minimum number of Rx slots (includes slot for GSO metadata). */ 87 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1) 88 89 /* Queue name is interface name with "-qNNN" appended */ 90 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6) 91 92 /* IRQ name is queue name with "-tx" or "-rx" appended */ 93 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3) 94 95 static DECLARE_WAIT_QUEUE_HEAD(module_wq); 96 97 struct netfront_stats { 98 u64 packets; 99 u64 bytes; 100 struct u64_stats_sync syncp; 101 }; 102 103 struct netfront_info; 104 105 struct netfront_queue { 106 unsigned int id; /* Queue ID, 0-based */ 107 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */ 108 struct netfront_info *info; 109 110 struct bpf_prog __rcu *xdp_prog; 111 112 struct napi_struct napi; 113 114 /* Split event channels support, tx_* == rx_* when using 115 * single event channel. 116 */ 117 unsigned int tx_evtchn, rx_evtchn; 118 unsigned int tx_irq, rx_irq; 119 /* Only used when split event channels support is enabled */ 120 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */ 121 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */ 122 123 spinlock_t tx_lock; 124 struct xen_netif_tx_front_ring tx; 125 int tx_ring_ref; 126 127 /* 128 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries 129 * are linked from tx_skb_freelist through skb_entry.link. 130 * 131 * NB. Freelist index entries are always going to be less than 132 * PAGE_OFFSET, whereas pointers to skbs will always be equal or 133 * greater than PAGE_OFFSET: we use this property to distinguish 134 * them. 135 */ 136 union skb_entry { 137 struct sk_buff *skb; 138 unsigned long link; 139 } tx_skbs[NET_TX_RING_SIZE]; 140 grant_ref_t gref_tx_head; 141 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; 142 struct page *grant_tx_page[NET_TX_RING_SIZE]; 143 unsigned tx_skb_freelist; 144 145 spinlock_t rx_lock ____cacheline_aligned_in_smp; 146 struct xen_netif_rx_front_ring rx; 147 int rx_ring_ref; 148 149 struct timer_list rx_refill_timer; 150 151 struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; 152 grant_ref_t gref_rx_head; 153 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; 154 155 struct page_pool *page_pool; 156 struct xdp_rxq_info xdp_rxq; 157 }; 158 159 struct netfront_info { 160 struct list_head list; 161 struct net_device *netdev; 162 163 struct xenbus_device *xbdev; 164 165 /* Multi-queue support */ 166 struct netfront_queue *queues; 167 168 /* Statistics */ 169 struct netfront_stats __percpu *rx_stats; 170 struct netfront_stats __percpu *tx_stats; 171 172 /* XDP state */ 173 bool netback_has_xdp_headroom; 174 bool netfront_xdp_enabled; 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 static void skb_entry_set_link(union skb_entry *list, unsigned short id) 185 { 186 list->link = id; 187 } 188 189 static int skb_entry_is_link(const union skb_entry *list) 190 { 191 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link)); 192 return (unsigned long)list->skb < PAGE_OFFSET; 193 } 194 195 /* 196 * Access macros for acquiring freeing slots in tx_skbs[]. 197 */ 198 199 static void add_id_to_freelist(unsigned *head, union skb_entry *list, 200 unsigned short id) 201 { 202 skb_entry_set_link(&list[id], *head); 203 *head = id; 204 } 205 206 static unsigned short get_id_from_freelist(unsigned *head, 207 union skb_entry *list) 208 { 209 unsigned int id = *head; 210 *head = list[id].link; 211 return id; 212 } 213 214 static int xennet_rxidx(RING_IDX idx) 215 { 216 return idx & (NET_RX_RING_SIZE - 1); 217 } 218 219 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue, 220 RING_IDX ri) 221 { 222 int i = xennet_rxidx(ri); 223 struct sk_buff *skb = queue->rx_skbs[i]; 224 queue->rx_skbs[i] = NULL; 225 return skb; 226 } 227 228 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue, 229 RING_IDX ri) 230 { 231 int i = xennet_rxidx(ri); 232 grant_ref_t ref = queue->grant_rx_ref[i]; 233 queue->grant_rx_ref[i] = GRANT_INVALID_REF; 234 return ref; 235 } 236 237 #ifdef CONFIG_SYSFS 238 static const struct attribute_group xennet_dev_group; 239 #endif 240 241 static bool xennet_can_sg(struct net_device *dev) 242 { 243 return dev->features & NETIF_F_SG; 244 } 245 246 247 static void rx_refill_timeout(struct timer_list *t) 248 { 249 struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer); 250 napi_schedule(&queue->napi); 251 } 252 253 static int netfront_tx_slot_available(struct netfront_queue *queue) 254 { 255 return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) < 256 (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1); 257 } 258 259 static void xennet_maybe_wake_tx(struct netfront_queue *queue) 260 { 261 struct net_device *dev = queue->info->netdev; 262 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id); 263 264 if (unlikely(netif_tx_queue_stopped(dev_queue)) && 265 netfront_tx_slot_available(queue) && 266 likely(netif_running(dev))) 267 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id)); 268 } 269 270 271 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue) 272 { 273 struct sk_buff *skb; 274 struct page *page; 275 276 skb = __netdev_alloc_skb(queue->info->netdev, 277 RX_COPY_THRESHOLD + NET_IP_ALIGN, 278 GFP_ATOMIC | __GFP_NOWARN); 279 if (unlikely(!skb)) 280 return NULL; 281 282 page = page_pool_dev_alloc_pages(queue->page_pool); 283 if (unlikely(!page)) { 284 kfree_skb(skb); 285 return NULL; 286 } 287 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE); 288 289 /* Align ip header to a 16 bytes boundary */ 290 skb_reserve(skb, NET_IP_ALIGN); 291 skb->dev = queue->info->netdev; 292 293 return skb; 294 } 295 296 297 static void xennet_alloc_rx_buffers(struct netfront_queue *queue) 298 { 299 RING_IDX req_prod = queue->rx.req_prod_pvt; 300 int notify; 301 int err = 0; 302 303 if (unlikely(!netif_carrier_ok(queue->info->netdev))) 304 return; 305 306 for (req_prod = queue->rx.req_prod_pvt; 307 req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE; 308 req_prod++) { 309 struct sk_buff *skb; 310 unsigned short id; 311 grant_ref_t ref; 312 struct page *page; 313 struct xen_netif_rx_request *req; 314 315 skb = xennet_alloc_one_rx_buffer(queue); 316 if (!skb) { 317 err = -ENOMEM; 318 break; 319 } 320 321 id = xennet_rxidx(req_prod); 322 323 BUG_ON(queue->rx_skbs[id]); 324 queue->rx_skbs[id] = skb; 325 326 ref = gnttab_claim_grant_reference(&queue->gref_rx_head); 327 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); 328 queue->grant_rx_ref[id] = ref; 329 330 page = skb_frag_page(&skb_shinfo(skb)->frags[0]); 331 332 req = RING_GET_REQUEST(&queue->rx, req_prod); 333 gnttab_page_grant_foreign_access_ref_one(ref, 334 queue->info->xbdev->otherend_id, 335 page, 336 0); 337 req->id = id; 338 req->gref = ref; 339 } 340 341 queue->rx.req_prod_pvt = req_prod; 342 343 /* Try again later if there are not enough requests or skb allocation 344 * failed. 345 * Enough requests is quantified as the sum of newly created slots and 346 * the unconsumed slots at the backend. 347 */ 348 if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN || 349 unlikely(err)) { 350 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10)); 351 return; 352 } 353 354 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify); 355 if (notify) 356 notify_remote_via_irq(queue->rx_irq); 357 } 358 359 static int xennet_open(struct net_device *dev) 360 { 361 struct netfront_info *np = netdev_priv(dev); 362 unsigned int num_queues = dev->real_num_tx_queues; 363 unsigned int i = 0; 364 struct netfront_queue *queue = NULL; 365 366 if (!np->queues) 367 return -ENODEV; 368 369 for (i = 0; i < num_queues; ++i) { 370 queue = &np->queues[i]; 371 napi_enable(&queue->napi); 372 373 spin_lock_bh(&queue->rx_lock); 374 if (netif_carrier_ok(dev)) { 375 xennet_alloc_rx_buffers(queue); 376 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1; 377 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)) 378 napi_schedule(&queue->napi); 379 } 380 spin_unlock_bh(&queue->rx_lock); 381 } 382 383 netif_tx_start_all_queues(dev); 384 385 return 0; 386 } 387 388 static void xennet_tx_buf_gc(struct netfront_queue *queue) 389 { 390 RING_IDX cons, prod; 391 unsigned short id; 392 struct sk_buff *skb; 393 bool more_to_do; 394 395 BUG_ON(!netif_carrier_ok(queue->info->netdev)); 396 397 do { 398 prod = queue->tx.sring->rsp_prod; 399 rmb(); /* Ensure we see responses up to 'rp'. */ 400 401 for (cons = queue->tx.rsp_cons; cons != prod; cons++) { 402 struct xen_netif_tx_response *txrsp; 403 404 txrsp = RING_GET_RESPONSE(&queue->tx, cons); 405 if (txrsp->status == XEN_NETIF_RSP_NULL) 406 continue; 407 408 id = txrsp->id; 409 skb = queue->tx_skbs[id].skb; 410 if (unlikely(gnttab_query_foreign_access( 411 queue->grant_tx_ref[id]) != 0)) { 412 pr_alert("%s: warning -- grant still in use by backend domain\n", 413 __func__); 414 BUG(); 415 } 416 gnttab_end_foreign_access_ref( 417 queue->grant_tx_ref[id], GNTMAP_readonly); 418 gnttab_release_grant_reference( 419 &queue->gref_tx_head, queue->grant_tx_ref[id]); 420 queue->grant_tx_ref[id] = GRANT_INVALID_REF; 421 queue->grant_tx_page[id] = NULL; 422 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id); 423 dev_kfree_skb_irq(skb); 424 } 425 426 queue->tx.rsp_cons = prod; 427 428 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do); 429 } while (more_to_do); 430 431 xennet_maybe_wake_tx(queue); 432 } 433 434 struct xennet_gnttab_make_txreq { 435 struct netfront_queue *queue; 436 struct sk_buff *skb; 437 struct page *page; 438 struct xen_netif_tx_request *tx; /* Last request */ 439 unsigned int size; 440 }; 441 442 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset, 443 unsigned int len, void *data) 444 { 445 struct xennet_gnttab_make_txreq *info = data; 446 unsigned int id; 447 struct xen_netif_tx_request *tx; 448 grant_ref_t ref; 449 /* convenient aliases */ 450 struct page *page = info->page; 451 struct netfront_queue *queue = info->queue; 452 struct sk_buff *skb = info->skb; 453 454 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs); 455 tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); 456 ref = gnttab_claim_grant_reference(&queue->gref_tx_head); 457 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); 458 459 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id, 460 gfn, GNTMAP_readonly); 461 462 queue->tx_skbs[id].skb = skb; 463 queue->grant_tx_page[id] = page; 464 queue->grant_tx_ref[id] = ref; 465 466 tx->id = id; 467 tx->gref = ref; 468 tx->offset = offset; 469 tx->size = len; 470 tx->flags = 0; 471 472 info->tx = tx; 473 info->size += tx->size; 474 } 475 476 static struct xen_netif_tx_request *xennet_make_first_txreq( 477 struct netfront_queue *queue, struct sk_buff *skb, 478 struct page *page, unsigned int offset, unsigned int len) 479 { 480 struct xennet_gnttab_make_txreq info = { 481 .queue = queue, 482 .skb = skb, 483 .page = page, 484 .size = 0, 485 }; 486 487 gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info); 488 489 return info.tx; 490 } 491 492 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset, 493 unsigned int len, void *data) 494 { 495 struct xennet_gnttab_make_txreq *info = data; 496 497 