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