drivers/net/xen-netback/xenbus.c

   1 /*
   2  * Xenbus code for netif backend
   3  *
   4  * Copyright (C) 2005 Rusty Russell <rusty@rustcorp.com.au>
   5  * Copyright (C) 2005 XenSource Ltd
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation; either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, see <http://www.gnu.org/licenses/>.
  19 */
  20
  21 #include "common.h"
  22 #include <linux/vmalloc.h>
  23 #include <linux/rtnetlink.h>
  24
  25 struct backend_info {
  26         struct xenbus_device *dev;
  27         struct xenvif *vif;
  28
  29         /* This is the state that will be reflected in xenstore when any
  30          * active hotplug script completes.
  31          */
  32         enum xenbus_state state;
  33
  34         enum xenbus_state frontend_state;
  35         struct xenbus_watch hotplug_status_watch;
  36         u8 have_hotplug_status_watch:1;
  37 };
  38
  39 static int connect_rings(struct backend_info *be, struct xenvif_queue *queue);
  40 static void connect(struct backend_info *be);
  41 static int read_xenbus_vif_flags(struct backend_info *be);
  42 static void backend_create_xenvif(struct backend_info *be);
  43 static void unregister_hotplug_status_watch(struct backend_info *be);
  44 static void set_backend_state(struct backend_info *be,
  45                               enum xenbus_state state);
  46
  47 #ifdef CONFIG_DEBUG_FS
  48 struct dentry *xen_netback_dbg_root = NULL;
  49
  50 static int xenvif_read_io_ring(struct seq_file *m, void *v)
  51 {
  52         struct xenvif_queue *queue = m->private;
  53         struct xen_netif_tx_back_ring *tx_ring = &queue->tx;
  54         struct xen_netif_rx_back_ring *rx_ring = &queue->rx;
  55
  56         if (tx_ring->sring) {
  57                 struct xen_netif_tx_sring *sring = tx_ring->sring;
  58
  59                 seq_printf(m, "Queue %d\nTX: nr_ents %u\n", queue->id,
  60                            tx_ring->nr_ents);
  61                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  62                            sring->req_prod,
  63                            sring->req_prod - sring->rsp_prod,
  64                            tx_ring->req_cons,
  65                            tx_ring->req_cons - sring->rsp_prod,
  66                            sring->req_event,
  67                            sring->req_event - sring->rsp_prod);
  68                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n",
  69                            sring->rsp_prod,
  70                            tx_ring->rsp_prod_pvt,
  71                            tx_ring->rsp_prod_pvt - sring->rsp_prod,
  72                            sring->rsp_event,
  73                            sring->rsp_event - sring->rsp_prod);
  74                 seq_printf(m, "pending prod %u pending cons %u nr_pending_reqs %u\n",
  75                            queue->pending_prod,
  76                            queue->pending_cons,
  77                            nr_pending_reqs(queue));
  78                 seq_printf(m, "dealloc prod %u dealloc cons %u dealloc_queue %u\n\n",
  79                            queue->dealloc_prod,
  80                            queue->dealloc_cons,
  81                            queue->dealloc_prod - queue->dealloc_cons);
  82         }
  83
  84         if (rx_ring->sring) {
  85                 struct xen_netif_rx_sring *sring = rx_ring->sring;
  86
  87                 seq_printf(m, "RX: nr_ents %u\n", rx_ring->nr_ents);
  88                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  89                            sring->req_prod,
  90                            sring->req_prod - sring->rsp_prod,
  91                            rx_ring->req_cons,
  92                            rx_ring->req_cons - sring->rsp_prod,
  93                            sring->req_event,
  94                            sring->req_event - sring->rsp_prod);
  95                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n\n",
  96                            sring->rsp_prod,
  97                            rx_ring->rsp_prod_pvt,
  98                            rx_ring->rsp_prod_pvt - sring->rsp_prod,
  99                            sring->rsp_event,
 100                            sring->rsp_event - sring->rsp_prod);
 101         }
 102
 103         seq_printf(m, "NAPI state: %lx NAPI weight: %d TX queue len %u\n"
 104                    "Credit timer_pending: %d, credit: %lu, usec: %lu\n"
 105                    "remaining: %lu, expires: %lu, now: %lu\n",
 106                    queue->napi.state, queue->napi.weight,
 107                    skb_queue_len(&queue->tx_queue),
 108                    timer_pending(&queue->credit_timeout),
 109                    queue->credit_bytes,
 110                    queue->credit_usec,
 111                    queue->remaining_credit,