info->tx->flags |= XEN_NETTXF_more_data; 498 skb_get(info->skb); 499 xennet_tx_setup_grant(gfn, offset, len, data); 500 } 501 502 static struct xen_netif_tx_request *xennet_make_txreqs( 503 struct netfront_queue *queue, struct xen_netif_tx_request *tx, 504 struct sk_buff *skb, struct page *page, 505 unsigned int offset, unsigned int len) 506 { 507 struct xennet_gnttab_make_txreq info = { 508 .queue = queue, 509 .skb = skb, 510 .tx = tx, 511 }; 512 513 /* Skip unused frames from start of page */ 514 page += offset >> PAGE_SHIFT; 515 offset &= ~PAGE_MASK; 516 517 while (len) { 518 info.page = page; 519 info.size = 0; 520 521 gnttab_foreach_grant_in_range(page, offset, len, 522 xennet_make_one_txreq, 523 &info); 524 525 page++; 526 offset = 0; 527 len -= info.size; 528 } 529 530 return info.tx; 531 } 532 533 /* 534 * Count how many ring slots are required to send this skb. Each frag 535 * might be a compound page. 536 */ 537 static int xennet_count_skb_slots(struct sk_buff *skb) 538 { 539 int i, frags = skb_shinfo(skb)->nr_frags; 540 int slots; 541 542 slots = gnttab_count_grant(offset_in_page(skb->data), 543 skb_headlen(skb)); 544 545 for (i = 0; i < frags; i++) { 546 skb_frag_t *frag = skb_shinfo(skb)->frags + i; 547 unsigned long size = skb_frag_size(frag); 548 unsigned long offset = skb_frag_off(frag); 549 550 /* Skip unused frames from start of page */ 551 offset &= ~PAGE_MASK; 552 553 slots += gnttab_count_grant(offset, size); 554 } 555 556 return slots; 557 } 558 559 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb, 560 struct net_device *sb_dev) 561 { 562 unsigned int num_queues = dev->real_num_tx_queues; 563 u32 hash; 564 u16 queue_idx; 565 566 /* First, check if there is only one queue */ 567 if (num_queues == 1) { 568 queue_idx = 0; 569 } else { 570 hash = skb_get_hash(skb); 571 queue_idx = hash % num_queues; 572 } 573 574 return queue_idx; 575 } 576 577 static int xennet_xdp_xmit_one(struct net_device *dev, 578 struct netfront_queue *queue, 579 struct xdp_frame *xdpf) 580 { 581 struct netfront_info *np = netdev_priv(dev); 582 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); 583 int notify; 584 585 xennet_make_first_txreq(queue, NULL, 586 virt_to_page(xdpf->data), 587 offset_in_page(xdpf->data), 588 xdpf->len); 589 590 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); 591 if (notify) 592 notify_remote_via_irq(queue->tx_irq); 593 594 u64_stats_update_begin(&tx_stats->syncp); 595 tx_stats->bytes += xdpf->len; 596 tx_stats->packets++; 597 u64_stats_update_end(&tx_stats->syncp); 598 599 xennet_tx_buf_gc(queue); 600 601 return 0; 602 } 603 604 static int xennet_xdp_xmit(struct net_device *dev, int n, 605 struct xdp_frame **frames, u32 flags) 606 { 607 unsigned int num_queues = dev->real_num_tx_queues; 608 struct netfront_info *np = netdev_priv(dev); 609 struct netfront_queue *queue = NULL; 610 unsigned long irq_flags; 611 int nxmit = 0; 612 int i; 613 614 if (unlikely(flags & ~XDP_XMIT_FLAGS_MASK)) 615 return -EINVAL; 616 617 queue = &np->queues[smp_processor_id() % num_queues]; 618 619 spin_lock_irqsave(&queue->tx_lock, irq_flags); 620 for (i = 0; i < n; i++) { 621 struct xdp_frame *xdpf = frames[i]; 622 623 if (!xdpf) 624 continue; 625 if (xennet_xdp_xmit_one(dev, queue, xdpf)) 626 break; 627 nxmit++; 628 } 629 spin_unlock_irqrestore(&queue->tx_lock, irq_flags); 630 631 return nxmit; 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_init_buff(xdp, XEN_PAGE_SIZE - XDP_PACKET_HEADROOM, 866 &queue->xdp_rxq); 867 xdp_prepare_buff(xdp, page_address(pdata), XDP_PACKET_HEADROOM, 868 len, false); 869 870 act = bpf_prog_run_xdp(prog, xdp); 871 switch (act) { 872 case XDP_TX: 873 get_page(pdata); 874 xdpf = xdp_convert_buff_to_frame(xdp); 875 err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0); 876 if (unlikely(!err)) 877 xdp_return_frame_rx_napi(xdpf); 878 else 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 int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp) 1482 { 1483 switch (xdp->command) { 1484 case XDP_SETUP_PROG: 1485 return xennet_xdp_set(dev, xdp->prog, xdp->extack); 1486 default: 1487 return -EINVAL; 1488 } 1489 } 1490 1491 static const struct net_device_ops xennet_netdev_ops = { 1492 .ndo_open = xennet_open, 1493 .ndo_stop = xennet_close, 1494 .ndo_start_xmit = xennet_start_xmit, 1495 .ndo_change_mtu = xennet_change_mtu, 1496 .ndo_get_stats64 = xennet_get_stats64, 1497 .ndo_set_mac_address = eth_mac_addr, 1498 .ndo_validate_addr = eth_validate_addr, 1499 .ndo_fix_features = xennet_fix_features, 1500 .ndo_set_features = xennet_set_features, 1501 .ndo_select_queue = xennet_select_queue, 1502 .ndo_bpf = xennet_xdp, 1503 .ndo_xdp_xmit = xennet_xdp_xmit, 1504 #ifdef CONFIG_NET_POLL_CONTROLLER 1505 .ndo_poll_controller = xennet_poll_controller, 1506 #endif 1507 }; 1508 1509 static void xennet_free_netdev(struct net_device *netdev) 1510 { 1511 struct netfront_info *np = netdev_priv(netdev); 1512 1513 free_percpu(np->rx_stats); 1514 free_percpu(np->tx_stats); 1515 free_netdev(netdev); 1516 } 1517 1518 static struct net_device *xennet_create_dev(struct xenbus_device *dev) 1519 { 1520 int err; 1521 struct net_device *netdev; 1522 struct netfront_info *np; 1523 1524 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues); 1525 if (!netdev) 1526 return ERR_PTR(-ENOMEM); 1527 1528 np = netdev_priv(netdev); 1529 np->xbdev = dev; 1530 1531 np->queues = NULL; 1532 1533 err = -ENOMEM; 1534 np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); 1535 if (np->rx_stats == NULL) 1536 goto exit; 1537 np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); 1538 if (np->tx_stats == NULL) 1539 goto exit; 1540 1541 netdev->netdev_ops = &xennet_netdev_ops; 1542 1543 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | 1544 NETIF_F_GSO_ROBUST; 1545 netdev->hw_features = NETIF_F_SG | 1546 NETIF_F_IPV6_CSUM | 1547 NETIF_F_TSO | NETIF_F_TSO6; 1548 1549 /* 1550 * Assume that all hw features are available for now. This set 1551 * will be adjusted by the call to netdev_update_features() in 1552 * xennet_connect() which is the earliest point where we can 1553 * negotiate with the backend regarding supported features. 