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 int 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         struct netdev_queue *dev_queue;
  56
  57         if (tx_ring->sring) {
  58                 struct xen_netif_tx_sring *sring = tx_ring->sring;
  59
  60                 seq_printf(m, "Queue %d\nTX: nr_ents %u\n", queue->id,
  61                            tx_ring->nr_ents);
  62                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  63                            sring->req_prod,
  64                            sring->req_prod - sring->rsp_prod,
  65                            tx_ring->req_cons,
  66                            tx_ring->req_cons - sring->rsp_prod,
  67                            sring->req_event,
  68                            sring->req_event - sring->rsp_prod);
  69                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n",
  70                            sring->rsp_prod,
  71                            tx_ring->rsp_prod_pvt,
  72                            tx_ring->rsp_prod_pvt - sring->rsp_prod,
  73                            sring->rsp_event,
  74                            sring->rsp_event - sring->rsp_prod);
  75                 seq_printf(m, "pending prod %u pending cons %u nr_pending_reqs %u\n",
  76                            queue->pending_prod,
  77                            queue->pending_cons,
  78                            nr_pending_reqs(queue));
  79                 seq_printf(m, "dealloc prod %u dealloc cons %u dealloc_queue %u\n\n",
  80                            queue->dealloc_prod,
  81                            queue->dealloc_cons,
  82                            queue->dealloc_prod - queue->dealloc_cons);
  83         }
  84
  85         if (rx_ring->sring) {
  86                 struct xen_netif_rx_sring *sring = rx_ring->sring;
  87
  88                 seq_printf(m, "RX: nr_ents %u\n", rx_ring->nr_ents);
  89                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  90                            sring->req_prod,
  91                            sring->req_prod - sring->rsp_prod,
  92                            rx_ring->req_cons,
  93                            rx_ring->req_cons - sring->rsp_prod,
  94                            sring->req_event,
  95                            sring->req_event - sring->rsp_prod);
  96                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n\n",
  97                            sring->rsp_prod,
  98                            rx_ring->rsp_prod_pvt,
  99                            rx_ring->rsp_prod_pvt - sring->rsp_prod,
 100                            sring->rsp_event,
 101                            sring->rsp_event - sring->rsp_prod);
 102         }
 103
 104         seq_printf(m, "NAPI state: %lx NAPI weight: %d TX queue len %u\n"
 105                    "Credit timer_pending: %d, credit: %lu, usec: %lu\n"
 106                    "remaining: %lu, expires: %lu, now: %lu\n",
 107                    queue->napi.state, queue->napi.weight,
 108                    skb_queue_len(&queue->tx_queue),
 109                    timer_pending(&queue->credit_timeout),
 110                    queue->credit_bytes,
 111                    queue->credit_usec,
 112                    queue->remaining_credit,
 113                    queue->credit_timeout.expires,
 114                    jiffies);
 115
 116         dev_queue = netdev_get_tx_queue(queue->vif->dev, queue->id);
 117
 118         seq_printf(m, "\nRx internal queue: len %u max %u pkts %u %s\n",
 119                    queue->rx_queue_len, queue->rx_queue_max,
 120                    skb_queue_len(&queue->rx_queue),
 121                    netif_tx_queue_stopped(dev_queue) ? "stopped" : "running");
 122
 123         return 0;
 124 }
 125
 126 #define XENVIF_KICK_STR "kick"
 127 #define BUFFER_SIZE     32
 128
 129 static ssize_t
 130 xenvif_write_io_ring(struct file *filp, const char __user *buf, size_t count,
 131                      loff_t *ppos)
 132 {
 133         struct xenvif_queue *queue =
 134                 ((struct seq_file *)filp->private_data)->private;
 135         int len;
 136         char write[BUFFER_SIZE];
 137
 138         /* don't allow partial writes and check the length */
