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 static void read_xenbus_frontend_xdp(struct backend_info *be,
 397                                       struct xenbus_device *dev)
 398 {
 399         struct xenvif *vif = be->vif;
 400         u16 headroom;
 401         int err;
 402
 403         err = xenbus_scanf(XBT_NIL, dev->otherend,
 404                            "xdp-headroom", "%hu", &headroom);
 405         if (err != 1) {
 406                 vif->xdp_headroom = 0;
 407                 return;
 408         }
 409         if (headroom > XEN_NETIF_MAX_XDP_HEADROOM)
 410                 headroom = XEN_NETIF_MAX_XDP_HEADROOM;
 411         vif->xdp_headroom = headroom;
 412 }
 413
 414 /**
 415  * Callback received when the frontend's state changes.
 416  */
 417 static void frontend_changed(struct xenbus_device *dev,
 418                              enum xenbus_state frontend_state)
 419 {
 420         struct backend_info *be = dev_get_drvdata(&dev->dev);
 421
 422         pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state));
 423
 424         be->frontend_state = frontend_state;
 425
 426         switch (frontend_state) {
 427         case XenbusStateInitialising:
 428                 set_backend_state(be, XenbusStateInitWait);
 429                 break;
 430
 431         case XenbusStateInitialised:
 432                 break;
 433
 434         case XenbusStateConnected:
 435                 set_backend_state(be, XenbusStateConnected);
 436                 break;
 437
 438         case XenbusStateReconfiguring:
 439                 read_xenbus_frontend_xdp(be, dev);
 440                 xenbus_switch_state(dev, XenbusStateReconfigured);
 441                 break;
 442
 443         case XenbusStateClosing:
 444                 set_backend_state(be, XenbusStateClosing);
 445                 break;
 446
 447         case XenbusStateClosed:
 448                 set_backend_state(be, XenbusStateClosed);
 449                 if (xenbus_dev_is_online(dev))
 450                         break;
 451                 fallthrough;    /* if not online */
 452         case XenbusStateUnknown:
 453                 set_backend_state(be, XenbusStateClosed);
 454                 device_unregister(&dev->dev);
 455                 break;
 456
 457         default:
 458                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 459                                  frontend_state);
 460                 break;
 461         }
 462 }
 463
 464
 465 static void xen_net_read_rate(struct xenbus_device *dev,
 466                               unsigned long *bytes, unsigned long *usec)
 467 {
 468         char *s, *e;
 469         unsigned long b, u;
 470         char *ratestr;
 471
 472         /* Default to unlimited bandwidth. */
 473         *bytes = ~0UL;
 474         *usec = 0;
 475
 476         ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL);
 477         if (IS_ERR(ratestr))
 478                 return;
 479
 480         s = ratestr;
 481         b = simple_strtoul(s, &e, 10);
 482         if ((s == e) || (*e != ','))
 483                 goto fail;
 484
 485         s = e + 1;
 486         u = simple_strtoul(s, &e, 10);
 487         if ((s == e) || (*e != '\0'))
 488                 goto fail;
 489
 490         *bytes = b;
 491         *usec = u;
 492
 493         kfree(ratestr);
 494         return;
 495
 496  fail:
 497         pr_warn("Failed to parse network rate limit. Traffic unlimited.\n");
 498         kfree(ratestr);
 499 }
 500
 501 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
 502 {
 503         char *s, *e, *macstr;
 504         int i;
 505
 506         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
 507         if (IS_ERR(macstr))
 508                 return PTR_ERR(macstr);
 509
 510         for (i = 0; i < ETH_ALEN; i++) {
 511                 mac[i] = simple_strtoul(s, &e, 16);
 512                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
 513                         kfree(macstr);
 514                         return -ENOENT;
 515                 }
 516                 s = e+1;
 517         }
 518
 519         kfree(macstr);
 520         return 0;
