drivers/net/xen-netback/xenbus.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Xenbus code for netif backend
   4  *
   5  * Copyright (C) 2005 Rusty Russell <[email protected]>
   6  * Copyright (C) 2005 XenSource Ltd
   7 */
   8
   9 #include "common.h"
  10 #include <linux/vmalloc.h>
  11 #include <linux/rtnetlink.h>
  12
  13 static int connect_data_rings(struct backend_info *be,
  14                               struct xenvif_queue *queue);
  15 static void connect(struct backend_info *be);
  16 static int read_xenbus_vif_flags(struct backend_info *be);
  17 static int backend_create_xenvif(struct backend_info *be);
  18 static void unregister_hotplug_status_watch(struct backend_info *be);
  19 static void xen_unregister_watchers(struct xenvif *vif);
  20 static void set_backend_state(struct backend_info *be,
  21                               enum xenbus_state state);
  22
  23 #ifdef CONFIG_DEBUG_FS
  24 struct dentry *xen_netback_dbg_root = NULL;
  25
  26 static int xenvif_read_io_ring(struct seq_file *m, void *v)
  27 {
  28         struct xenvif_queue *queue = m->private;
  29         struct xen_netif_tx_back_ring *tx_ring = &queue->tx;
  30         struct xen_netif_rx_back_ring *rx_ring = &queue->rx;
  31         struct netdev_queue *dev_queue;
  32
  33         if (tx_ring->sring) {
  34                 struct xen_netif_tx_sring *sring = tx_ring->sring;
  35
  36                 seq_printf(m, "Queue %d\nTX: nr_ents %u\n", queue->id,
  37                            tx_ring->nr_ents);
  38                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  39                            sring->req_prod,
  40                            sring->req_prod - sring->rsp_prod,
  41                            tx_ring->req_cons,
  42                            tx_ring->req_cons - sring->rsp_prod,
  43                            sring->req_event,
  44                            sring->req_event - sring->rsp_prod);
  45                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n",
  46                            sring->rsp_prod,
  47                            tx_ring->rsp_prod_pvt,
  48                            tx_ring->rsp_prod_pvt - sring->rsp_prod,
  49                            sring->rsp_event,
  50                            sring->rsp_event - sring->rsp_prod);
  51                 seq_printf(m, "pending prod %u pending cons %u nr_pending_reqs %u\n",
  52                            queue->pending_prod,
  53                            queue->pending_cons,
  54                            nr_pending_reqs(queue));
  55                 seq_printf(m, "dealloc prod %u dealloc cons %u dealloc_queue %u\n\n",
  56                            queue->dealloc_prod,
  57                            queue->dealloc_cons,
  58                            queue->dealloc_prod - queue->dealloc_cons);
  59         }
  60
  61         if (rx_ring->sring) {
  62                 struct xen_netif_rx_sring *sring = rx_ring->sring;
  63
  64                 seq_printf(m, "RX: nr_ents %u\n", rx_ring->nr_ents);
  65                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  66                            sring->req_prod,
  67                            sring->req_prod - sring->rsp_prod,
  68                            rx_ring->req_cons,
  69                            rx_ring->req_cons - sring->rsp_prod,
  70                            sring->req_event,
  71                            sring->req_event - sring->rsp_prod);
  72                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n\n",
  73                            sring->rsp_prod,
  74                            rx_ring->rsp_prod_pvt,
  75                            rx_ring->rsp_prod_pvt - sring->rsp_prod,
  76                            sring->rsp_event,
  77                            sring->rsp_event - sring->rsp_prod);
  78         }
  79
  80         seq_printf(m, "NAPI state: %lx NAPI weight: %d TX queue len %u\n"
  81                    "Credit timer_pending: %d, credit: %lu, usec: %lu\n"
  82                    "remaining: %lu, expires: %lu, now: %lu\n",
  83                    queue->napi.state, queue->napi.weight,
  84                    skb_queue_len(&queue->tx_queue),
  85                    timer_pending(&queue->credit_timeout),
  86                    queue->credit_bytes,
  87                    queue->credit_usec,
  88                    queue->remaining_credit,
  89                    queue->credit_timeout.expires,
  90                    jiffies);
  91
  92         dev_queue = netdev_get_tx_queue(queue->vif->dev, queue->id);
  93
  94         seq_printf(m, "\nRx internal queue: len %u max %u pkts %u %s\n",
  95                    queue->rx_queue_len, queue->rx_queue_max,
  96                    skb_queue_len(&queue->rx_queue),
  97                    netif_tx_queue_stopped(dev_queue) ? "stopped" : "running");
  98
  99         return 0;
 100 }
 101
 102 #define XENVIF_KICK_STR "kick"
 103 #define BUFFER_SIZE     32
 104
 105 static ssize_t
 106 xenvif_write_io_ring(struct file *filp, const char __user *buf, size_t count,
 107                      loff_t *ppos)
 108 {
 109         struct xenvif_queue *queue =
 110                 ((struct seq_file *)filp->private_data)->private;