1554 */ 1555 netdev->features |= netdev->hw_features; 1556 1557 netdev->ethtool_ops = &xennet_ethtool_ops; 1558 netdev->min_mtu = ETH_MIN_MTU; 1559 netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE; 1560 SET_NETDEV_DEV(netdev, &dev->dev); 1561 1562 np->netdev = netdev; 1563 np->netfront_xdp_enabled = false; 1564 1565 netif_carrier_off(netdev); 1566 1567 do { 1568 xenbus_switch_state(dev, XenbusStateInitialising); 1569 err = wait_event_timeout(module_wq, 1570 xenbus_read_driver_state(dev->otherend) != 1571 XenbusStateClosed && 1572 xenbus_read_driver_state(dev->otherend) != 1573 XenbusStateUnknown, XENNET_TIMEOUT); 1574 } while (!err); 1575 1576 return netdev; 1577 1578 exit: 1579 xennet_free_netdev(netdev); 1580 return ERR_PTR(err); 1581 } 1582 1583 /* 1584 * Entry point to this code when a new device is created. Allocate the basic 1585 * structures and the ring buffers for communication with the backend, and 1586 * inform the backend of the appropriate details for those. 1587 */ 1588 static int netfront_probe(struct xenbus_device *dev, 1589 const struct xenbus_device_id *id) 1590 { 1591 int err; 1592 struct net_device *netdev; 1593 struct netfront_info *info; 1594 1595 netdev = xennet_create_dev(dev); 1596 if (IS_ERR(netdev)) { 1597 err = PTR_ERR(netdev); 1598 xenbus_dev_fatal(dev, err, "creating netdev"); 1599 return err; 1600 } 1601 1602 info = netdev_priv(netdev); 1603 dev_set_drvdata(&dev->dev, info); 1604 #ifdef CONFIG_SYSFS 1605 info->netdev->sysfs_groups[0] = &xennet_dev_group; 1606 #endif 1607 1608 return 0; 1609 } 1610 1611 static void xennet_end_access(int ref, void *page) 1612 { 1613 /* This frees the page as a side-effect */ 1614 if (ref != GRANT_INVALID_REF) 1615 gnttab_end_foreign_access(ref, 0, (unsigned long)page); 1616 } 1617 1618 static void xennet_disconnect_backend(struct netfront_info *info) 1619 { 1620 unsigned int i = 0; 1621 unsigned int num_queues = info->netdev->real_num_tx_queues; 1622 1623 netif_carrier_off(info->netdev); 1624 1625 for (i = 0; i < num_queues && info->queues; ++i) { 1626 struct netfront_queue *queue = &info->queues[i]; 1627 1628 del_timer_sync(&queue->rx_refill_timer); 1629 1630 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq)) 1631 unbind_from_irqhandler(queue->tx_irq, queue); 1632 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) { 1633 unbind_from_irqhandler(queue->tx_irq, queue); 1634 unbind_from_irqhandler(queue->rx_irq, queue); 1635 } 1636 queue->tx_evtchn = queue->rx_evtchn = 0; 1637 queue->tx_irq = queue->rx_irq = 0; 1638 1639 if (netif_running(info->netdev)) 1640 napi_synchronize(&queue->napi); 1641 1642 xennet_release_tx_bufs(queue); 1643 xennet_release_rx_bufs(queue); 1644 gnttab_free_grant_references(queue->gref_tx_head); 1645 gnttab_free_grant_references(queue->gref_rx_head); 1646 1647 /* End access and free the pages */ 1648 xennet_end_access(queue->tx_ring_ref, queue->tx.sring); 1649 xennet_end_access(queue->rx_ring_ref, queue->rx.sring); 1650 1651 queue->tx_ring_ref = GRANT_INVALID_REF; 1652 queue->rx_ring_ref = GRANT_INVALID_REF; 1653 queue->tx.sring = NULL; 1654 queue->rx.sring = NULL; 1655 1656 page_pool_destroy(queue->page_pool); 1657 } 1658 } 1659 1660 /* 1661 * We are reconnecting to the backend, due to a suspend/resume, or a backend 1662 * driver restart. We tear down our netif structure and recreate it, but 1663 * leave the device-layer structures intact so that this is transparent to the 1664 * rest of the kernel. 1665 */ 1666 static int netfront_resume(struct xenbus_device *dev) 1667 { 1668 struct netfront_info *info = dev_get_drvdata(&dev->dev); 1669 1670 dev_dbg(&dev->dev, "%s\n", dev->nodename); 1671 1672 xennet_disconnect_backend(info); 1673 return 0; 1674 } 1675 1676 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) 1677 { 1678 char *s, *e, *macstr; 1679 int i; 1680 1681 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); 1682 if (IS_ERR(macstr)) 1683 return PTR_ERR(macstr); 1684 1685 for (i = 0; i < ETH_ALEN; i++) { 1686 mac[i] = simple_strtoul(s, &e, 16); 1687 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { 1688 kfree(macstr); 1689 return -ENOENT; 1690 } 1691 s = e+1; 1692 } 1693 1694 kfree(macstr); 1695 return 0; 1696 } 1697 1698 static int setup_netfront_single(struct netfront_queue *queue) 1699 { 1700 int err; 1701 1702 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); 1703 if (err < 0) 1704 goto fail; 1705 1706 err = bind_evtchn_to_irqhandler(queue->tx_evtchn, 1707 xennet_interrupt, 1708 0, queue->info->netdev->name, queue); 1709 if (err < 0) 1710 goto bind_fail; 1711 queue->rx_evtchn = queue->tx_evtchn; 1712 queue->rx_irq = queue->tx_irq = err; 1713 1714 return 0; 1715 1716 bind_fail: 1717 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); 1718 queue->tx_evtchn = 0; 1719 fail: 1720 return err; 1721 } 1722 1723 static int setup_netfront_split(struct netfront_queue *queue) 1724 { 1725 int err; 1726 1727 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); 1728 