 521 }
 522
 523 static void xen_net_rate_changed(struct xenbus_watch *watch,
 524                                  const char *path, const char *token)
 525 {
 526         struct xenvif *vif = container_of(watch, struct xenvif, credit_watch);
 527         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 528         unsigned long   credit_bytes;
 529         unsigned long   credit_usec;
 530         unsigned int queue_index;
 531
 532         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 533         for (queue_index = 0; queue_index < vif->num_queues; queue_index++) {
 534                 struct xenvif_queue *queue = &vif->queues[queue_index];
 535
 536                 queue->credit_bytes = credit_bytes;
 537                 queue->credit_usec = credit_usec;
 538                 if (!mod_timer_pending(&queue->credit_timeout, jiffies) &&
 539                         queue->remaining_credit > queue->credit_bytes) {
 540                         queue->remaining_credit = queue->credit_bytes;
 541                 }
 542         }
 543 }
 544
 545 static int xen_register_credit_watch(struct xenbus_device *dev,
 546                                      struct xenvif *vif)
 547 {
 548         int err = 0;
 549         char *node;
 550         unsigned maxlen = strlen(dev->nodename) + sizeof("/rate");
 551
 552         if (vif->credit_watch.node)
 553                 return -EADDRINUSE;
 554
 555         node = kmalloc(maxlen, GFP_KERNEL);
 556         if (!node)
 557                 return -ENOMEM;
 558         snprintf(node, maxlen, "%s/rate", dev->nodename);
 559         vif->credit_watch.node = node;
 560         vif->credit_watch.will_handle = NULL;
 561         vif->credit_watch.callback = xen_net_rate_changed;
 562         err = register_xenbus_watch(&vif->credit_watch);
 563         if (err) {
 564                 pr_err("Failed to set watcher %s\n", vif->credit_watch.node);
 565                 kfree(node);
 566                 vif->credit_watch.node = NULL;
 567                 vif->credit_watch.will_handle = NULL;
 568                 vif->credit_watch.callback = NULL;
 569         }
 570         return err;
 571 }
 572
 573 static void xen_unregister_credit_watch(struct xenvif *vif)
 574 {
 575         if (vif->credit_watch.node) {
 576                 unregister_xenbus_watch(&vif->credit_watch);
 577                 kfree(vif->credit_watch.node);
 578                 vif->credit_watch.node = NULL;
 579         }
 580 }
 581
 582 static void xen_mcast_ctrl_changed(struct xenbus_watch *watch,
 583                                    const char *path, const char *token)
 584 {
 585         struct xenvif *vif = container_of(watch, struct xenvif,
 586                                           mcast_ctrl_watch);
 587         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 588
 589         vif->multicast_control = !!xenbus_read_unsigned(dev->otherend,
 590                                         "request-multicast-control", 0);
 591 }
 592
 593 static int xen_register_mcast_ctrl_watch(struct xenbus_device *dev,
 594                                          struct xenvif *vif)
 595 {
 596         int err = 0;
 597         char *node;
 598         unsigned maxlen = strlen(dev->otherend) +
 599                 sizeof("/request-multicast-control");
 600
 601         if (vif->mcast_ctrl_watch.node) {
 602                 pr_err_ratelimited("Watch is already registered\n");
 603                 return -EADDRINUSE;
 604         }
 605
 606         node = kmalloc(maxlen, GFP_KERNEL);
 607         if (!node) {
 608                 pr_err("Failed to allocate memory for watch\n");
 609                 return -ENOMEM;
 610         }
 611         snprintf(node, maxlen, "%s/request-multicast-control",
 612                  dev->otherend);
 613         vif->mcast_ctrl_watch.node = node;
 614         vif->mcast_ctrl_watch.will_handle = NULL;
 615         vif->mcast_ctrl_watch.callback = xen_mcast_ctrl_changed;
 616         err = register_xenbus_watch(&vif->mcast_ctrl_watch);
 617         if (err) {
 618                 pr_err("Failed to set watcher %s\n",
 619                        vif->mcast_ctrl_watch.node);
