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 drops = 0; 612 int i, err; 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 err = xennet_xdp_xmit_one(dev, queue, xdpf); 626 if (err) { 627 xdp_return_frame_rx_napi(xdpf); 628 drops++; 629 } 630 } 631 spin_unlock_irqrestore(&queue->tx_lock, irq_flags); 632 633 return n - drops; 634 } 635 636 637 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1) 638 639 static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) 640 { 641 struct netfront_info *np = netdev_priv(dev); 642 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); 643 struct xen_netif_tx_request *tx, *first_tx; 644 unsigned int i; 645 int notify; 646 int slots; 647 struct page *page; 648 unsigned int offset; 649 unsigned int len; 650 unsigned long flags; 651 struct netfront_queue *queue = NULL; 652 unsigned int num_queues = dev->real_num_tx_queues; 653 u16 queue_index; 654 struct sk_buff *nskb; 655 656 /* Drop the packet if no queues are set up */ 657 if (num_queues < 1) 658 goto drop; 659 /* Determine which queue to transmit this SKB on */ 660 queue_index = skb_get_queue_mapping(skb); 661 queue = &np->queues[queue_index]; 662 663 /* If skb->len is too big for wire format, drop skb and alert 664 * user about misconfiguration. 665 */ 666 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { 667 net_alert_ratelimited( 668 "xennet: skb->len = %u, too big for wire format\n", 669 skb->len); 670 goto drop; 671 } 672 673 slots = xennet_count_skb_slots(skb); 674 if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) { 675 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n", 676 slots, skb->len); 677 if (skb_linearize(skb)) 678 goto drop; 679 } 680 681 page = virt_to_page(skb->data); 682 offset = offset_in_page(skb->data); 683 684 /* The first req should be at least ETH_HLEN size or the packet will be 685 * dropped by netback. 686 */ 687 if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) { 688 nskb = skb_copy(skb, GFP_ATOMIC); 689 if (!nskb) 690 goto drop; 691 dev_consume_skb_any(skb); 692 skb = nskb; 693 page = virt_to_page(skb->data); 694 offset = offset_in_page(skb->data); 695 } 696 697 len = skb_headlen(skb); 698 699 spin_lock_irqsave(&queue->tx_lock, flags); 700 701 if (unlikely(!netif_carrier_ok(dev) || 702 (slots > 1 && !xennet_can_sg(dev)) || 703 netif_needs_gso(skb, netif_skb_features(skb)))) { 704 spin_unlock_irqrestore(&queue->tx_lock, flags); 705 goto drop; 706 } 707 708 /* First request for the linear area. */ 709 first_tx = tx = xennet_make_first_txreq(queue, skb, 710 page, offset, len); 711 offset += tx->size; 712 if (offset == PAGE_SIZE) { 713 page++; 714 offset = 0; 715 } 716 len -= tx->size; 717 718 if (skb->ip_summed == CHECKSUM_PARTIAL) 719 /* local packet? */ 720 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated; 721 else if (skb->ip_summed == CHECKSUM_UNNECESSARY) 722 /* remote but checksummed. */ 723 tx->flags |= XEN_NETTXF_data_validated; 724 725 /* Optional extra info after the first request. */ 726 if (skb_shinfo(skb)->gso_size) { 727 struct xen_netif_extra_info *gso; 728 729 gso = (struct xen_netif_extra_info *) 730 RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); 731 732 tx->flags |= XEN_NETTXF_extra_info; 733 734 gso->u.gso.size = skb_shinfo(skb)->gso_size; 735 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ? 736 XEN_NETIF_GSO_TYPE_TCPV6 : 737 XEN_NETIF_GSO_TYPE_TCPV4; 738 gso->u.gso.pad = 0; 739 gso->u.gso.features = 0; 740 741 gso->type = XEN_NETIF_EXTRA_TYPE_GSO; 742 gso->flags = 0; 743 } 744 745 /* Requests for the rest of the linear area. */ 746 tx = xennet_make_txreqs(queue, tx, skb, page, offset, len); 747 748 /* Requests for all the frags. */ 749 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 750 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 751 tx = xennet_make_txreqs(queue, tx, skb, skb_frag_page(frag), 752 skb_frag_off(frag), 753 skb_frag_size(frag)); 754 } 755 756 /* First request has the packet length. */ 757 first_tx->size = skb->len; 758 759 /* timestamp packet in software */ 760 skb_tx_timestamp(skb); 761 762 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); 763 if (notify) 764 notify_remote_via_irq(queue->tx_irq); 765 766 u64_stats_update_begin(&tx_stats->syncp); 767 tx_stats->bytes += skb->len; 768 tx_stats->packets++; 769 u64_stats_update_end(&tx_stats->syncp); 770 771 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ 772 xennet_tx_buf_gc(queue); 773 774 if (!netfront_tx_slot_available(queue)) 775 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id)); 776 777 spin_unlock_irqrestore(&queue->tx_lock, flags); 778 779 return NETDEV_TX_OK; 780 781 drop: 782 dev->stats.tx_dropped++; 783 dev_kfree_skb_any(skb); 784 return NETDEV_TX_OK; 785 } 786 787 static int xennet_close(struct net_device *dev) 788 { 789 struct netfront_info *np = netdev_priv(dev); 790 unsigned int num_queues = dev->real_num_tx_queues; 791 unsigned int i; 792 struct netfront_queue *queue; 793 netif_tx_stop_all_queues(np->netdev); 794 for (i = 0; i < num_queues; ++i) { 795 queue = &np->queues[i]; 796 napi_disable(&queue->napi); 797 } 798 return 0; 799 } 800 801 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb, 802 grant_ref_t ref) 803 { 804 int new = xennet_rxidx(queue->rx.req_prod_pvt); 805 806 BUG_ON(queue->rx_skbs[new]); 807 queue->rx_skbs[new] = skb; 808 queue->grant_rx_ref[new] = ref; 809 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new; 810 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref; 811 queue->rx.req_prod_pvt++; 812 } 813 814 static int xennet_get_extras(struct netfront_queue *queue, 815 struct xen_netif_extra_info *extras, 816 RING_IDX rp) 817 818 { 819 struct xen_netif_extra_info *extra; 820 struct device *dev = &queue->info->netdev->dev; 821 RING_IDX cons = queue->rx.rsp_cons; 822 int err = 0; 823 824 do { 825 struct sk_buff *skb; 826 grant_ref_t ref; 827 828 if (unlikely(cons + 1 == rp)) { 829 if (net_ratelimit()) 830 dev_warn(dev, "Missing extra info\n"); 831 err = -EBADR; 832 break; 833 } 834 835 extra = (struct xen_netif_extra_info *) 836 RING_GET_RESPONSE(&queue->rx, ++cons); 837 838 if (unlikely(!extra->type || 839 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { 840 if (net_ratelimit()) 841 dev_warn(dev, "Invalid extra type: %d\n", 842 extra->type); 843 err = -EINVAL; 844 } else { 845 memcpy(&extras[extra->type - 1], extra, 846 sizeof(*extra)); 847 } 848 849 skb = xennet_get_rx_skb(queue, cons); 850 ref = xennet_get_rx_ref(queue, cons); 851 xennet_move_rx_slot(queue, skb, ref); 852 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); 853 854 queue->rx.rsp_cons = cons; 855 return err; 856 } 857 858 static u32 xennet_run_xdp(struct netfront_queue *queue, struct page *pdata, 859 struct xen_netif_rx_response *rx, struct bpf_prog *prog, 860 struct xdp_buff *xdp, bool *need_xdp_flush) 861 { 862 struct xdp_frame *xdpf; 863 u32 len = rx->status; 864 u32 act; 865 int err; 866 867 xdp->data_hard_start = page_address(pdata); 868 xdp->data = xdp->data_hard_start + XDP_PACKET_HEADROOM; 869 xdp_set_data_meta_invalid(xdp); 870 xdp->data_end = xdp->data + len; 871 xdp->rxq = &queue->xdp_rxq; 872 xdp->frame_sz = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM; 873 874 act = bpf_prog_run_xdp(prog, xdp); 875 switch (act) { 876 case XDP_TX: 877 get_page(pdata); 878 xdpf = xdp_convert_buff_to_frame(xdp); 879 err = xennet_xdp_xmit(queue->info->netdev, 1, &xdpf, 0); 880 if (unlikely(err < 0)) 881 trace_xdp_exception(queue->info->netdev, prog, act); 882 break; 883 case XDP_REDIRECT: 884 get_page(pdata); 885 err = xdp_do_redirect(queue->info->netdev, xdp, prog); 886 *need_xdp_flush = true; 887 if (unlikely(err)) 888 trace_xdp_exception(queue->info->netdev, prog, act); 889 break; 890 case XDP_PASS: 891 case XDP_DROP: 892 break; 893 894 case XDP_ABORTED: 895 trace_xdp_exception(queue->info->netdev, prog, act); 896 break; 897 898 default: 899 bpf_warn_invalid_xdp_action(act); 900 } 901 902 return act; 903 } 904 905 static int xennet_get_responses(struct netfront_queue *queue, 906 struct netfront_rx_info *rinfo, RING_IDX rp, 907 struct sk_buff_head *list, 908 bool *need_xdp_flush) 909 { 910 struct xen_netif_rx_response *rx = &rinfo->rx; 911 int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD); 912 RING_IDX cons = queue->rx.rsp_cons; 913 struct sk_buff *skb = xennet_get_rx_skb(queue, cons); 914 struct xen_netif_extra_info *extras = rinfo->extras; 915 grant_ref_t ref = xennet_get_rx_ref(queue, cons); 916 struct device *dev = &queue->info->netdev->dev; 917 struct bpf_prog *xdp_prog; 918 struct xdp_buff xdp; 919 unsigned long ret; 920 int slots = 1; 921 int err = 0; 922 u32 verdict; 923 924 if (rx->flags & XEN_NETRXF_extra_info) { 925 err = xennet_get_extras(queue, extras, rp); 926 if (!err) { 927 if (extras[XEN_NETIF_EXTRA_TYPE_XDP - 1].type) { 928 struct xen_netif_extra_info *xdp; 929 930 xdp = &extras[XEN_NETIF_EXTRA_TYPE_XDP - 1]; 931 rx->offset = xdp->u.xdp.headroom; 932 } 933 } 934 cons = queue->rx.rsp_cons; 935 } 936 937 for (;;) { 938 if (unlikely(rx->status < 0 || 939 rx->offset + rx->status > XEN_PAGE_SIZE)) { 940 if (net_ratelimit()) 941 dev_warn(dev, "rx->offset: %u, size: %d\n", 942 rx->offset, rx->status); 943 xennet_move_rx_slot(queue, skb, ref); 944 err = -EINVAL; 945 goto next; 946 } 947 948 /* 949 * This definitely indicates a bug, either in this driver or in 950 * the backend driver. In future this should flag the bad 951 * situation to the system controller to reboot the backend. 952 */ 953 if (ref == GRANT_INVALID_REF) { 954 if (net_ratelimit()) 955 dev_warn(dev, "Bad rx response id %d.\n", 956 rx->id); 957 err = -EINVAL; 958 goto next; 959 } 960 961 ret = gnttab_end_foreign_access_ref(ref, 0); 962 BUG_ON(!ret); 963 964 gnttab_release_grant_reference(&queue->gref_rx_head, ref); 965 966 rcu_read_lock(); 967 xdp_prog = rcu_dereference(queue->xdp_prog); 968 if (xdp_prog) { 969 if (!(rx->flags & XEN_NETRXF_more_data)) { 970 /* currently only a single page contains data */ 971 verdict = xennet_run_xdp(queue, 972 skb_frag_page(&skb_shinfo(skb)->frags[0]), 973 rx, xdp_prog, &xdp, need_xdp_flush); 974 if (verdict != XDP_PASS) 975 err = -EINVAL; 976 } else { 977 /* drop the frame */ 978 err = -EINVAL; 979 } 980 } 981 rcu_read_unlock(); 982 next: 983 __skb_queue_tail(list, skb); 984 if (!(rx->flags & XEN_NETRXF_more_data)) 985 break; 986 987 if (cons + slots == rp) { 988 if (net_ratelimit()) 989 dev_warn(dev, "Need more slots\n"); 990 err = -ENOENT; 991 break; 992 } 993 994 rx = RING_GET_RESPONSE(&queue->rx, cons + slots); 995 skb = xennet_get_rx_skb(queue, cons + slots); 996 ref = xennet_get_rx_ref(queue, cons + slots); 997 slots++; 998 } 999 1000 if (unlikely(slots > max)) { 1001 if (net_ratelimit()) 1002 dev_warn(dev, "Too many slots\n"); 1003 err = -E2BIG; 1004 } 1005 1006 if (unlikely(err)) 1007 queue->rx.rsp_cons = cons + slots; 1008 1009 return err; 1010 } 1011 1012 static int xennet_set_skb_gso(struct sk_buff *skb, 1013 struct xen_netif_extra_info *gso) 1014 { 1015 if (!gso->u.gso.size) { 1016 if (net_ratelimit()) 1017 pr_warn("GSO size must not be zero\n"); 1018 return -EINVAL; 1019 } 1020 1021 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 && 1022 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) { 1023 if (net_ratelimit()) 1024 pr_warn("Bad GSO type %d\n", gso->u.gso.type); 1025 return -EINVAL; 1026 } 1027 1028 skb_shinfo(skb)->gso_size = gso->u.gso.size; 1029 skb_shinfo(skb)->gso_type = 1030 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ? 1031 SKB_GSO_TCPV4 : 1032 SKB_GSO_TCPV6; 1033 1034 /* Header must be checked, and gso_segs computed. */ 1035 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 1036 skb_shinfo(skb)->gso_segs = 0; 1037 1038 return 0; 1039 } 1040 1041 static int xennet_fill_frags(struct netfront_queue *queue, 1042 struct sk_buff *skb, 1043 struct sk_buff_head *list) 1044 { 1045 RING_IDX cons = queue->rx.rsp_cons; 1046 struct sk_buff *nskb; 1047 1048 while ((nskb = __skb_dequeue(list))) { 1049 struct xen_netif_rx_response *rx = 1050 RING_GET_RESPONSE(&queue->rx, ++cons); 1051 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; 1052 1053 if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) { 1054 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to; 1055 1056 BUG_ON(pull_to < skb_headlen(skb)); 1057 __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); 1058 } 1059 if (unlikely(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS)) { 1060 queue->rx.rsp_cons = ++cons + skb_queue_len(list); 1061 kfree_skb(nskb); 1062 return -ENOENT; 1063 } 1064 1065 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, 1066 skb_frag_page(nfrag), 1067 rx->offset, rx->status, PAGE_SIZE); 1068 1069 skb_shinfo(nskb)->nr_frags = 0; 1070 kfree_skb(nskb); 1071 } 1072 1073 queue->rx.rsp_cons = cons; 1074 1075 return 0; 1076 } 1077 1078 static int checksum_setup(struct net_device *dev, struct sk_buff *skb) 1079 { 1080 bool recalculate_partial_csum = false; 1081 1082 /* 1083 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy 1084 * peers can fail to set NETRXF_csum_blank when sending a GSO 1085 * frame. In this case force the SKB to CHECKSUM_PARTIAL and 1086 * recalculate the partial checksum. 