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 xen_unregister_watchers(struct xenvif *vif);
  45 static void set_backend_state(struct backend_info *be,
  46                               enum xenbus_state state);
  47
  48 #ifdef CONFIG_DEBUG_FS
  49 struct dentry *xen_netback_dbg_root = NULL;
  50
  51 static int xenvif_read_io_ring(struct seq_file *m, void *v)
  52 {
  53         struct xenvif_queue *queue = m->private;
  54         struct xen_netif_tx_back_ring *tx_ring = &queue->tx;
  55         struct xen_netif_rx_back_ring *rx_ring = &queue->rx;
  56         struct netdev_queue *dev_queue;
  57
  58         if (tx_ring->sring) {
  59                 struct xen_netif_tx_sring *sring = tx_ring->sring;
  60
  61                 seq_printf(m, "Queue %d\nTX: nr_ents %u\n", queue->id,
  62                            tx_ring->nr_ents);
  63                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  64                            sring->req_prod,
  65                            sring->req_prod - sring->rsp_prod,
  66                            tx_ring->req_cons,
  67                            tx_ring->req_cons - sring->rsp_prod,
  68                            sring->req_event,
  69                            sring->req_event - sring->rsp_prod);
  70                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n",
  71                            sring->rsp_prod,
  72                            tx_ring->rsp_prod_pvt,
  73                            tx_ring->rsp_prod_pvt - sring->rsp_prod,
  74                            sring->rsp_event,
  75                            sring->rsp_event - sring->rsp_prod);
  76                 seq_printf(m, "pending prod %u pending cons %u nr_pending_reqs %u\n",
  77                            queue->pending_prod,
  78                            queue->pending_cons,
  79                            nr_pending_reqs(queue));
  80                 seq_printf(m, "dealloc prod %u dealloc cons %u dealloc_queue %u\n\n",
  81                            queue->dealloc_prod,
  82                            queue->dealloc_cons,
  83                            queue->dealloc_prod - queue->dealloc_cons);
  84         }
  85
  86         if (rx_ring->sring) {
  87                 struct xen_netif_rx_sring *sring = rx_ring->sring;
  88
  89                 seq_printf(m, "RX: nr_ents %u\n", rx_ring->nr_ents);
  90                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  91                            sring->req_prod,
  92                            sring->req_prod - sring->rsp_prod,
  93                            rx_ring->req_cons,
  94                            rx_ring->req_cons - sring->rsp_prod,
  95                            sring->req_event,
  96                            sring->req_event - sring->rsp_prod);
  97                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n\n",
  98                            sring->rsp_prod,
  99                            rx_ring->rsp_prod_pvt,
 100                            rx_ring->rsp_prod_pvt - sring->rsp_prod,
 101                            sring->rsp_event,
 102                            sring->rsp_event - sring->rsp_prod);
 103         }
 104
 105         seq_printf(m, "NAPI state: %lx NAPI weight: %d TX queue len %u\n"
 106                    "Credit timer_pending: %d, credit: %lu, usec: %lu\n"
 107                    "remaining: %lu, expires: %lu, now: %lu\n",
 108                    queue->napi.state, queue->napi.weight,
 109                    skb_queue_len(&queue->tx_queue),
 110                    timer_pending(&queue->credit_timeout),
 111                    queue->credit_bytes,
 112                    queue->credit_usec,
 113                    queue->remaining_credit,
 114                    queue->credit_timeout.expires,
 115                    jiffies);
 116
 117         dev_queue = netdev_get_tx_queue(queue->vif->dev, queue->id);
 118
 119         seq_printf(m, "\nRx internal queue: len %u max %u pkts %u %s\n",
 120                    queue->rx_queue_len, queue->rx_queue_max,
 121                    skb_queue_len(&queue->rx_queue),
 122                    netif_tx_queue_stopped(dev_queue) ? "stopped" : "running");
 123
 124         return 0;
 125 }
 126
 127 #define XENVIF_KICK_STR "kick"
 128 #define BUFFER_SIZE     32
 129
 130 static ssize_t
 131 xenvif_write_io_ring(struct file *filp, const char __user *buf, size_t count,
 132                      loff_t *ppos)
 133 {