if (err < 0) 1729 goto fail; 1730 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn); 1731 if (err < 0) 1732 goto alloc_rx_evtchn_fail; 1733 1734 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), 1735 "%s-tx", queue->name); 1736 err = bind_evtchn_to_irqhandler(queue->tx_evtchn, 1737 xennet_tx_interrupt, 1738 0, queue->tx_irq_name, queue); 1739 if (err < 0) 1740 goto bind_tx_fail; 1741 queue->tx_irq = err; 1742 1743 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), 1744 "%s-rx", queue->name); 1745 err = bind_evtchn_to_irqhandler(queue->rx_evtchn, 1746 xennet_rx_interrupt, 1747 0, queue->rx_irq_name, queue); 1748 if (err < 0) 1749 goto bind_rx_fail; 1750 queue->rx_irq = err; 1751 1752 return 0; 1753 1754 bind_rx_fail: 1755 unbind_from_irqhandler(queue->tx_irq, queue); 1756 queue->tx_irq = 0; 1757 bind_tx_fail: 1758 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn); 1759 queue->rx_evtchn = 0; 1760 alloc_rx_evtchn_fail: 1761 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); 1762 queue->tx_evtchn = 0; 1763 fail: 1764 return err; 1765 } 1766 1767 static int setup_netfront(struct xenbus_device *dev, 1768 struct netfront_queue *queue, unsigned int feature_split_evtchn) 1769 { 1770 struct xen_netif_tx_sring *txs; 1771 struct xen_netif_rx_sring *rxs; 1772 grant_ref_t gref; 1773 int err; 1774 1775 queue->tx_ring_ref = GRANT_INVALID_REF; 1776 queue->rx_ring_ref = GRANT_INVALID_REF; 1777 queue->rx.sring = NULL; 1778 queue->tx.sring = NULL; 1779 1780 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); 1781 if (!txs) { 1782 err = -ENOMEM; 1783 xenbus_dev_fatal(dev, err, "allocating tx ring page"); 1784 goto fail; 1785 } 1786 SHARED_RING_INIT(txs); 1787 FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE); 1788 1789 err = xenbus_grant_ring(dev, txs, 1, &gref); 1790 if (err < 0) 1791 goto grant_tx_ring_fail; 1792 queue->tx_ring_ref = gref; 1793 1794 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); 1795 if (!rxs) { 1796 err = -ENOMEM; 1797 xenbus_dev_fatal(dev, err, "allocating rx ring page"); 1798 goto alloc_rx_ring_fail; 1799 } 1800 SHARED_RING_INIT(rxs); 1801 FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE); 1802 1803 err = xenbus_grant_ring(dev, rxs, 1, &gref); 1804 if (err < 0) 1805 goto grant_rx_ring_fail; 1806 queue->rx_ring_ref = gref; 1807 1808 if (feature_split_evtchn) 1809 err = setup_netfront_split(queue); 1810 /* setup single event channel if 1811 * a) feature-split-event-channels == 0 1812 * b) feature-split-event-channels == 1 but failed to setup 1813 */ 1814 if (!feature_split_evtchn || err) 1815 err = setup_netfront_single(queue); 1816 1817 if (err) 1818 goto alloc_evtchn_fail; 1819 1820 return 0; 1821 1822 /* If we fail to setup netfront, it is safe to just revoke access to 1823 * granted pages because backend is not accessing it at this point. 1824 */ 1825 alloc_evtchn_fail: 1826 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0); 1827 grant_rx_ring_fail: 1828 free_page((unsigned long)rxs); 1829 alloc_rx_ring_fail: 1830 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0); 1831 grant_tx_ring_fail: 1832 free_page((unsigned long)txs); 1833 fail: 1834 return err; 1835 } 1836 1837 /* Queue-specific initialisation 1838 * This used to be done in xennet_create_dev() but must now 1839 * be run per-queue. 1840 */ 1841 static int xennet_init_queue(struct netfront_queue *queue) 1842 { 1843 unsigned short i; 1844 int err = 0; 1845 char *devid; 1846 1847 spin_lock_init(&queue->tx_lock); 1848 spin_lock_init(&queue->rx_lock); 1849 1850 timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0); 1851 1852 devid = strrchr(queue->info->xbdev->nodename, '/') + 1; 1853 snprintf(queue->name, sizeof(queue->name), "vif%s-q%u", 1854 devid, queue->id); 1855 1856 /* Initialise tx_skbs as a free chain containing every entry. */ 1857 queue->tx_skb_freelist = 0; 1858 for (i = 0; i < NET_TX_RING_SIZE; i++) { 1859 skb_entry_set_link(&queue->tx_skbs[i], i+1); 1860 queue->grant_tx_ref[i] = GRANT_INVALID_REF; 1861 queue->grant_tx_page[i] = NULL; 1862 } 1863 1864 /* Clear out rx_skbs */ 1865 for (i = 0; i < NET_RX_RING_SIZE; i++) { 1866 queue->rx_skbs[i] = NULL; 1867 queue->grant_rx_ref[i] = GRANT_INVALID_REF; 1868 } 1869 1870 /* A grant for every tx ring slot */ 1871 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, 1872 &queue->gref_tx_head) < 0) { 1873 pr_alert("can't alloc tx grant refs\n"); 1874 err = -ENOMEM; 1875 goto exit; 1876 } 1877 1878 /* A grant for every rx ring slot */ 1879 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE, 1880 &queue->gref_rx_head) < 0) { 1881 pr_alert("can't alloc rx grant refs\n"); 1882 err = -ENOMEM; 1883 goto exit_free_tx; 1884 } 1885 1886 return 0; 1887 1888 exit_free_tx: 1889 gnttab_free_grant_references(queue->gref_tx_head); 1890 exit: 1891 return err; 1892 } 1893 1894 static int write_queue_xenstore_keys(struct netfront_queue *queue, 1895 struct xenbus_transaction *xbt, int write_hierarchical) 1896 { 1897 /* Write the queue-specific keys into XenStore in the traditional 1898 * way for a single queue, or in a queue subkeys for multiple 1899 * queues. 