1087 */ 1088 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { 1089 struct netfront_info *np = netdev_priv(dev); 1090 atomic_inc(&np->rx_gso_checksum_fixup); 1091 skb->ip_summed = CHECKSUM_PARTIAL; 1092 recalculate_partial_csum = true; 1093 } 1094 1095 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ 1096 if (skb->ip_summed != CHECKSUM_PARTIAL) 1097 return 0; 1098 1099 return skb_checksum_setup(skb, recalculate_partial_csum); 1100 } 1101 1102 static int handle_incoming_queue(struct netfront_queue *queue, 1103 struct sk_buff_head *rxq) 1104 { 1105 struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats); 1106 int packets_dropped = 0; 1107 struct sk_buff *skb; 1108 1109 while ((skb = __skb_dequeue(rxq)) != NULL) { 1110 int pull_to = NETFRONT_SKB_CB(skb)->pull_to; 1111 1112 if (pull_to > skb_headlen(skb)) 1113 __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); 1114 1115 /* Ethernet work: Delayed to here as it peeks the header. */ 1116 skb->protocol = eth_type_trans(skb, queue->info->netdev); 1117 skb_reset_network_header(skb); 1118 1119 if (checksum_setup(queue->info->netdev, skb)) { 1120 kfree_skb(skb); 1121 packets_dropped++; 1122 queue->info->netdev->stats.rx_errors++; 1123 continue; 1124 } 1125 1126 u64_stats_update_begin(&rx_stats->syncp); 1127 rx_stats->packets++; 1128 rx_stats->bytes += skb->len; 1129 u64_stats_update_end(&rx_stats->syncp); 1130 1131 /* Pass it up. */ 1132 napi_gro_receive(&queue->napi, skb); 1133 } 1134 1135 return packets_dropped; 1136 } 1137 1138 static int xennet_poll(struct napi_struct *napi, int budget) 1139 { 1140 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi); 1141 struct net_device *dev = queue->info->netdev; 1142 struct sk_buff *skb; 1143 struct netfront_rx_info rinfo; 1144 struct xen_netif_rx_response *rx = &rinfo.rx; 1145 struct xen_netif_extra_info *extras = rinfo.extras; 1146 RING_IDX i, rp; 1147 int work_done; 1148 struct sk_buff_head rxq; 1149 struct sk_buff_head errq; 1150 struct sk_buff_head tmpq; 1151 int err; 1152 bool need_xdp_flush = false; 1153 1154 spin_lock(&queue->rx_lock); 1155 1156 skb_queue_head_init(&rxq); 1157 skb_queue_head_init(&errq); 1158 skb_queue_head_init(&tmpq); 1159 1160 rp = queue->rx.sring->rsp_prod; 1161 rmb(); /* Ensure we see queued responses up to 'rp'. */ 1162 1163 i = queue->rx.rsp_cons; 1164 work_done = 0; 1165 while ((i != rp) && (work_done < budget)) { 1166 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx)); 1167 memset(extras, 0, sizeof(rinfo.extras)); 1168 1169 err = xennet_get_responses(queue, &rinfo, rp, &tmpq, 1170 &need_xdp_flush); 1171 1172 if (unlikely(err)) { 1173 err: 1174 while ((skb = __skb_dequeue(&tmpq))) 1175 __skb_queue_tail(&errq, skb); 1176 dev->stats.rx_errors++; 1177 i = queue->rx.rsp_cons; 1178 continue; 1179 } 1180 1181 skb = __skb_dequeue(&tmpq); 1182 1183 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { 1184 struct xen_netif_extra_info *gso; 1185 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; 1186 1187 if (unlikely(xennet_set_skb_gso(skb, gso))) { 1188 __skb_queue_head(&tmpq, skb); 1189 queue->rx.rsp_cons += skb_queue_len(&tmpq); 1190 goto err; 1191 } 1192 } 1193 1194 NETFRONT_SKB_CB(skb)->pull_to = rx->status; 1195 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) 1196 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; 1197 1198 skb_frag_off_set(&skb_shinfo(skb)->frags[0], rx->offset); 1199 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status); 1200 skb->data_len = rx->status; 1201 skb->len += rx->status; 1202 1203 if (unlikely(xennet_fill_frags(queue, skb, &tmpq))) 1204 goto err; 1205 1206 if (rx->flags & XEN_NETRXF_csum_blank) 1207 skb->ip_summed = CHECKSUM_PARTIAL; 1208 else if (rx->flags & XEN_NETRXF_data_validated) 1209 skb->ip_summed = CHECKSUM_UNNECESSARY; 1210 1211 __skb_queue_tail(&rxq, skb); 1212 1213 i = ++queue->rx.rsp_cons; 1214 work_done++; 1215 } 1216 if (need_xdp_flush) 1217 xdp_do_flush(); 1218 1219 __skb_queue_purge(&errq); 1220 1221 work_done -= handle_incoming_queue(queue, &rxq); 1222 1223 xennet_alloc_rx_buffers(queue); 1224 1225 if (work_done < budget) { 1226 int more_to_do = 0; 1227 1228 napi_complete_done(napi, work_done); 1229 1230 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do); 1231 if (more_to_do) 1232 napi_schedule(napi); 1233 } 1234 1235 spin_unlock(&queue->rx_lock); 1236 1237 return work_done; 1238 } 1239 1240 static int xennet_change_mtu(struct net_device *dev, int mtu) 1241 { 1242 int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN; 1243 1244 if (mtu > max) 1245 return -EINVAL; 1246 dev->mtu = mtu; 1247 return 0; 1248 } 1249 1250 static void xennet_get_stats64(struct net_device *dev, 1251 struct rtnl_link_stats64 *tot) 1252 { 1253 struct netfront_info *np = netdev_priv(dev); 1254 int cpu; 1255 1256 for_each_possible_cpu(cpu) { 1257 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu); 1258 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu); 1259 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 1260 unsigned int start; 1261 1262 do { 1263 start = u64_stats_fetch_begin_irq(&tx_stats->syncp); 1264 tx_packets = tx_stats->packets; 1265 tx_bytes = tx_stats->bytes; 1266 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start)); 1267 1268 do { 1269 start = u64_stats_fetch_begin_irq(&rx_stats->syncp); 1270 rx_packets = rx_stats->packets; 1271 rx_bytes = rx_stats->bytes; 1272 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start)); 1273 1274 tot->rx_packets += rx_packets; 1275 tot->tx_packets += tx_packets; 1276 tot->rx_bytes += rx_bytes; 1277 tot->tx_bytes += tx_bytes; 1278 } 1279 1280 tot->rx_errors = dev->stats.rx_errors; 1281 tot->tx_dropped = dev->stats.tx_dropped; 1282 } 1283 1284 static void xennet_release_tx_bufs(struct netfront_queue *queue) 1285 { 1286 struct sk_buff *skb; 1287 int i; 1288 1289 for (i = 0; i < NET_TX_RING_SIZE; i++) { 1290 /* Skip over entries which are actually freelist references */ 1291 if (skb_entry_is_link(&queue->tx_skbs[i])) 1292 continue; 1293 1294 skb = queue->tx_skbs[i].skb; 1295 get_page(queue->grant_tx_page[i]); 1296 gnttab_end_foreign_access(queue->grant_tx_ref[i], 1297 GNTMAP_readonly, 1298 (unsigned long)page_address(queue->grant_tx_page[i])); 1299 queue->grant_tx_page[i] = NULL; 1300 queue->grant_tx_ref[i] = GRANT_INVALID_REF; 1301 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i); 1302 dev_kfree_skb_irq(skb); 1303 } 1304 } 1305 1306 static void xennet_release_rx_bufs(struct netfront_queue *queue) 1307 { 1308 int id, ref; 1309 1310 spin_lock_bh(&queue->rx_lock); 1311 1312 for (id = 0; id < NET_RX_RING_SIZE; id++) { 1313 struct sk_buff *skb; 1314 struct page *page; 1315 1316 skb = queue->rx_skbs[id]; 1317 if (!skb) 1318 continue; 1319 1320 ref = queue->grant_rx_ref[id]; 1321 if (ref == GRANT_INVALID_REF) 1322 continue; 1323 1324 page = skb_frag_page(&skb_shinfo(skb)->frags[0]); 1325 1326 /* gnttab_end_foreign_access() needs a page ref until 1327 * foreign access is ended (which may be deferred). 