 134         struct xenvif_queue *queue =
 135                 ((struct seq_file *)filp->private_data)->private;
 136         int len;
 137         char write[BUFFER_SIZE];
 138
 139         /* don't allow partial writes and check the length */
 140         if (*ppos != 0)
 141                 return 0;
 142         if (count >= sizeof(write))
 143                 return -ENOSPC;
 144
 145         len = simple_write_to_buffer(write,
 146                                      sizeof(write) - 1,
 147                                      ppos,
 148                                      buf,
 149                                      count);
 150         if (len < 0)
 151                 return len;
 152
 153         write[len] = '\0';
 154
 155         if (!strncmp(write, XENVIF_KICK_STR, sizeof(XENVIF_KICK_STR) - 1))
 156                 xenvif_interrupt(0, (void *)queue);
 157         else {
 158                 pr_warn("Unknown command to io_ring_q%d. Available: kick\n",
 159                         queue->id);
 160                 count = -EINVAL;
 161         }
 162         return count;
 163 }
 164
 165 static int xenvif_dump_open(struct inode *inode, struct file *filp)
 166 {
 167         int ret;
 168         void *queue = NULL;
 169
 170         if (inode->i_private)
 171                 queue = inode->i_private;
 172         ret = single_open(filp, xenvif_read_io_ring, queue);
 173         filp->f_mode |= FMODE_PWRITE;
 174         return ret;
 175 }
 176
 177 static const struct file_operations xenvif_dbg_io_ring_ops_fops = {
 178         .owner = THIS_MODULE,
 179         .open = xenvif_dump_open,
 180         .read = seq_read,
 181         .llseek = seq_lseek,
 182         .release = single_release,
 183         .write = xenvif_write_io_ring,
 184 };
 185
 186 static void xenvif_debugfs_addif(struct xenvif *vif)
 187 {
 188         struct dentry *pfile;
 189         int i;
 190
 191         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 192                 return;
 193
 194         vif->xenvif_dbg_root = debugfs_create_dir(vif->dev->name,
 195                                                   xen_netback_dbg_root);
 196         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) {
 197                 for (i = 0; i < vif->num_queues; ++i) {
 198                         char filename[sizeof("io_ring_q") + 4];
 199
 200                         snprintf(filename, sizeof(filename), "io_ring_q%d", i);
 201                         pfile = debugfs_create_file(filename,
 202                                                     S_IRUSR | S_IWUSR,
 203                                                     vif->xenvif_dbg_root,
 204                                                     &vif->queues[i],
 205                                                     &xenvif_dbg_io_ring_ops_fops);
 206                         if (IS_ERR_OR_NULL(pfile))
 207                                 pr_warn("Creation of io_ring file returned %ld!\n",
 208                                         PTR_ERR(pfile));
 209                 }
 210         } else
 211                 netdev_warn(vif->dev,
 212                             "Creation of vif debugfs dir returned %ld!\n",
 213                             PTR_ERR(vif->xenvif_dbg_root));
 214 }
 215
 216 static void xenvif_debugfs_delif(struct xenvif *vif)
 217 {
 218         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 219                 return;
 220
 221         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root))
 222                 debugfs_remove_recursive(vif->xenvif_dbg_root);
 223         vif->xenvif_dbg_root = NULL;
 224 }
 225 #endif /* CONFIG_DEBUG_FS */
 226
 227 static int netback_remove(struct xenbus_device *dev)
 228 {
 229         struct backend_info *be = dev_get_drvdata(&dev->dev);
 230
 231         set_backend_state(be, XenbusStateClosed);
 232
 233         unregister_hotplug_status_watch(be);
 234         if (be->vif) {
 235                 kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE);
 236                 xen_unregister_watchers(be->vif);
 237                 xenbus_rm(XBT_NIL, dev->nodename, "hotplug-status");
 238                 xenvif_free(be->vif);
 239                 be->vif = NULL;
 240         }
 241         kfree(be);
 242         dev_set_drvdata(&dev->dev, NULL);
 243         return 0;
 244 }
 245
 246
 247 /**
 248  * Entry point to this code when a new device is created.  Allocate the basic
 249  * structures and switch to InitWait.