 111         int len;
 112         char write[BUFFER_SIZE];
 113
 114         /* don't allow partial writes and check the length */
 115         if (*ppos != 0)
 116                 return 0;
 117         if (count >= sizeof(write))
 118                 return -ENOSPC;
 119
 120         len = simple_write_to_buffer(write,
 121                                      sizeof(write) - 1,
 122                                      ppos,
 123                                      buf,
 124                                      count);
 125         if (len < 0)
 126                 return len;
 127
 128         write[len] = '\0';
 129
 130         if (!strncmp(write, XENVIF_KICK_STR, sizeof(XENVIF_KICK_STR) - 1))
 131                 xenvif_interrupt(0, (void *)queue);
 132         else {
 133                 pr_warn("Unknown command to io_ring_q%d. Available: kick\n",
 134                         queue->id);
 135                 count = -EINVAL;
 136         }
 137         return count;
 138 }
 139
 140 static int xenvif_io_ring_open(struct inode *inode, struct file *filp)
 141 {
 142         int ret;
 143         void *queue = NULL;
 144
 145         if (inode->i_private)
 146                 queue = inode->i_private;
 147         ret = single_open(filp, xenvif_read_io_ring, queue);
 148         filp->f_mode |= FMODE_PWRITE;
 149         return ret;
 150 }
 151
 152 static const struct file_operations xenvif_dbg_io_ring_ops_fops = {
 153         .owner = THIS_MODULE,
 154         .open = xenvif_io_ring_open,
 155         .read = seq_read,
 156         .llseek = seq_lseek,
 157         .release = single_release,
 158         .write = xenvif_write_io_ring,
 159 };
 160
 161 static int xenvif_ctrl_show(struct seq_file *m, void *v)
 162 {
 163         struct xenvif *vif = m->private;
 164
 165         xenvif_dump_hash_info(vif, m);
 166
 167         return 0;
 168 }
 169 DEFINE_SHOW_ATTRIBUTE(xenvif_ctrl);
 170
 171 static void xenvif_debugfs_addif(struct xenvif *vif)
 172 {
 173         struct dentry *pfile;
 174         int i;
 175
 176         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 177                 return;
 178
 179         vif->xenvif_dbg_root = debugfs_create_dir(vif->dev->name,
 180                                                   xen_netback_dbg_root);
 181         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) {
 182                 for (i = 0; i < vif->num_queues; ++i) {
 183                         char filename[sizeof("io_ring_q") + 4];
 184
 185                         snprintf(filename, sizeof(filename), "io_ring_q%d", i);
 186                         pfile = debugfs_create_file(filename,
 187                                                     0600,
 188                                                     vif->xenvif_dbg_root,
 189                                                     &vif->queues[i],
 190                                                     &xenvif_dbg_io_ring_ops_fops);
 191                         if (IS_ERR_OR_NULL(pfile))
 192                                 pr_warn("Creation of io_ring file returned %ld!\n",
 193                                         PTR_ERR(pfile));
 194                 }
 195
 196                 if (vif->ctrl_irq) {
 197                         pfile = debugfs_create_file("ctrl",
 198                                                     0400,
 199                                                     vif->xenvif_dbg_root,
 200                                                     vif,
 201                                                     &xenvif_ctrl_fops);
 202                         if (IS_ERR_OR_NULL(pfile))
 203                                 pr_warn("Creation of ctrl file returned %ld!\n",
 204                                         PTR_ERR(pfile));
 205                 }
 206         } else
 207                 netdev_warn(vif->dev,
 208                             "Creation of vif debugfs dir returned %ld!\n",
 209                             PTR_ERR(vif->xenvif_dbg_root));
 210 }
 211
 212 static void xenvif_debugfs_delif(struct xenvif *vif)
 213 {
 214         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 215                 return;
 216
 217         debugfs_remove_recursive(vif->xenvif_dbg_root);
 218         vif->xenvif_dbg_root = NULL;
 219 }
 220 #endif /* CONFIG_DEBUG_FS */
 221
 222 static int netback_remove(struct xenbus_device *dev)
 223 {
 224         struct backend_info *be = dev_get_drvdata(&dev->dev);
 225
 226         set_backend_state(be, XenbusStateClosed);
 227
 228         unregister_hotplug_status_watch(be);
 229         if (be->vif) {
 230                 kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE);
 231                 xen_unregister_watchers(be->vif);
 232                 xenbus_rm(XBT_NIL, dev->nodename, "hotplug-status");
 233                 xenvif_free(be->vif);
 234                 be->vif = NULL;
 235         }
 236         kfree(be->hotplug_script);
 237         kfree(be);
 238         dev_set_drvdata(&dev->dev, NULL);
 239         return 0;
 240 }
 241
 242
 243 /**
 244  * Entry point to this code when a new device is created.  Allocate the basic
 245  * structures and switch to InitWait.