1328 */ 1329 get_page(page); 1330 gnttab_end_foreign_access(ref, 0, 1331 (unsigned long)page_address(page)); 1332 queue->grant_rx_ref[id] = GRANT_INVALID_REF; 1333 1334 kfree_skb(skb); 1335 } 1336 1337 spin_unlock_bh(&queue->rx_lock); 1338 } 1339 1340 static netdev_features_t xennet_fix_features(struct net_device *dev, 1341 netdev_features_t features) 1342 { 1343 struct netfront_info *np = netdev_priv(dev); 1344 1345 if (features & NETIF_F_SG && 1346 !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0)) 1347 features &= ~NETIF_F_SG; 1348 1349 if (features & NETIF_F_IPV6_CSUM && 1350 !xenbus_read_unsigned(np->xbdev->otherend, 1351 "feature-ipv6-csum-offload", 0)) 1352 features &= ~NETIF_F_IPV6_CSUM; 1353 1354 if (features & NETIF_F_TSO && 1355 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0)) 1356 features &= ~NETIF_F_TSO; 1357 1358 if (features & NETIF_F_TSO6 && 1359 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0)) 1360 features &= ~NETIF_F_TSO6; 1361 1362 return features; 1363 } 1364 1365 static int xennet_set_features(struct net_device *dev, 1366 netdev_features_t features) 1367 { 1368 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { 1369 netdev_info(dev, "Reducing MTU because no SG offload"); 1370 dev->mtu = ETH_DATA_LEN; 1371 } 1372 1373 return 0; 1374 } 1375 1376 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id) 1377 { 1378 struct netfront_queue *queue = dev_id; 1379 unsigned long flags; 1380 1381 spin_lock_irqsave(&queue->tx_lock, flags); 1382 xennet_tx_buf_gc(queue); 1383 spin_unlock_irqrestore(&queue->tx_lock, flags); 1384 1385 return IRQ_HANDLED; 1386 } 1387 1388 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id) 1389 { 1390 struct netfront_queue *queue = dev_id; 1391 struct net_device *dev = queue->info->netdev; 1392 1393 if (likely(netif_carrier_ok(dev) && 1394 RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))) 1395 napi_schedule(&queue->napi); 1396 1397 return IRQ_HANDLED; 1398 } 1399 1400 static irqreturn_t xennet_interrupt(int irq, void *dev_id) 1401 { 1402 xennet_tx_interrupt(irq, dev_id); 1403 xennet_rx_interrupt(irq, dev_id); 1404 return IRQ_HANDLED; 1405 } 1406 1407 #ifdef CONFIG_NET_POLL_CONTROLLER 1408 static void xennet_poll_controller(struct net_device *dev) 1409 { 1410 /* Poll each queue */ 1411 struct netfront_info *info = netdev_priv(dev); 1412 unsigned int num_queues = dev->real_num_tx_queues; 1413 unsigned int i; 1414 for (i = 0; i < num_queues; ++i) 1415 xennet_interrupt(0, &info->queues[i]); 1416 } 1417 #endif 1418 1419 #define NETBACK_XDP_HEADROOM_DISABLE 0 1420 #define NETBACK_XDP_HEADROOM_ENABLE 1 1421 1422 static int talk_to_netback_xdp(struct netfront_info *np, int xdp) 1423 { 1424 int err; 1425 unsigned short headroom; 1426 1427 headroom = xdp ? XDP_PACKET_HEADROOM : 0; 1428 err = xenbus_printf(XBT_NIL, np->xbdev->nodename, 1429 "xdp-headroom", "%hu", 1430 headroom); 1431 if (err) 1432 pr_warn("Error writing xdp-headroom\n"); 1433 1434 return err; 1435 } 1436 1437 static int xennet_xdp_set(struct net_device *dev, struct bpf_prog *prog, 1438 struct netlink_ext_ack *extack) 1439 { 1440 unsigned long max_mtu = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM; 1441 struct netfront_info *np = netdev_priv(dev); 1442 struct bpf_prog *old_prog; 1443 unsigned int i, err; 1444 1445 if (dev->mtu > max_mtu) { 1446 netdev_warn(dev, "XDP requires MTU less than %lu\n", max_mtu); 1447 return -EINVAL; 1448 } 1449 1450 if (!np->netback_has_xdp_headroom) 1451 return 0; 1452 1453 xenbus_switch_state(np->xbdev, XenbusStateReconfiguring); 1454 1455 err = talk_to_netback_xdp(np, prog ? NETBACK_XDP_HEADROOM_ENABLE : 1456 NETBACK_XDP_HEADROOM_DISABLE); 1457 if (err) 1458 return err; 1459 1460 /* avoid the race with XDP headroom adjustment */ 1461 wait_event(module_wq, 1462 xenbus_read_driver_state(np->xbdev->otherend) == 1463 XenbusStateReconfigured); 1464 np->netfront_xdp_enabled = true; 1465 1466 old_prog = rtnl_dereference(np->queues[0].xdp_prog); 1467 1468 if (prog) 1469 bpf_prog_add(prog, dev->real_num_tx_queues); 1470 1471 for (i = 0; i < dev->real_num_tx_queues; ++i) 1472 rcu_assign_pointer(np->queues[i].xdp_prog, prog); 1473 1474 if (old_prog) 1475 for (i = 0; i < dev->real_num_tx_queues; ++i) 1476 bpf_prog_put(old_prog); 1477 1478 xenbus_switch_state(np->xbdev, XenbusStateConnected); 1479 1480 return 0; 1481 } 1482 1483 static int xennet_xdp(struct net_device *dev, struct netdev_bpf *xdp) 1484 { 1485 switch (xdp->command) { 1486 case XDP_SETUP_PROG: 1487 return xennet_xdp_set(dev, xdp->prog, xdp->extack); 1488 default: 1489 return -EINVAL; 1490 } 1491 } 1492 1493 static const struct net_device_ops xennet_netdev_ops = { 1494 .ndo_open = xennet_open, 1495 .ndo_stop = xennet_close, 1496 .ndo_start_xmit = xennet_start_xmit, 1497 .ndo_change_mtu = xennet_change_mtu, 1498 .ndo_get_stats64 = xennet_get_stats64, 1499 .ndo_set_mac_address = eth_mac_addr, 1500 .ndo_validate_addr = eth_validate_addr, 1501 .ndo_fix_features = xennet_fix_features, 1502 .ndo_set_features = xennet_set_features, 1503 .ndo_select_queue = xennet_select_queue, 1504 .ndo_bpf = xennet_xdp, 1505 .ndo_xdp_xmit = xennet_xdp_xmit, 1506 #ifdef CONFIG_NET_POLL_CONTROLLER 1507 .ndo_poll_controller = xennet_poll_controller, 1508 #endif 1509 }; 1510 1511 static void xennet_free_netdev(struct net_device *netdev) 1512 { 1513 struct netfront_info *np = netdev_priv(netdev); 1514 1515 free_percpu(np->rx_stats); 1516 free_percpu(np->tx_stats); 1517 free_netdev(netdev); 1518 } 1519 1520 static struct net_device *xennet_create_dev(struct xenbus_device *dev) 1521 { 1522 int err; 1523 struct net_device *netdev; 1524 struct netfront_info *np; 1525 1526 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues); 1527 if (!netdev) 1528 return ERR_PTR(-ENOMEM); 1529 1530 np = netdev_priv(netdev); 1531 np->xbdev = dev; 1532 1533 np->queues = NULL; 1534 1535 err = -ENOMEM; 1536 np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); 1537 if (np->rx_stats == NULL) 1538 goto exit; 1539 np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); 1540 if (np->tx_stats == NULL) 1541 goto exit; 1542 1543 netdev->netdev_ops = &xennet_netdev_ops; 1544 1545 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | 1546 NETIF_F_GSO_ROBUST; 1547 netdev->hw_features = NETIF_F_SG | 1548 NETIF_F_IPV6_CSUM | 1549 NETIF_F_TSO | NETIF_F_TSO6; 1550 1551 /* 1552 * Assume that all hw features are available for now. This set 1553 * will be adjusted by the call to netdev_update_features() in 1554 * xennet_connect() which is the earliest point where we can 1555 * negotiate with the backend regarding supported features. 