 139         if (*ppos != 0)
 140                 return 0;
 141         if (count >= sizeof(write))
 142                 return -ENOSPC;
 143
 144         len = simple_write_to_buffer(write,
 145                                      sizeof(write) - 1,
 146                                      ppos,
 147                                      buf,
 148                                      count);
 149         if (len < 0)
 150                 return len;
 151
 152         write[len] = '\0';
 153
 154         if (!strncmp(write, XENVIF_KICK_STR, sizeof(XENVIF_KICK_STR) - 1))
 155                 xenvif_interrupt(0, (void *)queue);
 156         else {
 157                 pr_warn("Unknown command to io_ring_q%d. Available: kick\n",
 158                         queue->id);
 159                 count = -EINVAL;
 160         }
 161         return count;
 162 }
 163
 164 static int xenvif_dump_open(struct inode *inode, struct file *filp)
 165 {
 166         int ret;
 167         void *queue = NULL;
 168
 169         if (inode->i_private)
 170                 queue = inode->i_private;
 171         ret = single_open(filp, xenvif_read_io_ring, queue);
 172         filp->f_mode |= FMODE_PWRITE;
 173         return ret;
 174 }
 175
 176 static const struct file_operations xenvif_dbg_io_ring_ops_fops = {
 177         .owner = THIS_MODULE,
 178         .open = xenvif_dump_open,
 179         .read = seq_read,
 180         .llseek = seq_lseek,
 181         .release = single_release,
 182         .write = xenvif_write_io_ring,
 183 };
 184
 185 static void xenvif_debugfs_addif(struct xenvif *vif)
 186 {
 187         struct dentry *pfile;
 188         int i;
 189
 190         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 191                 return;
 192
 193         vif->xenvif_dbg_root = debugfs_create_dir(vif->dev->name,
 194                                                   xen_netback_dbg_root);
 195         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) {
 196                 for (i = 0; i < vif->num_queues; ++i) {
 197                         char filename[sizeof("io_ring_q") + 4];
 198
 199                         snprintf(filename, sizeof(filename), "io_ring_q%d", i);
 200                         pfile = debugfs_create_file(filename,
 201                                                     S_IRUSR | S_IWUSR,
 202                                                     vif->xenvif_dbg_root,
 203                                                     &vif->queues[i],
 204                                                     &xenvif_dbg_io_ring_ops_fops);
 205                         if (IS_ERR_OR_NULL(pfile))
 206                                 pr_warn("Creation of io_ring file returned %ld!\n",
 207                                         PTR_ERR(pfile));
 208                 }
 209         } else
 210                 netdev_warn(vif->dev,
 211                             "Creation of vif debugfs dir returned %ld!\n",
 212                             PTR_ERR(vif->xenvif_dbg_root));
 213 }
 214
 215 static void xenvif_debugfs_delif(struct xenvif *vif)
 216 {
 217         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 218                 return;
 219
 220         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root))
 221                 debugfs_remove_recursive(vif->xenvif_dbg_root);
 222         vif->xenvif_dbg_root = NULL;
 223 }
 224 #endif /* CONFIG_DEBUG_FS */
 225
 226 static int netback_remove(struct xenbus_device *dev)
 227 {
 228         struct backend_info *be = dev_get_drvdata(&dev->dev);
 229
 230         set_backend_state(be, XenbusStateClosed);
 231
 232         unregister_hotplug_status_watch(be);
 233         if (be->vif) {
 234                 kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE);
 235                 xenbus_rm(XBT_NIL, dev->nodename, "hotplug-status");
 236                 xenvif_free(be->vif);
 237                 be->vif = NULL;
 238         }
 239         kfree(be);
 240         dev_set_drvdata(&dev->dev, NULL);
 241         return 0;
 242 }
 243
 244
 245 /**
 246  * Entry point to this code when a new device is created.  Allocate the basic
 247  * structures and switch to InitWait.