 246  */
 247 static int netback_probe(struct xenbus_device *dev,
 248                          const struct xenbus_device_id *id)
 249 {
 250         const char *message;
 251         struct xenbus_transaction xbt;
 252         int err;
 253         int sg;
 254         const char *script;
 255         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 256                                           GFP_KERNEL);
 257         if (!be) {
 258                 xenbus_dev_fatal(dev, -ENOMEM,
 259                                  "allocating backend structure");
 260                 return -ENOMEM;
 261         }
 262
 263         be->dev = dev;
 264         dev_set_drvdata(&dev->dev, be);
 265
 266         be->state = XenbusStateInitialising;
 267         err = xenbus_switch_state(dev, XenbusStateInitialising);
 268         if (err)
 269                 goto fail;
 270
 271         sg = 1;
 272
 273         do {
 274                 err = xenbus_transaction_start(&xbt);
 275                 if (err) {
 276                         xenbus_dev_fatal(dev, err, "starting transaction");
 277                         goto fail;
 278                 }
 279
 280                 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg);
 281                 if (err) {
 282                         message = "writing feature-sg";
 283                         goto abort_transaction;
 284                 }
 285
 286                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4",
 287                                     "%d", sg);
 288                 if (err) {
 289                         message = "writing feature-gso-tcpv4";
 290                         goto abort_transaction;
 291                 }
 292
 293                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6",
 294                                     "%d", sg);
 295                 if (err) {
 296                         message = "writing feature-gso-tcpv6";
 297                         goto abort_transaction;
 298                 }
 299
 300                 /* We support partial checksum setup for IPv6 packets */
 301                 err = xenbus_printf(xbt, dev->nodename,
 302                                     "feature-ipv6-csum-offload",
 303                                     "%d", 1);
 304                 if (err) {
 305                         message = "writing feature-ipv6-csum-offload";
 306                         goto abort_transaction;
 307                 }
 308
 309                 /* We support rx-copy path. */
 310                 err = xenbus_printf(xbt, dev->nodename,
 311                                     "feature-rx-copy", "%d", 1);
 312                 if (err) {
 313                         message = "writing feature-rx-copy";
 314                         goto abort_transaction;
 315                 }
 316
 317                 /*
 318                  * We don't support rx-flip path (except old guests who don't
 319                  * grok this feature flag).
 320                  */
 321                 err = xenbus_printf(xbt, dev->nodename,
 322                                     "feature-rx-flip", "%d", 0);
 323                 if (err) {
 324                         message = "writing feature-rx-flip";
 325                         goto abort_transaction;
 326                 }
 327
 328                 /* We support dynamic multicast-control. */
 329                 err = xenbus_printf(xbt, dev->nodename,
 330                                     "feature-multicast-control", "%d", 1);
 331                 if (err) {
 332                         message = "writing feature-multicast-control";
 333                         goto abort_transaction;
 334                 }
 335
 336                 err = xenbus_printf(xbt, dev->nodename,
 337                                     "feature-dynamic-multicast-control",
 338                                     "%d", 1);
 339                 if (err) {
 340                         message = "writing feature-dynamic-multicast-control";
 341                         goto abort_transaction;
 342                 }
 343
 344                 err = xenbus_transaction_end(xbt, 0);
 345         } while (err == -EAGAIN);
 346
 347         if (err) {
 348                 xenbus_dev_fatal(dev, err, "completing transaction");
 349                 goto fail;
 350         }
 351
 352         /*
 353          * Split event channels support, this is optional so it is not
 354          * put inside the above loop.
 355          */
 356         err = xenbus_printf(XBT_NIL, dev->nodename,
 357                             "feature-split-event-channels",
 358                             "%u", separate_tx_rx_irq);
 359         if (err)
 360                 pr_debug("Error writing feature-split-event-channels\n");
 361
 362         /* Multi-queue support: This is an optional feature. */
 363         err = xenbus_printf(XBT_NIL, dev->nodename,
 364                             "multi-queue-max-queues", "%u", xenvif_max_queues);
 365         if (err)
 366                 pr_debug("Error writing multi-queue-max-queues\n");
 367
 368         err = xenbus_printf(XBT_NIL, dev->nodename,
 369                             "feature-ctrl-ring",
 370                             "%u", true);
 371         if (err)
 372                 pr_debug("Error writing feature-ctrl-ring\n");
 373
 374         script = xenbus_read(XBT_NIL, dev->nodename, "script", NULL);
 375         if (IS_ERR(script)) {
 376                 err = PTR_ERR(script);
 377                 xenbus_dev_fatal(dev, err, "reading script");
 378                 goto fail;
 379         }
 380
 381         be->hotplug_script = script;
 382
 383
 384         /* This kicks hotplug scripts, so do it immediately. */
 385         err = backend_create_xenvif(be);
 386         if (err)
 387                 goto fail;
 388
 389         return 0;
 390
 391 abort_transaction:
 392         xenbus_transaction_end(xbt, 1);
 393         xenbus_dev_fatal(dev, err, "%s", message);
 394 fail:
 395         pr_debug("failed\n");
 396         netback_remove(dev);
 397         return err;
 398 }
 399
 400
 401 /*
 402  * Handle the creation of the hotplug script environment.  We add the script
 403  * and vif variables to the environment, for the benefit of the vif-* hotplug
 404  * scripts.