 250  */
 251 static int netback_probe(struct xenbus_device *dev,
 252                          const struct xenbus_device_id *id)
 253 {
 254         const char *message;
 255         struct xenbus_transaction xbt;
 256         int err;
 257         int sg;
 258         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 259                                           GFP_KERNEL);
 260         if (!be) {
 261                 xenbus_dev_fatal(dev, -ENOMEM,
 262                                  "allocating backend structure");
 263                 return -ENOMEM;
 264         }
 265
 266         be->dev = dev;
 267         dev_set_drvdata(&dev->dev, be);
 268
 269         sg = 1;
 270
 271         do {
 272                 err = xenbus_transaction_start(&xbt);
 273                 if (err) {
 274                         xenbus_dev_fatal(dev, err, "starting transaction");
 275                         goto fail;
 276                 }
 277
 278                 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg);
 279                 if (err) {
 280                         message = "writing feature-sg";
 281                         goto abort_transaction;
 282                 }
 283
 284                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4",
 285                                     "%d", sg);
 286                 if (err) {
 287                         message = "writing feature-gso-tcpv4";
 288                         goto abort_transaction;
 289                 }
 290
 291                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6",
 292                                     "%d", sg);
 293                 if (err) {
 294                         message = "writing feature-gso-tcpv6";
 295                         goto abort_transaction;
 296                 }
 297
 298                 /* We support partial checksum setup for IPv6 packets */
 299                 err = xenbus_printf(xbt, dev->nodename,
 300                                     "feature-ipv6-csum-offload",
 301                                     "%d", 1);
 302                 if (err) {
 303                         message = "writing feature-ipv6-csum-offload";
 304                         goto abort_transaction;
 305                 }
 306
 307                 /* We support rx-copy path. */
 308                 err = xenbus_printf(xbt, dev->nodename,
 309                                     "feature-rx-copy", "%d", 1);
 310                 if (err) {
 311                         message = "writing feature-rx-copy";
 312                         goto abort_transaction;
 313                 }
 314
 315                 /*
 316                  * We don't support rx-flip path (except old guests who don't
 317                  * grok this feature flag).
 318                  */
 319                 err = xenbus_printf(xbt, dev->nodename,
 320                                     "feature-rx-flip", "%d", 0);
 321                 if (err) {
 322                         message = "writing feature-rx-flip";
 323                         goto abort_transaction;
 324                 }
 325
 326                 err = xenbus_transaction_end(xbt, 0);
 327         } while (err == -EAGAIN);
 328
 329         if (err) {
 330                 xenbus_dev_fatal(dev, err, "completing transaction");
 331                 goto fail;
 332         }
 333
 334         /*
 335          * Split event channels support, this is optional so it is not
 336          * put inside the above loop.
 337          */
 338         err = xenbus_printf(XBT_NIL, dev->nodename,
 339                             "feature-split-event-channels",
 340                             "%u", separate_tx_rx_irq);
 341         if (err)
 342                 pr_debug("Error writing feature-split-event-channels\n");
 343
 344         /* Multi-queue support: This is an optional feature. */
 345         err = xenbus_printf(XBT_NIL, dev->nodename,
 346                             "multi-queue-max-queues", "%u", xenvif_max_queues);
 347         if (err)
 348                 pr_debug("Error writing multi-queue-max-queues\n");
 349
 350         err = xenbus_switch_state(dev, XenbusStateInitWait);
 351         if (err)
 352                 goto fail;
 353
 354         be->state = XenbusStateInitWait;
 355
 356         /* This kicks hotplug scripts, so do it immediately. */
 357         err = backend_create_xenvif(be);
 358         if (err)
 359                 goto fail;
 360
 361         return 0;
 362
 363 abort_transaction:
 364         xenbus_transaction_end(xbt, 1);
 365         xenbus_dev_fatal(dev, err, "%s", message);
 366 fail:
 367         pr_debug("failed\n");
 368         netback_remove(dev);
 369         return err;
 370 }
 371
 372
 373 /*
 374  * Handle the creation of the hotplug script environment.  We add the script
 375  * and vif variables to the environment, for the benefit of the vif-* hotplug
 376  * scripts.
