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 #include <linux/module.h> 33 #include <linux/kernel.h> 34 #include <linux/netdevice.h> 35 #include <linux/etherdevice.h> 36 #include <linux/skbuff.h> 37 #include <linux/ethtool.h> 38 #include <linux/if_ether.h> 39 #include <linux/tcp.h> 40 #include <linux/udp.h> 41 #include <linux/moduleparam.h> 42 #include <linux/mm.h> 43 #include <linux/slab.h> 44 #include <net/ip.h> 45 46 #include <xen/xen.h> 47 #include <xen/xenbus.h> 48 #include <xen/events.h> 49 #include <xen/page.h> 50 #include <xen/grant_table.h> 51 52 #include <xen/interface/io/netif.h> 53 #include <xen/interface/memory.h> 54 #include <xen/interface/grant_table.h> 55 56 static const struct ethtool_ops xennet_ethtool_ops; 57 58 struct netfront_cb { 59 struct page *page; 60 unsigned offset; 61 }; 62 63 #define NETFRONT_SKB_CB(skb) ((struct netfront_cb *)((skb)->cb)) 64 65 #define RX_COPY_THRESHOLD 256 66 67 #define GRANT_INVALID_REF 0 68 69 #define NET_TX_RING_SIZE __CONST_RING_SIZE(xen_netif_tx, PAGE_SIZE) 70 #define NET_RX_RING_SIZE __CONST_RING_SIZE(xen_netif_rx, PAGE_SIZE) 71 #define TX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256) 72 73 struct netfront_info { 74 struct list_head list; 75 struct net_device *netdev; 76 77 struct napi_struct napi; 78 79 unsigned int evtchn; 80 struct xenbus_device *xbdev; 81 82 spinlock_t tx_lock; 83 struct xen_netif_tx_front_ring tx; 84 int tx_ring_ref; 85 86 /* 87 * {tx,rx}_skbs store outstanding skbuffs. Free tx_skb entries 88 * are linked from tx_skb_freelist through skb_entry.link. 89 * 90 * NB. Freelist index entries are always going to be less than 91 * PAGE_OFFSET, whereas pointers to skbs will always be equal or 92 * greater than PAGE_OFFSET: we use this property to distinguish 93 * them. 94 */ 95 union skb_entry { 96 struct sk_buff *skb; 97 unsigned long link; 98 } tx_skbs[NET_TX_RING_SIZE]; 99 grant_ref_t gref_tx_head; 100 grant_ref_t grant_tx_ref[NET_TX_RING_SIZE]; 101 unsigned tx_skb_freelist; 102 103 spinlock_t rx_lock ____cacheline_aligned_in_smp; 104 struct xen_netif_rx_front_ring rx; 105 int rx_ring_ref; 106 107 /* Receive-ring batched refills. */ 108 #define RX_MIN_TARGET 8 109 #define RX_DFL_MIN_TARGET 64 110 #define RX_MAX_TARGET min_t(int, NET_RX_RING_SIZE, 256) 111 unsigned rx_min_target, rx_max_target, rx_target; 112 struct sk_buff_head rx_batch; 113 114 struct timer_list rx_refill_timer; 115 116 struct sk_buff *rx_skbs[NET_RX_RING_SIZE]; 117 grant_ref_t gref_rx_head; 118 grant_ref_t grant_rx_ref[NET_RX_RING_SIZE]; 119 120 unsigned long rx_pfn_array[NET_RX_RING_SIZE]; 121 struct multicall_entry rx_mcl[NET_RX_RING_SIZE+1]; 122 struct mmu_update rx_mmu[NET_RX_RING_SIZE]; 123 124 /* Statistics */ 125 int rx_gso_checksum_fixup; 126 }; 127 128 struct netfront_rx_info { 129 struct xen_netif_rx_response rx; 130 struct xen_netif_extra_info extras[XEN_NETIF_EXTRA_TYPE_MAX - 1]; 131 }; 132 133 static void skb_entry_set_link(union skb_entry *list, unsigned short id) 134 { 135 list->link = id; 136 } 137 138 static int skb_entry_is_link(const union skb_entry *list) 139 { 140 BUILD_BUG_ON(sizeof(list->skb) != sizeof(list->link)); 141 return (unsigned long)list->skb < PAGE_OFFSET; 142 } 143 144 /* 145 * Access macros for acquiring freeing slots in tx_skbs[]. 146 */ 147 148 static void add_id_to_freelist(unsigned *head, union skb_entry *list, 149 unsigned short id) 150 { 151 skb_entry_set_link(&list[id], *head); 152 *head = id; 153 } 154 155 static unsigned short get_id_from_freelist(unsigned *head, 156 union skb_entry *list) 157 { 158 unsigned int id = *head; 159 *head = list[id].link; 160 return id; 161 } 162 163 static int xennet_rxidx(RING_IDX idx) 164 { 165 return idx & (NET_RX_RING_SIZE - 1); 166 } 167 168 static struct sk_buff *xennet_get_rx_skb(struct netfront_info *np, 169 RING_IDX ri) 170 { 171 int i = xennet_rxidx(ri); 172 struct sk_buff *skb = np->rx_skbs[i]; 173 np->rx_skbs[i] = NULL; 174 return skb; 175 } 176 177 static grant_ref_t xennet_get_rx_ref(struct netfront_info *np, 178 RING_IDX ri) 179 { 180 int i = xennet_rxidx(ri); 181 grant_ref_t ref = np->grant_rx_ref[i]; 182 np->grant_rx_ref[i] = GRANT_INVALID_REF; 183 return ref; 184 } 185 186 #ifdef CONFIG_SYSFS 187 static int xennet_sysfs_addif(struct net_device *netdev); 188 static void xennet_sysfs_delif(struct net_device *netdev); 189 #else /* !CONFIG_SYSFS */ 190 #define xennet_sysfs_addif(dev) (0) 191 #define xennet_sysfs_delif(dev) do { } while (0) 192 #endif 193 194 static int xennet_can_sg(struct net_device *dev) 195 { 196 return dev->features & NETIF_F_SG; 197 } 198 199 200 static void rx_refill_timeout(unsigned long data) 201 { 202 struct net_device *dev = (struct net_device *)data; 203 struct netfront_info *np = netdev_priv(dev); 204 napi_schedule(&np->napi); 205 } 206 207 static int netfront_tx_slot_available(struct netfront_info *np) 208 { 209 return (np->tx.req_prod_pvt - np->tx.rsp_cons) < 210 (TX_MAX_TARGET - MAX_SKB_FRAGS - 2); 211 } 212 213 static void xennet_maybe_wake_tx(struct net_device *dev) 214 { 215 struct netfront_info *np = netdev_priv(dev); 216 217 if (unlikely(netif_queue_stopped(dev)) && 218 netfront_tx_slot_available(np) && 219 likely(netif_running(dev))) 220 netif_wake_queue(dev); 221 } 222 223 static void xennet_alloc_rx_buffers(struct net_device *dev) 224 { 225 unsigned short id; 226 struct netfront_info *np = netdev_priv(dev); 227 struct sk_buff *skb; 228 struct page *page; 229 int i, batch_target, notify; 230 RING_IDX req_prod = np->rx.req_prod_pvt; 231 grant_ref_t ref; 232 unsigned long pfn; 233 void *vaddr; 234 struct xen_netif_rx_request *req; 235 236 if (unlikely(!netif_carrier_ok(dev))) 237 return; 238 239 /* 240 * Allocate skbuffs greedily, even though we batch updates to the 241 * receive ring. This creates a less bursty demand on the memory 242 * allocator, so should reduce the chance of failed allocation requests 243 * both for ourself and for other kernel subsystems. 