 112                    queue->credit_timeout.expires,
 113                    jiffies);
 114
 115         return 0;
 116 }
 117
 118 #define XENVIF_KICK_STR "kick"
 119 #define BUFFER_SIZE     32
 120
 121 static ssize_t
 122 xenvif_write_io_ring(struct file *filp, const char __user *buf, size_t count,
 123                      loff_t *ppos)
 124 {
 125         struct xenvif_queue *queue =
 126                 ((struct seq_file *)filp->private_data)->private;
 127         int len;
 128         char write[BUFFER_SIZE];
 129
 130         /* don't allow partial writes and check the length */
 131         if (*ppos != 0)
 132                 return 0;
 133         if (count >= sizeof(write))
 134                 return -ENOSPC;
 135
 136         len = simple_write_to_buffer(write,
 137                                      sizeof(write) - 1,
 138                                      ppos,
 139                                      buf,
 140                                      count);
 141         if (len < 0)
 142                 return len;
 143
 144         write[len] = '\0';
 145
 146         if (!strncmp(write, XENVIF_KICK_STR, sizeof(XENVIF_KICK_STR) - 1))
 147                 xenvif_interrupt(0, (void *)queue);
 148         else {
 149                 pr_warn("Unknown command to io_ring_q%d. Available: kick\n",
 150                         queue->id);
 151                 count = -EINVAL;
 152         }
 153         return count;
 154 }
 155
 156 static int xenvif_dump_open(struct inode *inode, struct file *filp)
 157 {
 158         int ret;
 159         void *queue = NULL;
 160
 161         if (inode->i_private)
 162                 queue = inode->i_private;
 163         ret = single_open(filp, xenvif_read_io_ring, queue);
 164         filp->f_mode |= FMODE_PWRITE;
 165         return ret;
 166 }
 167
 168 static const struct file_operations xenvif_dbg_io_ring_ops_fops = {
 169         .owner = THIS_MODULE,
 170         .open = xenvif_dump_open,
 171         .read = seq_read,
 172         .llseek = seq_lseek,
 173         .release = single_release,
 174         .write = xenvif_write_io_ring,
 175 };
 176
 177 static void xenvif_debugfs_addif(struct xenvif_queue *queue)
 178 {
 179         struct dentry *pfile;
 180         struct xenvif *vif = queue->vif;
 181         int i;
 182
 183         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 184                 return;
 185
 186         vif->xenvif_dbg_root = debugfs_create_dir(vif->dev->name,
 187                                                   xen_netback_dbg_root);
 188         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) {
 189                 for (i = 0; i < vif->num_queues; ++i) {
 190                         char filename[sizeof("io_ring_q") + 4];
 191
 192                         snprintf(filename, sizeof(filename), "io_ring_q%d", i);
 193                         pfile = debugfs_create_file(filename,
 194                                                     S_IRUSR | S_IWUSR,
 195                                                     vif->xenvif_dbg_root,
 196                                                     &vif->queues[i],
 197                                                     &xenvif_dbg_io_ring_ops_fops);
 198                         if (IS_ERR_OR_NULL(pfile))
 199                                 pr_warn("Creation of io_ring file returned %ld!\n",
 200                                         PTR_ERR(pfile));
 201                 }
 202         } else
 203                 netdev_warn(vif->dev,
 204                             "Creation of vif debugfs dir returned %ld!\n",
 205                             PTR_ERR(vif->xenvif_dbg_root));
 206 }
 207
 208 static void xenvif_debugfs_delif(struct xenvif *vif)
 209 {
 210         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 211                 return;
 212
 213         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root))
 214                 debugfs_remove_recursive(vif->xenvif_dbg_root);
 215         vif->xenvif_dbg_root = NULL;
 216 }
 217 #endif /* CONFIG_DEBUG_FS */
 218
 219 static int netback_remove(struct xenbus_device *dev)
 220 {
 221         struct backend_info *be = dev_get_drvdata(&dev->dev);
 222
 223         set_backend_state(be, XenbusStateClosed);
 224
 225         unregister_hotplug_status_watch(be);
 226         if (be->vif) {
 227                 kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE);
 228                 xenbus_rm(XBT_NIL, dev->nodename, "hotplug-status");
 229                 xenvif_free(be->vif);
 230                 be->vif = NULL;
 231         }
 232         kfree(be);
 233         dev_set_drvdata(&dev->dev, NULL);
 234         return 0;
 235 }
 236
 237
 238 /**
 239  * Entry point to this code when a new device is created.  Allocate the basic
 240  * structures and switch to InitWait.
