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