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 48 #include <xen/xen.h> 49 #include <xen/xenbus.h> 50 #include <xen/events.h> 51 #include <xen/page.h> 52 #include <xen/platform_pci.h> 53 #include <xen/grant_table.h> 54 55 #include <xen/interface/io/netif.h> 56 #include <xen/interface/memory.h> 57 #include <xen/interface/grant_table.h> 58 59 /* Module parameters */ 60 #define MAX_QUEUES_DEFAULT 8 61 static unsigned int xennet_max_queues; 62 module_param_named(max_queues, xennet_max_queues, uint, 0644); 63 MODULE_PARM_DESC(max_queues, 64 "Maximum number of queues per virtual interface"); 65 66 static const struct ethtool_ops xennet_ethtool_ops; 67 68 struct netfront_cb { 69 int pull_to; 70 }; 71 72 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb)) 73 74 #define RX_COPY_THRESHOLD 256 75 76 #define GRANT_INVALID_REF 0 77 78 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, XEN_PAGE_SIZE) 79 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, XEN_PAGE_SIZE) 80 81 /* Minimum number of Rx slots (includes slot for GSO metadata). */ 82 #define NET_RX_SLOTS_MIN (XEN_NETIF_NR_SLOTS_MIN + 1) 83 84 /* Queue name is interface name with "-qNNN" appended */ 85 #define QUEUE_NAME_SIZE (IFNAMSIZ + 6) 86 87 /* IRQ name is queue name with "-tx" or "-rx" appended */ 88 #define IRQ_NAME_SIZE (QUEUE_NAME_SIZE + 3) 89 90 static DECLARE_WAIT_QUEUE_HEAD(module_load_q); 91 static DECLARE_WAIT_QUEUE_HEAD(module_unload_q); 92 93 struct netfront_stats { 94 u64 packets; 95 u64 bytes; 96 struct u64_stats_sync syncp; 97 }; 98 99 struct netfront_info; 100 101 struct netfront_queue { 102 unsigned int id; /* Queue ID, 0-based */ 103 char name[QUEUE_NAME_SIZE]; /* DEVNAME-qN */ 104 struct netfront_info *info; 105 106 struct napi_struct napi; 107 108 /* Split event channels support, tx_* == rx_* when using 109 * single event channel. 110 */ 111 unsigned int tx_evtchn, rx_evtchn; 112 unsigned int tx_irq, rx_irq; 113 /* Only used when split event channels support is enabled */ 114 char tx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-tx */ 115 char rx_irq_name[IRQ_NAME_SIZE]; /* DEVNAME-qN-rx */ 116 117 spinlock_t tx_lock; 118 struct xen_netif_tx_front_ring tx; 119 int tx_ring_ref; 120 121 /* 122 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries 123 * are linked from tx_skb_freelist through skb_entry.link. 124 * 125 * NB. Freelist index entries are always going to be less than 126 * PAGE_OFFSET, whereas pointers to skbs will always be equal or 127 * greater than PAGE_OFFSET: we use this property to distinguish 128 * them. 129 */ 130 union skb_entry { 131 struct sk_buff *skb; 132 unsigned long link; 133 } tx_skbs[NET_TX_RING_SIZE]; 134 grant_ref_t gref_tx_head; 135 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; 136 struct page *grant_tx_page[NET_TX_RING_SIZE]; 137 unsigned tx_skb_freelist; 138 139 spinlock_t rx_lock ____cacheline_aligned_in_smp; 140 struct xen_netif_rx_front_ring rx; 141 int rx_ring_ref; 142 143 struct timer_list rx_refill_timer; 144 145 struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; 146 grant_ref_t gref_rx_head; 147 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; 148 }; 149 150 struct netfront_info { 151 struct list_head list; 152 struct net_device *netdev; 153 154 struct xenbus_device *xbdev; 155 156 /* Multi-queue support */ 157 struct netfront_queue *queues; 158 159 /* Statistics */ 160 struct netfront_stats __percpu *rx_stats; 161 struct netfront_stats __percpu *tx_stats; 162 163 atomic_t rx_gso_checksum_fixup; 164 }; 165 166 struct netfront_rx_info { 167 struct xen_netif_rx_response rx; 168 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; 169 }; 170 171 static void skb_entry_set_link(union skb_entry *list, unsigned short id) 172 { 173 list->link = id; 174 } 175 176 static int skb_entry_is_link(const union skb_entry *list) 177 { 178 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link)); 179 return (unsigned long)list->skb < PAGE_OFFSET; 180 } 181 182 /* 183 * Access macros for acquiring freeing slots in tx_skbs[]. 184 */ 185 186 static void add_id_to_freelist(unsigned *head, union skb_entry *list, 187 unsigned short id) 188 { 189 skb_entry_set_link(&list[id], *head); 190 *head = id; 191 } 192 193 static unsigned short get_id_from_freelist(unsigned *head, 194 union skb_entry *list) 195 { 196 unsigned int id = *head; 197 *head = list[id].link; 198 return id; 199 } 200 201 static int xennet_rxidx(RING_IDX idx) 202 { 203 return idx & (NET_RX_RING_SIZE - 1); 204 } 205 206 static struct sk_buff *xennet_get_rx_skb(struct netfront_queue *queue, 207 RING_IDX ri) 208 { 209 int i = xennet_rxidx(ri); 210 struct sk_buff *skb = queue->rx_skbs[i]; 211 queue->rx_skbs[i] = NULL; 212 return skb; 213 } 214 215 static grant_ref_t xennet_get_rx_ref(struct netfront_queue *queue, 216 RING_IDX ri) 217 { 218 int i = xennet_rxidx(ri); 219 grant_ref_t ref = queue->grant_rx_ref[i]; 220 queue->grant_rx_ref[i] = GRANT_INVALID_REF; 221 return ref; 222 } 223 224 #ifdef CONFIG_SYSFS 225 static const struct attribute_group xennet_dev_group; 226 #endif 227 228 static bool xennet_can_sg(struct net_device *dev) 229 { 230 return dev->features & NETIF_F_SG; 231 } 232 233 234 static void rx_refill_timeout(struct timer_list *t) 235 { 236 struct netfront_queue *queue = from_timer(queue, t, rx_refill_timer); 237 napi_schedule(&queue->napi); 238 } 239 240 static int netfront_tx_slot_available(struct netfront_queue *queue) 241 { 242 return (queue->tx.req_prod_pvt - queue->tx.rsp_cons) < 243 (NET_TX_RING_SIZE - XEN_NETIF_NR_SLOTS_MIN - 1); 244 } 245 246 static void xennet_maybe_wake_tx(struct netfront_queue *queue) 247 { 248 struct net_device *dev = queue->info->netdev; 249 struct netdev_queue *dev_queue = netdev_get_tx_queue(dev, queue->id); 250 251 if (unlikely(netif_tx_queue_stopped(dev_queue)) && 252 netfront_tx_slot_available(queue) && 253 likely(netif_running(dev))) 254 netif_tx_wake_queue(netdev_get_tx_queue(dev, queue->id)); 255 } 256 257 258 static struct sk_buff *xennet_alloc_one_rx_buffer(struct netfront_queue *queue) 259 { 260 struct sk_buff *skb; 261 struct page *page; 262 263 skb = __netdev_alloc_skb(queue->info->netdev, 264 RX_COPY_THRESHOLD + NET_IP_ALIGN, 265 GFP_ATOMIC | __GFP_NOWARN); 266 if (unlikely(!skb)) 267 return NULL; 268 269 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN); 270 if (!page) { 271 kfree_skb(skb); 272 return NULL; 273 } 274 skb_add_rx_frag(skb, 0, page, 0, 0, PAGE_SIZE); 275 276 /* Align ip header to a 16 bytes boundary */ 277 skb_reserve(skb, NET_IP_ALIGN); 278 skb->dev = queue->info->netdev; 279 280 return skb; 281 } 282 283 284 static void xennet_alloc_rx_buffers(struct netfront_queue *queue) 285 { 286 RING_IDX req_prod = queue->rx.req_prod_pvt; 287 int notify; 288 int err = 0; 289 290 if (unlikely(!netif_carrier_ok(queue->info->netdev))) 291 return; 292 293 for (req_prod = queue->rx.req_prod_pvt; 294 req_prod - queue->rx.rsp_cons < NET_RX_RING_SIZE; 295 req_prod++) { 296 struct sk_buff *skb; 297 unsigned short id; 298 grant_ref_t ref; 299 struct page *page; 300 struct xen_netif_rx_request *req; 301 302 skb = xennet_alloc_one_rx_buffer(queue); 303 if (!skb) { 304 err = -ENOMEM; 305 break; 306 } 307 308 id = xennet_rxidx(req_prod); 309 310 BUG_ON(queue->rx_skbs[id]); 311 queue->rx_skbs[id] = skb; 312 313 ref = gnttab_claim_grant_reference(&queue->gref_rx_head); 314 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); 315 queue->grant_rx_ref[id] = ref; 316 317 page = skb_frag_page(&skb_shinfo(skb)->frags[0]); 318 319 req = RING_GET_REQUEST(&queue->rx, req_prod); 320 gnttab_page_grant_foreign_access_ref_one(ref, 321 queue->info->xbdev->otherend_id, 322 page, 323 0); 324 req->id = id; 325 req->gref = ref; 326 } 327 328 queue->rx.req_prod_pvt = req_prod; 329 330 /* Try again later if there are not enough requests or skb allocation 331 * failed. 