 241  */
 242 static int netback_probe(struct xenbus_device *dev,
 243                          const struct xenbus_device_id *id)
 244 {
 245         const char *message;
 246         struct xenbus_transaction xbt;
 247         int err;
 248         int sg;
 249         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 250                                           GFP_KERNEL);
 251         if (!be) {
 252                 xenbus_dev_fatal(dev, -ENOMEM,
 253                                  "allocating backend structure");
 254                 return -ENOMEM;
 255         }
 256
 257         be->dev = dev;
 258         dev_set_drvdata(&dev->dev, be);
 259
 260         sg = 1;
 261
 262         do {
 263                 err = xenbus_transaction_start(&xbt);
 264                 if (err) {
 265                         xenbus_dev_fatal(dev, err, "starting transaction");
 266                         goto fail;
 267                 }
 268
 269                 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg);
 270                 if (err) {
 271                         message = "writing feature-sg";
 272                         goto abort_transaction;
 273                 }
 274
 275                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4",
 276                                     "%d", sg);
 277                 if (err) {
 278                         message = "writing feature-gso-tcpv4";
 279                         goto abort_transaction;
 280                 }
 281
 282                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6",
 283                                     "%d", sg);
 284                 if (err) {
 285                         message = "writing feature-gso-tcpv6";
 286                         goto abort_transaction;
 287                 }
 288
 289                 /* We support partial checksum setup for IPv6 packets */
 290                 err = xenbus_printf(xbt, dev->nodename,
 291                                     "feature-ipv6-csum-offload",
 292                                     "%d", 1);
 293                 if (err) {
 294                         message = "writing feature-ipv6-csum-offload";
 295                         goto abort_transaction;
 296                 }
 297
 298                 /* We support rx-copy path. */
 299                 err = xenbus_printf(xbt, dev->nodename,
 300                                     "feature-rx-copy", "%d", 1);
 301                 if (err) {
 302                         message = "writing feature-rx-copy";
 303                         goto abort_transaction;
 304                 }
 305
 306                 /*
 307                  * We don't support rx-flip path (except old guests who don't
 308                  * grok this feature flag).
 309                  */
 310                 err = xenbus_printf(xbt, dev->nodename,
 311                                     "feature-rx-flip", "%d", 0);
 312                 if (err) {
 313                         message = "writing feature-rx-flip";
 314                         goto abort_transaction;
 315                 }
 316
 317                 err = xenbus_transaction_end(xbt, 0);
 318         } while (err == -EAGAIN);
 319
 320         if (err) {
 321                 xenbus_dev_fatal(dev, err, "completing transaction");
 322                 goto fail;
 323         }
 324
 325         /*
 326          * Split event channels support, this is optional so it is not
 327          * put inside the above loop.