 405  */
 406 static int netback_uevent(struct xenbus_device *xdev,
 407                           struct kobj_uevent_env *env)
 408 {
 409         struct backend_info *be = dev_get_drvdata(&xdev->dev);
 410
 411         if (!be)
 412                 return 0;
 413
 414         if (add_uevent_var(env, "script=%s", be->hotplug_script))
 415                 return -ENOMEM;
 416
 417         if (!be->vif)
 418                 return 0;
 419
 420         return add_uevent_var(env, "vif=%s", be->vif->dev->name);
 421 }
 422
 423
 424 static int backend_create_xenvif(struct backend_info *be)
 425 {
 426         int err;
 427         long handle;
 428         struct xenbus_device *dev = be->dev;
 429         struct xenvif *vif;
 430
 431         if (be->vif != NULL)
 432                 return 0;
 433
 434         err = xenbus_scanf(XBT_NIL, dev->nodename, "handle", "%li", &handle);
 435         if (err != 1) {
 436                 xenbus_dev_fatal(dev, err, "reading handle");
 437                 return (err < 0) ? err : -EINVAL;
 438         }
 439
 440         vif = xenvif_alloc(&dev->dev, dev->otherend_id, handle);
 441         if (IS_ERR(vif)) {
 442                 err = PTR_ERR(vif);
 443                 xenbus_dev_fatal(dev, err, "creating interface");
 444                 return err;
 445         }
 446         be->vif = vif;
 447         vif->be = be;
 448
 449         kobject_uevent(&dev->dev.kobj, KOBJ_ONLINE);
 450         return 0;
 451 }
 452
 453 static void backend_disconnect(struct backend_info *be)
 454 {
 455         struct xenvif *vif = be->vif;
 456
 457         if (vif) {
 458                 unsigned int num_queues = vif->num_queues;
 459                 unsigned int queue_index;
 460
 461                 xen_unregister_watchers(vif);
 462 #ifdef CONFIG_DEBUG_FS
 463                 xenvif_debugfs_delif(vif);
 464 #endif /* CONFIG_DEBUG_FS */
 465                 xenvif_disconnect_data(vif);
 466
 467                 /* At this point some of the handlers may still be active
 468                  * so we need to have additional synchronization here.
 469                  */
 470                 vif->num_queues = 0;
 471                 synchronize_net();
 472
 473                 for (queue_index = 0; queue_index < num_queues; ++queue_index)
 474                         xenvif_deinit_queue(&vif->queues[queue_index]);
 475
 476                 vfree(vif->queues);
 477                 vif->queues = NULL;
 478
 479                 xenvif_disconnect_ctrl(vif);
 480         }
 481 }
 482
 483 static void backend_connect(struct backend_info *be)
 484 {
 485         if (be->vif)
 486                 connect(be);
 487 }
 488
 489 static inline void backend_switch_state(struct backend_info *be,
 490                                         enum xenbus_state state)
 491 {
 492         struct xenbus_device *dev = be->dev;
 493
 494         pr_debug("%s -> %s\n", dev->nodename, xenbus_strstate(state));
 495         be->state = state;
 496
 497         /* If we are waiting for a hotplug script then defer the
 498          * actual xenbus state change.
 499          */
 500         if (!be->have_hotplug_status_watch)
 501                 xenbus_switch_state(dev, state);
 502 }
 503
 504 /* Handle backend state transitions:
 505  *
 506  * The backend state starts in Initialising and the following transitions are
 507  * allowed.
 508  *
 509  * Initialising -> InitWait -> Connected
 510  *          \
 511  *           \        ^    \         |
 512  *            \       |     \        |
 513  *             \      |      \       |
 514  *              \     |       \      |
 515  *               \    |        \     |
 516  *                \   |         \    |
 517  *                 V  |          V   V
 518  *
 519  *                  Closed  <-> Closing
 520  *
 521  * The state argument specifies the eventual state of the backend and the
 522  * function transitions to that state via the shortest path.
 523  */
 524 static void set_backend_state(struct backend_info *be,
 525                               enum xenbus_state state)
 526 {
 527         while (be->state != state) {
 528                 switch (be->state) {
 529                 case XenbusStateInitialising:
 530                         switch (state) {
 531                         case XenbusStateInitWait:
 532                         case XenbusStateConnected:
 533                         case XenbusStateClosing:
 534                                 backend_switch_state(be, XenbusStateInitWait);
 535                                 break;
 536                         case XenbusStateClosed:
 537                                 backend_switch_state(be, XenbusStateClosed);
 538                                 break;
 539                         default:
 540                                 BUG();
 541                         }
 542                         break;
 543                 case XenbusStateClosed:
 544                         switch (state) {
 545                         case XenbusStateInitWait:
 546                         case XenbusStateConnected:
 547                                 backend_switch_state(be, XenbusStateInitWait);
 548                                 break;
 549                         case XenbusStateClosing:
 550                                 backend_switch_state(be, XenbusStateClosing);
 551                                 break;
 552                         default:
 553                                 BUG();
 554                         }
 555                         break;
 556                 case XenbusStateInitWait:
 557                         switch (state) {
 558                         case XenbusStateConnected:
 559                                 backend_connect(be);
 560                                 backend_switch_state(be, XenbusStateConnected);
 561                                 break;
 562                         case XenbusStateClosing:
 563                         case XenbusStateClosed:
 564                                 backend_switch_state(be, XenbusStateClosing);
 565                                 break;
 566                         default:
 567                                 BUG();
 568                         }
 569                         break;
 570                 case XenbusStateConnected:
 571                         switch (state) {
 572                         case XenbusStateInitWait:
 573                         case XenbusStateClosing:
 574                         case XenbusStateClosed:
 575                                 backend_disconnect(be);
 576                                 backend_switch_state(be, XenbusStateClosing);
 577                                 break;
 578                         default:
 579                                 BUG();
 580                         }
 581                         break;
 582                 case XenbusStateClosing:
 583                         switch (state) {
 584                         case XenbusStateInitWait:
 585                         case XenbusStateConnected:
 586                         case XenbusStateClosed:
 587                                 backend_switch_state(be, XenbusStateClosed);
 588                                 break;
 589                         default:
 590                                 BUG();
 591                         }
 592                         break;
 593                 default:
 594                         BUG();
 595                 }
 596         }
 597 }
 598
 599 /**
 600  * Callback received when the frontend's state changes.