 620                 kfree(node);
 621                 vif->mcast_ctrl_watch.node = NULL;
 622                 vif->mcast_ctrl_watch.will_handle = NULL;
 623                 vif->mcast_ctrl_watch.callback = NULL;
 624         }
 625         return err;
 626 }
 627
 628 static void xen_unregister_mcast_ctrl_watch(struct xenvif *vif)
 629 {
 630         if (vif->mcast_ctrl_watch.node) {
 631                 unregister_xenbus_watch(&vif->mcast_ctrl_watch);
 632                 kfree(vif->mcast_ctrl_watch.node);
 633                 vif->mcast_ctrl_watch.node = NULL;
 634         }
 635 }
 636
 637 static void xen_register_watchers(struct xenbus_device *dev,
 638                                   struct xenvif *vif)
 639 {
 640         xen_register_credit_watch(dev, vif);
 641         xen_register_mcast_ctrl_watch(dev, vif);
 642 }
 643
 644 static void xen_unregister_watchers(struct xenvif *vif)
 645 {
 646         xen_unregister_mcast_ctrl_watch(vif);
 647         xen_unregister_credit_watch(vif);
 648 }
 649
 650 static void unregister_hotplug_status_watch(struct backend_info *be)
 651 {
 652         if (be->have_hotplug_status_watch) {
 653                 unregister_xenbus_watch(&be->hotplug_status_watch);
 654                 kfree(be->hotplug_status_watch.node);
 655         }
 656         be->have_hotplug_status_watch = 0;
 657 }
 658
 659 static void hotplug_status_changed(struct xenbus_watch *watch,
 660                                    const char *path,
 661                                    const char *token)
 662 {
 663         struct backend_info *be = container_of(watch,
 664                                                struct backend_info,
 665                                                hotplug_status_watch);
 666         char *str;
 667         unsigned int len;
 668
 669         str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len);
 670         if (IS_ERR(str))
 671                 return;
 672         if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) {
 673                 /* Complete any pending state change */
 674                 xenbus_switch_state(be->dev, be->state);
 675
 676                 /* Not interested in this watch anymore. */
 677                 unregister_hotplug_status_watch(be);
 678                 xenbus_rm(XBT_NIL, be->dev->nodename, "hotplug-status");
 679         }
 680         kfree(str);
 681 }
 682
 683 static int connect_ctrl_ring(struct backend_info *be)
 684 {
 685         struct xenbus_device *dev = be->dev;
 686         struct xenvif *vif = be->vif;
 687         unsigned int val;
 688         grant_ref_t ring_ref;
 689         unsigned int evtchn;
 690         int err;
 691
 692         err = xenbus_scanf(XBT_NIL, dev->otherend,
 693                            "ctrl-ring-ref", "%u", &val);
 694         if (err < 0)
 695                 goto done; /* The frontend does not have a control ring */
 696
 697         ring_ref = val;
 698
 699         err = xenbus_scanf(XBT_NIL, dev->otherend,
 700                            "event-channel-ctrl", "%u", &val);
 701         if (err < 0) {
 702                 xenbus_dev_fatal(dev, err,
 703                                  "reading %s/event-channel-ctrl",
 704                                  dev->otherend);
 705                 goto fail;
 706         }
 707
 708         evtchn = val;
 709
 710         err = xenvif_connect_ctrl(vif, ring_ref, evtchn);
 711         if (err) {
 712                 xenbus_dev_fatal(dev, err,
 713                                  "mapping shared-frame %u port %u",
 714                                  ring_ref, evtchn);
 715                 goto fail;
 716         }
 717
 718 done:
 719         return 0;
 720
 721 fail:
 722         return err;
 723 }
 724
 725 static void connect(struct backend_info *be)
 726 {
 727         int err;
 728         struct xenbus_device *dev = be->dev;
 729         unsigned long credit_bytes, credit_usec;
 730         unsigned int queue_index;
 731         unsigned int requested_num_queues;
 732         struct xenvif_queue *queue;
 733
 734         /* Check whether the frontend requested multiple queues
 735          * and read the number requested.