244 */ 245 batch_target = np->rx_target - (req_prod - np->rx.rsp_cons); 246 for (i = skb_queue_len(&np->rx_batch); i < batch_target; i++) { 247 skb = __netdev_alloc_skb(dev, RX_COPY_THRESHOLD + NET_IP_ALIGN, 248 GFP_ATOMIC | __GFP_NOWARN); 249 if (unlikely(!skb)) 250 goto no_skb; 251 252 /* Align ip header to a 16 bytes boundary */ 253 skb_reserve(skb, NET_IP_ALIGN); 254 255 page = alloc_page(GFP_ATOMIC | __GFP_NOWARN); 256 if (!page) { 257 kfree_skb(skb); 258 no_skb: 259 /* Any skbuffs queued for refill? Force them out. */ 260 if (i != 0) 261 goto refill; 262 /* Could not allocate any skbuffs. Try again later. */ 263 mod_timer(&np->rx_refill_timer, 264 jiffies + (HZ/10)); 265 break; 266 } 267 268 skb_shinfo(skb)->frags[0].page = page; 269 skb_shinfo(skb)->nr_frags = 1; 270 __skb_queue_tail(&np->rx_batch, skb); 271 } 272 273 /* Is the batch large enough to be worthwhile? */ 274 if (i < (np->rx_target/2)) { 275 if (req_prod > np->rx.sring->req_prod) 276 goto push; 277 return; 278 } 279 280 /* Adjust our fill target if we risked running out of buffers. */ 281 if (((req_prod - np->rx.sring->rsp_prod) < (np->rx_target / 4)) && 282 ((np->rx_target *= 2) > np->rx_max_target)) 283 np->rx_target = np->rx_max_target; 284 285 refill: 286 for (i = 0; ; i++) { 287 skb = __skb_dequeue(&np->rx_batch); 288 if (skb == NULL) 289 break; 290 291 skb->dev = dev; 292 293 id = xennet_rxidx(req_prod + i); 294 295 BUG_ON(np->rx_skbs[id]); 296 np->rx_skbs[id] = skb; 297 298 ref = gnttab_claim_grant_reference(&np->gref_rx_head); 299 BUG_ON((signed short)ref < 0); 300 np->grant_rx_ref[id] = ref; 301 302 pfn = page_to_pfn(skb_shinfo(skb)->frags[0].page); 303 vaddr = page_address(skb_shinfo(skb)->frags[0].page); 304 305 req = RING_GET_REQUEST(&np->rx, req_prod + i); 306 gnttab_grant_foreign_access_ref(ref, 307 np->xbdev->otherend_id, 308 pfn_to_mfn(pfn), 309 0); 310 311 req->id = id; 312 req->gref = ref; 313 } 314 315 wmb(); /* barrier so backend seens requests */ 316 317 /* Above is a suitable barrier to ensure backend will see requests. */ 318 np->rx.req_prod_pvt = req_prod + i; 319 push: 320 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->rx, notify); 321 if (notify) 322 notify_remote_via_irq(np->netdev->irq); 323 } 324 325 static int xennet_open(struct net_device *dev) 326 { 327 struct netfront_info *np = netdev_priv(dev); 328 329 napi_enable(&np->napi); 330 331 spin_lock_bh(&np->rx_lock); 332 if (netif_carrier_ok(dev)) { 333 xennet_alloc_rx_buffers(dev); 334 np->rx.sring->rsp_event = np->rx.rsp_cons + 1; 335 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx)) 336 napi_schedule(&np->napi); 337 } 338 spin_unlock_bh(&np->rx_lock); 339 340 netif_start_queue(dev); 341 342 return 0; 343 } 344 345 static void xennet_tx_buf_gc(struct net_device *dev) 346 { 347 RING_IDX cons, prod; 348 unsigned short id; 349 struct netfront_info *np = netdev_priv(dev); 350 struct sk_buff *skb; 351 352 BUG_ON(!netif_carrier_ok(dev)); 353 354 do { 355 prod = np->tx.sring->rsp_prod; 356 rmb(); /* Ensure we see responses up to 'rp'. */ 357 358 for (cons = np->tx.rsp_cons; cons != prod; cons++) { 359 struct xen_netif_tx_response *txrsp; 360 361 txrsp = RING_GET_RESPONSE(&np->tx, cons); 362 if (txrsp->status == NETIF_RSP_NULL) 363 continue; 364 365 id = txrsp->id; 366 skb = np->tx_skbs[id].skb; 367 if (unlikely(gnttab_query_foreign_access( 368 np->grant_tx_ref[id]) != 0)) { 369 printk(KERN_ALERT "xennet_tx_buf_gc: warning " 370 "-- grant still in use by backend " 371 "domain.\n"); 372 BUG(); 373 } 374 gnttab_end_foreign_access_ref( 375 np->grant_tx_ref[id], GNTMAP_readonly); 376 gnttab_release_grant_reference( 377 &np->gref_tx_head, np->grant_tx_ref[id]); 378 np->grant_tx_ref[id] = GRANT_INVALID_REF; 379 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, id); 380 dev_kfree_skb_irq(skb); 381 } 382 383 np->tx.rsp_cons = prod; 384 385 /* 386 * Set a new event, then check for race with update of tx_cons. 387 * Note that it is essential to schedule a callback, no matter 388 * how few buffers are pending. Even if there is space in the 389 * transmit ring, higher layers may be blocked because too much 390 * data is outstanding: in such cases notification from Xen is 391 * likely to be the only kick that we'll get. 392 */ 393 np->tx.sring->rsp_event = 394 prod + ((np->tx.sring->req_prod - prod) >> 1) + 1; 395 mb(); /* update shared area */ 396 } while ((cons == prod) && (prod != np->tx.sring->rsp_prod)); 397 398 xennet_maybe_wake_tx(dev); 399 } 400 401 static void xennet_make_frags(struct sk_buff *skb, struct net_device *dev, 402 struct xen_netif_tx_request *tx) 403 { 404 struct netfront_info *np = netdev_priv(dev); 405 char *data = skb->data; 406 unsigned long mfn; 407 RING_IDX prod = np->tx.req_prod_pvt; 408 int frags = skb_shinfo(skb)->nr_frags; 409 unsigned int offset = offset_in_page(data); 410 unsigned int len = skb_headlen(skb); 411 unsigned int id; 412 grant_ref_t ref; 413 int i; 414 415 /* While the header overlaps a page boundary (including being 416 larger than a page), split it it into page-sized chunks. */ 417 while (len > PAGE_SIZE - offset) { 418 tx->size = PAGE_SIZE - offset; 419 tx->flags |= NETTXF_more_data; 420 len -= tx->size; 421 data += tx->size; 422 offset = 0; 423 424 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); 425 np->tx_skbs[id].skb = skb_get(skb); 426 tx = RING_GET_REQUEST(&np->tx, prod++); 427 tx->id = id; 428 ref = gnttab_claim_grant_reference(&np->gref_tx_head); 429 BUG_ON((signed short)ref < 0); 430 431 mfn = virt_to_mfn(data); 432 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id, 433 mfn, GNTMAP_readonly); 434 435 tx->gref = np->grant_tx_ref[id] = ref; 436 tx->offset = offset; 437 tx->size = len; 438 tx->flags = 0; 439 } 440 441 /* Grant backend access to each skb fragment page. */ 442 for (i = 0; i < frags; i++) { 443 skb_frag_t *frag = skb_shinfo(skb)->frags + i; 444 445 tx->flags |= NETTXF_more_data; 446 447 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); 448 np->tx_skbs[id].skb = skb_get(skb); 449 tx = RING_GET_REQUEST(&np->tx, prod++); 450 tx->id = id; 451 ref = gnttab_claim_grant_reference(&np->gref_tx_head); 452 BUG_ON((signed short)ref < 0); 453 454 mfn = pfn_to_mfn(page_to_pfn(frag->page)); 455 gnttab_grant_foreign_access_ref(ref, np->xbdev->otherend_id, 456 mfn, GNTMAP_readonly); 457 458 tx->gref = np->grant_tx_ref[id] = ref; 459 tx->offset = frag->page_offset; 460 tx->size = frag->size; 461 tx->flags = 0; 462 } 463 464 np->tx.req_prod_pvt = prod; 465 } 466 467 static int xennet_start_xmit(struct sk_buff *skb, struct net_device *dev) 468 { 469 unsigned short id; 470 struct netfront_info *np = netdev_priv(dev); 471 struct xen_netif_tx_request *tx; 472 struct xen_netif_extra_info *extra; 473 char *data = skb->data; 474 RING_IDX i; 475 grant_ref_t ref; 476 unsigned long mfn; 477 int notify; 478 int frags = skb_shinfo(skb)->nr_frags; 479 unsigned int offset = offset_in_page(data); 480 unsigned int len = skb_headlen(skb); 481 482 frags += DIV_ROUND_UP(offset + len, PAGE_SIZE); 483 if (unlikely(frags > MAX_SKB_FRAGS + 1)) { 484 printk(KERN_ALERT "xennet: skb rides the rocket: %d frags\n", 485 frags); 486 dump_stack(); 487 goto drop; 488 } 489 490 spin_lock_irq(&np->tx_lock); 491 492 if (unlikely(!netif_carrier_ok(dev) || 493 (frags > 1 && !xennet_can_sg(dev)) || 494 netif_needs_gso(skb, netif_skb_features(skb)))) { 495 spin_unlock_irq(&np->tx_lock); 496 goto drop; 497 } 498 499 i = np->tx.req_prod_pvt; 500 501 id = get_id_from_freelist(&np->tx_skb_freelist, np->tx_skbs); 502 np->tx_skbs[id].skb = skb; 503 504 tx = RING_GET_REQUEST(&np->tx, i); 505 506 tx->id = id; 507 ref = gnttab_claim_grant_reference(&np->gref_tx_head); 508 BUG_ON((signed short)ref < 0); 509 mfn = virt_to_mfn(data); 510 gnttab_grant_foreign_access_ref( 511 ref, np->xbdev->otherend_id, mfn, GNTMAP_readonly); 512 tx->gref = np->grant_tx_ref[id] = ref; 513 tx->offset = offset; 514 tx->size = len; 515 extra = NULL; 516 517 tx->flags = 0; 518 if (skb->ip_summed == CHECKSUM_PARTIAL) 519 /* local packet? */ 520 tx->flags |= NETTXF_csum_blank | NETTXF_data_validated; 521 else if (skb->ip_summed == CHECKSUM_UNNECESSARY) 522 /* remote but checksummed. */ 523 tx->flags |= NETTXF_data_validated; 524 525 if (skb_shinfo(skb)->gso_size) { 526 struct xen_netif_extra_info *gso; 527 528 gso = (struct xen_netif_extra_info *) 529 RING_GET_REQUEST(&np->tx, ++i); 530 531 if (extra) 532 extra->flags |= XEN_NETIF_EXTRA_FLAG_MORE; 533 else 534 tx->flags |= NETTXF_extra_info; 535 536 gso->u.gso.size = skb_shinfo(skb)->gso_size; 537 gso->u.gso.type = XEN_NETIF_GSO_TYPE_TCPV4; 538 gso->u.gso.pad = 0; 539 gso->u.gso.features = 0; 540 541 gso->type = XEN_NETIF_EXTRA_TYPE_GSO; 542 gso->flags = 0; 543 extra = gso; 544 } 545 546 np->tx.req_prod_pvt = i + 1; 547 548 xennet_make_frags(skb, dev, tx); 549 tx->size = skb->len; 550 551 RING_PUSH_REQUESTS_AND_CHECK_NOTIFY(&np->tx, notify); 552 if (notify) 553 notify_remote_via_irq(np->netdev->irq); 554 555 dev->stats.tx_bytes += skb->len; 556 dev->stats.tx_packets++; 557 558 /* Note: It is not safe to access skb after xennet_tx_buf_gc()! */ 559 xennet_tx_buf_gc(dev); 560 561 if (!netfront_tx_slot_available(np)) 562 netif_stop_queue(dev); 563 564 spin_unlock_irq(&np->tx_lock); 565 566 return NETDEV_TX_OK; 567 568 drop: 569 dev->stats.tx_dropped++; 570 dev_kfree_skb(skb); 571 return NETDEV_TX_OK; 572 } 573 574 static int xennet_close(struct net_device *dev) 575 { 576 struct netfront_info *np = netdev_priv(dev); 577 netif_stop_queue(np->netdev); 578 napi_disable(&np->napi); 579 return 0; 580 } 581 582 static void xennet_move_rx_slot(struct netfront_info *np, struct sk_buff *skb, 583 grant_ref_t ref) 584 { 585 int new = xennet_rxidx(np->rx.req_prod_pvt); 586 587 BUG_ON(np->rx_skbs[new]); 588 np->rx_skbs[new] = skb; 589 np->grant_rx_ref[new] = ref; 590 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->id = new; 591 RING_GET_REQUEST(&np->rx, np->rx.req_prod_pvt)->gref = ref; 592 np->rx.req_prod_pvt++; 593 } 594 595 static int xennet_get_extras(struct netfront_info *np, 596 struct xen_netif_extra_info *extras, 597 RING_IDX rp) 598 599 { 600 struct xen_netif_extra_info *extra; 601 struct device *dev = &np->netdev->dev; 602 RING_IDX cons = np->rx.rsp_cons; 603 int err = 0; 604 605 do { 606 struct sk_buff *skb; 607 grant_ref_t ref; 608 609 if (unlikely(cons + 1 == rp)) { 610 if (net_ratelimit()) 611 dev_warn(dev, "Missing extra info\n"); 612 err = -EBADR; 613 break; 614 } 615 616 extra = (struct xen_netif_extra_info *) 617 RING_GET_RESPONSE(&np->rx, ++cons); 618 619 if (unlikely(!extra->type || 620 extra->type >= XEN_NETIF_EXTRA_TYPE_MAX)) { 621 if (net_ratelimit()) 622 dev_warn(dev, "Invalid extra type: %d\n", 623 extra->type); 624 err = -EINVAL; 625 } else { 626 memcpy(&extras[extra->type - 1], extra, 627 sizeof(*extra)); 628 } 629 630 skb = xennet_get_rx_skb(np, cons); 631 ref = xennet_get_rx_ref(np, cons); 632 xennet_move_rx_slot(np, skb, ref); 633 } while (extra->flags & XEN_NETIF_EXTRA_FLAG_MORE); 634 635 np->rx.rsp_cons = cons; 636 return err; 637 } 638 639 static int xennet_get_responses(struct netfront_info *np, 640 struct netfront_rx_info *rinfo, RING_IDX rp, 641 struct sk_buff_head *list) 642 { 643 struct xen_netif_rx_response *rx = &rinfo->rx; 644 struct xen_netif_extra_info *extras = rinfo->extras; 645 struct device *dev = &np->netdev->dev; 646 RING_IDX cons = np->rx.rsp_cons; 647 struct sk_buff *skb = xennet_get_rx_skb(np, cons); 648 grant_ref_t ref = xennet_get_rx_ref(np, cons); 649 int max = MAX_SKB_FRAGS + (rx->status <= RX_COPY_THRESHOLD); 650 int frags = 1; 651 int err = 0; 652 unsigned long ret; 653 654 if (rx->flags & NETRXF_extra_info) { 655 err = xennet_get_extras(np, extras, rp); 656 cons = np->rx.rsp_cons; 657 } 658 659 for (;;) { 660 if (unlikely(rx->status < 0 || 661 rx->offset + rx->status > PAGE_SIZE)) { 662 if (net_ratelimit()) 663 dev_warn(dev, "rx->offset: %x, size: %u\n", 664 rx->offset, rx->status); 665 xennet_move_rx_slot(np, skb, ref); 666 err = -EINVAL; 667 goto next; 668 } 669 670 /* 671 * This definitely indicates a bug, either in this driver or in 672 * the backend driver. In future this should flag the bad 673 * situation to the system controller to reboot the backed. 674 */ 675 if (ref == GRANT_INVALID_REF) { 676 if (net_ratelimit()) 677 dev_warn(dev, "Bad rx response id %d.\n", 678 rx->id); 679 err = -EINVAL; 680 goto next; 681 } 682 683 ret = gnttab_end_foreign_access_ref(ref, 0); 684 BUG_ON(!ret); 685 686 gnttab_release_grant_reference(&np->gref_rx_head, ref); 687 688 __skb_queue_tail(list, skb); 689 690 next: 691 if (!(rx->flags & NETRXF_more_data)) 692 break; 693 694 if (cons + frags == rp) { 695 if (net_ratelimit()) 696 dev_warn(dev, "Need more frags\n"); 697 err = -ENOENT; 698 break; 699 } 700 701 rx = RING_GET_RESPONSE(&np->rx, cons + frags); 702 skb = xennet_get_rx_skb(np, cons + frags); 703 ref = xennet_get_rx_ref(np, cons + frags); 704 frags++; 705 } 706 707 if (unlikely(frags > max)) { 708 if (net_ratelimit()) 709 dev_warn(dev, "Too many frags\n"); 710 err = -E2BIG; 711 } 712 713 if (unlikely(err)) 714 np->rx.rsp_cons = cons + frags; 715 716 return err; 717 } 718 719 static int xennet_set_skb_gso(struct sk_buff *skb, 720 struct xen_netif_extra_info *gso) 721 { 722 if (!gso->u.gso.size) { 723 if (net_ratelimit()) 724 printk(KERN_WARNING "GSO size must not be zero.\n"); 725 return -EINVAL; 726 } 727 728 /* Currently only TCPv4 S.O. is supported. */ 729 if (gso->u.gso.type != XEN_NETIF_GSO_TYPE_TCPV4) { 730 if (net_ratelimit()) 731 printk(KERN_WARNING "Bad GSO type %d.