 601  */
 602 static void frontend_changed(struct xenbus_device *dev,
 603                              enum xenbus_state frontend_state)
 604 {
 605         struct backend_info *be = dev_get_drvdata(&dev->dev);
 606
 607         pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state));
 608
 609         be->frontend_state = frontend_state;
 610
 611         switch (frontend_state) {
 612         case XenbusStateInitialising:
 613                 set_backend_state(be, XenbusStateInitWait);
 614                 break;
 615
 616         case XenbusStateInitialised:
 617                 break;
 618
 619         case XenbusStateConnected:
 620                 set_backend_state(be, XenbusStateConnected);
 621                 break;
 622
 623         case XenbusStateClosing:
 624                 set_backend_state(be, XenbusStateClosing);
 625                 break;
 626
 627         case XenbusStateClosed:
 628                 set_backend_state(be, XenbusStateClosed);
 629                 if (xenbus_dev_is_online(dev))
 630                         break;
 631                 /* fall through - if not online */
 632         case XenbusStateUnknown:
 633                 set_backend_state(be, XenbusStateClosed);
 634                 device_unregister(&dev->dev);
 635                 break;
 636
 637         default:
 638                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 639                                  frontend_state);
 640                 break;
 641         }
 642 }
 643
 644
 645 static void xen_net_read_rate(struct xenbus_device *dev,
 646                               unsigned long *bytes, unsigned long *usec)
 647 {
 648         char *s, *e;
 649         unsigned long b, u;
 650         char *ratestr;
 651
 652         /* Default to unlimited bandwidth. */
 653         *bytes = ~0UL;
 654         *usec = 0;
 655
 656         ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL);
 657         if (IS_ERR(ratestr))
 658                 return;
 659
 660         s = ratestr;
 661         b = simple_strtoul(s, &e, 10);
 662         if ((s == e) || (*e != ','))
 663                 goto fail;
 664
 665         s = e + 1;
 666         u = simple_strtoul(s, &e, 10);
 667         if ((s == e) || (*e != '\0'))
 668                 goto fail;
 669
 670         *bytes = b;
 671         *usec = u;
 672
 673         kfree(ratestr);
 674         return;
 675
 676  fail:
 677         pr_warn("Failed to parse network rate limit. Traffic unlimited.\n");
 678         kfree(ratestr);
 679 }
 680
 681 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
 682 {
 683         char *s, *e, *macstr;
 684         int i;
 685
 686         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
 687         if (IS_ERR(macstr))
 688                 return PTR_ERR(macstr);
 689
 690         for (i = 0; i < ETH_ALEN; i++) {
 691                 mac[i] = simple_strtoul(s, &e, 16);
 692                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
 693                         kfree(macstr);
 694                         return -ENOENT;
 695                 }
 696                 s = e+1;
 697         }
 698
 699         kfree(macstr);
 700         return 0;
 701 }
 702
 703 static void xen_net_rate_changed(struct xenbus_watch *watch,
 704                                  const char *path, const char *token)
 705 {
 706         struct xenvif *vif = container_of(watch, struct xenvif, credit_watch);
 707         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 708         unsigned long   credit_bytes;
 709         unsigned long   credit_usec;
 710         unsigned int queue_index;
 711
 712         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 713         for (queue_index = 0; queue_index < vif->num_queues; queue_index++) {
 714                 struct xenvif_queue *queue = &vif->queues[queue_index];
 715
 716                 queue->credit_bytes = credit_bytes;
 717                 queue->credit_usec = credit_usec;
 718                 if (!mod_timer_pending(&queue->credit_timeout, jiffies) &&
 719                         queue->remaining_credit > queue->credit_bytes) {
 720                         queue->remaining_credit = queue->credit_bytes;
 721                 }
 722         }
 723 }
 724
 725 static int xen_register_credit_watch(struct xenbus_device *dev,
 726                                      struct xenvif *vif)
 727 {
 728         int err = 0;
 729         char *node;
 730         unsigned maxlen = strlen(dev->nodename) + sizeof("/rate");
 731
 732         if (vif->credit_watch.node)
 733                 return -EADDRINUSE;
 734
 735         node = kmalloc(maxlen, GFP_KERNEL);
 736         if (!node)
 737                 return -ENOMEM;
 738         snprintf(node, maxlen, "%s/rate", dev->nodename);
 739         vif->credit_watch.node = node;
 740         vif->credit_watch.callback = xen_net_rate_changed;
 741         err = register_xenbus_watch(&vif->credit_watch);
 742         if (err) {
 743                 pr_err("Failed to set watcher %s\n", vif->credit_watch.node);
 744                 kfree(node);
 745                 vif->credit_watch.node = NULL;
 746                 vif->credit_watch.callback = NULL;
 747         }
 748         return err;
 749 }
 750
 751 static void xen_unregister_credit_watch(struct xenvif *vif)
 752 {
 753         if (vif->credit_watch.node) {
 754                 unregister_xenbus_watch(&vif->credit_watch);
 755                 kfree(vif->credit_watch.node);
 756                 vif->credit_watch.node = NULL;
 757         }
 758 }
 759
 760 static void xen_mcast_ctrl_changed(struct xenbus_watch *watch,