 248  */
 249 static int netback_probe(struct xenbus_device *dev,
 250                          const struct xenbus_device_id *id)
 251 {
 252         const char *message;
 253         struct xenbus_transaction xbt;
 254         int err;
 255         int sg;
 256         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 257                                           GFP_KERNEL);
 258         if (!be) {
 259                 xenbus_dev_fatal(dev, -ENOMEM,
 260                                  "allocating backend structure");
 261                 return -ENOMEM;
 262         }
 263
 264         be->dev = dev;
 265         dev_set_drvdata(&dev->dev, be);
 266
 267         sg = 1;
 268
 269         do {
 270                 err = xenbus_transaction_start(&xbt);
 271                 if (err) {
 272                         xenbus_dev_fatal(dev, err, "starting transaction");
 273                         goto fail;
 274                 }
 275
 276                 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg);
 277                 if (err) {
 278                         message = "writing feature-sg";
 279                         goto abort_transaction;
 280                 }
 281
 282                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4",
 283                                     "%d", sg);
 284                 if (err) {
 285                         message = "writing feature-gso-tcpv4";
 286                         goto abort_transaction;
 287                 }
 288
 289                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6",
 290                                     "%d", sg);
 291                 if (err) {
 292                         message = "writing feature-gso-tcpv6";
 293                         goto abort_transaction;
 294                 }
 295
 296                 /* We support partial checksum setup for IPv6 packets */
 297                 err = xenbus_printf(xbt, dev->nodename,
 298                                     "feature-ipv6-csum-offload",
 299                                     "%d", 1);
 300                 if (err) {
 301                         message = "writing feature-ipv6-csum-offload";
 302                         goto abort_transaction;
 303                 }
 304
 305                 /* We support rx-copy path. */
 306                 err = xenbus_printf(xbt, dev->nodename,
 307                                     "feature-rx-copy", "%d", 1);
 308                 if (err) {
 309                         message = "writing feature-rx-copy";
 310                         goto abort_transaction;
 311                 }
 312
 313                 /*
 314                  * We don't support rx-flip path (except old guests who don't
 315                  * grok this feature flag).
 316                  */
 317                 err = xenbus_printf(xbt, dev->nodename,
 318                                     "feature-rx-flip", "%d", 0);
 319                 if (err) {
 320                         message = "writing feature-rx-flip";
 321                         goto abort_transaction;
 322                 }
 323
 324                 err = xenbus_transaction_end(xbt, 0);
 325         } while (err == -EAGAIN);
 326
 327         if (err) {
 328                 xenbus_dev_fatal(dev, err, "completing transaction");
 329                 goto fail;
 330         }
 331
 332         /*
 333          * Split event channels support, this is optional so it is not
 334          * put inside the above loop.
 335          */
 336         err = xenbus_printf(XBT_NIL, dev->nodename,
 337                             "feature-split-event-channels",
 338                             "%u", separate_tx_rx_irq);
 339         if (err)
 340                 pr_debug("Error writing feature-split-event-channels\n");
 341
 342         /* Multi-queue support: This is an optional feature. */
 343         err = xenbus_printf(XBT_NIL, dev->nodename,
 344                             "multi-queue-max-queues", "%u", xenvif_max_queues);
 345         if (err)
 346                 pr_debug("Error writing multi-queue-max-queues\n");
 347
 348         err = xenbus_switch_state(dev, XenbusStateInitWait);
 349         if (err)
 350                 goto fail;
 351
 352         be->state = XenbusStateInitWait;
 353
 354         /* This kicks hotplug scripts, so do it immediately. */
 355         err = backend_create_xenvif(be);
 356         if (err)
 357                 goto fail;
 358
 359         return 0;
 360
 361 abort_transaction:
 362         xenbus_transaction_end(xbt, 1);
 363         xenbus_dev_fatal(dev, err, "%s", message);
 364 fail:
 365         pr_debug("failed\n");
 366         netback_remove(dev);
 367         return err;
 368 }
 369
 370
 371 /*
 372  * Handle the creation of the hotplug script environment.  We add the script
 373  * and vif variables to the environment, for the benefit of the vif-* hotplug
 374  * scripts.