\n", gso->u.gso.type); 732 return -EINVAL; 733 } 734 735 skb_shinfo(skb)->gso_size = gso->u.gso.size; 736 skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4; 737 738 /* Header must be checked, and gso_segs computed. */ 739 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY; 740 skb_shinfo(skb)->gso_segs = 0; 741 742 return 0; 743 } 744 745 static RING_IDX xennet_fill_frags(struct netfront_info *np, 746 struct sk_buff *skb, 747 struct sk_buff_head *list) 748 { 749 struct skb_shared_info *shinfo = skb_shinfo(skb); 750 int nr_frags = shinfo->nr_frags; 751 RING_IDX cons = np->rx.rsp_cons; 752 skb_frag_t *frag = shinfo->frags + nr_frags; 753 struct sk_buff *nskb; 754 755 while ((nskb = __skb_dequeue(list))) { 756 struct xen_netif_rx_response *rx = 757 RING_GET_RESPONSE(&np->rx, ++cons); 758 759 frag->page = skb_shinfo(nskb)->frags[0].page; 760 frag->page_offset = rx->offset; 761 frag->size = rx->status; 762 763 skb->data_len += rx->status; 764 765 skb_shinfo(nskb)->nr_frags = 0; 766 kfree_skb(nskb); 767 768 frag++; 769 nr_frags++; 770 } 771 772 shinfo->nr_frags = nr_frags; 773 return cons; 774 } 775 776 static int checksum_setup(struct net_device *dev, struct sk_buff *skb) 777 { 778 struct iphdr *iph; 779 unsigned char *th; 780 int err = -EPROTO; 781 int recalculate_partial_csum = 0; 782 783 /* 784 * A GSO SKB must be CHECKSUM_PARTIAL. However some buggy 785 * peers can fail to set NETRXF_csum_blank when sending a GSO 786 * frame. In this case force the SKB to CHECKSUM_PARTIAL and 787 * recalculate the partial checksum. 788 */ 789 if (skb->ip_summed != CHECKSUM_PARTIAL && skb_is_gso(skb)) { 790 struct netfront_info *np = netdev_priv(dev); 791 np->rx_gso_checksum_fixup++; 792 skb->ip_summed = CHECKSUM_PARTIAL; 793 recalculate_partial_csum = 1; 794 } 795 796 /* A non-CHECKSUM_PARTIAL SKB does not require setup. */ 797 if (skb->ip_summed != CHECKSUM_PARTIAL) 798 return 0; 799 800 if (skb->protocol != htons(ETH_P_IP)) 801 goto out; 802 803 iph = (void *)skb->data; 804 th = skb->data + 4 * iph->ihl; 805 if (th >= skb_tail_pointer(skb)) 806 goto out; 807 808 skb->csum_start = th - skb->head; 809 switch (iph->protocol) { 810 case IPPROTO_TCP: 811 skb->csum_offset = offsetof(struct tcphdr, check); 812 813 if (recalculate_partial_csum) { 814 struct tcphdr *tcph = (struct tcphdr *)th; 815 tcph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 816 skb->len - iph->ihl*4, 817 IPPROTO_TCP, 0); 818 } 819 break; 820 case IPPROTO_UDP: 821 skb->csum_offset = offsetof(struct udphdr, check); 822 823 if (recalculate_partial_csum) { 824 struct udphdr *udph = (struct udphdr *)th; 825 udph->check = ~csum_tcpudp_magic(iph->saddr, iph->daddr, 826 skb->len - iph->ihl*4, 827 IPPROTO_UDP, 0); 828 } 829 break; 830 default: 831 if (net_ratelimit()) 832 printk(KERN_ERR "Attempting to checksum a non-" 833 "TCP/UDP packet, dropping a protocol" 834 " %d packet", iph->protocol); 835 goto out; 836 } 837 838 if ((th + skb->csum_offset + 2) > skb_tail_pointer(skb)) 839 goto out; 840 841 err = 0; 842 843 out: 844 return err; 845 } 846 847 static int handle_incoming_queue(struct net_device *dev, 848 struct sk_buff_head *rxq) 849 { 850 int packets_dropped = 0; 851 struct sk_buff *skb; 852 853 while ((skb = __skb_dequeue(rxq)) != NULL) { 854 struct page *page = NETFRONT_SKB_CB(skb)->page; 855 void *vaddr = page_address(page); 856 unsigned offset = NETFRONT_SKB_CB(skb)->offset; 857 858 memcpy(skb->data, vaddr + offset, 859 skb_headlen(skb)); 860 861 if (page != skb_shinfo(skb)->frags[0].page) 862 __free_page(page); 863 864 /* Ethernet work: Delayed to here as it peeks the header. */ 865 skb->protocol = eth_type_trans(skb, dev); 866 867 if (checksum_setup(dev, skb)) { 868 kfree_skb(skb); 869 packets_dropped++; 870 dev->stats.rx_errors++; 871 continue; 872 } 873 874 dev->stats.rx_packets++; 875 dev->stats.rx_bytes += skb->len; 876 877 /* Pass it up. */ 878 netif_receive_skb(skb); 879 } 880 881 return packets_dropped; 882 } 883 884 static int xennet_poll(struct napi_struct *napi, int budget) 885 { 886 struct netfront_info *np = container_of(napi, struct netfront_info, napi); 887 struct net_device *dev = np->netdev; 888 struct sk_buff *skb; 889 struct netfront_rx_info rinfo; 890 struct xen_netif_rx_response *rx = &rinfo.rx; 891 struct xen_netif_extra_info *extras = rinfo.extras; 892 RING_IDX i, rp; 893 int work_done; 894 struct sk_buff_head rxq; 895 struct sk_buff_head errq; 896 struct sk_buff_head tmpq; 897 unsigned long flags; 898 unsigned int len; 899 int err; 900 901 spin_lock(&np->rx_lock); 902 903 skb_queue_head_init(&rxq); 904 skb_queue_head_init(&errq); 905 skb_queue_head_init(&tmpq); 906 907 rp = np->rx.sring->rsp_prod; 908 rmb(); /* Ensure we see queued responses up to 'rp'. */ 909 910 i = np->rx.rsp_cons; 911 work_done = 0; 912 while ((i != rp) && (work_done < budget)) { 913 memcpy(rx, RING_GET_RESPONSE(&np->rx, i), sizeof(*rx)); 914 memset(extras, 0, sizeof(rinfo.extras)); 915 916 err = xennet_get_responses(np, &rinfo, rp, &tmpq); 917 918 if (unlikely(err)) { 919 err: 920 while ((skb = __skb_dequeue(&tmpq))) 921 __skb_queue_tail(&errq, skb); 922 dev->stats.rx_errors++; 923 i = np->rx.rsp_cons; 924 continue; 925 } 926 927 skb = __skb_dequeue(&tmpq); 928 929 if (extras[XEN_NETIF_EXTRA_TYPE_GSO - 1].type) { 930 struct xen_netif_extra_info *gso; 931 gso = &extras[XEN_NETIF_EXTRA_TYPE_GSO - 1]; 932 933 if (unlikely(xennet_set_skb_gso(skb, gso))) { 934 __skb_queue_head(&tmpq, skb); 935 np->rx.rsp_cons += skb_queue_len(&tmpq); 936 goto err; 937 } 938 } 939 940 NETFRONT_SKB_CB(skb)->page = skb_shinfo(skb)->frags[0].page; 941 NETFRONT_SKB_CB(skb)->offset = rx->offset; 942 943 len = rx->status; 944 if (len > RX_COPY_THRESHOLD) 945 len = RX_COPY_THRESHOLD; 946 skb_put(skb, len); 947 948 if (rx->status > len) { 949 skb_shinfo(skb)->frags[0].page_offset = 950 rx->offset + len; 951 skb_shinfo(skb)->frags[0].size = rx->status - len; 952 skb->data_len = rx->status - len; 953 } else { 954 skb_shinfo(skb)->frags[0].page = NULL; 955 skb_shinfo(skb)->nr_frags = 0; 956 } 957 958 i = xennet_fill_frags(np, skb, &tmpq); 959 960 /* 961 * Truesize approximates the size of true data plus 962 * any supervisor overheads. Adding hypervisor 963 * overheads has been shown to significantly reduce 964 * achievable bandwidth with the default receive 965 * buffer size. It is therefore not wise to account 966 * for it here. 967 * 968 * After alloc_skb(RX_COPY_THRESHOLD), truesize is set 969 * to RX_COPY_THRESHOLD + the supervisor 970 * overheads. Here, we add the size of the data pulled 971 * in xennet_fill_frags(). 