 377  */
 378 static int netback_uevent(struct xenbus_device *xdev,
 379                           struct kobj_uevent_env *env)
 380 {
 381         struct backend_info *be = dev_get_drvdata(&xdev->dev);
 382         char *val;
 383
 384         val = xenbus_read(XBT_NIL, xdev->nodename, "script", NULL);
 385         if (IS_ERR(val)) {
 386                 int err = PTR_ERR(val);
 387                 xenbus_dev_fatal(xdev, err, "reading script");
 388                 return err;
 389         } else {
 390                 if (add_uevent_var(env, "script=%s", val)) {
 391                         kfree(val);
 392                         return -ENOMEM;
 393                 }
 394                 kfree(val);
 395         }
 396
 397         if (!be || !be->vif)
 398                 return 0;
 399
 400         return add_uevent_var(env, "vif=%s", be->vif->dev->name);
 401 }
 402
 403
 404 static int backend_create_xenvif(struct backend_info *be)
 405 {
 406         int err;
 407         long handle;
 408         struct xenbus_device *dev = be->dev;
 409         struct xenvif *vif;
 410
 411         if (be->vif != NULL)
 412                 return 0;
 413
 414         err = xenbus_scanf(XBT_NIL, dev->nodename, "handle", "%li", &handle);
 415         if (err != 1) {
 416                 xenbus_dev_fatal(dev, err, "reading handle");
 417                 return (err < 0) ? err : -EINVAL;
 418         }
 419
 420         vif = xenvif_alloc(&dev->dev, dev->otherend_id, handle);
 421         if (IS_ERR(vif)) {
 422                 err = PTR_ERR(vif);
 423                 xenbus_dev_fatal(dev, err, "creating interface");
 424                 return err;
 425         }
 426         be->vif = vif;
 427
 428         kobject_uevent(&dev->dev.kobj, KOBJ_ONLINE);
 429         return 0;
 430 }
 431
 432 static void backend_disconnect(struct backend_info *be)
 433 {
 434         if (be->vif) {
 435                 xen_unregister_watchers(be->vif);
 436 #ifdef CONFIG_DEBUG_FS
 437                 xenvif_debugfs_delif(be->vif);
 438 #endif /* CONFIG_DEBUG_FS */
 439                 xenvif_disconnect(be->vif);
 440         }
 441 }
 442
 443 static void backend_connect(struct backend_info *be)
 444 {
 445         if (be->vif)
 446                 connect(be);
 447 }
 448
 449 static inline void backend_switch_state(struct backend_info *be,
 450                                         enum xenbus_state state)
 451 {
 452         struct xenbus_device *dev = be->dev;
 453
 454         pr_debug("%s -> %s\n", dev->nodename, xenbus_strstate(state));
 455         be->state = state;
 456
 457         /* If we are waiting for a hotplug script then defer the
 458          * actual xenbus state change.
 459          */
 460         if (!be->have_hotplug_status_watch)
 461                 xenbus_switch_state(dev, state);
 462 }
 463
 464 /* Handle backend state transitions:
 465  *
 466  * The backend state starts in InitWait and the following transitions are
 467  * allowed.
 468  *
 469  * InitWait -> Connected
 470  *
 471  *    ^    \         |
 472  *    |     \        |
 473  *    |      \       |
 474  *    |       \      |
 475  *    |        \     |
 476  *    |         \    |
 477  *    |          V   V
 478  *
 479  *  Closed  <-> Closing
 480  *
 481  * The state argument specifies the eventual state of the backend and the
 482  * function transitions to that state via the shortest path.
 483  */
 484 static void set_backend_state(struct backend_info *be,
 485                               enum xenbus_state state)
 486 {
 487         while (be->state != state) {
 488                 switch (be->state) {
 489                 case XenbusStateClosed:
 490                         switch (state) {
 491                         case XenbusStateInitWait:
 492                         case XenbusStateConnected:
 493                                 pr_info("%s: prepare for reconnect\n",
 494                                         be->dev->nodename);
 495                                 backend_switch_state(be, XenbusStateInitWait);
 496                                 break;
 497                         case XenbusStateClosing:
 498                                 backend_switch_state(be, XenbusStateClosing);
 499                                 break;
 500                         default:
 501                                 BUG();
 502                         }
 503                         break;
 504                 case XenbusStateInitWait:
 505                         switch (state) {
 506                         case XenbusStateConnected:
 507                                 backend_connect(be);
 508                                 backend_switch_state(be, XenbusStateConnected);
 509                                 break;
 510                         case XenbusStateClosing:
 511                         case XenbusStateClosed:
 512                                 backend_switch_state(be, XenbusStateClosing);
 513                                 break;
 514                         default:
 515                                 BUG();
 516                         }
 517                         break;
 518                 case XenbusStateConnected:
 519                         switch (state) {
 520                         case XenbusStateInitWait:
 521                         case XenbusStateClosing:
 522                         case XenbusStateClosed:
 523                                 backend_disconnect(be);
 524                                 backend_switch_state(be, XenbusStateClosing);
 525                                 break;
 526                         default:
 527                                 BUG();
 528                         }
 529                         break;
 530                 case XenbusStateClosing:
 531                         switch (state) {
 532                         case XenbusStateInitWait:
 533                         case XenbusStateConnected:
 534                         case XenbusStateClosed:
 535                                 backend_switch_state(be, XenbusStateClosed);
 536                                 break;
 537                         default:
 538                                 BUG();
 539                         }
 540                         break;
 541                 default:
 542                         BUG();
 543                 }
 544         }
 545 }
 546
 547 /**
 548  * Callback received when the frontend's state changes.