 375  */
 376 static int netback_uevent(struct xenbus_device *xdev,
 377                           struct kobj_uevent_env *env)
 378 {
 379         struct backend_info *be = dev_get_drvdata(&xdev->dev);
 380         char *val;
 381
 382         val = xenbus_read(XBT_NIL, xdev->nodename, "script", NULL);
 383         if (IS_ERR(val)) {
 384                 int err = PTR_ERR(val);
 385                 xenbus_dev_fatal(xdev, err, "reading script");
 386                 return err;
 387         } else {
 388                 if (add_uevent_var(env, "script=%s", val)) {
 389                         kfree(val);
 390                         return -ENOMEM;
 391                 }
 392                 kfree(val);
 393         }
 394
 395         if (!be || !be->vif)
 396                 return 0;
 397
 398         return add_uevent_var(env, "vif=%s", be->vif->dev->name);
 399 }
 400
 401
 402 static int backend_create_xenvif(struct backend_info *be)
 403 {
 404         int err;
 405         long handle;
 406         struct xenbus_device *dev = be->dev;
 407
 408         if (be->vif != NULL)
 409                 return 0;
 410
 411         err = xenbus_scanf(XBT_NIL, dev->nodename, "handle", "%li", &handle);
 412         if (err != 1) {
 413                 xenbus_dev_fatal(dev, err, "reading handle");
 414                 return (err < 0) ? err : -EINVAL;
 415         }
 416
 417         be->vif = xenvif_alloc(&dev->dev, dev->otherend_id, handle);
 418         if (IS_ERR(be->vif)) {
 419                 err = PTR_ERR(be->vif);
 420                 be->vif = NULL;
 421                 xenbus_dev_fatal(dev, err, "creating interface");
 422                 return err;
 423         }
 424
 425         kobject_uevent(&dev->dev.kobj, KOBJ_ONLINE);
 426         return 0;
 427 }
 428
 429 static void backend_disconnect(struct backend_info *be)
 430 {
 431         if (be->vif) {
 432 #ifdef CONFIG_DEBUG_FS
 433                 xenvif_debugfs_delif(be->vif);
 434 #endif /* CONFIG_DEBUG_FS */
 435                 xenvif_disconnect(be->vif);
 436         }
 437 }
 438
 439 static void backend_connect(struct backend_info *be)
 440 {
 441         if (be->vif)
 442                 connect(be);
 443 }
 444
 445 static inline void backend_switch_state(struct backend_info *be,
 446                                         enum xenbus_state state)
 447 {
 448         struct xenbus_device *dev = be->dev;
 449
 450         pr_debug("%s -> %s\n", dev->nodename, xenbus_strstate(state));
 451         be->state = state;
 452
 453         /* If we are waiting for a hotplug script then defer the
 454          * actual xenbus state change.
 455          */
 456         if (!be->have_hotplug_status_watch)
 457                 xenbus_switch_state(dev, state);
 458 }
 459
 460 /* Handle backend state transitions:
 461  *
 462  * The backend state starts in InitWait and the following transitions are
 463  * allowed.
 464  *
 465  * InitWait -> Connected
 466  *
 467  *    ^    \         |
 468  *    |     \        |
 469  *    |      \       |
 470  *    |       \      |
 471  *    |        \     |
 472  *    |         \    |
 473  *    |          V   V
 474  *
 475  *  Closed  <-> Closing
 476  *
 477  * The state argument specifies the eventual state of the backend and the
 478  * function transitions to that state via the shortest path.