1556 */ 1557 netdev->features |= netdev->hw_features; 1558 1559 netdev->ethtool_ops = &xennet_ethtool_ops; 1560 netdev->min_mtu = ETH_MIN_MTU; 1561 netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE; 1562 SET_NETDEV_DEV(netdev, &dev->dev); 1563 1564 np->netdev = netdev; 1565 np->netfront_xdp_enabled = false; 1566 1567 netif_carrier_off(netdev); 1568 1569 do { 1570 xenbus_switch_state(dev, XenbusStateInitialising); 1571 err = wait_event_timeout(module_wq, 1572 xenbus_read_driver_state(dev->otherend) != 1573 XenbusStateClosed && 1574 xenbus_read_driver_state(dev->otherend) != 1575 XenbusStateUnknown, XENNET_TIMEOUT); 1576 } while (!err); 1577 1578 return netdev; 1579 1580 exit: 1581 xennet_free_netdev(netdev); 1582 return ERR_PTR(err); 1583 } 1584 1585 /** 1586 * Entry point to this code when a new device is created. Allocate the basic 1587 * structures and the ring buffers for communication with the backend, and 1588 * inform the backend of the appropriate details for those. 1589 */ 1590 static int netfront_probe(struct xenbus_device *dev, 1591 const struct xenbus_device_id *id) 1592 { 1593 int err; 1594 struct net_device *netdev; 1595 struct netfront_info *info; 1596 1597 netdev = xennet_create_dev(dev); 1598 if (IS_ERR(netdev)) { 1599 err = PTR_ERR(netdev); 1600 xenbus_dev_fatal(dev, err, "creating netdev"); 1601 return err; 1602 } 1603 1604 info = netdev_priv(netdev); 1605 dev_set_drvdata(&dev->dev, info); 1606 #ifdef CONFIG_SYSFS 1607 info->netdev->sysfs_groups[0] = &xennet_dev_group; 1608 #endif 1609 1610 return 0; 1611 } 1612 1613 static void xennet_end_access(int ref, void *page) 1614 { 1615 /* This frees the page as a side-effect */ 1616 if (ref != GRANT_INVALID_REF) 1617 gnttab_end_foreign_access(ref, 0, (unsigned long)page); 1618 } 1619 1620 static void xennet_disconnect_backend(struct netfront_info *info) 1621 { 1622 unsigned int i = 0; 1623 unsigned int num_queues = info->netdev->real_num_tx_queues; 1624 1625 netif_carrier_off(info->netdev); 1626 1627 for (i = 0; i < num_queues && info->queues; ++i) { 1628 struct netfront_queue *queue = &info->queues[i]; 1629 1630 del_timer_sync(&queue->rx_refill_timer); 1631 1632 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq)) 1633 unbind_from_irqhandler(queue->tx_irq, queue); 1634 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) { 1635 unbind_from_irqhandler(queue->tx_irq, queue); 1636 unbind_from_irqhandler(queue->rx_irq, queue); 1637 } 1638 queue->tx_evtchn = queue->rx_evtchn = 0; 1639 queue->tx_irq = queue->rx_irq = 0; 1640 1641 if (netif_running(info->netdev)) 1642 napi_synchronize(&queue->napi); 1643 1644 xennet_release_tx_bufs(queue); 1645 xennet_release_rx_bufs(queue); 1646 gnttab_free_grant_references(queue->gref_tx_head); 1647 gnttab_free_grant_references(queue->gref_rx_head); 1648 1649 /* End access and free the pages */ 1650 xennet_end_access(queue->tx_ring_ref, queue->tx.sring); 1651 xennet_end_access(queue->rx_ring_ref, queue->rx.sring); 1652 1653 queue->tx_ring_ref = GRANT_INVALID_REF; 1654 queue->rx_ring_ref = GRANT_INVALID_REF; 1655 queue->tx.sring = NULL; 1656 queue->rx.sring = NULL; 1657 1658 page_pool_destroy(queue->page_pool); 1659 } 1660 } 1661 1662 /** 1663 * We are reconnecting to the backend, due to a suspend/resume, or a backend 1664 * driver restart. We tear down our netif structure and recreate it, but 1665 * leave the device-layer structures intact so that this is transparent to the 1666 * rest of the kernel. 1667 */ 1668 static int netfront_resume(struct xenbus_device *dev) 1669 { 1670 struct netfront_info *info = dev_get_drvdata(&dev->dev); 1671 1672 dev_dbg(&dev->dev, "%s\n", dev->nodename); 1673 1674 xennet_disconnect_backend(info); 1675 return 0; 1676 } 1677 1678 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) 1679 { 1680 char *s, *e, *macstr; 1681 int i; 1682 1683 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); 1684 if (IS_ERR(macstr)) 1685 return PTR_ERR(macstr); 1686 1687 for (i = 0; i < ETH_ALEN; i++) { 1688 mac[i] = simple_strtoul(s, &e, 16); 1689 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { 1690 kfree(macstr); 1691 return -ENOENT; 1692 } 1693 s = e+1; 1694 } 1695 1696 kfree(macstr); 1697 return 0; 1698 } 1699 1700 static int setup_netfront_single(struct netfront_queue *queue) 1701 { 1702 int err; 1703 1704 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); 1705 if (err < 0) 1706 goto fail; 1707 1708 err = bind_evtchn_to_irqhandler(queue->tx_evtchn, 1709 xennet_interrupt, 1710 0, queue->info->netdev->name, queue); 1711 if (err < 0) 1712 goto bind_fail; 1713 queue->rx_evtchn = queue->tx_evtchn; 1714 queue->rx_irq = queue->tx_irq = err; 1715 1716 return 0; 1717 1718 bind_fail: 1719 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); 1720 queue->tx_evtchn = 0; 1721 fail: 1722 return err; 1723 } 1724 1725 static int setup_netfront_split(struct netfront_queue *queue) 1726 { 1727 int err; 1728 1729 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); 1730 if (err < 0) 1731 goto fail; 1732 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn); 1733 if (err < 0) 1734 goto alloc_rx_evtchn_fail; 1735 1736 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), 1737 "%s-tx", queue->name); 1738 err = bind_evtchn_to_irqhandler(queue->tx_evtchn, 1739 xennet_tx_interrupt, 1740 0, queue->tx_irq_name, queue); 1741 if (err < 0) 1742 goto bind_tx_fail; 1743 queue->tx_irq = err; 1744 1745 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), 1746 "%s-rx", queue->name); 1747 err = bind_evtchn_to_irqhandler(queue->rx_evtchn, 1748 xennet_rx_interrupt, 1749 0, queue->rx_irq_name, queue); 1750 if (err < 0) 1751 goto bind_rx_fail; 1752 queue->rx_irq = err; 1753 1754 return 0; 1755 1756 bind_rx_fail: 1757 unbind_from_irqhandler(queue->tx_irq, queue); 1758 queue->tx_irq = 0; 1759 bind_tx_fail: 1760 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn); 1761 queue->rx_evtchn = 0; 1762 alloc_rx_evtchn_fail: 1763 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); 1764 queue->tx_evtchn = 0; 1765 fail: 1766 return err; 1767 } 1768 1769 static int setup_netfront(struct xenbus_device *dev, 1770 struct netfront_queue *queue, unsigned int feature_split_evtchn) 1771 { 1772 struct xen_netif_tx_sring *txs; 1773 struct xen_netif_rx_sring *rxs; 1774 grant_ref_t gref; 1775 int err; 1776 1777 queue->tx_ring_ref = GRANT_INVALID_REF; 1778 queue->rx_ring_ref = GRANT_INVALID_REF; 1779 queue->rx.sring = NULL; 1780 queue->tx.sring = NULL; 1781 1782 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); 1783 if (!txs) { 1784 err = -ENOMEM; 1785 xenbus_dev_fatal(dev, err, "allocating tx ring page"); 1786 goto fail; 1787 } 1788 SHARED_RING_INIT(txs); 1789 FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE); 1790 1791 err = xenbus_grant_ring(dev, txs, 1, &gref); 1792 if (err < 0) 1793 goto grant_tx_ring_fail; 1794 queue->tx_ring_ref = gref; 1795 1796 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); 1797 if (!rxs) { 1798 err = -ENOMEM; 1799 xenbus_dev_fatal(dev, err, "allocating rx ring page"); 1800 goto alloc_rx_ring_fail; 1801 } 1802 SHARED_RING_INIT(rxs); 1803 FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE); 1804 1805 err = xenbus_grant_ring(dev, rxs, 1, &gref); 1806 if (err < 0) 1807 goto grant_rx_ring_fail; 1808 queue->rx_ring_ref = gref; 1809 1810 if (feature_split_evtchn) 1811 err = setup_netfront_split(queue); 1812 /* setup single event channel if 1813 * a) feature-split-event-channels == 0 1814 * b) feature-split-event-channels == 1 but failed to setup 1815 */ 1816 if (!feature_split_evtchn || (feature_split_evtchn && err)) 1817 err = setup_netfront_single(queue); 1818 1819 if (err) 1820 goto alloc_evtchn_fail; 1821 1822 return 0; 1823 1824 /* If we fail to setup netfront, it is safe to just revoke access to 1825 * granted pages because backend is not accessing it at this point. 