332 * Enough requests is quantified as the sum of newly created slots and 333 * the unconsumed slots at the backend. 334 */ 335 if (req_prod - queue->rx.rsp_cons < NET_RX_SLOTS_MIN || 336 unlikely(err)) { 337 mod_timer(&queue->rx_refill_timer, jiffies + (HZ/10)); 338 return; 339 } 340 341 wmb(); /* barrier so backend seens requests */ 342 343 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->rx, notify); 344 if (notify) 345 notify_remote_via_irq(queue->rx_irq); 346 } 347 348 static int xennet_open(struct net_device *dev) 349 { 350 struct netfront_info *np = netdev_priv(dev); 351 unsigned int num_queues = dev->real_num_tx_queues; 352 unsigned int i = 0; 353 struct netfront_queue *queue = NULL; 354 355 if (!np->queues) 356 return -ENODEV; 357 358 for (i = 0; i < num_queues; ++i) { 359 queue = &np->queues[i]; 360 napi_enable(&queue->napi); 361 362 spin_lock_bh(&queue->rx_lock); 363 if (netif_carrier_ok(dev)) { 364 xennet_alloc_rx_buffers(queue); 365 queue->rx.sring->rsp_event = queue->rx.rsp_cons + 1; 366 if (RING_HAS_UNCONSUMED_RESPONSES(&queue->rx)) 367 napi_schedule(&queue->napi); 368 } 369 spin_unlock_bh(&queue->rx_lock); 370 } 371 372 netif_tx_start_all_queues(dev); 373 374 return 0; 375 } 376 377 static void xennet_tx_buf_gc(struct netfront_queue *queue) 378 { 379 RING_IDX cons, prod; 380 unsigned short id; 381 struct sk_buff *skb; 382 bool more_to_do; 383 384 BUG_ON(!netif_carrier_ok(queue->info->netdev)); 385 386 do { 387 prod = queue->tx.sring->rsp_prod; 388 rmb(); /* Ensure we see responses up to 'rp'. */ 389 390 for (cons = queue->tx.rsp_cons; cons != prod; cons++) { 391 struct xen_netif_tx_response *txrsp; 392 393 txrsp = RING_GET_RESPONSE(&queue->tx, cons); 394 if (txrsp->status == XEN_NETIF_RSP_NULL) 395 continue; 396 397 id = txrsp->id; 398 skb = queue->tx_skbs[id].skb; 399 if (unlikely(gnttab_query_foreign_access( 400 queue->grant_tx_ref[id]) != 0)) { 401 pr_alert("%s: warning -- grant still in use by backend domain\n", 402 __func__); 403 BUG(); 404 } 405 gnttab_end_foreign_access_ref( 406 queue->grant_tx_ref[id], GNTMAP_readonly); 407 gnttab_release_grant_reference( 408 &queue->gref_tx_head, queue->grant_tx_ref[id]); 409 queue->grant_tx_ref[id] = GRANT_INVALID_REF; 410 queue->grant_tx_page[id] = NULL; 411 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, id); 412 dev_kfree_skb_irq(skb); 413 } 414 415 queue->tx.rsp_cons = prod; 416 417 RING_FINAL_CHECK_FOR_RESPONSES(&queue->tx, more_to_do); 418 } while (more_to_do); 419 420 xennet_maybe_wake_tx(queue); 421 } 422 423 struct xennet_gnttab_make_txreq { 424 struct netfront_queue *queue; 425 struct sk_buff *skb; 426 struct page *page; 427 struct xen_netif_tx_request *tx; /* Last request */ 428 unsigned int size; 429 }; 430 431 static void xennet_tx_setup_grant(unsigned long gfn, unsigned int offset, 432 unsigned int len, void *data) 433 { 434 struct xennet_gnttab_make_txreq *info = data; 435 unsigned int id; 436 struct xen_netif_tx_request *tx; 437 grant_ref_t ref; 438 /* convenient aliases */ 439 struct page *page = info->page; 440 struct netfront_queue *queue = info->queue; 441 struct sk_buff *skb = info->skb; 442 443 id = get_id_from_freelist(&queue->tx_skb_freelist, queue->tx_skbs); 444 tx = RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); 445 ref = gnttab_claim_grant_reference(&queue->gref_tx_head); 446 WARN_ON_ONCE(IS_ERR_VALUE((unsigned long)(int)ref)); 447 448 gnttab_grant_foreign_access_ref(ref, queue->info->xbdev->otherend_id, 449 gfn, GNTMAP_readonly); 450 451 queue->tx_skbs[id].skb = skb; 452 queue->grant_tx_page[id] = page; 453 queue->grant_tx_ref[id] = ref; 454 455 tx->id = id; 456 tx->gref = ref; 457 tx->offset = offset; 458 tx->size = len; 459 tx->flags = 0; 460 461 info->tx = tx; 462 info->size += tx->size; 463 } 464 465 static struct xen_netif_tx_request *xennet_make_first_txreq( 466 struct netfront_queue *queue, struct sk_buff *skb, 467 struct page *page, unsigned int offset, unsigned int len) 468 { 469 struct xennet_gnttab_make_txreq info = { 470 .queue = queue, 471 .skb = skb, 472 .page = page, 473 .size = 0, 474 }; 475 476 gnttab_for_one_grant(page, offset, len, xennet_tx_setup_grant, &info); 477 478 return info.tx; 479 } 480 481 static void xennet_make_one_txreq(unsigned long gfn, unsigned int offset, 482 unsigned int len, void *data) 483 { 484 struct xennet_gnttab_make_txreq *info = data; 485 486 info->tx->flags |= XEN_NETTXF_more_data; 487 skb_get(info->skb); 488 xennet_tx_setup_grant(gfn, offset, len, data); 489 } 490 491 static struct xen_netif_tx_request *xennet_make_txreqs( 492 struct netfront_queue *queue, struct xen_netif_tx_request *tx, 493 struct sk_buff *skb, struct page *page, 494 unsigned int offset, unsigned int len) 495 { 496 struct xennet_gnttab_make_txreq info = { 497 .queue = queue, 498 .skb = skb, 499 .tx = tx, 500 }; 501 502 /* Skip unused frames from start of page */ 503 page += offset >> PAGE_SHIFT; 504 offset &= ~PAGE_MASK; 505 506 while (len) { 507 info.page = page; 508 info.size = 0; 509 510 gnttab_foreach_grant_in_range(page, offset, len, 511 xennet_make_one_txreq, 512 &info); 513 514 page++; 515 offset = 0; 516 len -= info.size; 517 } 518 519 return info.tx; 520 } 521 522 /* 523 * Count how many ring slots are required to send this skb. Each frag 524 * might be a compound page. 525 */ 526 static int xennet_count_skb_slots(struct sk_buff *skb) 527 { 528 int i, frags = skb_shinfo(skb)->nr_frags; 529 int slots; 530 531 slots = gnttab_count_grant(offset_in_page(skb->data), 532 skb_headlen(skb)); 533 534 for (i = 0; i < frags; i++) { 535 skb_frag_t *frag = skb_shinfo(skb)->frags + i; 536 unsigned long size = skb_frag_size(frag); 537 unsigned long offset = frag->page_offset; 538 539 /* Skip unused frames from start of page */ 540 offset &= ~PAGE_MASK; 541 542 slots += gnttab_count_grant(offset, size); 543 } 544 545 return slots; 546 } 547 548 static u16 xennet_select_queue(struct net_device *dev, struct sk_buff *skb, 549 struct net_device *sb_dev, 550 select_queue_fallback_t fallback) 551 { 552 unsigned int num_queues = dev->real_num_tx_queues; 553 u32 hash; 554 u16 queue_idx; 555 556 /* First, check if there is only one queue */ 557 if (num_queues == 1) { 558 queue_idx = 0; 559 } else { 560 hash = skb_get_hash(skb); 561 queue_idx = hash % num_queues; 562 } 563 564 return queue_idx; 565 } 566 567 #define MAX_XEN_SKB_FRAGS (65536 / XEN_PAGE_SIZE + 1) 568 569 static netdev_tx_t xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) 570 { 571 struct netfront_info *np = netdev_priv(dev); 572 struct netfront_stats *tx_stats = this_cpu_ptr(np->tx_stats); 573 struct xen_netif_tx_request *tx, *first_tx; 574 unsigned int i; 575 int notify; 576 int slots; 577 struct page *page; 578 unsigned int offset; 579 unsigned int len; 580 unsigned long flags; 581 struct netfront_queue *queue = NULL; 582 unsigned int num_queues = dev->real_num_tx_queues; 583 u16 queue_index; 584 struct sk_buff *nskb; 585 586 /* Drop the packet if no queues are set up */ 587 if (num_queues < 1) 588 goto drop; 589 /* Determine which queue to transmit this SKB on */ 590 queue_index = skb_get_queue_mapping(skb); 591 queue = &np->queues[queue_index]; 592 593 /* If skb->len is too big for wire format, drop skb and alert 594 * user about misconfiguration. 