1900 */ 1901 struct xenbus_device *dev = queue->info->xbdev; 1902 int err; 1903 const char *message; 1904 char *path; 1905 size_t pathsize; 1906 1907 /* Choose the correct place to write the keys */ 1908 if (write_hierarchical) { 1909 pathsize = strlen(dev->nodename) + 10; 1910 path = kzalloc(pathsize, GFP_KERNEL); 1911 if (!path) { 1912 err = -ENOMEM; 1913 message = "out of memory while writing ring references"; 1914 goto error; 1915 } 1916 snprintf(path, pathsize, "%s/queue-%u", 1917 dev->nodename, queue->id); 1918 } else { 1919 path = (char *)dev->nodename; 1920 } 1921 1922 /* Write ring references */ 1923 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u", 1924 queue->tx_ring_ref); 1925 if (err) { 1926 message = "writing tx-ring-ref"; 1927 goto error; 1928 } 1929 1930 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u", 1931 queue->rx_ring_ref); 1932 if (err) { 1933 message = "writing rx-ring-ref"; 1934 goto error; 1935 } 1936 1937 /* Write event channels; taking into account both shared 1938 * and split event channel scenarios. 1939 */ 1940 if (queue->tx_evtchn == queue->rx_evtchn) { 1941 /* Shared event channel */ 1942 err = xenbus_printf(*xbt, path, 1943 "event-channel", "%u", queue->tx_evtchn); 1944 if (err) { 1945 message = "writing event-channel"; 1946 goto error; 1947 } 1948 } else { 1949 /* Split event channels */ 1950 err = xenbus_printf(*xbt, path, 1951 "event-channel-tx", "%u", queue->tx_evtchn); 1952 if (err) { 1953 message = "writing event-channel-tx"; 1954 goto error; 1955 } 1956 1957 err = xenbus_printf(*xbt, path, 1958 "event-channel-rx", "%u", queue->rx_evtchn); 1959 if (err) { 1960 message = "writing event-channel-rx"; 1961 goto error; 1962 } 1963 } 1964 1965 if (write_hierarchical) 1966 kfree(path); 1967 return 0; 1968 1969 error: 1970 if (write_hierarchical) 1971 kfree(path); 1972 xenbus_dev_fatal(dev, err, "%s", message); 1973 return err; 1974 } 1975 1976 static void xennet_destroy_queues(struct netfront_info *info) 1977 { 1978 unsigned int i; 1979 1980 for (i = 0; i < info->netdev->real_num_tx_queues; i++) { 1981 struct netfront_queue *queue = &info->queues[i]; 1982 1983 if (netif_running(info->netdev)) 1984 napi_disable(&queue->napi); 1985 netif_napi_del(&queue->napi); 1986 } 1987 1988 kfree(info->queues); 1989 info->queues = NULL; 1990 } 1991 1992 1993 1994 static int xennet_create_page_pool(struct netfront_queue *queue) 1995 { 1996 int err; 1997 struct page_pool_params pp_params = { 1998 .order = 0, 1999 .flags = 0, 2000 .pool_size = NET_RX_RING_SIZE, 2001 .nid = NUMA_NO_NODE, 2002 .dev = &queue->info->netdev->dev, 2003 .offset = XDP_PACKET_HEADROOM, 2004 .max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM, 2005 }; 2006 2007 queue->page_pool = page_pool_create(&pp_params); 2008 if (IS_ERR(queue->page_pool)) { 2009 err = PTR_ERR(queue->page_pool); 2010 queue->page_pool = NULL; 2011 return err; 2012 } 2013 2014 err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev, 2015 queue->id, 0); 2016 if (err) { 2017 netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n"); 2018 goto err_free_pp; 2019 } 2020 2021 err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq, 2022 MEM_TYPE_PAGE_POOL, queue->page_pool); 2023 if (err) { 2024 netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n"); 2025 goto err_unregister_rxq; 2026 } 2027 return 0; 2028 2029 err_unregister_rxq: 2030 xdp_rxq_info_unreg(&queue->xdp_rxq); 2031 err_free_pp: 2032 page_pool_destroy(queue->page_pool); 2033 queue->page_pool = NULL; 2034 return err; 2035 } 2036 2037 static int xennet_create_queues(struct netfront_info *info, 2038 unsigned int *num_queues) 2039 { 2040 unsigned int i; 2041 int ret; 2042 2043 info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue), 2044 GFP_KERNEL); 2045 if (!info->queues) 2046 return -ENOMEM; 2047 2048 for (i = 0; i < *num_queues; i++) { 2049 struct netfront_queue *queue = &info->queues[i]; 2050 2051 queue->id = i; 2052 queue->info = info; 2053 2054 ret = xennet_init_queue(queue); 2055 if (ret < 0) { 2056 dev_warn(&info->xbdev->dev, 2057 "only created %d queues\n", i); 2058 *num_queues = i; 2059 break; 2060 } 2061 2062 /* use page pool recycling instead of buddy allocator */ 2063 ret = xennet_create_page_pool(queue); 2064 if (ret < 0) { 2065 dev_err(&info->xbdev->dev, "can't allocate page pool\n"); 2066 *num_queues = i; 2067 return ret; 2068 } 2069 2070 netif_napi_add(queue->info->netdev, &queue->napi, 2071 xennet_poll, 64); 2072 if (netif_running(info->netdev)) 2073 napi_enable(&queue->napi); 2074 } 2075 2076 netif_set_real_num_tx_queues(info->netdev, *num_queues); 2077 2078 if (*num_queues == 0) { 2079 dev_err(&info->xbdev->dev, "no queues\n"); 2080 return -EINVAL; 2081 } 2082 return 0; 2083 } 2084 2085 /* Common code used when first setting up, and when resuming. */ 2086 static int talk_to_netback(struct xenbus_device *dev, 2087 struct netfront_info *info) 2088 { 2089 const char *message; 2090 struct xenbus_transaction xbt; 2091 int err; 2092 unsigned int feature_split_evtchn; 2093 unsigned int i = 0; 2094 unsigned int max_queues = 0; 2095 struct netfront_queue *queue = NULL; 2096 unsigned int num_queues = 1; 2097 2098 info->netdev->irq = 0; 2099 2100 /* Check if backend supports multiple queues */ 2101 max_queues = xenbus_read_unsigned(info->xbdev->otherend, 2102 "multi-queue-max-queues", 1); 2103 num_queues = min(max_queues, xennet_max_queues); 2104 2105 /* Check feature-split-event-channels */ 2106 feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend, 2107 "feature-split-event-channels", 0); 2108 2109 /* Read mac addr. */ 2110 err = xen_net_read_mac(dev, info->netdev->dev_addr); 2111 if (err) { 2112 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); 2113 goto out_unlocked; 2114 } 2115 2116 info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend, 2117 "feature-xdp-headroom", 0); 2118 if (info->netback_has_xdp_headroom) { 2119 /* set the current xen-netfront xdp state */ 2120 err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ? 