 761                                    const char *path, const char *token)
 762 {
 763         struct xenvif *vif = container_of(watch, struct xenvif,
 764                                           mcast_ctrl_watch);
 765         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 766
 767         vif->multicast_control = !!xenbus_read_unsigned(dev->otherend,
 768                                         "request-multicast-control", 0);
 769 }
 770
 771 static int xen_register_mcast_ctrl_watch(struct xenbus_device *dev,
 772                                          struct xenvif *vif)
 773 {
 774         int err = 0;
 775         char *node;
 776         unsigned maxlen = strlen(dev->otherend) +
 777                 sizeof("/request-multicast-control");
 778
 779         if (vif->mcast_ctrl_watch.node) {
 780                 pr_err_ratelimited("Watch is already registered\n");
 781                 return -EADDRINUSE;
 782         }
 783
 784         node = kmalloc(maxlen, GFP_KERNEL);
 785         if (!node) {
 786                 pr_err("Failed to allocate memory for watch\n");
 787                 return -ENOMEM;
 788         }
 789         snprintf(node, maxlen, "%s/request-multicast-control",
 790                  dev->otherend);
 791         vif->mcast_ctrl_watch.node = node;
 792         vif->mcast_ctrl_watch.callback = xen_mcast_ctrl_changed;
 793         err = register_xenbus_watch(&vif->mcast_ctrl_watch);
 794         if (err) {
 795                 pr_err("Failed to set watcher %s\n",
 796                        vif->mcast_ctrl_watch.node);
 797                 kfree(node);
 798                 vif->mcast_ctrl_watch.node = NULL;
 799                 vif->mcast_ctrl_watch.callback = NULL;
 800         }
 801         return err;
 802 }
 803
 804 static void xen_unregister_mcast_ctrl_watch(struct xenvif *vif)
 805 {
 806         if (vif->mcast_ctrl_watch.node) {
 807                 unregister_xenbus_watch(&vif->mcast_ctrl_watch);
 808                 kfree(vif->mcast_ctrl_watch.node);
 809                 vif->mcast_ctrl_watch.node = NULL;
 810         }
 811 }
 812
 813 static void xen_register_watchers(struct xenbus_device *dev,
 814                                   struct xenvif *vif)
 815 {
 816         xen_register_credit_watch(dev, vif);
 817         xen_register_mcast_ctrl_watch(dev, vif);
 818 }
 819
 820 static void xen_unregister_watchers(struct xenvif *vif)
 821 {
 822         xen_unregister_mcast_ctrl_watch(vif);
 823         xen_unregister_credit_watch(vif);
 824 }
 825
 826 static void unregister_hotplug_status_watch(struct backend_info *be)
 827 {
 828         if (be->have_hotplug_status_watch) {
 829                 unregister_xenbus_watch(&be->hotplug_status_watch);
 830                 kfree(be->hotplug_status_watch.node);
 831         }
 832         be->have_hotplug_status_watch = 0;
 833 }
 834
 835 static void hotplug_status_changed(struct xenbus_watch *watch,
 836                                    const char *path,
 837                                    const char *token)
 838 {
 839         struct backend_info *be = container_of(watch,
 840                                                struct backend_info,
 841                                                hotplug_status_watch);
 842         char *str;
 843         unsigned int len;
 844
 845         str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len);
 846         if (IS_ERR(str))
 847                 return;
 848         if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) {
 849                 /* Complete any pending state change */
 850                 xenbus_switch_state(be->dev, be->state);
 851
 852                 /* Not interested in this watch anymore. */
 853                 unregister_hotplug_status_watch(be);
 854         }
 855         kfree(str);
 856 }
 857
 858 static int connect_ctrl_ring(struct backend_info *be)
 859 {
 860         struct xenbus_device *dev = be->dev;
 861         struct xenvif *vif = be->vif;
 862         unsigned int val;
 863         grant_ref_t ring_ref;
 864         unsigned int evtchn;
 865         int err;
 866
 867         err = xenbus_scanf(XBT_NIL, dev->otherend,
 868                            "ctrl-ring-ref", "%u", &val);
 869         if (err < 0)
 870                 goto done; /* The frontend does not have a control ring */
 871
 872         ring_ref = val;
 873
 874         err = xenbus_scanf(XBT_NIL, dev->otherend,
 875                            "event-channel-ctrl", "%u", &val);
 876         if (err < 0) {
 877                 xenbus_dev_fatal(dev, err,
 878                                  "reading %s/event-channel-ctrl",
 879                                  dev->otherend);
 880                 goto fail;
 881         }
 882
 883         evtchn = val;
 884
 885         err = xenvif_connect_ctrl(vif, ring_ref, evtchn);
 886         if (err) {
 887                 xenbus_dev_fatal(dev, err,
 888                                  "mapping shared-frame %u port %u",
 889                                  ring_ref, evtchn);
 890                 goto fail;
 891         }
 892
 893 done:
 894         return 0;
 895
 896 fail:
 897         return err;
 898 }
 899
 900 static void connect(struct backend_info *be)
 901 {
 902         int err;
 903         struct xenbus_device *dev = be->dev;
 904         unsigned long credit_bytes, credit_usec;
 905         unsigned int queue_index;
 906         unsigned int requested_num_queues;
 907         struct xenvif_queue *queue;
 908
 909         /* Check whether the frontend requested multiple queues
 910          * and read the number requested.