 736          */
 737         requested_num_queues = xenbus_read_unsigned(dev->otherend,
 738                                         "multi-queue-num-queues", 1);
 739         if (requested_num_queues > xenvif_max_queues) {
 740                 /* buggy or malicious guest */
 741                 xenbus_dev_fatal(dev, -EINVAL,
 742                                  "guest requested %u queues, exceeding the maximum of %u.",
 743                                  requested_num_queues, xenvif_max_queues);
 744                 return;
 745         }
 746
 747         err = xen_net_read_mac(dev, be->vif->fe_dev_addr);
 748         if (err) {
 749                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
 750                 return;
 751         }
 752
 753         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 754         xen_unregister_watchers(be->vif);
 755         xen_register_watchers(dev, be->vif);
 756         read_xenbus_vif_flags(be);
 757
 758         err = connect_ctrl_ring(be);
 759         if (err) {
 760                 xenbus_dev_fatal(dev, err, "connecting control ring");
 761                 return;
 762         }
 763
 764         /* Use the number of queues requested by the frontend */
 765         be->vif->queues = vzalloc(array_size(requested_num_queues,
 766                                              sizeof(struct xenvif_queue)));
 767         if (!be->vif->queues) {
 768                 xenbus_dev_fatal(dev, -ENOMEM,
 769                                  "allocating queues");
 770                 return;
 771         }
 772
 773         be->vif->num_queues = requested_num_queues;
 774         be->vif->stalled_queues = requested_num_queues;
 775
 776         for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) {
 777                 queue = &be->vif->queues[queue_index];
 778                 queue->vif = be->vif;
 779                 queue->id = queue_index;
 780                 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
 781                                 be->vif->dev->name, queue->id);
 782
 783                 err = xenvif_init_queue(queue);
 784                 if (err) {
 785                         /* xenvif_init_queue() cleans up after itself on
 786                          * failure, but we need to clean up any previously
 787                          * initialised queues. Set num_queues to i so that
 788                          * earlier queues can be destroyed using the regular
 789                          * disconnect logic.
 790                          */
 791                         be->vif->num_queues = queue_index;
 792                         goto err;
 793                 }
 794
 795                 queue->credit_bytes = credit_bytes;
 796                 queue->remaining_credit = credit_bytes;
 797                 queue->credit_usec = credit_usec;
 798
 799                 err = connect_data_rings(be, queue);
 800                 if (err) {
 801                         /* connect_data_rings() cleans up after itself on
 802                          * failure, but we need to clean up after
 803                          * xenvif_init_queue() here, and also clean up any
 804                          * previously initialised queues.
 805                          */
 806                         xenvif_deinit_queue(queue);
 807                         be->vif->num_queues = queue_index;
 808                         goto err;
 809                 }
 810         }
 811
 812 #ifdef CONFIG_DEBUG_FS
 813         xenvif_debugfs_addif(be->vif);
 814 #endif /* CONFIG_DEBUG_FS */
 815
 816         /* Initialisation completed, tell core driver the number of
 817          * active queues.
 818          */
 819         rtnl_lock();
 820         netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues);
 821         netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues);
 822         rtnl_unlock();
 823
 824         xenvif_carrier_on(be->vif);
 825
 826         unregister_hotplug_status_watch(be);
 827         err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch, NULL,
 828                                    hotplug_status_changed,
 829                                    "%s/%s", dev->nodename, "hotplug-status");
 830         if (!err)
 831                 be->have_hotplug_status_watch = 1;
 832
 833         netif_tx_wake_all_queues(be->vif->dev);
 834
 835         return;
 836
 837 err:
 838         if (be->vif->num_queues > 0)
 839                 xenvif_disconnect_data(be->vif); /* Clean up existing queues */
 840         for (queue_index = 0; queue_index < be->vif->num_queues; ++queue_index)
 841                 xenvif_deinit_queue(&be->vif->queues[queue_index]);
 842         vfree(be->vif->queues);
 843         be->vif->queues = NULL;
 844         be->vif->num_queues = 0;
 845         xenvif_disconnect_ctrl(be->vif);
 846         return;
 847 }
 848
 849
 850 static int connect_data_rings(struct backend_info *be,
 851                               struct xenvif_queue *queue)
 852 {
 853         struct xenbus_device *dev = be->dev;
 854         unsigned int num_queues = queue->vif->num_queues;
 855         unsigned long tx_ring_ref, rx_ring_ref;
 856         unsigned int tx_evtchn, rx_evtchn;
 857         int err;
 858         char *xspath;
 859         size_t xspathsize;
 860         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
 861
 862         /* If the frontend requested 1 queue, or we have fallen back
 863          * to single queue due to lack of frontend support for multi-
 864          * queue, expect the remaining XenStore keys in the toplevel
 865          * directory. Otherwise, expect them in a subdirectory called
 866          * queue-N.