 328          */
 329         err = xenbus_printf(XBT_NIL, dev->nodename,
 330                             "feature-split-event-channels",
 331                             "%u", separate_tx_rx_irq);
 332         if (err)
 333                 pr_debug("Error writing feature-split-event-channels\n");
 334
 335         /* Multi-queue support: This is an optional feature. */
 336         err = xenbus_printf(XBT_NIL, dev->nodename,
 337                             "multi-queue-max-queues", "%u", xenvif_max_queues);
 338         if (err)
 339                 pr_debug("Error writing multi-queue-max-queues\n");
 340
 341         err = xenbus_switch_state(dev, XenbusStateInitWait);
 342         if (err)
 343                 goto fail;
 344
 345         be->state = XenbusStateInitWait;
 346
 347         /* This kicks hotplug scripts, so do it immediately. */
 348         backend_create_xenvif(be);
 349
 350         return 0;
 351
 352 abort_transaction:
 353         xenbus_transaction_end(xbt, 1);
 354         xenbus_dev_fatal(dev, err, "%s", message);
 355 fail:
 356         pr_debug("failed\n");
 357         netback_remove(dev);
 358         return err;
 359 }
 360
 361
 362 /*
 363  * Handle the creation of the hotplug script environment.  We add the script
 364  * and vif variables to the environment, for the benefit of the vif-* hotplug
 365  * scripts.
 366  */
 367 static int netback_uevent(struct xenbus_device *xdev,
 368                           struct kobj_uevent_env *env)
 369 {
 370         struct backend_info *be = dev_get_drvdata(&xdev->dev);
 371         char *val;
 372
 373         val = xenbus_read(XBT_NIL, xdev->nodename, "script", NULL);
 374         if (IS_ERR(val)) {
 375                 int err = PTR_ERR(val);
 376                 xenbus_dev_fatal(xdev, err, "reading script");
 377                 return err;
 378         } else {
 379                 if (add_uevent_var(env, "script=%s", val)) {
 380                         kfree(val);
 381                         return -ENOMEM;
 382                 }
 383                 kfree(val);
 384         }
 385
 386         if (!be || !be->vif)
 387                 return 0;
 388
 389         return add_uevent_var(env, "vif=%s", be->vif->dev->name);
 390 }
 391
 392
 393 static void backend_create_xenvif(struct backend_info *be)
 394 {
 395         int err;
 396         long handle;
 397         struct xenbus_device *dev = be->dev;
 398
 399         if (be->vif != NULL)
 400                 return;
 401
 402         err = xenbus_scanf(XBT_NIL, dev->nodename, "handle", "%li", &handle);
 403         if (err != 1) {
 404                 xenbus_dev_fatal(dev, err, "reading handle");
 405                 return;
 406         }
 407
 408         be->vif = xenvif_alloc(&dev->dev, dev->otherend_id, handle);
 409         if (IS_ERR(be->vif)) {
 410                 err = PTR_ERR(be->vif);
 411                 be->vif = NULL;
 412                 xenbus_dev_fatal(dev, err, "creating interface");
 413                 return;
 414         }
 415
 416         kobject_uevent(&dev->dev.kobj, KOBJ_ONLINE);
 417 }
 418
 419 static void backend_disconnect(struct backend_info *be)
 420 {
 421         if (be->vif) {
 422 #ifdef CONFIG_DEBUG_FS
 423                 xenvif_debugfs_delif(be->vif);
 424 #endif /* CONFIG_DEBUG_FS */
 425                 xenvif_disconnect(be->vif);
 426         }
 427 }
 428
 429 static void backend_connect(struct backend_info *be)
 430 {
 431         if (be->vif)
 432                 connect(be);
 433 }
 434
 435 static inline void backend_switch_state(struct backend_info *be,
 436                                         enum xenbus_state state)
 437 {
 438         struct xenbus_device *dev = be->dev;
 439
 440         pr_debug("%s -> %s\n", dev->nodename, xenbus_strstate(state));
 441         be->state = state;
 442
 443         /* If we are waiting for a hotplug script then defer the
 444          * actual xenbus state change.
 445          */
 446         if (!be->have_hotplug_status_watch)
 447                 xenbus_switch_state(dev, state);
 448 }
 449
 450 /* Handle backend state transitions:
 451  *
 452  * The backend state starts in InitWait and the following transitions are
 453  * allowed.
 454  *
 455  * InitWait -> Connected
 456  *
 457  *    ^    \         |
 458  *    |     \        |
 459  *    |      \       |
 460  *    |       \      |
 461  *    |        \     |
 462  *    |         \    |
 463  *    |          V   V
 464  *
 465  *  Closed  <-> Closing
 466  *
 467  * The state argument specifies the eventual state of the backend and the
 468  * function transitions to that state via the shortest path.