972 * 973 * We also adjust for any unused space in the main 974 * data area by subtracting (RX_COPY_THRESHOLD - 975 * len). This is especially important with drivers 976 * which split incoming packets into header and data, 977 * using only 66 bytes of the main data area (see the 978 * e1000 driver for example.) On such systems, 979 * without this last adjustement, our achievable 980 * receive throughout using the standard receive 981 * buffer size was cut by 25%(!!!). 982 */ 983 skb->truesize += skb->data_len - (RX_COPY_THRESHOLD - len); 984 skb->len += skb->data_len; 985 986 if (rx->flags & NETRXF_csum_blank) 987 skb->ip_summed = CHECKSUM_PARTIAL; 988 else if (rx->flags & NETRXF_data_validated) 989 skb->ip_summed = CHECKSUM_UNNECESSARY; 990 991 __skb_queue_tail(&rxq, skb); 992 993 np->rx.rsp_cons = ++i; 994 work_done++; 995 } 996 997 __skb_queue_purge(&errq); 998 999 work_done -= handle_incoming_queue(dev, &rxq); 1000 1001 /* If we get a callback with very few responses, reduce fill target. */ 1002 /* NB. Note exponential increase, linear decrease. */ 1003 if (((np->rx.req_prod_pvt - np->rx.sring->rsp_prod) > 1004 ((3*np->rx_target) / 4)) && 1005 (--np->rx_target < np->rx_min_target)) 1006 np->rx_target = np->rx_min_target; 1007 1008 xennet_alloc_rx_buffers(dev); 1009 1010 if (work_done < budget) { 1011 int more_to_do = 0; 1012 1013 local_irq_save(flags); 1014 1015 RING_FINAL_CHECK_FOR_RESPONSES(&np->rx, more_to_do); 1016 if (!more_to_do) 1017 __napi_complete(napi); 1018 1019 local_irq_restore(flags); 1020 } 1021 1022 spin_unlock(&np->rx_lock); 1023 1024 return work_done; 1025 } 1026 1027 static int xennet_change_mtu(struct net_device *dev, int mtu) 1028 { 1029 int max = xennet_can_sg(dev) ? 65535 - ETH_HLEN : ETH_DATA_LEN; 1030 1031 if (mtu > max) 1032 return -EINVAL; 1033 dev->mtu = mtu; 1034 return 0; 1035 } 1036 1037 static void xennet_release_tx_bufs(struct netfront_info *np) 1038 { 1039 struct sk_buff *skb; 1040 int i; 1041 1042 for (i = 0; i < NET_TX_RING_SIZE; i++) { 1043 /* Skip over entries which are actually freelist references */ 1044 if (skb_entry_is_link(&np->tx_skbs[i])) 1045 continue; 1046 1047 skb = np->tx_skbs[i].skb; 1048 gnttab_end_foreign_access_ref(np->grant_tx_ref[i], 1049 GNTMAP_readonly); 1050 gnttab_release_grant_reference(&np->gref_tx_head, 1051 np->grant_tx_ref[i]); 1052 np->grant_tx_ref[i] = GRANT_INVALID_REF; 1053 add_id_to_freelist(&np->tx_skb_freelist, np->tx_skbs, i); 1054 dev_kfree_skb_irq(skb); 1055 } 1056 } 1057 1058 static void xennet_release_rx_bufs(struct netfront_info *np) 1059 { 1060 struct mmu_update *mmu = np->rx_mmu; 1061 struct multicall_entry *mcl = np->rx_mcl; 1062 struct sk_buff_head free_list; 1063 struct sk_buff *skb; 1064 unsigned long mfn; 1065 int xfer = 0, noxfer = 0, unused = 0; 1066 int id, ref; 1067 1068 dev_warn(&np->netdev->dev, "%s: fix me for copying receiver.\n", 1069 __func__); 1070 return; 1071 1072 skb_queue_head_init(&free_list); 1073 1074 spin_lock_bh(&np->rx_lock); 1075 1076 for (id = 0; id < NET_RX_RING_SIZE; id++) { 1077 ref = np->grant_rx_ref[id]; 1078 if (ref == GRANT_INVALID_REF) { 1079 unused++; 1080 continue; 1081 } 1082 1083 skb = np->rx_skbs[id]; 1084 mfn = gnttab_end_foreign_transfer_ref(ref); 1085 gnttab_release_grant_reference(&np->gref_rx_head, ref); 1086 np->grant_rx_ref[id] = GRANT_INVALID_REF; 1087 1088 if (0 == mfn) { 1089 skb_shinfo(skb)->nr_frags = 0; 1090 dev_kfree_skb(skb); 1091 noxfer++; 1092 continue; 1093 } 1094 1095 if (!xen_feature(XENFEAT_auto_translated_physmap)) { 1096 /* Remap the page. */ 1097 struct page *page = skb_shinfo(skb)->frags[0].page; 1098 unsigned long pfn = page_to_pfn(page); 1099 void *vaddr = page_address(page); 1100 1101 MULTI_update_va_mapping(mcl, (unsigned long)vaddr, 1102 mfn_pte(mfn, PAGE_KERNEL), 1103 0); 1104 mcl++; 1105 mmu->ptr = ((u64)mfn << PAGE_SHIFT) 1106 | MMU_MACHPHYS_UPDATE; 1107 mmu->val = pfn; 1108 mmu++; 1109 1110 set_phys_to_machine(pfn, mfn); 1111 } 1112 __skb_queue_tail(&free_list, skb); 1113 xfer++; 1114 } 1115 1116 dev_info(&np->netdev->dev, "%s: %d xfer, %d noxfer, %d unused\n", 1117 __func__, xfer, noxfer, unused); 1118 1119 if (xfer) { 1120 if (!xen_feature(XENFEAT_auto_translated_physmap)) { 1121 /* Do all the remapping work and M2P updates. */ 1122 MULTI_mmu_update(mcl, np->rx_mmu, mmu - np->rx_mmu, 1123 NULL, DOMID_SELF); 1124 mcl++; 1125 HYPERVISOR_multicall(np->rx_mcl, mcl - np->rx_mcl); 1126 } 1127 } 1128 1129 __skb_queue_purge(&free_list); 1130 1131 spin_unlock_bh(&np->rx_lock); 1132 } 1133 1134 static void xennet_uninit(struct net_device *dev) 1135 { 1136 struct netfront_info *np = netdev_priv(dev); 1137 xennet_release_tx_bufs(np); 1138 xennet_release_rx_bufs(np); 1139 gnttab_free_grant_references(np->gref_tx_head); 1140 gnttab_free_grant_references(np->gref_rx_head); 1141 } 1142 1143 static const struct net_device_ops xennet_netdev_ops = { 1144 .ndo_open = xennet_open, 1145 .ndo_uninit = xennet_uninit, 1146 .ndo_stop = xennet_close, 1147 .ndo_start_xmit = xennet_start_xmit, 1148 .ndo_change_mtu = xennet_change_mtu, 1149 .ndo_set_mac_address = eth_mac_addr, 1150 .ndo_validate_addr = eth_validate_addr, 1151 }; 1152 1153 static struct net_device * __devinit xennet_create_dev(struct xenbus_device *dev) 1154 { 1155 int i, err; 1156 struct net_device *netdev; 1157 struct netfront_info *np; 1158 1159 netdev = alloc_etherdev(sizeof(struct netfront_info)); 1160 if (!netdev) { 1161 printk(KERN_WARNING "%s> alloc_etherdev failed.