 867          */
 868         if (num_queues == 1) {
 869                 xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL);
 870                 if (!xspath) {
 871                         xenbus_dev_fatal(dev, -ENOMEM,
 872                                          "reading ring references");
 873                         return -ENOMEM;
 874                 }
 875                 strcpy(xspath, dev->otherend);
 876         } else {
 877                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
 878                 xspath = kzalloc(xspathsize, GFP_KERNEL);
 879                 if (!xspath) {
 880                         xenbus_dev_fatal(dev, -ENOMEM,
 881                                          "reading ring references");
 882                         return -ENOMEM;
 883                 }
 884                 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend,
 885                          queue->id);
 886         }
 887
 888         err = xenbus_gather(XBT_NIL, xspath,
 889                             "tx-ring-ref", "%lu", &tx_ring_ref,
 890                             "rx-ring-ref", "%lu", &rx_ring_ref, NULL);
 891         if (err) {
 892                 xenbus_dev_fatal(dev, err,
 893                                  "reading %s/ring-ref",
 894                                  xspath);
 895                 goto err;
 896         }
 897
 898         /* Try split event channels first, then single event channel. */
 899         err = xenbus_gather(XBT_NIL, xspath,
 900                             "event-channel-tx", "%u", &tx_evtchn,
 901                             "event-channel-rx", "%u", &rx_evtchn, NULL);
 902         if (err < 0) {
 903                 err = xenbus_scanf(XBT_NIL, xspath,
 904                                    "event-channel", "%u", &tx_evtchn);
 905                 if (err < 0) {
 906                         xenbus_dev_fatal(dev, err,
 907                                          "reading %s/event-channel(-tx/rx)",
 908                                          xspath);
 909                         goto err;
 910                 }
 911                 rx_evtchn = tx_evtchn;
 912         }
 913
 914         /* Map the shared frame, irq etc. */
 915         err = xenvif_connect_data(queue, tx_ring_ref, rx_ring_ref,
 916                                   tx_evtchn, rx_evtchn);
 917         if (err) {
 918                 xenbus_dev_fatal(dev, err,
 919                                  "mapping shared-frames %lu/%lu port tx %u rx %u",
 920                                  tx_ring_ref, rx_ring_ref,
 921                                  tx_evtchn, rx_evtchn);
 922                 goto err;
 923         }
 924
 925         err = 0;
 926 err: /* Regular return falls through with err == 0 */
 927         kfree(xspath);
 928         return err;
 929 }
 930
 931 static int read_xenbus_vif_flags(struct backend_info *be)
 932 {
 933         struct xenvif *vif = be->vif;
 934         struct xenbus_device *dev = be->dev;
 935         unsigned int rx_copy;
 936         int err;
 937
 938         err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u",
 939                            &rx_copy);
 940         if (err == -ENOENT) {
 941                 err = 0;
 942                 rx_copy = 0;
 943         }
 944         if (err < 0) {
 945                 xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy",
 946                                  dev->otherend);
 947                 return err;
 948         }
 949         if (!rx_copy)
 950                 return -EOPNOTSUPP;
 951
 952         if (!xenbus_read_unsigned(dev->otherend, "feature-rx-notify", 0)) {
 953                 /* - Reduce drain timeout to poll more frequently for
 954                  *   Rx requests.