 911          */
 912         requested_num_queues = xenbus_read_unsigned(dev->otherend,
 913                                         "multi-queue-num-queues", 1);
 914         if (requested_num_queues > xenvif_max_queues) {
 915                 /* buggy or malicious guest */
 916                 xenbus_dev_fatal(dev, -EINVAL,
 917                                  "guest requested %u queues, exceeding the maximum of %u.",
 918                                  requested_num_queues, xenvif_max_queues);
 919                 return;
 920         }
 921
 922         err = xen_net_read_mac(dev, be->vif->fe_dev_addr);
 923         if (err) {
 924                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
 925                 return;
 926         }
 927
 928         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 929         xen_unregister_watchers(be->vif);
 930         xen_register_watchers(dev, be->vif);
 931         read_xenbus_vif_flags(be);
 932
 933         err = connect_ctrl_ring(be);
 934         if (err) {
 935                 xenbus_dev_fatal(dev, err, "connecting control ring");
 936                 return;
 937         }
 938
 939         /* Use the number of queues requested by the frontend */
 940         be->vif->queues = vzalloc(array_size(requested_num_queues,
 941                                              sizeof(struct xenvif_queue)));
 942         if (!be->vif->queues) {
 943                 xenbus_dev_fatal(dev, -ENOMEM,
 944                                  "allocating queues");
 945                 return;
 946         }
 947
 948         be->vif->num_queues = requested_num_queues;
 949         be->vif->stalled_queues = requested_num_queues;
 950
 951         for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) {
 952                 queue = &be->vif->queues[queue_index];
 953                 queue->vif = be->vif;
 954                 queue->id = queue_index;
 955                 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
 956                                 be->vif->dev->name, queue->id);
 957
 958                 err = xenvif_init_queue(queue);
 959                 if (err) {
 960                         /* xenvif_init_queue() cleans up after itself on
 961                          * failure, but we need to clean up any previously
 962                          * initialised queues. Set num_queues to i so that
 963                          * earlier queues can be destroyed using the regular
 964                          * disconnect logic.
 965                          */
 966                         be->vif->num_queues = queue_index;
 967                         goto err;
 968                 }
 969
 970                 queue->credit_bytes = credit_bytes;
 971                 queue->remaining_credit = credit_bytes;
 972                 queue->credit_usec = credit_usec;
 973
 974                 err = connect_data_rings(be, queue);
 975                 if (err) {
 976                         /* connect_data_rings() cleans up after itself on
 977                          * failure, but we need to clean up after
 978                          * xenvif_init_queue() here, and also clean up any
 979                          * previously initialised queues.
 980                          */
 981                         xenvif_deinit_queue(queue);
 982                         be->vif->num_queues = queue_index;
 983                         goto err;
 984                 }
 985         }
 986
 987 #ifdef CONFIG_DEBUG_FS
 988         xenvif_debugfs_addif(be->vif);
 989 #endif /* CONFIG_DEBUG_FS */
 990
 991         /* Initialisation completed, tell core driver the number of
 992          * active queues.
 993          */
 994         rtnl_lock();
 995         netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues);
 996         netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues);
 997         rtnl_unlock();
 998
 999         xenvif_carrier_on(be->vif);
1000
1001         unregister_hotplug_status_watch(be);
1002         err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch,
1003                                    hotplug_status_changed,
1004                                    "%s/%s", dev->nodename, "hotplug-status");
1005         if (!err)
1006                 be->have_hotplug_status_watch = 1;
1007
1008         netif_tx_wake_all_queues(be->vif->dev);
1009
1010         return;
1011
1012 err:
1013         if (be->vif->num_queues > 0)
1014                 xenvif_disconnect_data(be->vif); /* Clean up existing queues */
1015         for (queue_index = 0; queue_index < be->vif->num_queues; ++queue_index)
1016                 xenvif_deinit_queue(&be->vif->queues[queue_index]);
1017         vfree(be->vif->queues);
1018         be->vif->queues = NULL;
1019         be->vif->num_queues = 0;
1020         xenvif_disconnect_ctrl(be->vif);
1021         return;
1022 }
1023
1024
1025 static int connect_data_rings(struct backend_info *be,
1026                               struct xenvif_queue *queue)
1027 {
1028         struct xenbus_device *dev = be->dev;
1029         unsigned int num_queues = queue->vif->num_queues;
1030         unsigned long tx_ring_ref, rx_ring_ref;
1031         unsigned int tx_evtchn, rx_evtchn;
1032         int err;
1033         char *xspath;
1034         size_t xspathsize;
1035         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1036
1037         /* If the frontend requested 1 queue, or we have fallen back
1038          * to single queue due to lack of frontend support for multi-
1039          * queue, expect the remaining XenStore keys in the toplevel
1040          * directory. Otherwise, expect them in a subdirectory called
1041          * queue-N.