 469  */
 470 static void set_backend_state(struct backend_info *be,
 471                               enum xenbus_state state)
 472 {
 473         while (be->state != state) {
 474                 switch (be->state) {
 475                 case XenbusStateClosed:
 476                         switch (state) {
 477                         case XenbusStateInitWait:
 478                         case XenbusStateConnected:
 479                                 pr_info("%s: prepare for reconnect\n",
 480                                         be->dev->nodename);
 481                                 backend_switch_state(be, XenbusStateInitWait);
 482                                 break;
 483                         case XenbusStateClosing:
 484                                 backend_switch_state(be, XenbusStateClosing);
 485                                 break;
 486                         default:
 487                                 BUG();
 488                         }
 489                         break;
 490                 case XenbusStateInitWait:
 491                         switch (state) {
 492                         case XenbusStateConnected:
 493                                 backend_connect(be);
 494                                 backend_switch_state(be, XenbusStateConnected);
 495                                 break;
 496                         case XenbusStateClosing:
 497                         case XenbusStateClosed:
 498                                 backend_switch_state(be, XenbusStateClosing);
 499                                 break;
 500                         default:
 501                                 BUG();
 502                         }
 503                         break;
 504                 case XenbusStateConnected:
 505                         switch (state) {
 506                         case XenbusStateInitWait:
 507                         case XenbusStateClosing:
 508                         case XenbusStateClosed:
 509                                 backend_disconnect(be);
 510                                 backend_switch_state(be, XenbusStateClosing);
 511                                 break;
 512                         default:
 513                                 BUG();
 514                         }
 515                         break;
 516                 case XenbusStateClosing:
 517                         switch (state) {
 518                         case XenbusStateInitWait:
 519                         case XenbusStateConnected:
 520                         case XenbusStateClosed:
 521                                 backend_switch_state(be, XenbusStateClosed);
 522                                 break;
 523                         default:
 524                                 BUG();
 525                         }
 526                         break;
 527                 default:
 528                         BUG();
 529                 }
 530         }
 531 }
 532
 533 /**
 534  * Callback received when the frontend's state changes.
 535  */
 536 static void frontend_changed(struct xenbus_device *dev,
 537                              enum xenbus_state frontend_state)
 538 {
 539         struct backend_info *be = dev_get_drvdata(&dev->dev);
 540
 541         pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state));
 542
 543         be->frontend_state = frontend_state;
 544
 545         switch (frontend_state) {
 546         case XenbusStateInitialising:
 547                 set_backend_state(be, XenbusStateInitWait);
 548                 break;
 549
 550         case XenbusStateInitialised:
 551                 break;
 552
 553         case XenbusStateConnected:
 554                 set_backend_state(be, XenbusStateConnected);
 555                 break;
 556
 557         case XenbusStateClosing:
 558                 set_backend_state(be, XenbusStateClosing);
 559                 break;
 560
 561         case XenbusStateClosed:
 562                 set_backend_state(be, XenbusStateClosed);
 563                 if (xenbus_dev_is_online(dev))
 564                         break;
 565                 /* fall through if not online */
 566         case XenbusStateUnknown:
 567                 set_backend_state(be, XenbusStateClosed);
 568                 device_unregister(&dev->dev);
 569                 break;
 570
 571         default:
 572                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 573                                  frontend_state);
 574                 break;
 575         }
 576 }
 577
 578
 579 static void xen_net_read_rate(struct xenbus_device *dev,
 580                               unsigned long *bytes, unsigned long *usec)
 581 {
 582         char *s, *e;
 583         unsigned long b, u;
 584         char *ratestr;
 585
 586         /* Default to unlimited bandwidth. */
 587         *bytes = ~0UL;
 588         *usec = 0;
 589
 590         ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL);
 591         if (IS_ERR(ratestr))
 592                 return;
 593