 549  */
 550 static void frontend_changed(struct xenbus_device *dev,
 551                              enum xenbus_state frontend_state)
 552 {
 553         struct backend_info *be = dev_get_drvdata(&dev->dev);
 554
 555         pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state));
 556
 557         be->frontend_state = frontend_state;
 558
 559         switch (frontend_state) {
 560         case XenbusStateInitialising:
 561                 set_backend_state(be, XenbusStateInitWait);
 562                 break;
 563
 564         case XenbusStateInitialised:
 565                 break;
 566
 567         case XenbusStateConnected:
 568                 set_backend_state(be, XenbusStateConnected);
 569                 break;
 570
 571         case XenbusStateClosing:
 572                 set_backend_state(be, XenbusStateClosing);
 573                 break;
 574
 575         case XenbusStateClosed:
 576                 set_backend_state(be, XenbusStateClosed);
 577                 if (xenbus_dev_is_online(dev))
 578                         break;
 579                 /* fall through if not online */
 580         case XenbusStateUnknown:
 581                 set_backend_state(be, XenbusStateClosed);
 582                 device_unregister(&dev->dev);
 583                 break;
 584
 585         default:
 586                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 587                                  frontend_state);
 588                 break;
 589         }
 590 }
 591
 592
 593 static void xen_net_read_rate(struct xenbus_device *dev,
 594                               unsigned long *bytes, unsigned long *usec)
 595 {
 596         char *s, *e;
 597         unsigned long b, u;
 598         char *ratestr;
 599
 600         /* Default to unlimited bandwidth. */
 601         *bytes = ~0UL;
 602         *usec = 0;
 603
 604         ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL);
 605         if (IS_ERR(ratestr))
 606                 return;
 607
 608         s = ratestr;
 609         b = simple_strtoul(s, &e, 10);
 610         if ((s == e) || (*e != ','))
 611                 goto fail;
 612
 613         s = e + 1;
 614         u = simple_strtoul(s, &e, 10);
 615         if ((s == e) || (*e != '\0'))
 616                 goto fail;
 617
 618         *bytes = b;
 619         *usec = u;
 620
 621         kfree(ratestr);
 622         return;
 623
 624  fail:
 625         pr_warn("Failed to parse network rate limit. Traffic unlimited.\n");
 626         kfree(ratestr);
 627 }
 628
 629 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
 630 {
 631         char *s, *e, *macstr;
 632         int i;
 633
 634         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
 635         if (IS_ERR(macstr))
 636                 return PTR_ERR(macstr);
 637
 638         for (i = 0; i < ETH_ALEN; i++) {
 639                 mac[i] = simple_strtoul(s, &e, 16);
 640                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
 641                         kfree(macstr);
 642                         return -ENOENT;
 643                 }
 644                 s = e+1;
 645         }
 646
 647         kfree(macstr);
 648         return 0;
 649 }
 650
 651 static void xen_net_rate_changed(struct xenbus_watch *watch,
 652                                 const char **vec, unsigned int len)
 653 {
 654         struct xenvif *vif = container_of(watch, struct xenvif, credit_watch);
 655         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 656         unsigned long   credit_bytes;
 657         unsigned long   credit_usec;
 658         unsigned int queue_index;
 659
 660         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 661         for (queue_index = 0; queue_index < vif->num_queues; queue_index++) {
 662                 struct xenvif_queue *queue = &vif->queues[queue_index];
 663
 664                 queue->credit_bytes = credit_bytes;
 665                 queue->credit_usec = credit_usec;
 666                 if (!mod_timer_pending(&queue->credit_timeout, jiffies) &&
 667                         queue->remaining_credit > queue->credit_bytes) {
 668                         queue->remaining_credit = queue->credit_bytes;
 669                 }
 670         }
 671 }
 672
 673 static int xen_register_watchers(struct xenbus_device *dev, struct xenvif *vif)
 674 {
 675         int err = 0;
 676         char *node;
 677         unsigned maxlen = strlen(dev->nodename) + sizeof("/rate");
 678
 679         node = kmalloc(maxlen, GFP_KERNEL);
 680         if (!node)
 681                 return -ENOMEM;
 682         snprintf(node, maxlen, "%s/rate", dev->nodename);
 683         vif->credit_watch.node = node;