\n", 1162 __func__); 1163 return ERR_PTR(-ENOMEM); 1164 } 1165 1166 np = netdev_priv(netdev); 1167 np->xbdev = dev; 1168 1169 spin_lock_init(&np->tx_lock); 1170 spin_lock_init(&np->rx_lock); 1171 1172 skb_queue_head_init(&np->rx_batch); 1173 np->rx_target = RX_DFL_MIN_TARGET; 1174 np->rx_min_target = RX_DFL_MIN_TARGET; 1175 np->rx_max_target = RX_MAX_TARGET; 1176 1177 init_timer(&np->rx_refill_timer); 1178 np->rx_refill_timer.data = (unsigned long)netdev; 1179 np->rx_refill_timer.function = rx_refill_timeout; 1180 1181 /* Initialise tx_skbs as a free chain containing every entry. */ 1182 np->tx_skb_freelist = 0; 1183 for (i = 0; i < NET_TX_RING_SIZE; i++) { 1184 skb_entry_set_link(&np->tx_skbs[i], i+1); 1185 np->grant_tx_ref[i] = GRANT_INVALID_REF; 1186 } 1187 1188 /* Clear out rx_skbs */ 1189 for (i = 0; i < NET_RX_RING_SIZE; i++) { 1190 np->rx_skbs[i] = NULL; 1191 np->grant_rx_ref[i] = GRANT_INVALID_REF; 1192 } 1193 1194 /* A grant for every tx ring slot */ 1195 if (gnttab_alloc_grant_references(TX_MAX_TARGET, 1196 &np->gref_tx_head) < 0) { 1197 printk(KERN_ALERT "#### netfront can't alloc tx grant refs\n"); 1198 err = -ENOMEM; 1199 goto exit; 1200 } 1201 /* A grant for every rx ring slot */ 1202 if (gnttab_alloc_grant_references(RX_MAX_TARGET, 1203 &np->gref_rx_head) < 0) { 1204 printk(KERN_ALERT "#### netfront can't alloc rx grant refs\n"); 1205 err = -ENOMEM; 1206 goto exit_free_tx; 1207 } 1208 1209 netdev->netdev_ops = &xennet_netdev_ops; 1210 1211 netif_napi_add(netdev, &np->napi, xennet_poll, 64); 1212 netdev->features = NETIF_F_IP_CSUM; 1213 1214 SET_ETHTOOL_OPS(netdev, &xennet_ethtool_ops); 1215 SET_NETDEV_DEV(netdev, &dev->dev); 1216 1217 np->netdev = netdev; 1218 1219 netif_carrier_off(netdev); 1220 1221 return netdev; 1222 1223 exit_free_tx: 1224 gnttab_free_grant_references(np->gref_tx_head); 1225 exit: 1226 free_netdev(netdev); 1227 return ERR_PTR(err); 1228 } 1229 1230 /** 1231 * Entry point to this code when a new device is created. Allocate the basic 1232 * structures and the ring buffers for communication with the backend, and 1233 * inform the backend of the appropriate details for those. 1234 */ 1235 static int __devinit netfront_probe(struct xenbus_device *dev, 1236 const struct xenbus_device_id *id) 1237 { 1238 int err; 1239 struct net_device *netdev; 1240 struct netfront_info *info; 1241 1242 netdev = xennet_create_dev(dev); 1243 if (IS_ERR(netdev)) { 1244 err = PTR_ERR(netdev); 1245 xenbus_dev_fatal(dev, err, "creating netdev"); 1246 return err; 1247 } 1248 1249 info = netdev_priv(netdev); 1250 dev_set_drvdata(&dev->dev, info); 1251 1252 err = register_netdev(info->netdev); 1253 if (err) { 1254 printk(KERN_WARNING "%s: register_netdev err=%d\n", 1255 __func__, err); 1256 goto fail; 1257 } 1258 1259 err = xennet_sysfs_addif(info->netdev); 1260 if (err) { 1261 unregister_netdev(info->netdev); 1262 printk(KERN_WARNING "%s: add sysfs failed err=%d\n", 1263 __func__, err); 1264 goto fail; 1265 } 1266 1267 return 0; 1268 1269 fail: 1270 free_netdev(netdev); 1271 dev_set_drvdata(&dev->dev, NULL); 1272 return err; 1273 } 1274 1275 static void xennet_end_access(int ref, void *page) 1276 { 1277 /* This frees the page as a side-effect */ 1278 if (ref != GRANT_INVALID_REF) 1279 gnttab_end_foreign_access(ref, 0, (unsigned long)page); 1280 } 1281 1282 static void xennet_disconnect_backend(struct netfront_info *info) 1283 { 1284 /* Stop old i/f to prevent errors whilst we rebuild the state. */ 1285 spin_lock_bh(&info->rx_lock); 1286 spin_lock_irq(&info->tx_lock); 1287 netif_carrier_off(info->netdev); 1288 spin_unlock_irq(&info->tx_lock); 1289 spin_unlock_bh(&info->rx_lock); 1290 1291 if (info->netdev->irq) 1292 unbind_from_irqhandler(info->netdev->irq, info->netdev); 1293 info->evtchn = info->netdev->irq = 0; 1294 1295 /* End access and free the pages */ 1296 xennet_end_access(info->tx_ring_ref, info->tx.sring); 1297 xennet_end_access(info->rx_ring_ref, info->rx.sring); 1298 1299 info->tx_ring_ref = GRANT_INVALID_REF; 1300 info->rx_ring_ref = GRANT_INVALID_REF; 1301 info->tx.sring = NULL; 1302 info->rx.sring = NULL; 1303 } 1304 1305 /** 1306 * We are reconnecting to the backend, due to a suspend/resume, or a backend 1307 * driver restart. We tear down our netif structure and recreate it, but 1308 * leave the device-layer structures intact so that this is transparent to the 1309 * rest of the kernel. 1310 */ 1311 static int netfront_resume(struct xenbus_device *dev) 1312 { 1313 struct netfront_info *info = dev_get_drvdata(&dev->dev); 1314 1315 dev_dbg(&dev->dev, "%s\n", dev->nodename); 1316 1317 xennet_disconnect_backend(info); 1318 return 0; 1319 } 1320 1321 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[]) 1322 { 1323 char *s, *e, *macstr; 1324 int i; 1325 1326 macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL); 1327 if (IS_ERR(macstr)) 1328 return PTR_ERR(macstr); 1329 1330 for (i = 0; i < ETH_ALEN; i++) { 1331 mac[i] = simple_strtoul(s, &e, 16); 1332 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) { 1333 kfree(macstr); 1334 return -ENOENT; 1335 } 1336 s = e+1; 1337 } 1338 1339 kfree(macstr); 1340 return 0; 1341 } 1342 1343 static irqreturn_t xennet_interrupt(int irq, void *dev_id) 1344 { 1345 struct net_device *dev = dev_id; 1346 struct netfront_info *np = netdev_priv(dev); 1347 unsigned long flags; 1348 1349 spin_lock_irqsave(&np->tx_lock, flags); 1350 1351 if (likely(netif_carrier_ok(dev))) { 1352 xennet_tx_buf_gc(dev); 1353 /* Under tx_lock: protects access to rx shared-ring indexes. */ 1354 if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx)) 1355 napi_schedule(&np->napi); 1356 } 1357 1358 spin_unlock_irqrestore(&np->tx_lock, flags); 1359 1360 return IRQ_HANDLED; 1361 } 1362 1363 static int setup_netfront(struct xenbus_device *dev, struct netfront_info *info) 1364 { 1365 struct xen_netif_tx_sring *txs; 1366 struct xen_netif_rx_sring *rxs; 1367 int err; 1368 struct net_device *netdev = info->netdev; 1369 1370 info->tx_ring_ref = GRANT_INVALID_REF; 1371 info->rx_ring_ref = GRANT_INVALID_REF; 1372 info->rx.sring = NULL; 1373 info->tx.sring = NULL; 1374 netdev->irq = 0; 1375 1376 err = xen_net_read_mac(dev, netdev->dev_addr); 1377 if (err) { 1378 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename); 1379 goto fail; 1380 } 1381 1382 txs = (struct xen_netif_tx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); 1383 if (!txs) { 1384 err = -ENOMEM; 1385 xenbus_dev_fatal(dev, err, "allocating tx ring page"); 1386 goto fail; 1387 } 1388 SHARED_RING_INIT(txs); 1389 FRONT_RING_INIT(&info->tx, txs, PAGE_SIZE); 1390 1391 err = xenbus_grant_ring(dev, virt_to_mfn(txs)); 1392 if (err < 0) { 1393 free_page((unsigned long)txs); 1394 goto fail; 1395 } 1396 1397 info->tx_ring_ref = err; 1398 rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH); 1399 if (!rxs) { 1400 err = -ENOMEM; 1401 xenbus_dev_fatal(dev, err, "allocating rx ring page"); 1402 goto fail; 1403 } 1404 SHARED_RING_INIT(rxs); 1405 FRONT_RING_INIT(&info->rx, rxs, PAGE_SIZE); 1406 1407 err = xenbus_grant_ring(dev, virt_to_mfn(rxs)); 1408 if (err < 0) { 1409 free_page((unsigned long)rxs); 1410 goto fail; 1411 } 1412 info->rx_ring_ref = err; 1413 1414 err = xenbus_alloc_evtchn(dev, &info->evtchn); 1415 if (err) 1416 goto fail; 1417 1418 err = bind_evtchn_to_irqhandler(info->evtchn, xennet_interrupt, 1419 IRQF_SAMPLE_RANDOM, netdev->name, 1420 netdev); 1421 if (err < 0) 1422 