 479  */
 480 static void set_backend_state(struct backend_info *be,
 481                               enum xenbus_state state)
 482 {
 483         while (be->state != state) {
 484                 switch (be->state) {
 485                 case XenbusStateClosed:
 486                         switch (state) {
 487                         case XenbusStateInitWait:
 488                         case XenbusStateConnected:
 489                                 pr_info("%s: prepare for reconnect\n",
 490                                         be->dev->nodename);
 491                                 backend_switch_state(be, XenbusStateInitWait);
 492                                 break;
 493                         case XenbusStateClosing:
 494                                 backend_switch_state(be, XenbusStateClosing);
 495                                 break;
 496                         default:
 497                                 BUG();
 498                         }
 499                         break;
 500                 case XenbusStateInitWait:
 501                         switch (state) {
 502                         case XenbusStateConnected:
 503                                 backend_connect(be);
 504                                 backend_switch_state(be, XenbusStateConnected);
 505                                 break;
 506                         case XenbusStateClosing:
 507                         case XenbusStateClosed:
 508                                 backend_switch_state(be, XenbusStateClosing);
 509                                 break;
 510                         default:
 511                                 BUG();
 512                         }
 513                         break;
 514                 case XenbusStateConnected:
 515                         switch (state) {
 516                         case XenbusStateInitWait:
 517                         case XenbusStateClosing:
 518                         case XenbusStateClosed:
 519                                 backend_disconnect(be);
 520                                 backend_switch_state(be, XenbusStateClosing);
 521                                 break;
 522                         default:
 523                                 BUG();
 524                         }
 525                         break;
 526                 case XenbusStateClosing:
 527                         switch (state) {
 528                         case XenbusStateInitWait:
 529                         case XenbusStateConnected:
 530                         case XenbusStateClosed:
 531                                 backend_switch_state(be, XenbusStateClosed);
 532                                 break;
 533                         default:
 534                                 BUG();
 535                         }
 536                         break;
 537                 default:
 538                         BUG();
 539                 }
 540         }
 541 }
 542
 543 /**
 544  * Callback received when the frontend's state changes.
 545  */
 546 static void frontend_changed(struct xenbus_device *dev,
 547                              enum xenbus_state frontend_state)
 548 {
 549         struct backend_info *be = dev_get_drvdata(&dev->dev);
 550
 551         pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state));
 552
 553         be->frontend_state = frontend_state;
 554
 555         switch (frontend_state) {
 556         case XenbusStateInitialising:
 557                 set_backend_state(be, XenbusStateInitWait);
 558                 break;
 559
 560         case XenbusStateInitialised:
 561                 break;
 562
 563         case XenbusStateConnected:
 564                 set_backend_state(be, XenbusStateConnected);
 565                 break;
 566
 567         case XenbusStateClosing:
 568                 set_backend_state(be, XenbusStateClosing);
 569                 break;
 570
 571         case XenbusStateClosed:
 572                 set_backend_state(be, XenbusStateClosed);
 573                 if (xenbus_dev_is_online(dev))
 574                         break;
 575                 /* fall through if not online */
 576         case XenbusStateUnknown:
 577                 set_backend_state(be, XenbusStateClosed);
 578                 device_unregister(&dev->dev);
 579                 break;
 580
 581         default:
 582                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 583                                  frontend_state);
 584                 break;
 585         }
 586 }
 587
 588
 589 static void xen_net_read_rate(struct xenbus_device *dev,
 590                               unsigned long *bytes, unsigned long *usec)
 591 {
 592         char *s, *e;
 593         unsigned long b, u;
 594         char *ratestr;
 595
 596         /* Default to unlimited bandwidth. */
 597         *bytes = ~0UL;
 598         *usec = 0;
 599
 600         ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL);
 601         if (IS_ERR(ratestr))
 602                 return;
 603
 604         s = ratestr;
 605         b = simple_strtoul(s, &e, 10);
 606         if ((s == e) || (*e != ','))
 607                 goto fail;
 608
 609         s = e + 1;
 610         u = simple_strtoul(s, &e, 10);
 611         if ((s == e) || (*e != '\0'))
 612                 goto fail;
 613
 614         *bytes = b;