 684         vif->credit_watch.callback = xen_net_rate_changed;
 685         err = register_xenbus_watch(&vif->credit_watch);
 686         if (err) {
 687                 pr_err("Failed to set watcher %s\n", vif->credit_watch.node);
 688                 kfree(node);
 689                 vif->credit_watch.node = NULL;
 690                 vif->credit_watch.callback = NULL;
 691         }
 692         return err;
 693 }
 694
 695 static void xen_unregister_watchers(struct xenvif *vif)
 696 {
 697         if (vif->credit_watch.node) {
 698                 unregister_xenbus_watch(&vif->credit_watch);
 699                 kfree(vif->credit_watch.node);
 700                 vif->credit_watch.node = NULL;
 701         }
 702 }
 703
 704 static void unregister_hotplug_status_watch(struct backend_info *be)
 705 {
 706         if (be->have_hotplug_status_watch) {
 707                 unregister_xenbus_watch(&be->hotplug_status_watch);
 708                 kfree(be->hotplug_status_watch.node);
 709         }
 710         be->have_hotplug_status_watch = 0;
 711 }
 712
 713 static void hotplug_status_changed(struct xenbus_watch *watch,
 714                                    const char **vec,
 715                                    unsigned int vec_size)
 716 {
 717         struct backend_info *be = container_of(watch,
 718                                                struct backend_info,
 719                                                hotplug_status_watch);
 720         char *str;
 721         unsigned int len;
 722
 723         str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len);
 724         if (IS_ERR(str))
 725                 return;
 726         if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) {
 727                 /* Complete any pending state change */
 728                 xenbus_switch_state(be->dev, be->state);
 729
 730                 /* Not interested in this watch anymore. */
 731                 unregister_hotplug_status_watch(be);
 732         }
 733         kfree(str);
 734 }
 735
 736 static void connect(struct backend_info *be)
 737 {
 738         int err;
 739         struct xenbus_device *dev = be->dev;
 740         unsigned long credit_bytes, credit_usec;
 741         unsigned int queue_index;
 742         unsigned int requested_num_queues;
 743         struct xenvif_queue *queue;
 744
 745         /* Check whether the frontend requested multiple queues
 746          * and read the number requested.
 747          */
 748         err = xenbus_scanf(XBT_NIL, dev->otherend,
 749                            "multi-queue-num-queues",
 750                            "%u", &requested_num_queues);
 751         if (err < 0) {
 752                 requested_num_queues = 1; /* Fall back to single queue */
 753         } else if (requested_num_queues > xenvif_max_queues) {
 754                 /* buggy or malicious guest */
 755                 xenbus_dev_fatal(dev, err,
 756                                  "guest requested %u queues, exceeding the maximum of %u.",
 757                                  requested_num_queues, xenvif_max_queues);
 758                 return;
 759         }
 760
 761         err = xen_net_read_mac(dev, be->vif->fe_dev_addr);
 762         if (err) {
 763                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
 764                 return;
 765         }
 766
 767         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 768         xen_register_watchers(dev, be->vif);
 769         read_xenbus_vif_flags(be);
 770
 771         /* Use the number of queues requested by the frontend */
 772         be->vif->queues = vzalloc(requested_num_queues *
 773                                   sizeof(struct xenvif_queue));
 774         be->vif->num_queues = requested_num_queues;
 775         be->vif->stalled_queues = requested_num_queues;
 776
 777         for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) {
 778                 queue = &be->vif->queues[queue_index];
 779                 queue->vif = be->vif;
 780                 queue->id = queue_index;
 781                 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
 782                                 be->vif->dev->name, queue->id);
 783
 784                 err = xenvif_init_queue(queue);
 785                 if (err) {
 786                         /* xenvif_init_queue() cleans up after itself on
 787                          * failure, but we need to clean up any previously
 788                          * initialised queues. Set num_queues to i so that
 789                          * earlier queues can be destroyed using the regular
 790                          * disconnect logic.