2121 NETBACK_XDP_HEADROOM_ENABLE : 2122 NETBACK_XDP_HEADROOM_DISABLE); 2123 if (err) 2124 goto out_unlocked; 2125 } 2126 2127 rtnl_lock(); 2128 if (info->queues) 2129 xennet_destroy_queues(info); 2130 2131 err = xennet_create_queues(info, &num_queues); 2132 if (err < 0) { 2133 xenbus_dev_fatal(dev, err, "creating queues"); 2134 kfree(info->queues); 2135 info->queues = NULL; 2136 goto out; 2137 } 2138 rtnl_unlock(); 2139 2140 /* Create shared ring, alloc event channel -- for each queue */ 2141 for (i = 0; i < num_queues; ++i) { 2142 queue = &info->queues[i]; 2143 err = setup_netfront(dev, queue, feature_split_evtchn); 2144 if (err) 2145 goto destroy_ring; 2146 } 2147 2148 again: 2149 err = xenbus_transaction_start(&xbt); 2150 if (err) { 2151 xenbus_dev_fatal(dev, err, "starting transaction"); 2152 goto destroy_ring; 2153 } 2154 2155 if (xenbus_exists(XBT_NIL, 2156 info->xbdev->otherend, "multi-queue-max-queues")) { 2157 /* Write the number of queues */ 2158 err = xenbus_printf(xbt, dev->nodename, 2159 "multi-queue-num-queues", "%u", num_queues); 2160 if (err) { 2161 message = "writing multi-queue-num-queues"; 2162 goto abort_transaction_no_dev_fatal; 2163 } 2164 } 2165 2166 if (num_queues == 1) { 2167 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */ 2168 if (err) 2169 goto abort_transaction_no_dev_fatal; 2170 } else { 2171 /* Write the keys for each queue */ 2172 for (i = 0; i < num_queues; ++i) { 2173 queue = &info->queues[i]; 2174 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */ 2175 if (err) 2176 goto abort_transaction_no_dev_fatal; 2177 } 2178 } 2179 2180 /* The remaining keys are not queue-specific */ 2181 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", 2182 1); 2183 if (err) { 2184 message = "writing request-rx-copy"; 2185 goto abort_transaction; 2186 } 2187 2188 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); 2189 if (err) { 2190 message = "writing feature-rx-notify"; 2191 goto abort_transaction; 2192 } 2193 2194 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); 2195 if (err) { 2196 message = "writing feature-sg"; 2197 goto abort_transaction; 2198 } 2199 2200 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); 2201 if (err) { 2202 message = "writing feature-gso-tcpv4"; 2203 goto abort_transaction; 2204 } 2205 2206 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1"); 2207 if (err) { 2208 message = "writing feature-gso-tcpv6"; 2209 goto abort_transaction; 2210 } 2211 2212 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload", 2213 "1"); 2214 if (err) { 2215 message = "writing feature-ipv6-csum-offload"; 2216 goto abort_transaction; 2217 } 2218 2219 err = xenbus_transaction_end(xbt, 0); 2220 if (err) { 2221 if (err == -EAGAIN) 2222 goto again; 2223 xenbus_dev_fatal(dev, err, "completing transaction"); 2224 goto destroy_ring; 2225 } 2226 2227 return 0; 2228 2229 abort_transaction: 2230 xenbus_dev_fatal(dev, err, "%s", message); 2231 abort_transaction_no_dev_fatal: 2232 xenbus_transaction_end(xbt, 1); 2233 destroy_ring: 2234 xennet_disconnect_backend(info); 2235 rtnl_lock(); 2236 xennet_destroy_queues(info); 2237 out: 2238 rtnl_unlock(); 2239 out_unlocked: 2240 device_unregister(&dev->dev); 2241 return err; 2242 } 2243 2244 static int xennet_connect(struct net_device *dev) 2245 { 2246 struct netfront_info *np = netdev_priv(dev); 2247 unsigned int num_queues = 0; 2248 int err; 2249 unsigned int j = 0; 2250 struct netfront_queue *queue = NULL; 2251 2252 if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) { 2253 dev_info(&dev->dev, 2254 "backend does not support copying receive path\n"); 2255 return -ENODEV; 2256 } 2257 2258 err = talk_to_netback(np->xbdev, np); 2259 if (err) 2260 return err; 2261 if (np->netback_has_xdp_headroom) 2262 pr_info("backend supports XDP headroom\n"); 2263 2264 /* talk_to_netback() sets the correct number of queues */ 2265 num_queues = dev->real_num_tx_queues; 2266 2267 if (dev->reg_state == NETREG_UNINITIALIZED) { 2268 err = register_netdev(dev); 2269 if (err) { 2270 pr_warn("%s: register_netdev err=%d\n", __func__, err); 2271 device_unregister(&np->xbdev->dev); 2272 return err; 2273 } 2274 } 2275 2276 rtnl_lock(); 2277 netdev_update_features(dev); 2278 rtnl_unlock(); 2279 2280 /* 2281 * All public and private state should now be sane. Get 2282 * ready to start sending and receiving packets and give the driver 2283 * domain a kick because we've probably just requeued some 2284 * packets. 2285 */ 2286 netif_carrier_on(np->netdev); 2287 for (j = 0; j < num_queues; ++j) { 2288 queue = &np->queues[j]; 2289 2290 notify_remote_via_irq(queue->tx_irq); 2291 if (queue->tx_irq != queue->rx_irq) 2292 notify_remote_via_irq(queue->rx_irq); 2293 2294 spin_lock_irq(&queue->tx_lock); 2295 xennet_tx_buf_gc(queue); 2296 spin_unlock_irq(&queue->tx_lock); 2297 2298 spin_lock_bh(&queue->rx_lock); 2299 xennet_alloc_rx_buffers(queue); 2300 spin_unlock_bh(&queue->rx_lock); 2301 } 2302 2303 return 0; 2304 } 2305 2306 /* 2307 * Callback received when the backend's state changes. 