1826 */ 1827 alloc_evtchn_fail: 1828 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0); 1829 grant_rx_ring_fail: 1830 free_page((unsigned long)rxs); 1831 alloc_rx_ring_fail: 1832 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0); 1833 grant_tx_ring_fail: 1834 free_page((unsigned long)txs); 1835 fail: 1836 return err; 1837 } 1838 1839 /* Queue-specific initialisation 1840 * This used to be done in xennet_create_dev() but must now 1841 * be run per-queue. 1842 */ 1843 static int xennet_init_queue(struct netfront_queue *queue) 1844 { 1845 unsigned short i; 1846 int err = 0; 1847 char *devid; 1848 1849 spin_lock_init(&queue->tx_lock); 1850 spin_lock_init(&queue->rx_lock); 1851 1852 timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0); 1853 1854 devid = strrchr(queue->info->xbdev->nodename, '/') + 1; 1855 snprintf(queue->name, sizeof(queue->name), "vif%s-q%u", 1856 devid, queue->id); 1857 1858 /* Initialise tx_skbs as a free chain containing every entry. */ 1859 queue->tx_skb_freelist = 0; 1860 for (i = 0; i < NET_TX_RING_SIZE; i++) { 1861 skb_entry_set_link(&queue->tx_skbs[i], i+1); 1862 queue->grant_tx_ref[i] = GRANT_INVALID_REF; 1863 queue->grant_tx_page[i] = NULL; 1864 } 1865 1866 /* Clear out rx_skbs */ 1867 for (i = 0; i < NET_RX_RING_SIZE; i++) { 1868 queue->rx_skbs[i] = NULL; 1869 queue->grant_rx_ref[i] = GRANT_INVALID_REF; 1870 } 1871 1872 /* A grant for every tx ring slot */ 1873 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, 1874 &queue->gref_tx_head) < 0) { 1875 pr_alert("can't alloc tx grant refs\n"); 1876 err = -ENOMEM; 1877 goto exit; 1878 } 1879 1880 /* A grant for every rx ring slot */ 1881 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE, 1882 &queue->gref_rx_head) < 0) { 1883 pr_alert("can't alloc rx grant refs\n"); 1884 err = -ENOMEM; 1885 goto exit_free_tx; 1886 } 1887 1888 return 0; 1889 1890 exit_free_tx: 1891 gnttab_free_grant_references(queue->gref_tx_head); 1892 exit: 1893 return err; 1894 } 1895 1896 static int write_queue_xenstore_keys(struct netfront_queue *queue, 1897 struct xenbus_transaction *xbt, int write_hierarchical) 1898 { 1899 /* Write the queue-specific keys into XenStore in the traditional 1900 * way for a single queue, or in a queue subkeys for multiple 1901 * queues. 1902 */ 1903 struct xenbus_device *dev = queue->info->xbdev; 1904 int err; 1905 const char *message; 1906 char *path; 1907 size_t pathsize; 1908 1909 /* Choose the correct place to write the keys */ 1910 if (write_hierarchical) { 1911 pathsize = strlen(dev->nodename) + 10; 1912 path = kzalloc(pathsize, GFP_KERNEL); 1913 if (!path) { 1914 err = -ENOMEM; 1915 message = "out of memory while writing ring references"; 1916 goto error; 1917 } 1918 snprintf(path, pathsize, "%s/queue-%u", 1919 dev->nodename, queue->id); 1920 } else { 1921 path = (char *)dev->nodename; 1922 } 1923 1924 /* Write ring references */ 1925 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u", 1926 queue->tx_ring_ref); 1927 if (err) { 1928 message = "writing tx-ring-ref"; 1929 goto error; 1930 } 1931 1932 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u", 1933 queue->rx_ring_ref); 1934 if (err) { 1935 message = "writing rx-ring-ref"; 1936 goto error; 1937 } 1938 1939 /* Write event channels; taking into account both shared 1940 * and split event channel scenarios. 1941 */ 1942 if (queue->tx_evtchn == queue->rx_evtchn) { 1943 /* Shared event channel */ 1944 err = xenbus_printf(*xbt, path, 1945 "event-channel", "%u", queue->tx_evtchn); 1946 if (err) { 1947 message = "writing event-channel"; 1948 goto error; 1949 } 1950 } else { 1951 /* Split event channels */ 1952 err = xenbus_printf(*xbt, path, 1953 "event-channel-tx", "%u", queue->tx_evtchn); 1954 if (err) { 1955 message = "writing event-channel-tx"; 1956 goto error; 1957 } 1958 1959 err = xenbus_printf(*xbt, path, 1960 "event-channel-rx", "%u", queue->rx_evtchn); 1961 if (err) { 1962 message = "writing event-channel-rx"; 1963 goto error; 1964 } 1965 } 1966 1967 if (write_hierarchical) 1968 kfree(path); 1969 return 0; 1970 1971 error: 1972 if (write_hierarchical) 1973 kfree(path); 1974 xenbus_dev_fatal(dev, err, "%s", message); 1975 return err; 1976 } 1977 1978 static void xennet_destroy_queues(struct netfront_info *info) 1979 { 1980 unsigned int i; 1981 1982 for (i = 0; i < info->netdev->real_num_tx_queues; i++) { 1983 struct netfront_queue *queue = &info->queues[i]; 1984 1985 if (netif_running(info->netdev)) 1986 napi_disable(&queue->napi); 1987 netif_napi_del(&queue->napi); 1988 } 1989 1990 kfree(info->queues); 1991 info->queues = NULL; 1992 } 1993 1994 1995 1996 static int xennet_create_page_pool(struct netfront_queue *queue) 1997 { 1998 int err; 1999 struct page_pool_params pp_params = { 2000 .order = 0, 2001 .flags = 0, 2002 .pool_size = NET_RX_RING_SIZE, 2003 .nid = NUMA_NO_NODE, 2004 .dev = &queue->info->netdev->dev, 2005 .offset = XDP_PACKET_HEADROOM, 2006 .max_len = XEN_PAGE_SIZE - XDP_PACKET_HEADROOM, 2007 }; 2008 2009 queue->page_pool = page_pool_create(&pp_params); 2010 if (IS_ERR(queue->page_pool)) { 2011 err = PTR_ERR(queue->page_pool); 2012 queue->page_pool = NULL; 2013 return err; 2014 } 2015 2016 err = xdp_rxq_info_reg(&queue->xdp_rxq, queue->info->netdev, 2017 queue->id); 2018 if (err) { 2019 netdev_err(queue->info->netdev, "xdp_rxq_info_reg failed\n"); 2020 goto err_free_pp; 2021 } 2022 2023 err = xdp_rxq_info_reg_mem_model(&queue->xdp_rxq, 2024 MEM_TYPE_PAGE_POOL, queue->page_pool); 2025 if (err) { 2026 netdev_err(queue->info->netdev, "xdp_rxq_info_reg_mem_model failed\n"); 2027 goto err_unregister_rxq; 2028 } 2029 return 0; 2030 2031 err_unregister_rxq: 2032 xdp_rxq_info_unreg(&queue->xdp_rxq); 2033 err_free_pp: 2034 page_pool_destroy(queue->page_pool); 2035 queue->page_pool = NULL; 2036 return err; 2037 } 2038 2039 static int xennet_create_queues(struct netfront_info *info, 2040 unsigned int *num_queues) 2041 { 2042 unsigned int i; 2043 int ret; 