 791                          */
 792                         be->vif->num_queues = queue_index;
 793                         goto err;
 794                 }
 795
 796                 queue->remaining_credit = credit_bytes;
 797                 queue->credit_usec = credit_usec;
 798
 799                 err = connect_rings(be, queue);
 800                 if (err) {
 801                         /* connect_rings() cleans up after itself on failure,
 802                          * but we need to clean up after xenvif_init_queue() here,
 803                          * and also clean up any previously initialised queues.
 804                          */
 805                         xenvif_deinit_queue(queue);
 806                         be->vif->num_queues = queue_index;
 807                         goto err;
 808                 }
 809         }
 810
 811 #ifdef CONFIG_DEBUG_FS
 812         xenvif_debugfs_addif(be->vif);
 813 #endif /* CONFIG_DEBUG_FS */
 814
 815         /* Initialisation completed, tell core driver the number of
 816          * active queues.
 817          */
 818         rtnl_lock();
 819         netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues);
 820         netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues);
 821         rtnl_unlock();
 822
 823         xenvif_carrier_on(be->vif);
 824
 825         unregister_hotplug_status_watch(be);
 826         err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch,
 827                                    hotplug_status_changed,
 828                                    "%s/%s", dev->nodename, "hotplug-status");
 829         if (!err)
 830                 be->have_hotplug_status_watch = 1;
 831
 832         netif_tx_wake_all_queues(be->vif->dev);
 833
 834         return;
 835
 836 err:
 837         if (be->vif->num_queues > 0)
 838                 xenvif_disconnect(be->vif); /* Clean up existing queues */
 839         vfree(be->vif->queues);
 840         be->vif->queues = NULL;
 841         be->vif->num_queues = 0;
 842         return;
 843 }
 844
 845
 846 static int connect_rings(struct backend_info *be, struct xenvif_queue *queue)
 847 {
 848         struct xenbus_device *dev = be->dev;
 849         unsigned int num_queues = queue->vif->num_queues;
 850         unsigned long tx_ring_ref, rx_ring_ref;
 851         unsigned int tx_evtchn, rx_evtchn;
 852         int err;
 853         char *xspath;
 854         size_t xspathsize;
 855         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
 856
 857         /* If the frontend requested 1 queue, or we have fallen back
 858          * to single queue due to lack of frontend support for multi-
 859          * queue, expect the remaining XenStore keys in the toplevel
 860          * directory. Otherwise, expect them in a subdirectory called
 861          * queue-N.
 862          */
 863         if (num_queues == 1) {
 864                 xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL);
 865                 if (!xspath) {
 866                         xenbus_dev_fatal(dev, -ENOMEM,
 867                                          "reading ring references");
 868                         return -ENOMEM;
 869                 }
 870                 strcpy(xspath, dev->otherend);
 871         } else {
 872                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
 873                 xspath = kzalloc(xspathsize, GFP_KERNEL);
 874                 if (!xspath) {
 875                         xenbus_dev_fatal(dev, -ENOMEM,
 876                                          "reading ring references");
 877                         return -ENOMEM;
 878                 }
 879                 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend,
 880                          queue->id);
 881         }
 882
 883         err = xenbus_gather(XBT_NIL, xspath,
 884                             "tx-ring-ref", "%lu", &tx_ring_ref,
 885                             "rx-ring-ref", "%lu", &rx_ring_ref, NULL);
 886         if (err) {
 887                 xenbus_dev_fatal(dev, err,
 888                                  "reading %s/ring-ref",
 889                                  xspath);
 890                 goto err;
 891         }
 892
 893         /* Try split event channels first, then single event channel. */
 894         err = xenbus_gather(XBT_NIL, xspath,
 895                             "event-channel-tx", "%u", &tx_evtchn,
 896                             "event-channel-rx", "%u", &rx_evtchn, NULL);
 897         if (err < 0) {
 898                 err = xenbus_scanf(XBT_NIL, xspath,
 899                                    "event-channel", "%u", &tx_evtchn);
 900                 if (err < 0) {
 901                         xenbus_dev_fatal(dev, err,