goto fail; 1423 netdev->irq = err; 1424 return 0; 1425 1426 fail: 1427 return err; 1428 } 1429 1430 /* Common code used when first setting up, and when resuming. */ 1431 static int talk_to_netback(struct xenbus_device *dev, 1432 struct netfront_info *info) 1433 { 1434 const char *message; 1435 struct xenbus_transaction xbt; 1436 int err; 1437 1438 /* Create shared ring, alloc event channel. */ 1439 err = setup_netfront(dev, info); 1440 if (err) 1441 goto out; 1442 1443 again: 1444 err = xenbus_transaction_start(&xbt); 1445 if (err) { 1446 xenbus_dev_fatal(dev, err, "starting transaction"); 1447 goto destroy_ring; 1448 } 1449 1450 err = xenbus_printf(xbt, dev->nodename, "tx-ring-ref", "%u", 1451 info->tx_ring_ref); 1452 if (err) { 1453 message = "writing tx ring-ref"; 1454 goto abort_transaction; 1455 } 1456 err = xenbus_printf(xbt, dev->nodename, "rx-ring-ref", "%u", 1457 info->rx_ring_ref); 1458 if (err) { 1459 message = "writing rx ring-ref"; 1460 goto abort_transaction; 1461 } 1462 err = xenbus_printf(xbt, dev->nodename, 1463 "event-channel", "%u", info->evtchn); 1464 if (err) { 1465 message = "writing event-channel"; 1466 goto abort_transaction; 1467 } 1468 1469 err = xenbus_printf(xbt, dev->nodename, "request-rx-copy", "%u", 1470 1); 1471 if (err) { 1472 message = "writing request-rx-copy"; 1473 goto abort_transaction; 1474 } 1475 1476 err = xenbus_printf(xbt, dev->nodename, "feature-rx-notify", "%d", 1); 1477 if (err) { 1478 message = "writing feature-rx-notify"; 1479 goto abort_transaction; 1480 } 1481 1482 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", 1); 1483 if (err) { 1484 message = "writing feature-sg"; 1485 goto abort_transaction; 1486 } 1487 1488 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4", "%d", 1); 1489 if (err) { 1490 message = "writing feature-gso-tcpv4"; 1491 goto abort_transaction; 1492 } 1493 1494 err = xenbus_transaction_end(xbt, 0); 1495 if (err) { 1496 if (err == -EAGAIN) 1497 goto again; 1498 xenbus_dev_fatal(dev, err, "completing transaction"); 1499 goto destroy_ring; 1500 } 1501 1502 return 0; 1503 1504 abort_transaction: 1505 xenbus_transaction_end(xbt, 1); 1506 xenbus_dev_fatal(dev, err, "%s", message); 1507 destroy_ring: 1508 xennet_disconnect_backend(info); 1509 out: 1510 return err; 1511 } 1512 1513 static int xennet_set_sg(struct net_device *dev, u32 data) 1514 { 1515 if (data) { 1516 struct netfront_info *np = netdev_priv(dev); 1517 int val; 1518 1519 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, "feature-sg", 1520 "%d", &val) < 0) 1521 val = 0; 1522 if (!val) 1523 return -ENOSYS; 1524 } else if (dev->mtu > ETH_DATA_LEN) 1525 dev->mtu = ETH_DATA_LEN; 1526 1527 return ethtool_op_set_sg(dev, data); 1528 } 1529 1530 static int xennet_set_tso(struct net_device *dev, u32 data) 1531 { 1532 if (data) { 1533 struct netfront_info *np = netdev_priv(dev); 1534 int val; 1535 1536 if (xenbus_scanf(XBT_NIL, np->xbdev->otherend, 1537 "feature-gso-tcpv4", "%d", &val) < 0) 1538 val = 0; 1539 if (!val) 1540 return -ENOSYS; 1541 } 1542 1543 return ethtool_op_set_tso(dev, data); 1544 } 1545 1546 static void xennet_set_features(struct net_device *dev) 1547 { 1548 /* Turn off all GSO bits except ROBUST. */ 1549 dev->features &= ~NETIF_F_GSO_MASK; 1550 dev->features |= NETIF_F_GSO_ROBUST; 1551 xennet_set_sg(dev, 0); 1552 1553 /* We need checksum offload to enable scatter/gather and TSO. */ 1554 if (!(dev->features & NETIF_F_IP_CSUM)) 1555 return; 1556 1557 if (!xennet_set_sg(dev, 1)) 1558 xennet_set_tso(dev, 1); 1559 } 1560 1561 static int xennet_connect(struct net_device *dev) 1562 { 1563 struct netfront_info *np = netdev_priv(dev); 1564 int i, requeue_idx, err; 1565 struct sk_buff *skb; 1566 grant_ref_t ref; 1567 struct xen_netif_rx_request *req; 1568 unsigned int feature_rx_copy; 1569 1570 err = xenbus_scanf(XBT_NIL, np->xbdev->otherend, 1571 "feature-rx-copy", "%u", &feature_rx_copy); 1572 if (err != 1) 1573 feature_rx_copy = 0; 1574 1575 if (!feature_rx_copy) { 1576 dev_info(&dev->dev, 1577 "backend does not support copying receive path\n"); 1578 return -ENODEV; 1579 } 1580 1581 err = talk_to_netback(np->xbdev, np); 1582 if (err) 1583 return err; 1584 1585 xennet_set_features(dev); 1586 1587 spin_lock_bh(&np->rx_lock); 1588 spin_lock_irq(&np->tx_lock); 1589 1590 /* Step 1: Discard all pending TX packet fragments. */ 1591 xennet_release_tx_bufs(np); 1592 1593 /* Step 2: Rebuild the RX buffer freelist and the RX ring itself. */ 1594 for (requeue_idx = 0, i = 0; i < NET_RX_RING_SIZE; i++) { 1595 if (!np->rx_skbs[i]) 1596 continue; 1597 1598 skb = np->rx_skbs[requeue_idx] = xennet_get_rx_skb(np, i); 1599 ref = np->grant_rx_ref[requeue_idx] = xennet_get_rx_ref(np, i); 1600 req = RING_GET_REQUEST(&np->rx, requeue_idx); 1601 1602 gnttab_grant_foreign_access_ref( 1603 ref, np->xbdev->otherend_id, 1604 pfn_to_mfn(page_to_pfn(skb_shinfo(skb)-> 1605 frags->page)), 1606 0); 1607 req->gref = ref; 1608 req->id = requeue_idx; 1609 1610 requeue_idx++; 1611 } 1612 1613 np->rx.req_prod_pvt = requeue_idx; 1614 1615 /* 1616 * Step 3: All public and private state should now be sane. Get 1617 * ready to start sending and receiving packets and give the driver 1618 * domain a kick because we've probably just requeued some 1619 * packets. 1620 */ 1621 netif_carrier_on(np->netdev); 1622 notify_remote_via_irq(np->netdev->irq); 1623 xennet_tx_buf_gc(dev); 1624 xennet_alloc_rx_buffers(dev); 1625 1626 spin_unlock_irq(&np->tx_lock); 1627 spin_unlock_bh(&np->rx_lock); 1628 1629 return 0; 1630 } 1631 1632 /** 1633 * Callback received when the backend's state changes. 