595 */ 596 if (unlikely(skb->len > XEN_NETIF_MAX_TX_SIZE)) { 597 net_alert_ratelimited( 598 "xennet: skb->len = %u, too big for wire format\n", 599 skb->len); 600 goto drop; 601 } 602 603 slots = xennet_count_skb_slots(skb); 604 if (unlikely(slots > MAX_XEN_SKB_FRAGS + 1)) { 605 net_dbg_ratelimited("xennet: skb rides the rocket: %d slots, %d bytes\n", 606 slots, skb->len); 607 if (skb_linearize(skb)) 608 goto drop; 609 } 610 611 page = virt_to_page(skb->data); 612 offset = offset_in_page(skb->data); 613 614 /* The first req should be at least ETH_HLEN size or the packet will be 615 * dropped by netback. 616 */ 617 if (unlikely(PAGE_SIZE - offset < ETH_HLEN)) { 618 nskb = skb_copy(skb, GFP_ATOMIC); 619 if (!nskb) 620 goto drop; 621 dev_consume_skb_any(skb); 622 skb = nskb; 623 page = virt_to_page(skb->data); 624 offset = offset_in_page(skb->data); 625 } 626 627 len = skb_headlen(skb); 628 629 spin_lock_irqsave(&queue->tx_lock, flags); 630 631 if (unlikely(!netif_carrier_ok(dev) || 632 (slots > 1 && !xennet_can_sg(dev)) || 633 netif_needs_gso(skb, netif_skb_features(skb)))) { 634 spin_unlock_irqrestore(&queue->tx_lock, flags); 635 goto drop; 636 } 637 638 /* First request for the linear area. */ 639 first_tx = tx = xennet_make_first_txreq(queue, skb, 640 page, offset, len); 641 offset += tx->size; 642 if (offset == PAGE_SIZE) { 643 page++; 644 offset = 0; 645 } 646 len -= tx->size; 647 648 if (skb->ip_summed == CHECKSUM_PARTIAL) 649 /* local packet? */ 650 tx->flags |= XEN_NETTXF_csum_blank | XEN_NETTXF_data_validated; 651 else if (skb->ip_summed == CHECKSUM_UNNECESSARY) 652 /* remote but checksummed. */ 653 tx->flags |= XEN_NETTXF_data_validated; 654 655 /* Optional extra info after the first request. */ 656 if (skb_shinfo(skb)->gso_size) { 657 struct xen_netif_extra_info *gso; 658 659 gso = (struct xen_netif_extra_info *) 660 RING_GET_REQUEST(&queue->tx, queue->tx.req_prod_pvt++); 661 662 tx->flags |= XEN_NETTXF_extra_info; 663 664 gso->u.gso.size = skb_shinfo(skb)->gso_size; 665 gso->u.gso.type = (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6) ? 666 XEN_NETIF_GSO_TYPE_TCPV6 : 667 XEN_NETIF_GSO_TYPE_TCPV4; 668 gso->u.gso.pad = 0; 669 gso->u.gso.features = 0; 670 671 gso->type = XEN_NETIF_EXTRA_TYPE_GSO; 672 gso->flags = 0; 673 } 674 675 /* Requests for the rest of the linear area. */ 676 tx = xennet_make_txreqs(queue, tx, skb, page, offset, len); 677 678 /* Requests for all the frags. */ 679 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 680 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 681 tx = xennet_make_txreqs(queue, tx, skb, 682 skb_frag_page(frag), frag->page_offset, 683 skb_frag_size(frag)); 684 } 685 686 /* First request has the packet length. */ 687 first_tx->size = skb->len; 688 689 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&queue->tx, notify); 690 if (notify) 691 notify_remote_via_irq(queue->tx_irq); 692 693 u64_stats_update_begin(&tx_stats->syncp); 694 tx_stats->bytes += skb->len; 695 tx_stats->packets++; 696 u64_stats_update_end(&tx_stats->syncp); 697 698 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ 699 xennet_tx_buf_gc(queue); 700 701 if (!netfront_tx_slot_available(queue)) 702 netif_tx_stop_queue(netdev_get_tx_queue(dev, queue->id)); 703 704 spin_unlock_irqrestore(&queue->tx_lock, flags); 705 706 return NETDEV_TX_OK; 707 708 drop: 709 dev->stats.tx_dropped++; 710 dev_kfree_skb_any(skb); 711 return NETDEV_TX_OK; 712 } 713 714 static int xennet_close(struct net_device *dev) 715 { 716 struct netfront_info *np = netdev_priv(dev); 717 unsigned int num_queues = dev->real_num_tx_queues; 718 unsigned int i; 719 struct netfront_queue *queue; 720 netif_tx_stop_all_queues(np->netdev); 721 for (i = 0; i < num_queues; ++i) { 722 queue = &np->queues[i]; 723 napi_disable(&queue->napi); 724 } 725 return 0; 726 } 727 728 static void xennet_move_rx_slot(struct netfront_queue *queue, struct sk_buff *skb, 729 grant_ref_t ref) 730 { 731 int new = xennet_rxidx(queue->rx.req_prod_pvt); 732 733 BUG_ON(queue->rx_skbs[new]); 734 queue->rx_skbs[new] = skb; 735 queue->grant_rx_ref[new] = ref; 736 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->id = new; 737 RING_GET_REQUEST(&queue->rx, queue->rx.req_prod_pvt)->gref = ref; 738 queue->rx.req_prod_pvt++; 739 } 740 741 static int xennet_get_extras(struct netfront_queue *queue, 742 struct xen_netif_extra_info *extras, 743 RING_IDX rp) 744 745 { 746 struct xen_netif_extra_info *extra; 747 struct device *dev = &queue->info->netdev->dev; 748 RING_IDX cons = queue->rx.rsp_cons; 749 int err = 0; 750 751 do { 752 struct sk_buff *skb; 753 grant_ref_t ref; 754 755 if (unlikely(cons + 1 == rp)) { 756 if (net_ratelimit()) 757 dev_warn(dev, "Missing extra info\n"); 758 err = -EBADR; 759 break; 760 } 761 762 extra = (struct xen_netif_extra_info *) 763 RING_GET_RESPONSE(&queue->rx, ++cons); 764 765 if (unlikely(!extra->type || 766 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { 767 if (net_ratelimit()) 768 dev_warn(dev, "Invalid extra type: %d\n", 769 extra->type); 770 err = -EINVAL; 771 } else { 772 memcpy(&extras[extra->type - 1], extra, 773 sizeof(*extra)); 774 } 775 776 skb = xennet_get_rx_skb(queue, cons); 777 ref = xennet_get_rx_ref(queue, cons); 778 xennet_move_rx_slot(queue, skb, ref); 779 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); 780 781 queue->rx.rsp_cons = cons; 782 return err; 783 } 784 785 static int xennet_get_responses(struct netfront_queue *queue, 786 struct netfront_rx_info *rinfo, RING_IDX rp, 787 struct sk_buff_head *list) 788 { 789 struct xen_netif_rx_response *rx = &rinfo->rx; 790 struct xen_netif_extra_info *extras = rinfo->extras; 791 struct device *dev = &queue->info->netdev->dev; 792 RING_IDX cons = queue->rx.rsp_cons; 793 struct sk_buff *skb = xennet_get_rx_skb(queue, cons); 794 grant_ref_t ref = xennet_get_rx_ref(queue, cons); 795 int max = XEN_NETIF_NR_SLOTS_MIN + (rx->status <= RX_COPY_THRESHOLD); 796 int slots = 1; 797 int err = 0; 798 unsigned long ret; 799 800 if (rx->flags & XEN_NETRXF_extra_info) { 801 err = xennet_get_extras(queue, extras, rp); 802 cons = queue->rx.rsp_cons; 803 } 804 805 for (;;) { 806 if (unlikely(rx->status < 0 || 807 rx->offset + rx->status > XEN_PAGE_SIZE)) { 808 if (net_ratelimit()) 809 dev_warn(dev, "rx->offset: %u, size: %d\n", 810 rx->offset, rx->status); 811 xennet_move_rx_slot(queue, skb, ref); 812 err = -EINVAL; 813 goto next; 814 } 815 816 /* 817 * This definitely indicates a bug, either in this driver or in 818 * the backend driver. In future this should flag the bad 819 * situation to the system controller to reboot the backend. 820 */ 821 if (ref == GRANT_INVALID_REF) { 822 if (net_ratelimit()) 823 dev_warn(dev, "Bad rx response id %d.\n", 824 rx->id); 825 err = -EINVAL; 826 goto next; 827 } 828 829 ret = gnttab_end_foreign_access_ref(ref, 0); 830 BUG_ON(!ret); 831 832 gnttab_release_grant_reference(&queue->gref_rx_head, ref); 833 834 __skb_queue_tail(list, skb); 835 836 next: 837 if (!(rx->flags & XEN_NETRXF_more_data)) 838 break; 839 840 if (cons + slots == rp) { 841 if (net_ratelimit()) 842 dev_warn(dev, "Need more slots\n"); 843 err = -ENOENT; 844 break; 845 } 846 847 rx = RING_GET_RESPONSE(&queue->rx, cons + slots); 848 skb = xennet_get_rx_skb(queue, cons + slots); 849 ref = xennet_get_rx_ref(queue, cons + slots); 850 slots++; 851 } 852 853 if (unlikely(slots > max)) { 854 if (net_ratelimit()) 855 dev_warn(dev, "Too many slots\n"); 856 err = -E2BIG; 857 } 858 859 if (unlikely(err)) 860 queue->rx.rsp_cons = cons + slots; 861 862 return err; 863 } 864 865 static int xennet_set_skb_gso(struct sk_buff *skb, 866 struct xen_netif_extra_info *gso) 867 { 868 if (!gso->u.gso.size) { 869 if (net_ratelimit()) 870 pr_warn("GSO size must not be zero\n"); 871 return -EINVAL; 872 } 873 874 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4 && 875 gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV6) { 876 if (net_ratelimit()) 877 pr_warn("Bad GSO type %d\n", gso->u.gso.type); 878 return -EINVAL; 879 } 880 881 skb_shinfo(skb)->gso_size = gso->u.gso.size; 882 skb_shinfo(skb)->gso_type = 883 (gso->u.gso.type == XEN_NETIF_GSO_TYPE_TCPV4) ? 