 615         *usec = u;
 616
 617         kfree(ratestr);
 618         return;
 619
 620  fail:
 621         pr_warn("Failed to parse network rate limit. Traffic unlimited.\n");
 622         kfree(ratestr);
 623 }
 624
 625 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
 626 {
 627         char *s, *e, *macstr;
 628         int i;
 629
 630         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
 631         if (IS_ERR(macstr))
 632                 return PTR_ERR(macstr);
 633
 634         for (i = 0; i < ETH_ALEN; i++) {
 635                 mac[i] = simple_strtoul(s, &e, 16);
 636                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
 637                         kfree(macstr);
 638                         return -ENOENT;
 639                 }
 640                 s = e+1;
 641         }
 642
 643         kfree(macstr);
 644         return 0;
 645 }
 646
 647 static void unregister_hotplug_status_watch(struct backend_info *be)
 648 {
 649         if (be->have_hotplug_status_watch) {
 650                 unregister_xenbus_watch(&be->hotplug_status_watch);
 651                 kfree(be->hotplug_status_watch.node);
 652         }
 653         be->have_hotplug_status_watch = 0;
 654 }
 655
 656 static void hotplug_status_changed(struct xenbus_watch *watch,
 657                                    const char **vec,
 658                                    unsigned int vec_size)
 659 {
 660         struct backend_info *be = container_of(watch,
 661                                                struct backend_info,
 662                                                hotplug_status_watch);
 663         char *str;
 664         unsigned int len;
 665
 666         str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len);
 667         if (IS_ERR(str))
 668                 return;
 669         if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) {
 670                 /* Complete any pending state change */
 671                 xenbus_switch_state(be->dev, be->state);
 672
 673                 /* Not interested in this watch anymore. */
 674                 unregister_hotplug_status_watch(be);
 675         }
 676         kfree(str);
 677 }
 678
 679 static void connect(struct backend_info *be)
 680 {
 681         int err;
 682         struct xenbus_device *dev = be->dev;
 683         unsigned long credit_bytes, credit_usec;
 684         unsigned int queue_index;
 685         unsigned int requested_num_queues;
 686         struct xenvif_queue *queue;
 687
 688         /* Check whether the frontend requested multiple queues
 689          * and read the number requested.
 690          */
 691         err = xenbus_scanf(XBT_NIL, dev->otherend,
 692                            "multi-queue-num-queues",
 693                            "%u", &requested_num_queues);
 694         if (err < 0) {
 695                 requested_num_queues = 1; /* Fall back to single queue */
 696         } else if (requested_num_queues > xenvif_max_queues) {
 697                 /* buggy or malicious guest */
 698                 xenbus_dev_fatal(dev, err,
 699                                  "guest requested %u queues, exceeding the maximum of %u.",
 700                                  requested_num_queues, xenvif_max_queues);
 701                 return;
 702         }
 703
 704         err = xen_net_read_mac(dev, be->vif->fe_dev_addr);
 705         if (err) {
 706                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
 707                 return;
 708         }
 709
 710         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 711         read_xenbus_vif_flags(be);
 712
 713         /* Use the number of queues requested by the frontend */
 714         be->vif->queues = vzalloc(requested_num_queues *
 715                                   sizeof(struct xenvif_queue));
 716         be->vif->num_queues = requested_num_queues;
 717         be->vif->stalled_queues = requested_num_queues;
 718
 719         for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) {
 720                 queue = &be->vif->queues[queue_index];
 721                 queue->vif = be->vif;
 722                 queue->id = queue_index;
 723                 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
 724                                 be->vif->dev->name, queue->id);
 725
 726                 err = xenvif_init_queue(queue);
 727                 if (err) {
 728                         /* xenvif_init_queue() cleans up after itself on
 729                          * failure, but we need to clean up any previously
 730                          * initialised queues. Set num_queues to i so that
 731                          * earlier queues can be destroyed using the regular
 732                          * disconnect logic.
 733                          */
 734                         be->vif->num_queues = queue_index;
 735                         goto err;
 736                 }
 737
 738                 queue->remaining_credit = credit_bytes;
 739
 740                 err = connect_rings(be, queue);
 741                 if (err) {
 742                         /* connect_rings() cleans up after itself on failure,
 743                          * but we need to clean up after xenvif_init_queue() here,
 744                          * and also clean up any previously initialised queues.