 594         s = ratestr;
 595         b = simple_strtoul(s, &e, 10);
 596         if ((s == e) || (*e != ','))
 597                 goto fail;
 598
 599         s = e + 1;
 600         u = simple_strtoul(s, &e, 10);
 601         if ((s == e) || (*e != '\0'))
 602                 goto fail;
 603
 604         *bytes = b;
 605         *usec = u;
 606
 607         kfree(ratestr);
 608         return;
 609
 610  fail:
 611         pr_warn("Failed to parse network rate limit. Traffic unlimited.\n");
 612         kfree(ratestr);
 613 }
 614
 615 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
 616 {
 617         char *s, *e, *macstr;
 618         int i;
 619
 620         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
 621         if (IS_ERR(macstr))
 622                 return PTR_ERR(macstr);
 623
 624         for (i = 0; i < ETH_ALEN; i++) {
 625                 mac[i] = simple_strtoul(s, &e, 16);
 626                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
 627                         kfree(macstr);
 628                         return -ENOENT;
 629                 }
 630                 s = e+1;
 631         }
 632
 633         kfree(macstr);
 634         return 0;
 635 }
 636
 637 static void unregister_hotplug_status_watch(struct backend_info *be)
 638 {
 639         if (be->have_hotplug_status_watch) {
 640                 unregister_xenbus_watch(&be->hotplug_status_watch);
 641                 kfree(be->hotplug_status_watch.node);
 642         }
 643         be->have_hotplug_status_watch = 0;
 644 }
 645
 646 static void hotplug_status_changed(struct xenbus_watch *watch,
 647                                    const char **vec,
 648                                    unsigned int vec_size)
 649 {
 650         struct backend_info *be = container_of(watch,
 651                                                struct backend_info,
 652                                                hotplug_status_watch);
 653         char *str;
 654         unsigned int len;
 655
 656         str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len);
 657         if (IS_ERR(str))
 658                 return;
 659         if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) {
 660                 /* Complete any pending state change */
 661                 xenbus_switch_state(be->dev, be->state);
 662
 663                 /* Not interested in this watch anymore. */
 664                 unregister_hotplug_status_watch(be);
 665         }
 666         kfree(str);
 667 }
 668
 669 static void connect(struct backend_info *be)
 670 {
 671         int err;
 672         struct xenbus_device *dev = be->dev;
 673         unsigned long credit_bytes, credit_usec;
 674         unsigned int queue_index;
 675         unsigned int requested_num_queues;
 676         struct xenvif_queue *queue;
 677
 678         /* Check whether the frontend requested multiple queues
 679          * and read the number requested.
 680          */
 681         err = xenbus_scanf(XBT_NIL, dev->otherend,
 682                            "multi-queue-num-queues",
 683                            "%u", &requested_num_queues);
 684         if (err < 0) {
 685                 requested_num_queues = 1; /* Fall back to single queue */
 686         } else if (requested_num_queues > xenvif_max_queues) {
 687                 /* buggy or malicious guest */
 688                 xenbus_dev_fatal(dev, err,
 689                                  "guest requested %u queues, exceeding the maximum of %u.",
 690                                  requested_num_queues, xenvif_max_queues);
 691                 return;
 692         }
 693
 694         err = xen_net_read_mac(dev, be->vif->fe_dev_addr);
 695         if (err) {
 696                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
 697                 return;
 698         }
 699
 700         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 701         read_xenbus_vif_flags(be);
 702
 703         /* Use the number of queues requested by the frontend */
 704         be->vif->queues = vzalloc(requested_num_queues *
 705                                   sizeof(struct xenvif_queue));
 706         be->vif->num_queues = requested_num_queues;
 707
 708         for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) {
 709                 queue = &be->vif->queues[queue_index];
 710                 queue->vif = be->vif;
 711                 queue->id = queue_index;
 712                 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
 713                                 be->vif->dev->name, queue->id);
 714
 715                 err = xenvif_init_queue(queue);
 716                 if (err) {
 717                         /* xenvif_init_queue() cleans up after itself on
 718                          * failure, but we need to clean up any previously
 719                          * initialised queues. Set num_queues to i so that
 720                          * earlier queues can be destroyed using the regular
 721                          * disconnect logic.