884 SKB_GSO_TCPV4 : 885 SKB_GSO_TCPV6; 886 887 /* Header must be checked, and gso_segs computed. */ 888 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 889 skb_shinfo(skb)->gso_segs = 0; 890 891 return 0; 892 } 893 894 static RING_IDX xennet_fill_frags(struct netfront_queue *queue, 895 struct sk_buff *skb, 896 struct sk_buff_head *list) 897 { 898 RING_IDX cons = queue->rx.rsp_cons; 899 struct sk_buff *nskb; 900 901 while ((nskb = __skb_dequeue(list))) { 902 struct xen_netif_rx_response *rx = 903 RING_GET_RESPONSE(&queue->rx, ++cons); 904 skb_frag_t *nfrag = &skb_shinfo(nskb)->frags[0]; 905 906 if (skb_shinfo(skb)->nr_frags == MAX_SKB_FRAGS) { 907 unsigned int pull_to = NETFRONT_SKB_CB(skb)->pull_to; 908 909 BUG_ON(pull_to <= skb_headlen(skb)); 910 __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); 911 } 912 BUG_ON(skb_shinfo(skb)->nr_frags >= MAX_SKB_FRAGS); 913 914 skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, 915 skb_frag_page(nfrag), 916 rx->offset, rx->status, PAGE_SIZE); 917 918 skb_shinfo(nskb)->nr_frags = 0; 919 kfree_skb(nskb); 920 } 921 922 return cons; 923 } 924 925 static int checksum_setup(struct net_device *dev, struct sk_buff *skb) 926 { 927 bool recalculate_partial_csum = false; 928 929 /* 930 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy 931 * peers can fail to set NETRXF_csum_blank when sending a GSO 932 * frame. In this case force the SKB to CHECKSUM_PARTIAL and 933 * recalculate the partial checksum. 934 */ 935 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { 936 struct netfront_info *np = netdev_priv(dev); 937 atomic_inc(&np->rx_gso_checksum_fixup); 938 skb->ip_summed = CHECKSUM_PARTIAL; 939 recalculate_partial_csum = true; 940 } 941 942 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ 943 if (skb->ip_summed != CHECKSUM_PARTIAL) 944 return 0; 945 946 return skb_checksum_setup(skb, recalculate_partial_csum); 947 } 948 949 static int handle_incoming_queue(struct netfront_queue *queue, 950 struct sk_buff_head *rxq) 951 { 952 struct netfront_stats *rx_stats = this_cpu_ptr(queue->info->rx_stats); 953 int packets_dropped = 0; 954 struct sk_buff *skb; 955 956 while ((skb = __skb_dequeue(rxq)) != NULL) { 957 int pull_to = NETFRONT_SKB_CB(skb)->pull_to; 958 959 if (pull_to > skb_headlen(skb)) 960 __pskb_pull_tail(skb, pull_to - skb_headlen(skb)); 961 962 /* Ethernet work: Delayed to here as it peeks the header. */ 963 skb->protocol = eth_type_trans(skb, queue->info->netdev); 964 skb_reset_network_header(skb); 965 966 if (checksum_setup(queue->info->netdev, skb)) { 967 kfree_skb(skb); 968 packets_dropped++; 969 queue->info->netdev->stats.rx_errors++; 970 continue; 971 } 972 973 u64_stats_update_begin(&rx_stats->syncp); 974 rx_stats->packets++; 975 rx_stats->bytes += skb->len; 976 u64_stats_update_end(&rx_stats->syncp); 977 978 /* Pass it up. */ 979 napi_gro_receive(&queue->napi, skb); 980 } 981 982 return packets_dropped; 983 } 984 985 static int xennet_poll(struct napi_struct *napi, int budget) 986 { 987 struct netfront_queue *queue = container_of(napi, struct netfront_queue, napi); 988 struct net_device *dev = queue->info->netdev; 989 struct sk_buff *skb; 990 struct netfront_rx_info rinfo; 991 struct xen_netif_rx_response *rx = &rinfo.rx; 992 struct xen_netif_extra_info *extras = rinfo.extras; 993 RING_IDX i, rp; 994 int work_done; 995 struct sk_buff_head rxq; 996 struct sk_buff_head errq; 997 struct sk_buff_head tmpq; 998 int err; 999 1000 spin_lock(&queue->rx_lock); 1001 1002 skb_queue_head_init(&rxq); 1003 skb_queue_head_init(&errq); 1004 skb_queue_head_init(&tmpq); 1005 1006 rp = queue->rx.sring->rsp_prod; 1007 rmb(); /* Ensure we see queued responses up to 'rp'. */ 1008 1009 i = queue->rx.rsp_cons; 1010 work_done = 0; 1011 while ((i != rp) && (work_done < budget)) { 1012 memcpy(rx, RING_GET_RESPONSE(&queue->rx, i), sizeof(*rx)); 1013 memset(extras, 0, sizeof(rinfo.extras)); 1014 1015 err = xennet_get_responses(queue, &rinfo, rp, &tmpq); 1016 1017 if (unlikely(err)) { 1018 err: 1019 while ((skb = __skb_dequeue(&tmpq))) 1020 __skb_queue_tail(&errq, skb); 1021 dev->stats.rx_errors++; 1022 i = queue->rx.rsp_cons; 1023 continue; 1024 } 1025 1026 skb = __skb_dequeue(&tmpq); 1027 1028 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { 1029 struct xen_netif_extra_info *gso; 1030 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; 1031 1032 if (unlikely(xennet_set_skb_gso(skb, gso))) { 1033 __skb_queue_head(&tmpq, skb); 1034 queue->rx.rsp_cons += skb_queue_len(&tmpq); 1035 goto err; 1036 } 1037 } 1038 1039 NETFRONT_SKB_CB(skb)->pull_to = rx->status; 1040 if (NETFRONT_SKB_CB(skb)->pull_to > RX_COPY_THRESHOLD) 1041 NETFRONT_SKB_CB(skb)->pull_to = RX_COPY_THRESHOLD; 1042 1043 skb_shinfo(skb)->frags[0].page_offset = rx->offset; 1044 skb_frag_size_set(&skb_shinfo(skb)->frags[0], rx->status); 1045 skb->data_len = rx->status; 1046 skb->len += rx->status; 1047 1048 i = xennet_fill_frags(queue, skb, &tmpq); 1049 1050 if (rx->flags & XEN_NETRXF_csum_blank) 1051 skb->ip_summed = CHECKSUM_PARTIAL; 1052 else if (rx->flags & XEN_NETRXF_data_validated) 1053 skb->ip_summed = CHECKSUM_UNNECESSARY; 1054 1055 __skb_queue_tail(&rxq, skb); 1056 1057 queue->rx.rsp_cons = ++i; 1058 work_done++; 1059 } 1060 1061 __skb_queue_purge(&errq); 1062 1063 work_done -= handle_incoming_queue(queue, &rxq); 1064 1065 xennet_alloc_rx_buffers(queue); 1066 1067 if (work_done < budget) { 1068 int more_to_do = 0; 1069 1070 napi_complete_done(napi, work_done); 1071 1072 RING_FINAL_CHECK_FOR_RESPONSES(&queue->rx, more_to_do); 1073 if (more_to_do) 1074 napi_schedule(napi); 1075 } 1076 1077 spin_unlock(&queue->rx_lock); 1078 1079 return work_done; 1080 } 1081 1082 static int xennet_change_mtu(struct net_device *dev, int mtu) 1083 { 1084 int max = xennet_can_sg(dev) ? XEN_NETIF_MAX_TX_SIZE : ETH_DATA_LEN; 1085 1086 if (mtu > max) 1087 return -EINVAL; 1088 dev->mtu = mtu; 1089 return 0; 1090 } 1091 1092 static void xennet_get_stats64(struct net_device *dev, 1093 struct rtnl_link_stats64 *tot) 1094 { 1095 struct netfront_info *np = netdev_priv(dev); 1096 int cpu; 1097 1098 for_each_possible_cpu(cpu) { 1099 struct netfront_stats *rx_stats = per_cpu_ptr(np->rx_stats, cpu); 1100 struct netfront_stats *tx_stats = per_cpu_ptr(np->tx_stats, cpu); 1101 u64 rx_packets, rx_bytes, tx_packets, tx_bytes; 1102 unsigned int start; 1103 1104 do { 1105 start = u64_stats_fetch_begin_irq(&tx_stats->syncp); 1106 tx_packets = tx_stats->packets; 1107 tx_bytes = tx_stats->bytes; 1108 } while (u64_stats_fetch_retry_irq(&tx_stats->syncp, start)); 1109 1110 do { 1111 start = u64_stats_fetch_begin_irq(&rx_stats->syncp); 1112 rx_packets = rx_stats->packets; 1113 rx_bytes = rx_stats->bytes; 1114 } while (u64_stats_fetch_retry_irq(&rx_stats->syncp, start)); 1115 1116 tot->rx_packets += rx_packets; 1117 tot->tx_packets += tx_packets; 1118 tot->rx_bytes += rx_bytes; 1119 tot->tx_bytes += tx_bytes; 1120 } 1121 1122 tot->rx_errors = dev->stats.rx_errors; 1123 tot->tx_dropped = dev->stats.tx_dropped; 1124 } 1125 1126 static void xennet_release_tx_bufs(struct netfront_queue *queue) 1127 { 1128 struct sk_buff *skb; 1129 int i; 1130 1131 for (i = 0; i < NET_TX_RING_SIZE; i++) { 1132 /* Skip over entries which are actually freelist references */ 1133 if (skb_entry_is_link(&queue->tx_skbs[i])) 1134 continue; 1135 1136 skb = queue->tx_skbs[i].skb; 1137 get_page(queue->grant_tx_page[i]); 1138 gnttab_end_foreign_access(queue->grant_tx_ref[i], 1139 GNTMAP_readonly, 1140 (unsigned long)page_address(queue->grant_tx_page[i])); 1141 queue->grant_tx_page[i] = NULL; 1142 queue->grant_tx_ref[i] = GRANT_INVALID_REF; 1143 add_id_to_freelist(&queue->tx_skb_freelist, queue->tx_skbs, i); 1144 dev_kfree_skb_irq(skb); 1145 } 1146 } 1147 1148 static void xennet_release_rx_bufs(struct netfront_queue *queue) 1149 { 1150 int id, ref; 1151 1152 spin_lock_bh(&queue->rx_lock); 1153 1154 for (id = 0; id < NET_RX_RING_SIZE; id++) { 1155 struct sk_buff *skb; 1156 struct page *page; 1157 1158 skb = queue->rx_skbs[id]; 1159 if (!skb) 1160 continue; 1161 1162 ref = queue->grant_rx_ref[id]; 1163 if (ref == GRANT_INVALID_REF) 1164 continue; 1165 1166 page = skb_frag_page(&skb_shinfo(skb)->frags[0]); 1167 1168 /* gnttab_end_foreign_access() needs a page ref until 1169 * foreign access is ended (which may be deferred). 