 745                          */
 746                         xenvif_deinit_queue(queue);
 747                         be->vif->num_queues = queue_index;
 748                         goto err;
 749                 }
 750         }
 751
 752 #ifdef CONFIG_DEBUG_FS
 753         xenvif_debugfs_addif(be->vif);
 754 #endif /* CONFIG_DEBUG_FS */
 755
 756         /* Initialisation completed, tell core driver the number of
 757          * active queues.
 758          */
 759         rtnl_lock();
 760         netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues);
 761         netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues);
 762         rtnl_unlock();
 763
 764         xenvif_carrier_on(be->vif);
 765
 766         unregister_hotplug_status_watch(be);
 767         err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch,
 768                                    hotplug_status_changed,
 769                                    "%s/%s", dev->nodename, "hotplug-status");
 770         if (!err)
 771                 be->have_hotplug_status_watch = 1;
 772
 773         netif_tx_wake_all_queues(be->vif->dev);
 774
 775         return;
 776
 777 err:
 778         if (be->vif->num_queues > 0)
 779                 xenvif_disconnect(be->vif); /* Clean up existing queues */
 780         vfree(be->vif->queues);
 781         be->vif->queues = NULL;
 782         be->vif->num_queues = 0;
 783         return;
 784 }
 785
 786
 787 static int connect_rings(struct backend_info *be, struct xenvif_queue *queue)
 788 {
 789         struct xenbus_device *dev = be->dev;
 790         unsigned int num_queues = queue->vif->num_queues;
 791         unsigned long tx_ring_ref, rx_ring_ref;
 792         unsigned int tx_evtchn, rx_evtchn;
 793         int err;
 794         char *xspath;
 795         size_t xspathsize;
 796         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
 797
 798         /* If the frontend requested 1 queue, or we have fallen back
 799          * to single queue due to lack of frontend support for multi-
 800          * queue, expect the remaining XenStore keys in the toplevel
 801          * directory. Otherwise, expect them in a subdirectory called
 802          * queue-N.
 803          */
 804         if (num_queues == 1) {
 805                 xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL);
 806                 if (!xspath) {
 807                         xenbus_dev_fatal(dev, -ENOMEM,
 808                                          "reading ring references");
 809                         return -ENOMEM;
 810                 }
 811                 strcpy(xspath, dev->otherend);
 812         } else {
 813                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
 814                 xspath = kzalloc(xspathsize, GFP_KERNEL);
 815                 if (!xspath) {
 816                         xenbus_dev_fatal(dev, -ENOMEM,
 817                                          "reading ring references");
 818                         return -ENOMEM;
 819                 }
 820                 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend,
 821                          queue->id);
 822         }
 823
 824         err = xenbus_gather(XBT_NIL, xspath,
 825                             "tx-ring-ref", "%lu", &tx_ring_ref,
 826                             "rx-ring-ref", "%lu", &rx_ring_ref, NULL);
 827         if (err) {
 828                 xenbus_dev_fatal(dev, err,
 829                                  "reading %s/ring-ref",
 830                                  xspath);
 831                 goto err;
 832         }
 833
 834         /* Try split event channels first, then single event channel. */
 835         err = xenbus_gather(XBT_NIL, xspath,
 836                             "event-channel-tx", "%u", &tx_evtchn,
 837                             "event-channel-rx", "%u", &rx_evtchn, NULL);
 838         if (err < 0) {
 839                 err = xenbus_scanf(XBT_NIL, xspath,
 840                                    "event-channel", "%u", &tx_evtchn);
 841                 if (err < 0) {
 842                         xenbus_dev_fatal(dev, err,
 843                                          "reading %s/event-channel(-tx/rx)",
 844                                          xspath);
 845                         goto err;
 846                 }