1634 */ 1635 static void netback_changed(struct xenbus_device *dev, 1636 enum xenbus_state backend_state) 1637 { 1638 struct netfront_info *np = dev_get_drvdata(&dev->dev); 1639 struct net_device *netdev = np->netdev; 1640 1641 dev_dbg(&dev->dev, "%s\n", xenbus_strstate(backend_state)); 1642 1643 switch (backend_state) { 1644 case XenbusStateInitialising: 1645 case XenbusStateInitialised: 1646 case XenbusStateReconfiguring: 1647 case XenbusStateReconfigured: 1648 case XenbusStateConnected: 1649 case XenbusStateUnknown: 1650 case XenbusStateClosed: 1651 break; 1652 1653 case XenbusStateInitWait: 1654 if (dev->state != XenbusStateInitialising) 1655 break; 1656 if (xennet_connect(netdev) != 0) 1657 break; 1658 xenbus_switch_state(dev, XenbusStateConnected); 1659 netif_notify_peers(netdev); 1660 break; 1661 1662 case XenbusStateClosing: 1663 xenbus_frontend_closed(dev); 1664 break; 1665 } 1666 } 1667 1668 static const struct xennet_stat { 1669 char name[ETH_GSTRING_LEN]; 1670 u16 offset; 1671 } xennet_stats[] = { 1672 { 1673 "rx_gso_checksum_fixup", 1674 offsetof(struct netfront_info, rx_gso_checksum_fixup) 1675 }, 1676 }; 1677 1678 static int xennet_get_sset_count(struct net_device *dev, int string_set) 1679 { 1680 switch (string_set) { 1681 case ETH_SS_STATS: 1682 return ARRAY_SIZE(xennet_stats); 1683 default: 1684 return -EINVAL; 1685 } 1686 } 1687 1688 static void xennet_get_ethtool_stats(struct net_device *dev, 1689 struct ethtool_stats *stats, u64 * data) 1690 { 1691 void *np = netdev_priv(dev); 1692 int i; 1693 1694 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) 1695 data[i] = *(int *)(np + xennet_stats[i].offset); 1696 } 1697 1698 static void xennet_get_strings(struct net_device *dev, u32 stringset, u8 * data) 1699 { 1700 int i; 1701 1702 switch (stringset) { 1703 case ETH_SS_STATS: 1704 for (i = 0; i < ARRAY_SIZE(xennet_stats); i++) 1705 memcpy(data + i * ETH_GSTRING_LEN, 1706 xennet_stats[i].name, ETH_GSTRING_LEN); 1707 break; 1708 } 1709 } 1710 1711 static const struct ethtool_ops xennet_ethtool_ops = 1712 { 1713 .set_tx_csum = ethtool_op_set_tx_csum, 1714 .set_sg = xennet_set_sg, 1715 .set_tso = xennet_set_tso, 1716 .get_link = ethtool_op_get_link, 1717 1718 .get_sset_count = xennet_get_sset_count, 1719 .get_ethtool_stats = xennet_get_ethtool_stats, 1720 .get_strings = xennet_get_strings, 1721 }; 1722 1723 #ifdef CONFIG_SYSFS 1724 static ssize_t show_rxbuf_min(struct device *dev, 1725 struct device_attribute *attr, char *buf) 1726 { 1727 struct net_device *netdev = to_net_dev(dev); 1728 struct netfront_info *info = netdev_priv(netdev); 1729 1730 return sprintf(buf, "%u\n", info->rx_min_target); 1731 } 1732 1733 static ssize_t store_rxbuf_min(struct device *dev, 1734 struct device_attribute *attr, 1735 const char *buf, size_t len) 1736 { 1737 struct net_device *netdev = to_net_dev(dev); 1738 struct netfront_info *np = netdev_priv(netdev); 1739 char *endp; 1740 unsigned long target; 1741 1742 if (!capable(CAP_NET_ADMIN)) 1743 return -EPERM; 1744 1745 target = simple_strtoul(buf, &endp, 0); 1746 if (endp == buf) 1747 return -EBADMSG; 1748 1749 if (target < RX_MIN_TARGET) 1750 target = RX_MIN_TARGET; 1751 if (target > RX_MAX_TARGET) 1752 target = RX_MAX_TARGET; 1753 1754 spin_lock_bh(&np->rx_lock); 1755 if (target > np->rx_max_target) 1756 np->rx_max_target = target; 1757 np->rx_min_target = target; 1758 if (target > np->rx_target) 1759 np->rx_target = target; 1760 1761 xennet_alloc_rx_buffers(netdev); 1762 1763 spin_unlock_bh(&np->rx_lock); 1764 return len; 1765 } 1766 1767 static ssize_t show_rxbuf_max(struct device *dev, 1768 struct device_attribute *attr, char *buf) 1769 { 1770 struct net_device *netdev = to_net_dev(dev); 1771 struct netfront_info *info = netdev_priv(netdev); 1772 1773 return sprintf(buf, "%u\n", info->rx_max_target); 1774 } 1775 1776 static ssize_t store_rxbuf_max(struct device *dev, 1777 struct device_attribute *attr, 1778 const char *buf, size_t len) 1779 { 1780 struct net_device *netdev = to_net_dev(dev); 1781 struct netfront_info *np = netdev_priv(netdev); 1782 char *endp; 1783 unsigned long target; 1784 1785 if (!capable(CAP_NET_ADMIN)) 1786 return -EPERM; 1787 1788 target = simple_strtoul(buf, &endp, 0); 1789 if (endp == buf) 1790 return -EBADMSG; 1791 1792 if (target < RX_MIN_TARGET) 1793 target = RX_MIN_TARGET; 1794 if (target > RX_MAX_TARGET) 1795 target = RX_MAX_TARGET; 1796 1797 spin_lock_bh(&np->rx_lock); 1798 if (target < np->rx_min_target) 1799 np->rx_min_target = target; 1800 np->rx_max_target = target; 1801 if (target < np->rx_target) 1802 np->rx_target = target; 1803 1804 xennet_alloc_rx_buffers(netdev); 1805 1806 spin_unlock_bh(&np->rx_lock); 1807 return len; 1808 } 1809 1810 static ssize_t show_rxbuf_cur(struct device *dev, 1811 struct device_attribute *attr, char *buf) 1812 { 1813 struct net_device *netdev = to_net_dev(dev); 1814 struct netfront_info *info = netdev_priv(netdev); 1815 1816 return sprintf(buf, "%u\n", info->rx_target); 1817 } 1818 1819 static struct device_attribute xennet_attrs[] = { 1820 __ATTR(rxbuf_min, S_IRUGO|S_IWUSR, show_rxbuf_min, store_rxbuf_min), 1821 __ATTR(rxbuf_max, S_IRUGO|S_IWUSR, show_rxbuf_max, store_rxbuf_max), 1822 __ATTR(rxbuf_cur, S_IRUGO, show_rxbuf_cur, NULL), 1823 }; 1824 1825 static int xennet_sysfs_addif(struct net_device *netdev) 1826 { 1827 int i; 1828 int err; 1829 1830 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) { 1831 err = device_create_file(&netdev->dev, 1832 &xennet_attrs[i]); 1833 if (err) 1834 goto fail; 1835 } 1836 return 0; 1837 1838 fail: 1839 while (--i >= 0) 1840 device_remove_file(&netdev->dev, &xennet_attrs[i]); 1841 return err; 1842 } 1843 1844 static void xennet_sysfs_delif(struct net_device *netdev) 1845 { 1846 int i; 1847 1848 for (i = 0; i < ARRAY_SIZE(xennet_attrs); i++) 1849 device_remove_file(&netdev->dev, &xennet_attrs[i]); 1850 } 1851 1852 #endif /* CONFIG_SYSFS */ 1853 1854 static struct xenbus_device_id netfront_ids[] = { 1855 { "vif" }, 1856 { "" } 1857 }; 1858 1859 1860 static int __devexit xennet_remove(struct xenbus_device *dev) 1861 { 1862 struct netfront_info *info = dev_get_drvdata(&dev->dev); 1863 1864 dev_dbg(&dev->dev, "%s\n", dev->nodename); 1865 1866 unregister_netdev(info->netdev); 1867 1868 xennet_disconnect_backend(info); 1869 1870 del_timer_sync(&info->rx_refill_timer); 1871 1872 xennet_sysfs_delif(info->netdev); 1873 1874 free_netdev(info->netdev); 1875 1876 return 0; 1877 } 1878 1879 static struct xenbus_driver netfront_driver = { 1880 .name = "vif", 1881 .owner = THIS_MODULE, 1882 .ids = netfront_ids, 1883 .probe = netfront_probe, 1884 .remove = __devexit_p(xennet_remove), 1885 .resume = netfront_resume, 1886 .otherend_changed = netback_changed, 1887 }; 1888 1889 static int __init netif_init(void) 1890 { 1891 if (!xen_domain()) 1892 return -ENODEV; 1893 1894 if (xen_initial_domain()) 1895 return 0; 1896 1897 printk(KERN_INFO "Initialising Xen virtual ethernet driver.\n"); 1898 1899 return xenbus_register_frontend(&netfront_driver); 1900 } 1901 module_init(netif_init); 1902 1903 1904 static void __exit netif_exit(void) 1905 { 1906 if (xen_initial_domain()) 1907 return; 1908 1909 xenbus_unregister_driver(&netfront_driver); 1910 } 1911 module_exit(netif_exit); 1912 1913 MODULE_DESCRIPTION("Xen virtual network device frontend"); 1914 MODULE_LICENSE("GPL"); 1915 MODULE_ALIAS("xen:vif"); 1916 MODULE_ALIAS("xennet"); 1917