1170 */ 1171 get_page(page); 1172 gnttab_end_foreign_access(ref, 0, 1173 (unsigned long)page_address(page)); 1174 queue->grant_rx_ref[id] = GRANT_INVALID_REF; 1175 1176 kfree_skb(skb); 1177 } 1178 1179 spin_unlock_bh(&queue->rx_lock); 1180 } 1181 1182 static netdev_features_t xennet_fix_features(struct net_device *dev, 1183 netdev_features_t features) 1184 { 1185 struct netfront_info *np = netdev_priv(dev); 1186 1187 if (features & NETIF_F_SG && 1188 !xenbus_read_unsigned(np->xbdev->otherend, "feature-sg", 0)) 1189 features &= ~NETIF_F_SG; 1190 1191 if (features & NETIF_F_IPV6_CSUM && 1192 !xenbus_read_unsigned(np->xbdev->otherend, 1193 "feature-ipv6-csum-offload", 0)) 1194 features &= ~NETIF_F_IPV6_CSUM; 1195 1196 if (features & NETIF_F_TSO && 1197 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv4", 0)) 1198 features &= ~NETIF_F_TSO; 1199 1200 if (features & NETIF_F_TSO6 && 1201 !xenbus_read_unsigned(np->xbdev->otherend, "feature-gso-tcpv6", 0)) 1202 features &= ~NETIF_F_TSO6; 1203 1204 return features; 1205 } 1206 1207 static int xennet_set_features(struct net_device *dev, 1208 netdev_features_t features) 1209 { 1210 if (!(features & NETIF_F_SG) && dev->mtu > ETH_DATA_LEN) { 1211 netdev_info(dev, "Reducing MTU because no SG offload"); 1212 dev->mtu = ETH_DATA_LEN; 1213 } 1214 1215 return 0; 1216 } 1217 1218 static irqreturn_t xennet_tx_interrupt(int irq, void *dev_id) 1219 { 1220 struct netfront_queue *queue = dev_id; 1221 unsigned long flags; 1222 1223 spin_lock_irqsave(&queue->tx_lock, flags); 1224 xennet_tx_buf_gc(queue); 1225 spin_unlock_irqrestore(&queue->tx_lock, flags); 1226 1227 return IRQ_HANDLED; 1228 } 1229 1230 static irqreturn_t xennet_rx_interrupt(int irq, void *dev_id) 1231 { 1232 struct netfront_queue *queue = dev_id; 1233 struct net_device *dev = queue->info->netdev; 1234 1235 if (likely(netif_carrier_ok(dev) && 1236 RING_HAS_UNCONSUMED_RESPONSES(&queue->rx))) 1237 napi_schedule(&queue->napi); 1238 1239 return IRQ_HANDLED; 1240 } 1241 1242 static irqreturn_t xennet_interrupt(int irq, void *dev_id) 1243 { 1244 xennet_tx_interrupt(irq, dev_id); 1245 xennet_rx_interrupt(irq, dev_id); 1246 return IRQ_HANDLED; 1247 } 1248 1249 #ifdef CONFIG_NET_POLL_CONTROLLER 1250 static void xennet_poll_controller(struct net_device *dev) 1251 { 1252 /* Poll each queue */ 1253 struct netfront_info *info = netdev_priv(dev); 1254 unsigned int num_queues = dev->real_num_tx_queues; 1255 unsigned int i; 1256 for (i = 0; i < num_queues; ++i) 1257 xennet_interrupt(0, &info->queues[i]); 1258 } 1259 #endif 1260 1261 static const struct net_device_ops xennet_netdev_ops = { 1262 .ndo_open = xennet_open, 1263 .ndo_stop = xennet_close, 1264 .ndo_start_xmit = xennet_start_xmit, 1265 .ndo_change_mtu = xennet_change_mtu, 1266 .ndo_get_stats64 = xennet_get_stats64, 1267 .ndo_set_mac_address = eth_mac_addr, 1268 .ndo_validate_addr = eth_validate_addr, 1269 .ndo_fix_features = xennet_fix_features, 1270 .ndo_set_features = xennet_set_features, 1271 .ndo_select_queue = xennet_select_queue, 1272 #ifdef CONFIG_NET_POLL_CONTROLLER 1273 .ndo_poll_controller = xennet_poll_controller, 1274 #endif 1275 }; 1276 1277 static void xennet_free_netdev(struct net_device *netdev) 1278 { 1279 struct netfront_info *np = netdev_priv(netdev); 1280 1281 free_percpu(np->rx_stats); 1282 free_percpu(np->tx_stats); 1283 free_netdev(netdev); 1284 } 1285 1286 static struct net_device *xennet_create_dev(struct xenbus_device *dev) 1287 { 1288 int err; 1289 struct net_device *netdev; 1290 struct netfront_info *np; 1291 1292 netdev = alloc_etherdev_mq(sizeof(struct netfront_info), xennet_max_queues); 1293 if (!netdev) 1294 return ERR_PTR(-ENOMEM); 1295 1296 np = netdev_priv(netdev); 1297 np->xbdev = dev; 1298 1299 np->queues = NULL; 1300 1301 err = -ENOMEM; 1302 np->rx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); 1303 if (np->rx_stats == NULL) 1304 goto exit; 1305 np->tx_stats = netdev_alloc_pcpu_stats(struct netfront_stats); 1306 if (np->tx_stats == NULL) 1307 goto exit; 1308 1309 netdev->netdev_ops = &xennet_netdev_ops; 1310 1311 netdev->features = NETIF_F_IP_CSUM | NETIF_F_RXCSUM | 1312 NETIF_F_GSO_ROBUST; 1313 netdev->hw_features = NETIF_F_SG | 1314 NETIF_F_IPV6_CSUM | 1315 NETIF_F_TSO | NETIF_F_TSO6; 1316 1317 /* 1318 * Assume that all hw features are available for now. This set 1319 * will be adjusted by the call to netdev_update_features() in 1320 * xennet_connect() which is the earliest point where we can 1321 * negotiate with the backend regarding supported features. 1322 */ 1323 netdev->features |= netdev->hw_features; 1324 1325 netdev->ethtool_ops = &xennet_ethtool_ops; 1326 netdev->min_mtu = ETH_MIN_MTU; 1327 netdev->max_mtu = XEN_NETIF_MAX_TX_SIZE; 1328 SET_NETDEV_DEV(netdev, &dev->dev); 1329 1330 np->netdev = netdev; 1331 1332 netif_carrier_off(netdev); 1333 1334 xenbus_switch_state(dev, XenbusStateInitialising); 1335 wait_event(module_load_q, 1336 xenbus_read_driver_state(dev->otherend) != 1337 XenbusStateClosed && 1338 xenbus_read_driver_state(dev->otherend) != 1339 XenbusStateUnknown); 1340 return netdev; 1341 1342 exit: 1343 xennet_free_netdev(netdev); 1344 return ERR_PTR(err); 1345 } 1346 1347 /** 1348 * Entry point to this code when a new device is created. Allocate the basic 1349 * structures and the ring buffers for communication with the backend, and 1350 * inform the backend of the appropriate details for those. 1351 */ 1352 static int netfront_probe(struct xenbus_device *dev, 1353 const struct xenbus_device_id *id) 1354 { 1355 int err; 1356 struct net_device *netdev; 1357 struct netfront_info *info; 1358 1359 netdev = xennet_create_dev(dev); 1360 if (IS_ERR(netdev)) { 1361 err = PTR_ERR(netdev); 1362 xenbus_dev_fatal(dev, err, "creating netdev"); 1363 return err; 1364 } 1365 1366 info = netdev_priv(netdev); 1367 dev_set_drvdata(&dev->dev, info); 1368 #ifdef CONFIG_SYSFS 1369 info->netdev->sysfs_groups[0] = &xennet_dev_group; 1370 #endif 1371 1372 return 0; 1373 } 1374 1375 static void xennet_end_access(int ref, void *page) 1376 { 1377 /* This frees the page as a side-effect */ 1378 if (ref != GRANT_INVALID_REF) 1379 gnttab_end_foreign_access(ref, 0, (unsigned long)page); 1380 } 1381 1382 static void xennet_disconnect_backend(struct netfront_info *info) 1383 { 1384 unsigned int i = 0; 1385 unsigned int num_queues = info->netdev->real_num_tx_queues; 1386 1387 netif_carrier_off(info->netdev); 1388 1389 for (i = 0; i < num_queues && info->queues; ++i) { 1390 struct netfront_queue *queue = &info->queues[i]; 1391 1392 del_timer_sync(&queue->rx_refill_timer); 1393 1394 if (queue->tx_irq && (queue->tx_irq == queue->rx_irq)) 1395 unbind_from_irqhandler(queue->tx_irq, queue); 1396 if (queue->tx_irq && (queue->tx_irq != queue->rx_irq)) { 1397 unbind_from_irqhandler(queue->tx_irq, queue); 1398 unbind_from_irqhandler(queue->rx_irq, queue); 1399 } 1400 queue->tx_evtchn = queue->rx_evtchn = 0; 1401 queue->tx_irq = queue->rx_irq = 0; 1402 1403 if (netif_running(info->netdev)) 1404 napi_synchronize(&queue->napi); 1405 1406 xennet_release_tx_bufs(queue); 1407 xennet_release_rx_bufs(queue); 1408 gnttab_free_grant_references(queue->gref_tx_head); 1409 gnttab_free_grant_references(queue->gref_rx_head); 1410 1411 /* End access and free the pages */ 1412 xennet_end_access(queue->tx_ring_ref, queue->tx.sring); 1413 xennet_end_access(queue->rx_ring_ref, queue->rx.sring); 1414 1415 queue->tx_ring_ref = GRANT_INVALID_REF; 1416 queue->rx_ring_ref = GRANT_INVALID_REF; 1417 queue->tx.sring = NULL; 1418 queue->rx.sring = NULL; 1419 } 1420 } 1421 1422 /** 1423 * We are reconnecting to the backend, due to a suspend/resume, or a backend 1424 * driver restart. We tear down our netif structure and recreate it, but 1425 * leave the device-layer structures intact so that this is transparent to the 1426 * rest of the kernel. 1427 */ 1428 static int netfront_resume(struct xenbus_device *dev) 1429 { 1430 struct netfront_info *info = dev_get_drvdata(&dev->dev); 1431 1432 dev_dbg(&dev->dev, "%s\n", dev->nodename); 1433 1434 xennet_disconnect_backend(info); 1435 return 0; 1436 } 1437 1438 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) 1439 { 1440 char *s, *e, *macstr; 1441 int i; 1442 1443 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); 1444 if (IS_ERR(macstr)) 1445 return PTR_ERR(macstr); 1446 1447 for (i = 0; i < ETH_ALEN; i++) { 1448 mac[i] = simple_strtoul(s, &e, 16); 1449 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { 1450 kfree(macstr); 1451 return -ENOENT; 1452 } 1453 s = e+1; 1454 } 1455 1456 kfree(macstr); 1457 return 0; 1458 } 1459 1460 static int setup_netfront_single(struct netfront_queue *queue) 1461 { 1462 int err; 1463 1464 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); 1465 if (err < 0) 1466 goto fail; 1467 1468 err = bind_evtchn_to_irqhandler(queue->tx_evtchn, 1469 xennet_interrupt, 1470 0, queue->info->netdev->name, queue); 1471 if (err < 0) 1472 goto bind_fail; 1473 queue->rx_evtchn = queue->tx_evtchn; 1474 queue->rx_irq = queue->tx_irq = err; 1475 1476 return 0; 1477 1478 bind_fail: 1479 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); 1480 queue->tx_evtchn = 0; 1481 fail: 1482 return err; 1483 } 1484 1485 static int setup_netfront_split(struct netfront_queue *queue) 1486 { 1487 int err; 1488 1489 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->tx_evtchn); 1490 if (err < 0) 1491 goto fail; 1492 err = xenbus_alloc_evtchn(queue->info->xbdev, &queue->rx_evtchn); 1493 if (err < 0) 1494 goto alloc_rx_evtchn_fail; 1495 1496 snprintf(queue->tx_irq_name, sizeof(queue->tx_irq_name), 1497 "%s-tx", queue->name); 1498 err = bind_evtchn_to_irqhandler(queue->tx_evtchn, 1499 xennet_tx_interrupt, 1500 0, queue->tx_irq_name, queue); 1501 if (err < 0) 1502 goto bind_tx_fail; 1503 queue->tx_irq = err; 1504 1505 snprintf(queue->rx_irq_name, sizeof(queue->rx_irq_name), 1506 "%s-rx", queue->name); 1507 err = bind_evtchn_to_irqhandler(queue->rx_evtchn, 1508 xennet_rx_interrupt, 1509 0, queue->rx_irq_name, queue); 1510 if (err < 0) 1511 goto bind_rx_fail; 1512 queue->rx_irq = err; 1513 1514 return 0; 1515 1516 bind_rx_fail: 1517 unbind_from_irqhandler(queue->tx_irq, queue); 1518 queue->tx_irq = 0; 1519 bind_tx_fail: 1520 xenbus_free_evtchn(queue->info->xbdev, queue->rx_evtchn); 1521 queue->rx_evtchn = 0; 1522 alloc_rx_evtchn_fail: 1523 xenbus_free_evtchn(queue->info->xbdev, queue->tx_evtchn); 1524 queue->tx_evtchn = 0; 1525 fail: 1526 return err; 1527 } 1528 1529 static int setup_netfront(struct xenbus_device *dev, 1530 struct netfront_queue *queue, unsigned int feature_split_evtchn) 1531 { 1532 struct xen_netif_tx_sring *txs; 1533 struct xen_netif_rx_sring *rxs; 1534 grant_ref_t gref; 1535 int err; 1536 1537 queue->tx_ring_ref = GRANT_INVALID_REF; 1538 queue->rx_ring_ref = GRANT_INVALID_REF; 1539 queue->rx.sring = NULL; 1540 queue->tx.sring = NULL; 1541 1542 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); 1543 if (!txs) { 1544 err = -ENOMEM; 1545 xenbus_dev_fatal(dev, err, "allocating tx ring page"); 1546 goto fail; 1547 } 1548 SHARED_RING_INIT(txs); 1549 FRONT_RING_INIT(&queue->tx, txs, XEN_PAGE_SIZE); 1550 1551 err = xenbus_grant_ring(dev, txs, 1, &gref); 1552 if (err < 0) 1553 goto grant_tx_ring_fail; 1554 queue->tx_ring_ref = gref; 1555 1556 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); 1557 if (!rxs) { 1558 err = -ENOMEM; 1559 xenbus_dev_fatal(dev, err, "allocating rx ring page"); 1560 goto alloc_rx_ring_fail; 1561 } 1562 SHARED_RING_INIT(rxs); 1563 FRONT_RING_INIT(&queue->rx, rxs, XEN_PAGE_SIZE); 1564 1565 err = xenbus_grant_ring(dev, rxs, 1, &gref); 1566 if (err < 0) 1567 goto grant_rx_ring_fail; 1568 queue->rx_ring_ref = gref; 1569 1570 if (feature_split_evtchn) 1571 err = setup_netfront_split(queue); 1572 /* setup single event channel if 1573 * a) feature-split-event-channels == 0 1574 * b) feature-split-event-channels == 1 but failed to setup 1575 */ 1576 if (!feature_split_evtchn || (feature_split_evtchn && err)) 1577 err = setup_netfront_single(queue); 1578 1579 if (err) 1580 goto alloc_evtchn_fail; 1581 1582 return 0; 1583 1584 /* If we fail to setup netfront, it is safe to just revoke access to 1585 * granted pages because backend is not accessing it at this point. 1586 */ 1587 alloc_evtchn_fail: 1588 gnttab_end_foreign_access_ref(queue->rx_ring_ref, 0); 1589 grant_rx_ring_fail: 1590 free_page((unsigned long)rxs); 1591 alloc_rx_ring_fail: 1592 gnttab_end_foreign_access_ref(queue->tx_ring_ref, 0); 1593 grant_tx_ring_fail: 1594 free_page((unsigned long)txs); 1595 fail: 1596 return err; 1597 } 1598 1599 /* Queue-specific initialisation 1600 * This used to be done in xennet_create_dev() but must now 1601 * be run per-queue. 1602 */ 1603 static int xennet_init_queue(struct netfront_queue *queue) 1604 { 1605 unsigned short i; 1606 int err = 0; 1607 char *devid; 1608 1609 spin_lock_init(&queue->tx_lock); 1610 spin_lock_init(&queue->rx_lock); 1611 1612 timer_setup(&queue->rx_refill_timer, rx_refill_timeout, 0); 1613 1614 devid = strrchr(queue->info->xbdev->nodename, '/') + 1; 1615 snprintf(queue->name, sizeof(queue->name), "vif%s-q%u", 1616 devid, queue->id); 1617 1618 /* Initialise tx_skbs as a free chain containing every entry. */ 1619 queue->tx_skb_freelist = 0; 1620 for (i = 0; i < NET_TX_RING_SIZE; i++) { 1621 skb_entry_set_link(&queue->tx_skbs[i], i+1); 1622 queue->grant_tx_ref[i] = GRANT_INVALID_REF; 1623 queue->grant_tx_page[i] = NULL; 1624 } 1625 1626 /* Clear out rx_skbs */ 1627 for (i = 0; i < NET_RX_RING_SIZE; i++) { 1628 queue->rx_skbs[i] = NULL; 1629 queue->grant_rx_ref[i] = GRANT_INVALID_REF; 1630 } 1631 1632 /* A grant for every tx ring slot */ 1633 if (gnttab_alloc_grant_references(NET_TX_RING_SIZE, 1634 &queue->gref_tx_head) < 0) { 1635 pr_alert("can't alloc tx grant refs\n"); 1636 err = -ENOMEM; 1637 goto exit; 1638 } 1639 1640 /* A grant for every rx ring slot */ 1641 if (gnttab_alloc_grant_references(NET_RX_RING_SIZE, 1642 &queue->gref_rx_head) < 0) { 1643 pr_alert("can't alloc rx grant refs\n"); 1644 err = -ENOMEM; 1645 goto exit_free_tx; 1646 } 1647 1648 return 0; 1649 1650 exit_free_tx: 1651 gnttab_free_grant_references(queue->gref_tx_head); 1652 exit: 1653 return err; 1654 } 1655 1656 static int write_queue_xenstore_keys(struct netfront_queue *queue, 1657 struct xenbus_transaction *xbt, int write_hierarchical) 1658 { 1659 /* Write the queue-specific keys into XenStore in the traditional 1660 * way for a single queue, or in a queue subkeys for multiple 1661 * queues. 