2308 */ 2309 static void netback_changed(struct xenbus_device *dev, 2310 enum xenbus_state backend_state) 2311 { 2312 struct netfront_info *np = dev_get_drvdata(&dev->dev); 2313 struct net_device *netdev = np->netdev; 2314 2315 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); 2316 2317 wake_up_all(&module_wq); 2318 2319 switch (backend_state) { 2320 case XenbusStateInitialising: 2321 case XenbusStateInitialised: 2322 case XenbusStateReconfiguring: 2323 case XenbusStateReconfigured: 2324 case XenbusStateUnknown: 2325 break; 2326 2327 case XenbusStateInitWait: 2328 if (dev->state != XenbusStateInitialising) 2329 break; 2330 if (xennet_connect(netdev) != 0) 2331 break; 2332 xenbus_switch_state(dev, XenbusStateConnected); 2333 break; 2334 2335 case XenbusStateConnected: 2336 netdev_notify_peers(netdev); 2337 break; 2338 2339 case XenbusStateClosed: 2340 if (dev->state == XenbusStateClosed) 2341 break; 2342 fallthrough; /* Missed the backend's CLOSING state */ 2343 case XenbusStateClosing: 2344 xenbus_frontend_closed(dev); 2345 break; 2346 } 2347 } 2348 2349 static const struct xennet_stat { 2350 char name[ETH_GSTRING_LEN]; 2351 u16 offset; 2352 } xennet_stats[] = { 2353 { 2354 "rx_gso_checksum_fixup", 2355 offsetof(struct netfront_info, rx_gso_checksum_fixup) 2356 }, 2357 }; 2358 2359 static int xennet_get_sset_count(struct net_device *dev, int string_set) 2360 { 2361 switch (string_set) { 2362 case ETH_SS_STATS: 2363 return ARRAY_SIZE(xennet_stats); 2364 default: 2365 return -EINVAL; 2366 } 2367 } 2368 2369 static void xennet_get_ethtool_stats(struct net_device *dev, 2370 struct ethtool_stats *stats, u64 * data) 2371 { 2372 void *np = netdev_priv(dev); 2373 int i; 2374 2375 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) 2376 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset)); 2377 } 2378 2379 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) 2380 { 2381 int i; 2382 2383 switch (stringset) { 2384 case ETH_SS_STATS: 2385 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) 2386 memcpy(data + i * ETH_GSTRING_LEN, 2387 xennet_stats[i].name, ETH_GSTRING_LEN); 2388 break; 2389 } 2390 } 2391 2392 static const struct ethtool_ops xennet_ethtool_ops = 2393 { 2394 .get_link = ethtool_op_get_link, 2395 2396 .get_sset_count = xennet_get_sset_count, 2397 .get_ethtool_stats = xennet_get_ethtool_stats, 2398 .get_strings = xennet_get_strings, 2399 .get_ts_info = ethtool_op_get_ts_info, 2400 }; 2401 2402 #ifdef CONFIG_SYSFS 2403 static ssize_t show_rxbuf(struct device *dev, 2404 struct device_attribute *attr, char *buf) 2405 { 2406 return sprintf(buf, "%lu\n", NET_RX_RING_SIZE); 2407 } 2408 2409 static ssize_t store_rxbuf(struct device *dev, 2410 struct device_attribute *attr, 2411 const char *buf, size_t len) 2412 { 2413 char *endp; 2414 2415 if (!capable(CAP_NET_ADMIN)) 2416 return -EPERM; 2417 2418 simple_strtoul(buf, &endp, 0); 2419 if (endp == buf) 2420 return -EBADMSG; 2421 2422 /* rxbuf_min and rxbuf_max are no longer configurable. */ 2423 2424 return len; 2425 } 2426 2427 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf); 2428 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf); 2429 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL); 2430 2431 static struct attribute *xennet_dev_attrs[] = { 2432 &dev_attr_rxbuf_min.attr, 2433 &dev_attr_rxbuf_max.attr, 2434 &dev_attr_rxbuf_cur.attr, 2435 NULL 2436 }; 2437 2438 static const struct attribute_group xennet_dev_group = { 2439 .attrs = xennet_dev_attrs 2440 }; 2441 #endif /* CONFIG_SYSFS */ 2442 2443 static void xennet_bus_close(struct xenbus_device *dev) 2444 { 2445 int ret; 2446 2447 if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed) 2448 return; 2449 do { 2450 xenbus_switch_state(dev, XenbusStateClosing); 2451 ret = wait_event_timeout(module_wq, 2452 xenbus_read_driver_state(dev->otherend) == 2453 XenbusStateClosing || 2454 xenbus_read_driver_state(dev->otherend) == 2455 XenbusStateClosed || 2456 xenbus_read_driver_state(dev->otherend) == 2457 XenbusStateUnknown, 2458 XENNET_TIMEOUT); 2459 } while (!ret); 2460 2461 if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed) 2462 return; 2463 2464 do { 2465 xenbus_switch_state(dev, XenbusStateClosed); 2466 ret = wait_event_timeout(module_wq, 2467 xenbus_read_driver_state(dev->otherend) == 2468 XenbusStateClosed || 2469 xenbus_read_driver_state(dev->otherend) == 2470 XenbusStateUnknown, 2471 XENNET_TIMEOUT); 2472 } while (!ret); 2473 } 2474 2475 static int xennet_remove(struct xenbus_device *dev) 2476 { 2477 struct netfront_info *info = dev_get_drvdata(&dev->dev); 2478 2479 xennet_bus_close(dev); 2480 xennet_disconnect_backend(info); 2481 2482 if (info->netdev->reg_state == NETREG_REGISTERED) 2483 unregister_netdev(info->netdev); 2484 2485 if (info->queues) { 2486 rtnl_lock(); 2487 xennet_destroy_queues(info); 2488 rtnl_unlock(); 2489 } 2490 xennet_free_netdev(info->netdev); 2491 2492 return 0; 2493 } 2494 2495 static const struct xenbus_device_id netfront_ids[] = { 2496 { "vif" }, 2497 { "" } 2498 }; 2499 2500 static struct xenbus_driver netfront_driver = { 2501 .ids = netfront_ids, 2502 .probe = netfront_probe, 2503 .remove = xennet_remove, 2504 .resume = netfront_resume, 2505 .otherend_changed = netback_changed, 2506 }; 2507 2508 static int __init netif_init(void) 2509 { 2510 if (!xen_domain()) 2511 return -ENODEV; 2512 2513 if (!xen_has_pv_nic_devices()) 2514 return -ENODEV; 2515 2516 pr_info("Initialising Xen virtual ethernet driver\n"); 2517 2518 /* Allow as many queues as there are CPUs inut max. 8 if user has not 2519 * specified a value. 2520 */ 2521 if (xennet_max_queues == 0) 2522 xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT, 2523 num_online_cpus()); 2524 2525 return xenbus_register_frontend(&netfront_driver); 2526 } 2527 module_init(netif_init); 2528 2529 2530 static void __exit netif_exit(void) 2531 { 2532 xenbus_unregister_driver(&netfront_driver); 2533 } 2534 module_exit(netif_exit); 2535 2536 MODULE_DESCRIPTION("Xen virtual network device frontend"); 2537 MODULE_LICENSE("GPL"); 2538 MODULE_ALIAS("xen:vif"); 2539 MODULE_ALIAS("xennet"); 2540