 902                                          "reading %s/event-channel(-tx/rx)",
 903                                          xspath);
 904                         goto err;
 905                 }
 906                 rx_evtchn = tx_evtchn;
 907         }
 908
 909         /* Map the shared frame, irq etc. */
 910         err = xenvif_connect(queue, tx_ring_ref, rx_ring_ref,
 911                              tx_evtchn, rx_evtchn);
 912         if (err) {
 913                 xenbus_dev_fatal(dev, err,
 914                                  "mapping shared-frames %lu/%lu port tx %u rx %u",
 915                                  tx_ring_ref, rx_ring_ref,
 916                                  tx_evtchn, rx_evtchn);
 917                 goto err;
 918         }
 919
 920         err = 0;
 921 err: /* Regular return falls through with err == 0 */
 922         kfree(xspath);
 923         return err;
 924 }
 925
 926 static int read_xenbus_vif_flags(struct backend_info *be)
 927 {
 928         struct xenvif *vif = be->vif;
 929         struct xenbus_device *dev = be->dev;
 930         unsigned int rx_copy;
 931         int err, val;
 932
 933         err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u",
 934                            &rx_copy);
 935         if (err == -ENOENT) {
 936                 err = 0;
 937                 rx_copy = 0;
 938         }
 939         if (err < 0) {
 940                 xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy",
 941                                  dev->otherend);
 942                 return err;
 943         }
 944         if (!rx_copy)
 945                 return -EOPNOTSUPP;
 946
 947         if (xenbus_scanf(XBT_NIL, dev->otherend,
 948                          "feature-rx-notify", "%d", &val) < 0)
 949                 val = 0;
 950         if (!val) {
 951                 /* - Reduce drain timeout to poll more frequently for
 952                  *   Rx requests.
 953                  * - Disable Rx stall detection.
 954                  */
 955                 be->vif->drain_timeout = msecs_to_jiffies(30);
 956                 be->vif->stall_timeout = 0;
 957         }
 958
 959         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-sg",
 960                          "%d", &val) < 0)
 961                 val = 0;
 962         vif->can_sg = !!val;
 963
 964         vif->gso_mask = 0;
 965         vif->gso_prefix_mask = 0;
 966
 967         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4",
 968                          "%d", &val) < 0)
 969                 val = 0;
 970         if (val)
 971                 vif->gso_mask |= GSO_BIT(TCPV4);
 972
 973         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4-prefix",
 974                          "%d", &val) < 0)
 975                 val = 0;
 976         if (val)
 977                 vif->gso_prefix_mask |= GSO_BIT(TCPV4);
 978
 979         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6",
 980                          "%d", &val) < 0)
 981                 val = 0;
 982         if (val)
 983                 vif->gso_mask |= GSO_BIT(TCPV6);
 984
 985         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6-prefix",
 986                          "%d", &val) < 0)
 987                 val = 0;
 988         if (val)
 989                 vif->gso_prefix_mask |= GSO_BIT(TCPV6);
 990
 991         if (vif->gso_mask & vif->gso_prefix_mask) {
 992                 xenbus_dev_fatal(dev, err,
 993                                  "%s: gso and gso prefix flags are not "
 994                                  "mutually exclusive",
 995                                  dev->otherend);
 996                 return -EOPNOTSUPP;
 997         }
 998
 999         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-no-csum-offload",
1000                          "%d", &val) < 0)
1001                 val = 0;
1002         vif->ip_csum = !val;
1003
1004         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-ipv6-csum-offload",
1005                          "%d", &val) < 0)
1006                 val = 0;
1007         vif->ipv6_csum = !!val;
1008
1009         return 0;
1010 }
1011
1012 static const struct xenbus_device_id netback_ids[] = {
1013         { "vif" },
1014         { "" }
1015 };
1016
1017 static struct xenbus_driver netback_driver = {
1018         .ids = netback_ids,
1019         .probe = netback_probe,
1020         .remove = netback_remove,
1021         .uevent = netback_uevent,
1022         .otherend_changed = frontend_changed,
1023 };
1024
1025 int xenvif_xenbus_init(void)
1026 {
1027         return xenbus_register_backend(&netback_driver);
1028 }
1029
1030 void xenvif_xenbus_fini(void)
1031 {
1032         return xenbus_unregister_driver(&netback_driver);
1033 }