1662 */ 1663 struct xenbus_device *dev = queue->info->xbdev; 1664 int err; 1665 const char *message; 1666 char *path; 1667 size_t pathsize; 1668 1669 /* Choose the correct place to write the keys */ 1670 if (write_hierarchical) { 1671 pathsize = strlen(dev->nodename) + 10; 1672 path = kzalloc(pathsize, GFP_KERNEL); 1673 if (!path) { 1674 err = -ENOMEM; 1675 message = "out of memory while writing ring references"; 1676 goto error; 1677 } 1678 snprintf(path, pathsize, "%s/queue-%u", 1679 dev->nodename, queue->id); 1680 } else { 1681 path = (char *)dev->nodename; 1682 } 1683 1684 /* Write ring references */ 1685 err = xenbus_printf(*xbt, path, "tx-ring-ref", "%u", 1686 queue->tx_ring_ref); 1687 if (err) { 1688 message = "writing tx-ring-ref"; 1689 goto error; 1690 } 1691 1692 err = xenbus_printf(*xbt, path, "rx-ring-ref", "%u", 1693 queue->rx_ring_ref); 1694 if (err) { 1695 message = "writing rx-ring-ref"; 1696 goto error; 1697 } 1698 1699 /* Write event channels; taking into account both shared 1700 * and split event channel scenarios. 1701 */ 1702 if (queue->tx_evtchn == queue->rx_evtchn) { 1703 /* Shared event channel */ 1704 err = xenbus_printf(*xbt, path, 1705 "event-channel", "%u", queue->tx_evtchn); 1706 if (err) { 1707 message = "writing event-channel"; 1708 goto error; 1709 } 1710 } else { 1711 /* Split event channels */ 1712 err = xenbus_printf(*xbt, path, 1713 "event-channel-tx", "%u", queue->tx_evtchn); 1714 if (err) { 1715 message = "writing event-channel-tx"; 1716 goto error; 1717 } 1718 1719 err = xenbus_printf(*xbt, path, 1720 "event-channel-rx", "%u", queue->rx_evtchn); 1721 if (err) { 1722 message = "writing event-channel-rx"; 1723 goto error; 1724 } 1725 } 1726 1727 if (write_hierarchical) 1728 kfree(path); 1729 return 0; 1730 1731 error: 1732 if (write_hierarchical) 1733 kfree(path); 1734 xenbus_dev_fatal(dev, err, "%s", message); 1735 return err; 1736 } 1737 1738 static void xennet_destroy_queues(struct netfront_info *info) 1739 { 1740 unsigned int i; 1741 1742 for (i = 0; i < info->netdev->real_num_tx_queues; i++) { 1743 struct netfront_queue *queue = &info->queues[i]; 1744 1745 if (netif_running(info->netdev)) 1746 napi_disable(&queue->napi); 1747 netif_napi_del(&queue->napi); 1748 } 1749 1750 kfree(info->queues); 1751 info->queues = NULL; 1752 } 1753 1754 static int xennet_create_queues(struct netfront_info *info, 1755 unsigned int *num_queues) 1756 { 1757 unsigned int i; 1758 int ret; 1759 1760 info->queues = kcalloc(*num_queues, sizeof(struct netfront_queue), 1761 GFP_KERNEL); 1762 if (!info->queues) 1763 return -ENOMEM; 1764 1765 for (i = 0; i < *num_queues; i++) { 1766 struct netfront_queue *queue = &info->queues[i]; 1767 1768 queue->id = i; 1769 queue->info = info; 1770 1771 ret = xennet_init_queue(queue); 1772 if (ret < 0) { 1773 dev_warn(&info->xbdev->dev, 1774 "only created %d queues\n", i); 1775 *num_queues = i; 1776 break; 1777 } 1778 1779 netif_napi_add(queue->info->netdev, &queue->napi, 1780 xennet_poll, 64); 1781 if (netif_running(info->netdev)) 1782 napi_enable(&queue->napi); 1783 } 1784 1785 netif_set_real_num_tx_queues(info->netdev, *num_queues); 1786 1787 if (*num_queues == 0) { 1788 dev_err(&info->xbdev->dev, "no queues\n"); 1789 return -EINVAL; 1790 } 1791 return 0; 1792 } 1793 1794 /* Common code used when first setting up, and when resuming. */ 1795 static int talk_to_netback(struct xenbus_device *dev, 1796 struct netfront_info *info) 1797 { 1798 const char *message; 1799 struct xenbus_transaction xbt; 1800 int err; 1801 unsigned int feature_split_evtchn; 1802 unsigned int i = 0; 1803 unsigned int max_queues = 0; 1804 struct netfront_queue *queue = NULL; 1805 unsigned int num_queues = 1; 1806 1807 info->netdev->irq = 0; 1808 1809 /* Check if backend supports multiple queues */ 1810 max_queues = xenbus_read_unsigned(info->xbdev->otherend, 1811 "multi-queue-max-queues", 1); 1812 num_queues = min(max_queues, xennet_max_queues); 1813 1814 /* Check feature-split-event-channels */ 1815 feature_split_evtchn = xenbus_read_unsigned(info->xbdev->otherend, 1816 "feature-split-event-channels", 0); 1817 1818 /* Read mac addr. */ 1819 err = xen_net_read_mac(dev, info->netdev->dev_addr); 1820 if (err) { 1821 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); 1822 goto out_unlocked; 1823 } 1824 1825 rtnl_lock(); 1826 if (info->queues) 1827 xennet_destroy_queues(info); 1828 1829 err = xennet_create_queues(info, &num_queues); 1830 if (err < 0) { 1831 xenbus_dev_fatal(dev, err, "creating queues"); 1832 kfree(info->queues); 1833 info->queues = NULL; 1834 goto out; 1835 } 1836 rtnl_unlock(); 1837 1838 /* Create shared ring, alloc event channel -- for each queue */ 1839 for (i = 0; i < num_queues; ++i) { 1840 queue = &info->queues[i]; 1841 err = setup_netfront(dev, queue, feature_split_evtchn); 1842 if (err) 1843 goto destroy_ring; 1844 } 1845 1846 again: 1847 err = xenbus_transaction_start(&xbt); 1848 if (err) { 1849 xenbus_dev_fatal(dev, err, "starting transaction"); 1850 goto destroy_ring; 1851 } 1852 1853 if (xenbus_exists(XBT_NIL, 1854 info->xbdev->otherend, "multi-queue-max-queues")) { 1855 /* Write the number of queues */ 1856 err = xenbus_printf(xbt, dev->nodename, 1857 "multi-queue-num-queues", "%u", num_queues); 1858 if (err) { 1859 message = "writing multi-queue-num-queues"; 1860 goto abort_transaction_no_dev_fatal; 1861 } 1862 } 1863 1864 if (num_queues == 1) { 1865 err = write_queue_xenstore_keys(&info->queues[0], &xbt, 0); /* flat */ 1866 if (err) 1867 goto abort_transaction_no_dev_fatal; 1868 } else { 1869 /* Write the keys for each queue */ 1870 for (i = 0; i < num_queues; ++i) { 1871 queue = &info->queues[i]; 1872 err = write_queue_xenstore_keys(queue, &xbt, 1); /* hierarchical */ 1873 if (err) 1874 goto abort_transaction_no_dev_fatal; 1875 } 1876 } 1877 1878 /* The remaining keys are not queue-specific */ 1879 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", 1880 1); 1881 if (err) { 1882 message = "writing request-rx-copy"; 1883 goto abort_transaction; 1884 } 1885 1886 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); 1887 if (err) { 1888 message = "writing feature-rx-notify"; 1889 goto abort_transaction; 1890 } 1891 1892 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); 1893 if (err) { 1894 message = "writing feature-sg"; 1895 goto abort_transaction; 1896 } 1897 1898 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); 1899 if (err) { 1900 message = "writing feature-gso-tcpv4"; 1901 goto abort_transaction; 1902 } 1903 1904 err = xenbus_write(xbt, dev->nodename, "feature-gso-tcpv6", "1"); 1905 if (err) { 1906 message = "writing feature-gso-tcpv6"; 1907 goto abort_transaction; 1908 } 1909 1910 err = xenbus_write(xbt, dev->nodename, "feature-ipv6-csum-offload", 1911 "1"); 1912 if (err) { 1913 message = "writing feature-ipv6-csum-offload"; 1914 goto abort_transaction; 1915 } 1916 1917 err = xenbus_transaction_end(xbt, 0); 1918 if (err) { 1919 if (err == -EAGAIN) 1920 goto again; 1921 xenbus_dev_fatal(dev, err, "completing transaction"); 1922 goto destroy_ring; 1923 } 1924 1925 return 0; 1926 1927 abort_transaction: 1928 xenbus_dev_fatal(dev, err, "%s", message); 1929 abort_transaction_no_dev_fatal: 1930 xenbus_transaction_end(xbt, 1); 1931 destroy_ring: 1932 xennet_disconnect_backend(info); 1933 rtnl_lock(); 1934 xennet_destroy_queues(info); 1935 out: 1936 rtnl_unlock(); 1937 out_unlocked: 1938 device_unregister(&dev->dev); 