2044 2045 info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue), 2046 GFP_KERNEL); 2047 if (!info->queues) 2048 return -ENOMEM; 2049 2050 for (i = 0; i < *num_queues; i++) { 2051 struct netfront_queue *queue = &info->queues[i]; 2052 2053 queue->id = i; 2054 queue->info = info; 2055 2056 ret = xennet_init_queue(queue); 2057 if (ret < 0) { 2058 dev_warn(&info->xbdev->dev, 2059 "only created %d queues\n", i); 2060 *num_queues = i; 2061 break; 2062 } 2063 2064 /* use page pool recycling instead of buddy allocator */ 2065 ret = xennet_create_page_pool(queue); 2066 if (ret < 0) { 2067 dev_err(&info->xbdev->dev, "can't allocate page pool\n"); 2068 *num_queues = i; 2069 return ret; 2070 } 2071 2072 netif_napi_add(queue->info->netdev, &queue->napi, 2073 xennet_poll, 64); 2074 if (netif_running(info->netdev)) 2075 napi_enable(&queue->napi); 2076 } 2077 2078 netif_set_real_num_tx_queues(info->netdev, *num_queues); 2079 2080 if (*num_queues == 0) { 2081 dev_err(&info->xbdev->dev, "no queues\n"); 2082 return -EINVAL; 2083 } 2084 return 0; 2085 } 2086 2087 /* Common code used when first setting up, and when resuming. */ 2088 static int talk_to_netback(struct xenbus_device *dev, 2089 struct netfront_info *info) 2090 { 2091 const char *message; 2092 struct xenbus_transaction xbt; 2093 int err; 2094 unsigned int feature_split_evtchn; 2095 unsigned int i = 0; 2096 unsigned int max_queues = 0; 2097 struct netfront_queue *queue = NULL; 2098 unsigned int num_queues = 1; 2099 2100 info->netdev->irq = 0; 2101 2102 /* Check if backend supports multiple queues */ 2103 max_queues = xenbus_read_unsigned(info->xbdev->otherend, 2104 "multi-queue-max-queues", 1); 2105 num_queues = min(max_queues, xennet_max_queues); 2106 2107 /* Check feature-split-event-channels */ 2108 feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend, 2109 "feature-split-event-channels", 0); 2110 2111 /* Read mac addr. */ 2112 err = xen_net_read_mac(dev, info->netdev->dev_addr); 2113 if (err) { 2114 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); 2115 goto out_unlocked; 2116 } 2117 2118 info->netback_has_xdp_headroom = xenbus_read_unsigned(info->xbdev->otherend, 2119 "feature-xdp-headroom", 0); 2120 if (info->netback_has_xdp_headroom) { 2121 /* set the current xen-netfront xdp state */ 2122 err = talk_to_netback_xdp(info, info->netfront_xdp_enabled ? 2123 NETBACK_XDP_HEADROOM_ENABLE : 2124 NETBACK_XDP_HEADROOM_DISABLE); 2125 if (err) 2126 goto out_unlocked; 2127 } 2128 2129 rtnl_lock(); 2130 if (info->queues) 2131 xennet_destroy_queues(info); 2132 2133 err = xennet_create_queues(info, &num_queues); 2134 if (err < 0) { 2135 xenbus_dev_fatal(dev, err, "creating queues"); 2136 kfree(info->queues); 2137 info->queues = NULL; 2138 goto out; 2139 } 2140 rtnl_unlock(); 2141 2142 /* Create shared ring, alloc event channel -- for each queue */ 2143 for (i = 0; i < num_queues; ++i) { 2144 queue = &info->queues[i]; 2145 err = setup_netfront(dev, queue, feature_split_evtchn); 2146 if (err) 2147 goto destroy_ring; 2148 } 2149 2150 again: 2151 err = xenbus_transaction_start(&xbt); 2152 if (err) { 2153 xenbus_dev_fatal(dev, err, "starting transaction"); 2154 goto destroy_ring; 2155 } 2156 2157 if (xenbus_exists(XBT_NIL, 2158 info->xbdev->otherend, "multi-queue-max-queues")) { 2159 /* Write the number of queues */ 2160 err = xenbus_printf(xbt, dev->nodename, 2161 "multi-queue-num-queues", "%u", num_queues); 2162 if (err) { 2163 message = "writing multi-queue-num-queues"; 2164 goto abort_transaction_no_dev_fatal; 2165 } 2166 } 2167 2168 if (num_queues == 1) { 2169 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */ 2170 if (err) 2171 goto abort_transaction_no_dev_fatal; 2172 } else { 2173 /* Write the keys for each queue */ 2174 for (i = 0; i < num_queues; ++i) { 2175 queue = &info->queues[i]; 2176 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */ 2177 if (err) 2178 goto abort_transaction_no_dev_fatal; 2179 } 2180 } 2181 2182 /* The remaining keys are not queue-specific */ 2183 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", 2184 1); 2185 if (err) { 2186 message = "writing request-rx-copy"; 2187 goto abort_transaction; 2188 } 2189 2190 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); 2191 if (err) { 2192 message = "writing feature-rx-notify"; 2193 goto abort_transaction; 2194 } 2195 2196 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); 2197 if (err) { 2198 message = "writing feature-sg"; 2199 goto abort_transaction; 2200 } 2201 2202 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); 2203 if (err) { 2204 message = "writing feature-gso-tcpv4"; 2205 goto abort_transaction; 2206 } 2207 2208 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1"); 2209 if (err) { 2210 message = "writing feature-gso-tcpv6"; 2211 goto abort_transaction; 2212 } 2213 2214 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload", 2215 "1"); 2216 if (err) { 2217 message = "writing feature-ipv6-csum-offload"; 2218 goto abort_transaction; 2219 } 2220 2221 err = xenbus_transaction_end(xbt, 0); 2222 if (err) { 2223 if (err == -EAGAIN) 2224 goto again; 2225 xenbus_dev_fatal(dev, err, "completing transaction"); 2226 goto destroy_ring; 2227 } 2228 2229 return 0; 2230 2231 abort_transaction: 2232 xenbus_dev_fatal(dev, err, "%s", message); 2233 abort_transaction_no_dev_fatal: 2234 xenbus_transaction_end(xbt, 1); 2235 destroy_ring: 2236 xennet_disconnect_backend(info); 2237 rtnl_lock(); 2238 xennet_destroy_queues(info); 2239 out: 2240 rtnl_unlock(); 2241 out_unlocked: 2242 device_unregister(&dev->dev); 2243 return err; 2244 } 2245 2246 static int xennet_connect(struct net_device *dev) 2247 { 2248 struct netfront_info *np = netdev_priv(dev); 2249 unsigned int num_queues = 0; 2250 int err; 2251 unsigned int j = 0; 2252 struct netfront_queue *queue = NULL; 2253 2254 if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) { 2255 dev_info(&dev->dev, 2256 "backend does not support copying receive path\n"); 2257 return -ENODEV; 2258 } 2259 2260 err = talk_to_netback(np->xbdev, np); 2261 if (err) 2262 return err; 2263 if (np->netback_has_xdp_headroom) 2264 pr_info("backend supports XDP headroom\n"); 2265 2266 /* talk_to_netback() sets the correct number of queues */ 2267 num_queues = dev->real_num_tx_queues; 2268 2269 if (dev->reg_state == NETREG_UNINITIALIZED) { 2270 err = register_netdev(dev); 2271 if (err) { 2272 pr_warn("%s: register_netdev err=%d\n", __func__, err); 2273 device_unregister(&np->xbdev->dev); 2274 return err; 2275 } 2276 } 2277 2278 rtnl_lock(); 2279 netdev_update_features(dev); 2280 rtnl_unlock(); 2281 2282 /* 2283 * All public and private state should now be sane. Get 2284 * ready to start sending and receiving packets and give the driver 2285 * domain a kick because we've probably just requeued some 2286 * packets. 