 722                          */
 723                         be->vif->num_queues = queue_index;
 724                         goto err;
 725                 }
 726
 727                 queue->remaining_credit = credit_bytes;
 728
 729                 err = connect_rings(be, queue);
 730                 if (err) {
 731                         /* connect_rings() cleans up after itself on failure,
 732                          * but we need to clean up after xenvif_init_queue() here,
 733                          * and also clean up any previously initialised queues.
 734                          */
 735                         xenvif_deinit_queue(queue);
 736                         be->vif->num_queues = queue_index;
 737                         goto err;
 738                 }
 739 #ifdef CONFIG_DEBUG_FS
 740                 xenvif_debugfs_addif(queue);
 741 #endif /* CONFIG_DEBUG_FS */
 742         }
 743
 744         /* Initialisation completed, tell core driver the number of
 745          * active queues.
 746          */
 747         rtnl_lock();
 748         netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues);
 749         netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues);
 750         rtnl_unlock();
 751
 752         xenvif_carrier_on(be->vif);
 753
 754         unregister_hotplug_status_watch(be);
 755         err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch,
 756                                    hotplug_status_changed,
 757                                    "%s/%s", dev->nodename, "hotplug-status");
 758         if (!err)
 759                 be->have_hotplug_status_watch = 1;
 760
 761         netif_tx_wake_all_queues(be->vif->dev);
 762
 763         return;
 764
 765 err:
 766         if (be->vif->num_queues > 0)
 767                 xenvif_disconnect(be->vif); /* Clean up existing queues */
 768         vfree(be->vif->queues);
 769         be->vif->queues = NULL;
 770         be->vif->num_queues = 0;
 771         return;
 772 }
 773
 774
 775 static int connect_rings(struct backend_info *be, struct xenvif_queue *queue)
 776 {
 777         struct xenbus_device *dev = be->dev;
 778         unsigned int num_queues = queue->vif->num_queues;
 779         unsigned long tx_ring_ref, rx_ring_ref;
 780         unsigned int tx_evtchn, rx_evtchn;
 781         int err;
 782         char *xspath;
 783         size_t xspathsize;
 784         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
 785
 786         /* If the frontend requested 1 queue, or we have fallen back
 787          * to single queue due to lack of frontend support for multi-
 788          * queue, expect the remaining XenStore keys in the toplevel
 789          * directory. Otherwise, expect them in a subdirectory called
 790          * queue-N.
 791          */
 792         if (num_queues == 1) {
 793                 xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL);
 794                 if (!xspath) {
 795                         xenbus_dev_fatal(dev, -ENOMEM,
 796                                          "reading ring references");
 797                         return -ENOMEM;
 798                 }
 799                 strcpy(xspath, dev->otherend);
 800         } else {
 801                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
 802                 xspath = kzalloc(xspathsize, GFP_KERNEL);
 803                 if (!xspath) {
 804                         xenbus_dev_fatal(dev, -ENOMEM,
 805                                          "reading ring references");
 806                         return -ENOMEM;
 807                 }
 808                 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend,
 809                          queue->id);
 810         }
 811
 812         err = xenbus_gather(XBT_NIL, xspath,
 813                             "tx-ring-ref", "%lu", &tx_ring_ref,
 814                             "rx-ring-ref", "%lu", &rx_ring_ref, NULL);
 815         if (err) {
 816                 xenbus_dev_fatal(dev, err,
 817                                  "reading %s/ring-ref",
 818                                  xspath);
 819                 goto err;
 820         }
 821
 822         /* Try split event channels first, then single event channel. */
 823         err = xenbus_gather(XBT_NIL, xspath,
 824                             "event-channel-tx", "%u", &tx_evtchn,
 825                             "event-channel-rx", "%u", &rx_evtchn, NULL);
 826         if (err < 0) {
 827                 err = xenbus_scanf(XBT_NIL, xspath,
 828                                    "event-channel", "%u", &tx_evtchn);
 829                 if (err < 0) {
 830                         xenbus_dev_fatal(dev, err,