1042          */
1043         if (num_queues == 1) {
1044                 xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL);
1045                 if (!xspath) {
1046                         xenbus_dev_fatal(dev, -ENOMEM,
1047                                          "reading ring references");
1048                         return -ENOMEM;
1049                 }
1050                 strcpy(xspath, dev->otherend);
1051         } else {
1052                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1053                 xspath = kzalloc(xspathsize, GFP_KERNEL);
1054                 if (!xspath) {
1055                         xenbus_dev_fatal(dev, -ENOMEM,
1056                                          "reading ring references");
1057                         return -ENOMEM;
1058                 }
1059                 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend,
1060                          queue->id);
1061         }
1062
1063         err = xenbus_gather(XBT_NIL, xspath,
1064                             "tx-ring-ref", "%lu", &tx_ring_ref,
1065                             "rx-ring-ref", "%lu", &rx_ring_ref, NULL);
1066         if (err) {
1067                 xenbus_dev_fatal(dev, err,
1068                                  "reading %s/ring-ref",
1069                                  xspath);
1070                 goto err;
1071         }
1072
1073         /* Try split event channels first, then single event channel. */
1074         err = xenbus_gather(XBT_NIL, xspath,
1075                             "event-channel-tx", "%u", &tx_evtchn,
1076                             "event-channel-rx", "%u", &rx_evtchn, NULL);
1077         if (err < 0) {
1078                 err = xenbus_scanf(XBT_NIL, xspath,
1079                                    "event-channel", "%u", &tx_evtchn);
1080                 if (err < 0) {
1081                         xenbus_dev_fatal(dev, err,
1082                                          "reading %s/event-channel(-tx/rx)",
1083                                          xspath);
1084                         goto err;
1085                 }
1086                 rx_evtchn = tx_evtchn;
1087         }
1088
1089         /* Map the shared frame, irq etc. */
1090         err = xenvif_connect_data(queue, tx_ring_ref, rx_ring_ref,
1091                                   tx_evtchn, rx_evtchn);
1092         if (err) {
1093                 xenbus_dev_fatal(dev, err,
1094                                  "mapping shared-frames %lu/%lu port tx %u rx %u",
1095                                  tx_ring_ref, rx_ring_ref,
1096                                  tx_evtchn, rx_evtchn);
1097                 goto err;
1098         }
1099
1100         err = 0;
1101 err: /* Regular return falls through with err == 0 */
1102         kfree(xspath);
1103         return err;
1104 }
1105
1106 static int read_xenbus_vif_flags(struct backend_info *be)
1107 {
1108         struct xenvif *vif = be->vif;
1109         struct xenbus_device *dev = be->dev;
1110         unsigned int rx_copy;
1111         int err;
1112
1113         err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u",
1114                            &rx_copy);
1115         if (err == -ENOENT) {
1116                 err = 0;
1117                 rx_copy = 0;
1118         }
1119         if (err < 0) {
1120                 xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy",
1121                                  dev->otherend);
1122                 return err;
1123         }
1124         if (!rx_copy)
1125                 return -EOPNOTSUPP;
1126
1127         if (!xenbus_read_unsigned(dev->otherend, "feature-rx-notify", 0)) {
1128                 /* - Reduce drain timeout to poll more frequently for
1129                  *   Rx requests.
1130                  * - Disable Rx stall detection.
1131                  */
1132                 be->vif->drain_timeout = msecs_to_jiffies(30);
1133                 be->vif->stall_timeout = 0;
1134         }
1135
1136         vif->can_sg = !!xenbus_read_unsigned(dev->otherend, "feature-sg", 0);
1137
1138         vif->gso_mask = 0;
1139
1140         if (xenbus_read_unsigned(dev->otherend, "feature-gso-tcpv4", 0))
1141                 vif->gso_mask |= GSO_BIT(TCPV4);
1142
1143         if (xenbus_read_unsigned(dev->otherend, "feature-gso-tcpv6", 0))
1144                 vif->gso_mask |= GSO_BIT(TCPV6);
1145
1146         vif->ip_csum = !xenbus_read_unsigned(dev->otherend,
1147                                              "feature-no-csum-offload", 0);
1148
1149         vif->ipv6_csum = !!xenbus_read_unsigned(dev->otherend,
1150                                                 "feature-ipv6-csum-offload", 0);
1151
1152         return 0;
1153 }
1154
1155 static const struct xenbus_device_id netback_ids[] = {
1156         { "vif" },
1157         { "" }
1158 };
1159
1160 static struct xenbus_driver netback_driver = {
1161         .ids = netback_ids,
1162         .probe = netback_probe,
1163         .remove = netback_remove,
1164         .uevent = netback_uevent,
1165         .otherend_changed = frontend_changed,
1166 };
1167
1168 int xenvif_xenbus_init(void)
1169 {
1170         return xenbus_register_backend(&netback_driver);
1171 }
1172
1173 void xenvif_xenbus_fini(void)
1174 {
1175         return xenbus_unregister_driver(&netback_driver);
1176 }