1939 return err; 1940 } 1941 1942 static int xennet_connect(struct net_device *dev) 1943 { 1944 struct netfront_info *np = netdev_priv(dev); 1945 unsigned int num_queues = 0; 1946 int err; 1947 unsigned int j = 0; 1948 struct netfront_queue *queue = NULL; 1949 1950 if (!xenbus_read_unsigned(np->xbdev->otherend, "feature-rx-copy", 0)) { 1951 dev_info(&dev->dev, 1952 "backend does not support copying receive path\n"); 1953 return -ENODEV; 1954 } 1955 1956 err = talk_to_netback(np->xbdev, np); 1957 if (err) 1958 return err; 1959 1960 /* talk_to_netback() sets the correct number of queues */ 1961 num_queues = dev->real_num_tx_queues; 1962 1963 if (dev->reg_state == NETREG_UNINITIALIZED) { 1964 err = register_netdev(dev); 1965 if (err) { 1966 pr_warn("%s: register_netdev err=%d\n", __func__, err); 1967 device_unregister(&np->xbdev->dev); 1968 return err; 1969 } 1970 } 1971 1972 rtnl_lock(); 1973 netdev_update_features(dev); 1974 rtnl_unlock(); 1975 1976 /* 1977 * All public and private state should now be sane. Get 1978 * ready to start sending and receiving packets and give the driver 1979 * domain a kick because we've probably just requeued some 1980 * packets. 1981 */ 1982 netif_carrier_on(np->netdev); 1983 for (j = 0; j < num_queues; ++j) { 1984 queue = &np->queues[j]; 1985 1986 notify_remote_via_irq(queue->tx_irq); 1987 if (queue->tx_irq != queue->rx_irq) 1988 notify_remote_via_irq(queue->rx_irq); 1989 1990 spin_lock_irq(&queue->tx_lock); 1991 xennet_tx_buf_gc(queue); 1992 spin_unlock_irq(&queue->tx_lock); 1993 1994 spin_lock_bh(&queue->rx_lock); 1995 xennet_alloc_rx_buffers(queue); 1996 spin_unlock_bh(&queue->rx_lock); 1997 } 1998 1999 return 0; 2000 } 2001 2002 /** 2003 * Callback received when the backend's state changes. 2004 */ 2005 static void netback_changed(struct xenbus_device *dev, 2006 enum xenbus_state backend_state) 2007 { 2008 struct netfront_info *np = dev_get_drvdata(&dev->dev); 2009 struct net_device *netdev = np->netdev; 2010 2011 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); 2012 2013 switch (backend_state) { 2014 case XenbusStateInitialising: 2015 case XenbusStateInitialised: 2016 case XenbusStateReconfiguring: 2017 case XenbusStateReconfigured: 2018 break; 2019 2020 case XenbusStateUnknown: 2021 wake_up_all(&module_unload_q); 2022 break; 2023 2024 case XenbusStateInitWait: 2025 if (dev->state != XenbusStateInitialising) 2026 break; 2027 if (xennet_connect(netdev) != 0) 2028 break; 2029 xenbus_switch_state(dev, XenbusStateConnected); 2030 break; 2031 2032 case XenbusStateConnected: 2033 netdev_notify_peers(netdev); 2034 break; 2035 2036 case XenbusStateClosed: 2037 wake_up_all(&module_unload_q); 2038 if (dev->state == XenbusStateClosed) 2039 break; 2040 /* Missed the backend's CLOSING state -- fallthrough */ 2041 case XenbusStateClosing: 2042 wake_up_all(&module_unload_q); 2043 xenbus_frontend_closed(dev); 2044 break; 2045 } 2046 } 2047 2048 static const struct xennet_stat { 2049 char name[ETH_GSTRING_LEN]; 2050 u16 offset; 2051 } xennet_stats[] = { 2052 { 2053 "rx_gso_checksum_fixup", 2054 offsetof(struct netfront_info, rx_gso_checksum_fixup) 2055 }, 2056 }; 2057 2058 static int xennet_get_sset_count(struct net_device *dev, int string_set) 2059 { 2060 switch (string_set) { 2061 case ETH_SS_STATS: 2062 return ARRAY_SIZE(xennet_stats); 2063 default: 2064 return -EINVAL; 2065 } 2066 } 2067 2068 static void xennet_get_ethtool_stats(struct net_device *dev, 2069 struct ethtool_stats *stats, u64 * data) 2070 { 2071 void *np = netdev_priv(dev); 2072 int i; 2073 2074 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) 2075 data[i] = atomic_read((atomic_t *)(np + xennet_stats[i].offset)); 2076 } 2077 2078 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) 2079 { 2080 int i; 2081 2082 switch (stringset) { 2083 case ETH_SS_STATS: 2084 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) 2085 memcpy(data + i * ETH_GSTRING_LEN, 2086 xennet_stats[i].name, ETH_GSTRING_LEN); 2087 break; 2088 } 2089 } 2090 2091 static const struct ethtool_ops xennet_ethtool_ops = 2092 { 2093 .get_link = ethtool_op_get_link, 2094 2095 .get_sset_count = xennet_get_sset_count, 2096 .get_ethtool_stats = xennet_get_ethtool_stats, 2097 .get_strings = xennet_get_strings, 2098 }; 2099 2100 #ifdef CONFIG_SYSFS 2101 static ssize_t show_rxbuf(struct device *dev, 2102 struct device_attribute *attr, char *buf) 2103 { 2104 return sprintf(buf, "%lu\n", NET_RX_RING_SIZE); 2105 } 2106 2107 static ssize_t store_rxbuf(struct device *dev, 2108 struct device_attribute *attr, 2109 const char *buf, size_t len) 2110 { 2111 char *endp; 2112 unsigned long target; 2113 2114 if (!capable(CAP_NET_ADMIN)) 2115 return -EPERM; 2116 2117 target = simple_strtoul(buf, &endp, 0); 2118 if (endp == buf) 2119 return -EBADMSG; 2120 2121 /* rxbuf_min and rxbuf_max are no longer configurable. */ 2122 2123 return len; 2124 } 2125 2126 static DEVICE_ATTR(rxbuf_min, 0644, show_rxbuf, store_rxbuf); 2127 static DEVICE_ATTR(rxbuf_max, 0644, show_rxbuf, store_rxbuf); 2128 static DEVICE_ATTR(rxbuf_cur, 0444, show_rxbuf, NULL); 2129 2130 static struct attribute *xennet_dev_attrs[] = { 2131 &dev_attr_rxbuf_min.attr, 2132 &dev_attr_rxbuf_max.attr, 2133 &dev_attr_rxbuf_cur.attr, 2134 NULL 2135 }; 2136 2137 static const struct attribute_group xennet_dev_group = { 2138 .attrs = xennet_dev_attrs 2139 }; 2140 #endif /* CONFIG_SYSFS */ 2141 2142 static int xennet_remove(struct xenbus_device *dev) 2143 { 2144 struct netfront_info *info = dev_get_drvdata(&dev->dev); 2145 2146 dev_dbg(&dev->dev, "%s\n", dev->nodename); 2147 2148 if (xenbus_read_driver_state(dev->otherend) != XenbusStateClosed) { 2149 xenbus_switch_state(dev, XenbusStateClosing); 2150 wait_event(module_unload_q, 2151 xenbus_read_driver_state(dev->otherend) == 2152 XenbusStateClosing || 2153 xenbus_read_driver_state(dev->otherend) == 2154 XenbusStateUnknown); 2155 2156 xenbus_switch_state(dev, XenbusStateClosed); 2157 wait_event(module_unload_q, 2158 xenbus_read_driver_state(dev->otherend) == 2159 XenbusStateClosed || 2160 xenbus_read_driver_state(dev->otherend) == 2161 XenbusStateUnknown); 2162 } 2163 2164 xennet_disconnect_backend(info); 2165 2166 if (info->netdev->reg_state == NETREG_REGISTERED) 2167 unregister_netdev(info->netdev); 2168 2169 if (info->queues) { 2170 rtnl_lock(); 2171 xennet_destroy_queues(info); 2172 rtnl_unlock(); 2173 } 2174 xennet_free_netdev(info->netdev); 2175 2176 return 0; 2177 } 2178 2179 static const struct xenbus_device_id netfront_ids[] = { 2180 { "vif" }, 2181 { "" } 2182 }; 2183 2184 static struct xenbus_driver netfront_driver = { 2185 .ids = netfront_ids, 2186 .probe = netfront_probe, 2187 .remove = xennet_remove, 2188 .resume = netfront_resume, 2189 .otherend_changed = netback_changed, 2190 }; 2191 2192 static int __init netif_init(void) 2193 { 2194 if (!xen_domain()) 2195 return -ENODEV; 2196 2197 if (!xen_has_pv_nic_devices()) 2198 return -ENODEV; 2199 2200 pr_info("Initialising Xen virtual ethernet driver\n"); 2201 2202 /* Allow as many queues as there are CPUs inut max. 8 if user has not 2203 * specified a value. 2204 */ 2205 if (xennet_max_queues == 0) 2206 xennet_max_queues = min_t(unsigned int, MAX_QUEUES_DEFAULT, 2207 num_online_cpus()); 2208 2209 return xenbus_register_frontend(&netfront_driver); 2210 } 2211 module_init(netif_init); 2212 2213 2214 static void __exit netif_exit(void) 2215 { 2216 xenbus_unregister_driver(&netfront_driver); 2217 } 2218 module_exit(netif_exit); 2219 2220 MODULE_DESCRIPTION("Xen virtual network device frontend"); 2221 MODULE_LICENSE("GPL"); 2222 MODULE_ALIAS("xen:vif"); 2223 MODULE_ALIAS("xennet"); 2224