 831                                          "reading %s/event-channel(-tx/rx)",
 832                                          xspath);
 833                         goto err;
 834                 }
 835                 rx_evtchn = tx_evtchn;
 836         }
 837
 838         /* Map the shared frame, irq etc. */
 839         err = xenvif_connect(queue, tx_ring_ref, rx_ring_ref,
 840                              tx_evtchn, rx_evtchn);
 841         if (err) {
 842                 xenbus_dev_fatal(dev, err,
 843                                  "mapping shared-frames %lu/%lu port tx %u rx %u",
 844                                  tx_ring_ref, rx_ring_ref,
 845                                  tx_evtchn, rx_evtchn);
 846                 goto err;
 847         }
 848
 849         err = 0;
 850 err: /* Regular return falls through with err == 0 */
 851         kfree(xspath);
 852         return err;
 853 }
 854
 855 static int read_xenbus_vif_flags(struct backend_info *be)
 856 {
 857         struct xenvif *vif = be->vif;
 858         struct xenbus_device *dev = be->dev;
 859         unsigned int rx_copy;
 860         int err, val;
 861
 862         err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u",
 863                            &rx_copy);
 864         if (err == -ENOENT) {
 865                 err = 0;
 866                 rx_copy = 0;
 867         }
 868         if (err < 0) {
 869                 xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy",
 870                                  dev->otherend);
 871                 return err;
 872         }
 873         if (!rx_copy)
 874                 return -EOPNOTSUPP;
 875
 876         if (vif->dev->tx_queue_len != 0) {
 877                 if (xenbus_scanf(XBT_NIL, dev->otherend,
 878                                  "feature-rx-notify", "%d", &val) < 0)
 879                         val = 0;
 880                 if (val)
 881                         vif->can_queue = 1;
 882                 else
 883                         /* Must be non-zero for pfifo_fast to work. */
 884                         vif->dev->tx_queue_len = 1;
 885         }
 886
 887         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-sg",
 888                          "%d", &val) < 0)
 889                 val = 0;
 890         vif->can_sg = !!val;
 891
 892         vif->gso_mask = 0;
 893         vif->gso_prefix_mask = 0;
 894
 895         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4",
 896                          "%d", &val) < 0)
 897                 val = 0;
 898         if (val)
 899                 vif->gso_mask |= GSO_BIT(TCPV4);
 900
 901         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4-prefix",
 902                          "%d", &val) < 0)
 903                 val = 0;
 904         if (val)
 905                 vif->gso_prefix_mask |= GSO_BIT(TCPV4);
 906
 907         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6",
 908                          "%d", &val) < 0)
 909                 val = 0;
 910         if (val)
 911                 vif->gso_mask |= GSO_BIT(TCPV6);
 912
 913         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6-prefix",
 914                          "%d", &val) < 0)
 915                 val = 0;
 916         if (val)
 917                 vif->gso_prefix_mask |= GSO_BIT(TCPV6);
 918
 919         if (vif->gso_mask & vif->gso_prefix_mask) {
 920                 xenbus_dev_fatal(dev, err,
 921                                  "%s: gso and gso prefix flags are not "
 922                                  "mutually exclusive",
 923                                  dev->otherend);
 924                 return -EOPNOTSUPP;
 925         }
 926
 927         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-no-csum-offload",
 928                          "%d", &val) < 0)
 929                 val = 0;
 930         vif->ip_csum = !val;
 931
 932         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-ipv6-csum-offload",
 933                          "%d", &val) < 0)
 934                 val = 0;
 935         vif->ipv6_csum = !!val;
 936
 937         return 0;
 938 }
 939
 940
 941 /* ** Driver Registration ** */
 942
 943
 944 static const struct xenbus_device_id netback_ids[] = {
 945         { "vif" },
 946         { "" }
 947 };
 948
 949
 950 static DEFINE_XENBUS_DRIVER(netback, ,
 951         .probe = netback_probe,
 952         .remove = netback_remove,
 953         .uevent = netback_uevent,
 954         .otherend_changed = frontend_changed,
 955 );
 956
 957 int xenvif_xenbus_init(void)
 958 {
 959         return xenbus_register_backend(&netback_driver);
 960 }
 961
 962 void xenvif_xenbus_fini(void)
 963 {
 964         return xenbus_unregister_driver(&netback_driver);
 965 }