2287 */ 2288 netif_carrier_on(np->netdev); 2289 for (j = 0; j < num_queues; ++j) { 2290 queue = &np->queues[j]; 2291 2292 notify_remote_via_irq(queue->tx_irq); 2293 if (queue->tx_irq != queue->rx_irq) 2294 notify_remote_via_irq(queue->rx_irq); 2295 2296 spin_lock_irq(&queue->tx_lock); 2297 xennet_tx_buf_gc(queue); 2298 spin_unlock_irq(&queue->tx_lock); 2299 2300 spin_lock_bh(&queue->rx_lock); 2301 xennet_alloc_rx_buffers(queue); 2302 spin_unlock_bh(&queue->rx_lock); 2303 } 2304 2305 return 0; 2306 } 2307 2308 /** 2309 * Callback received when the backend's state changes. 2310 */ 2311 static void netback_changed(struct xenbus_device *dev, 2312 enum xenbus_state backend_state) 2313 { 2314 struct netfront_info *np = dev_get_drvdata(&dev->dev); 2315 struct net_device *netdev = np->netdev; 2316 2317 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); 2318 2319 wake_up_all(&module_wq); 2320 2321 switch (backend_state) { 2322 case XenbusStateInitialising: 2323 case XenbusStateInitialised: 2324 case XenbusStateReconfiguring: 2325 case XenbusStateReconfigured: 2326 case XenbusStateUnknown: 2327 break; 2328 2329 case XenbusStateInitWait: 2330 if (dev->state != XenbusStateInitialising) 2331 break; 2332 if (xennet_connect(netdev) != 0) 2333 break; 2334 xenbus_switch_state(dev, XenbusStateConnected); 2335 break; 2336 2337 case XenbusStateConnected: 2338 netdev_notify_peers(netdev); 2339 break; 2340 2341 case XenbusStateClosed: 2342 if (dev->state == XenbusStateClosed) 2343 break; 2344 fallthrough; /* Missed the backend's CLOSING state */ 2345 case XenbusStateClosing: 2346 xenbus_frontend_closed(dev); 2347 break; 2348 } 2349 } 2350 2351 static const struct xennet_stat { 2352 char name[ETH_GSTRING_LEN]; 2353 u16 offset; 2354 } xennet_stats[] = { 2355 { 2356 "rx_gso_checksum_fixup", 2357 offsetof(struct netfront_info, rx_gso_checksum_fixup) 2358 }, 2359 }; 2360 2361 static int xennet_get_sset_count(struct net_device *dev, int string_set) 2362 { 2363 switch (string_set) { 2364 case ETH_SS_STATS: 2365 return ARRAY_SIZE(xennet_stats); 2366 default: 2367 return -EINVAL; 2368 } 2369 } 2370 2371 static void xennet_get_ethtool_stats(struct net_device *dev, 2372 struct ethtool_stats *stats, u64 * data) 2373 { 2374 void *np = netdev_priv(dev); 2375 int i; 2376 2377 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) 2378 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset)); 2379 } 2380 2381 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) 2382 { 2383 int i; 2384 2385 switch (stringset) { 2386 case ETH_SS_STATS: 2387 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) 2388 memcpy(data + i * ETH_GSTRING_LEN, 2389 xennet_stats[i].name, ETH_GSTRING_LEN); 2390 break; 2391 } 2392 } 2393 2394 static const struct ethtool_ops xennet_ethtool_ops = 2395 { 2396 .get_link = ethtool_op_get_link, 2397 2398 .get_sset_count = xennet_get_sset_count, 2399 .get_ethtool_stats = xennet_get_ethtool_stats, 2400 .get_strings = xennet_get_strings, 2401 .get_ts_info = ethtool_op_get_ts_info, 2402 }; 2403 2404 #ifdef CONFIG_SYSFS 2405 static ssize_t show_rxbuf(struct device *dev, 2406 struct device_attribute *attr, char *buf) 2407 { 2408 return sprintf(buf, "%lu\n", NET_RX_RING_SIZE); 2409 } 2410 2411 static ssize_t store_rxbuf(struct device *dev, 2412 struct device_attribute *attr, 2413 const char *buf, size_t len) 2414 { 2415 char *endp; 2416 unsigned long target; 2417 2418 if (!capable(CAP_NET_ADMIN)) 2419 return -EPERM; 2420 2421 target = simple_strtoul(buf, &endp, 0); 2422 if (endp == buf) 2423 return -EBADMSG; 2424 2425 /* rxbuf_min and rxbuf_max are no longer configurable. */ 2426 2427 return len; 2428 } 2429 2430 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf); 2431 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf); 2432 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL); 2433 2434 static struct attribute *xennet_dev_attrs[] = { 2435 &dev_attr_rxbuf_min.attr, 2436 &dev_attr_rxbuf_max.attr, 2437 &dev_attr_rxbuf_cur.attr, 2438 NULL 2439 }; 2440 2441 static const struct attribute_group xennet_dev_group = { 2442 .attrs = xennet_dev_attrs 2443 }; 2444 #endif /* CONFIG_SYSFS */ 2445 2446 static void xennet_bus_close(struct xenbus_device *dev) 2447 { 2448 int ret; 2449 2450 if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed) 2451 return; 2452 do { 2453 xenbus_switch_state(dev, XenbusStateClosing); 2454 ret = wait_event_timeout(module_wq, 2455 xenbus_read_driver_state(dev->otherend) == 2456 XenbusStateClosing || 2457 xenbus_read_driver_state(dev->otherend) == 2458 XenbusStateClosed || 2459 xenbus_read_driver_state(dev->otherend) == 2460 XenbusStateUnknown, 2461 XENNET_TIMEOUT); 2462 } while (!ret); 2463 2464 if (xenbus_read_driver_state(dev->otherend) == XenbusStateClosed) 2465 return; 2466 2467 do { 2468 xenbus_switch_state(dev, XenbusStateClosed); 2469 ret = wait_event_timeout(module_wq, 2470 xenbus_read_driver_state(dev->otherend) == 2471 XenbusStateClosed || 2472 xenbus_read_driver_state(dev->otherend) == 2473 XenbusStateUnknown, 2474 XENNET_TIMEOUT); 2475 } while (!ret); 2476 } 2477 2478 static int xennet_remove(struct xenbus_device *dev) 2479 { 2480 struct netfront_info *info = dev_get_drvdata(&dev->dev); 2481 2482 xennet_bus_close(dev); 2483 xennet_disconnect_backend(info); 2484 2485 if (info->netdev->reg_state == NETREG_REGISTERED) 2486 unregister_netdev(info->netdev); 2487 2488 if (info->queues) { 2489 rtnl_lock(); 2490 xennet_destroy_queues(info); 2491 rtnl_unlock(); 2492 } 2493 xennet_free_netdev(info->netdev); 2494 2495 return 0; 2496 } 2497 2498 static const struct xenbus_device_id netfront_ids[] = { 2499 { "vif" }, 2500 { "" } 2501 }; 2502 2503 static struct xenbus_driver netfront_driver = { 2504 .ids = netfront_ids, 2505 .probe = netfront_probe, 2506 .remove = xennet_remove, 2507 .resume = netfront_resume, 2508 .otherend_changed = netback_changed, 2509 }; 2510 2511 static int __init netif_init(void) 2512 { 2513 if (!xen_domain()) 2514 return -ENODEV; 2515 2516 if (!xen_has_pv_nic_devices()) 2517 return -ENODEV; 2518 2519 pr_info("Initialising Xen virtual ethernet driver\n"); 2520 2521 /* Allow as many queues as there are CPUs inut max. 8 if user has not 2522 * specified a value. 2523 */ 2524 if (xennet_max_queues == 0) 2525 xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT, 2526 num_online_cpus()); 2527 2528 return xenbus_register_frontend(&netfront_driver); 2529 } 2530 module_init(netif_init); 2531 2532 2533 static void __exit netif_exit(void) 2534 { 2535 xenbus_unregister_driver(&netfront_driver); 2536 } 2537 module_exit(netif_exit); 2538 2539 MODULE_DESCRIPTION("Xen virtual network device frontend"); 2540 MODULE_LICENSE("GPL"); 2541 MODULE_ALIAS("xen:vif"); 2542 MODULE_ALIAS("xennet"); 2543