 847                 rx_evtchn = tx_evtchn;
 848         }
 849
 850         /* Map the shared frame, irq etc. */
 851         err = xenvif_connect(queue, tx_ring_ref, rx_ring_ref,
 852                              tx_evtchn, rx_evtchn);
 853         if (err) {
 854                 xenbus_dev_fatal(dev, err,
 855                                  "mapping shared-frames %lu/%lu port tx %u rx %u",
 856                                  tx_ring_ref, rx_ring_ref,
 857                                  tx_evtchn, rx_evtchn);
 858                 goto err;
 859         }
 860
 861         err = 0;
 862 err: /* Regular return falls through with err == 0 */
 863         kfree(xspath);
 864         return err;
 865 }
 866
 867 static int read_xenbus_vif_flags(struct backend_info *be)
 868 {
 869         struct xenvif *vif = be->vif;
 870         struct xenbus_device *dev = be->dev;
 871         unsigned int rx_copy;
 872         int err, val;
 873
 874         err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u",
 875                            &rx_copy);
 876         if (err == -ENOENT) {
 877                 err = 0;
 878                 rx_copy = 0;
 879         }
 880         if (err < 0) {
 881                 xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy",
 882                                  dev->otherend);
 883                 return err;
 884         }
 885         if (!rx_copy)
 886                 return -EOPNOTSUPP;
 887
 888         if (xenbus_scanf(XBT_NIL, dev->otherend,
 889                          "feature-rx-notify", "%d", &val) < 0 || val == 0) {
 890                 xenbus_dev_fatal(dev, -EINVAL, "feature-rx-notify is mandatory");
 891                 return -EINVAL;
 892         }
 893
 894         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-sg",
 895                          "%d", &val) < 0)
 896                 val = 0;
 897         vif->can_sg = !!val;
 898
 899         vif->gso_mask = 0;
 900         vif->gso_prefix_mask = 0;
 901
 902         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4",
 903                          "%d", &val) < 0)
 904                 val = 0;
 905         if (val)
 906                 vif->gso_mask |= GSO_BIT(TCPV4);
 907
 908         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4-prefix",
 909                          "%d", &val) < 0)
 910                 val = 0;
 911         if (val)
 912                 vif->gso_prefix_mask |= GSO_BIT(TCPV4);
 913
 914         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6",
 915                          "%d", &val) < 0)
 916                 val = 0;
 917         if (val)
 918                 vif->gso_mask |= GSO_BIT(TCPV6);
 919
 920         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6-prefix",
 921                          "%d", &val) < 0)
 922                 val = 0;
 923         if (val)
 924                 vif->gso_prefix_mask |= GSO_BIT(TCPV6);
 925
 926         if (vif->gso_mask & vif->gso_prefix_mask) {
 927                 xenbus_dev_fatal(dev, err,
 928                                  "%s: gso and gso prefix flags are not "
 929                                  "mutually exclusive",
 930                                  dev->otherend);
 931                 return -EOPNOTSUPP;
 932         }
 933
 934         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-no-csum-offload",
 935                          "%d", &val) < 0)
 936                 val = 0;
 937         vif->ip_csum = !val;
 938
 939         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-ipv6-csum-offload",
 940                          "%d", &val) < 0)
 941                 val = 0;
 942         vif->ipv6_csum = !!val;
 943
 944         return 0;
 945 }
 946
 947 static const struct xenbus_device_id netback_ids[] = {
 948         { "vif" },
 949         { "" }
 950 };
 951
 952 static struct xenbus_driver netback_driver = {
 953         .ids = netback_ids,
 954         .probe = netback_probe,
 955         .remove = netback_remove,
 956         .uevent = netback_uevent,
 957         .otherend_changed = frontend_changed,
 958 };
 959
 960 int xenvif_xenbus_init(void)
 961 {
 962         return xenbus_register_backend(&netback_driver);
 963 }
 964
 965 void xenvif_xenbus_fini(void)
 966 {
 967         return xenbus_unregister_driver(&netback_driver);
 968 }