 955                  * - Disable Rx stall detection.
 956                  */
 957                 be->vif->drain_timeout = msecs_to_jiffies(30);
 958                 be->vif->stall_timeout = 0;
 959         }
 960
 961         vif->can_sg = !!xenbus_read_unsigned(dev->otherend, "feature-sg", 0);
 962
 963         vif->gso_mask = 0;
 964
 965         if (xenbus_read_unsigned(dev->otherend, "feature-gso-tcpv4", 0))
 966                 vif->gso_mask |= GSO_BIT(TCPV4);
 967
 968         if (xenbus_read_unsigned(dev->otherend, "feature-gso-tcpv6", 0))
 969                 vif->gso_mask |= GSO_BIT(TCPV6);
 970
 971         vif->ip_csum = !xenbus_read_unsigned(dev->otherend,
 972                                              "feature-no-csum-offload", 0);
 973
 974         vif->ipv6_csum = !!xenbus_read_unsigned(dev->otherend,
 975                                                 "feature-ipv6-csum-offload", 0);
 976
 977         read_xenbus_frontend_xdp(be, dev);
 978
 979         return 0;
 980 }
 981
 982 static int netback_remove(struct xenbus_device *dev)
 983 {
 984         struct backend_info *be = dev_get_drvdata(&dev->dev);
 985
 986         unregister_hotplug_status_watch(be);
 987         if (be->vif) {
 988                 kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE);
 989                 backend_disconnect(be);
 990                 xenvif_free(be->vif);
 991                 be->vif = NULL;
 992         }
 993         kfree(be->hotplug_script);
 994         kfree(be);
 995         dev_set_drvdata(&dev->dev, NULL);
 996         return 0;
 997 }
 998
 999 /**
1000  * Entry point to this code when a new device is created.  Allocate the basic
1001  * structures and switch to InitWait.
1002  */
1003 static int netback_probe(struct xenbus_device *dev,
1004                          const struct xenbus_device_id *id)
1005 {
1006         const char *message;
1007         struct xenbus_transaction xbt;
1008         int err;
1009         int sg;
1010         const char *script;
1011         struct backend_info *be = kzalloc(sizeof(*be), GFP_KERNEL);
1012
1013         if (!be) {
1014                 xenbus_dev_fatal(dev, -ENOMEM,
1015                                  "allocating backend structure");
1016                 return -ENOMEM;
1017         }
1018
1019         be->dev = dev;
1020         dev_set_drvdata(&dev->dev, be);
1021
1022         sg = 1;
1023
1024         do {
1025                 err = xenbus_transaction_start(&xbt);
1026                 if (err) {
1027                         xenbus_dev_fatal(dev, err, "starting transaction");
1028                         goto fail;
1029                 }
1030
1031                 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg);
1032                 if (err) {
1033                         message = "writing feature-sg";
1034                         goto abort_transaction;
1035                 }
1036
1037                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4",
1038                                     "%d", sg);
1039                 if (err) {
1040                         message = "writing feature-gso-tcpv4";
1041                         goto abort_transaction;
1042                 }
1043
1044                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6",
1045                                     "%d", sg);
1046                 if (err) {
1047                         message = "writing feature-gso-tcpv6";
1048                         goto abort_transaction;
1049                 }
1050
1051                 /* We support partial checksum setup for IPv6 packets */
1052                 err = xenbus_printf(xbt, dev->nodename,
1053                                     "feature-ipv6-csum-offload",
1054                                     "%d", 1);
1055                 if (err) {
1056                         message = "writing feature-ipv6-csum-offload";
1057                         goto abort_transaction;
1058                 }
1059
1060                 /* We support rx-copy path. */
1061                 err = xenbus_printf(xbt, dev->nodename,
1062                                     "feature-rx-copy", "%d", 1);
1063                 if (err) {
1064                         message = "writing feature-rx-copy";
1065                         goto abort_transaction;
1066                 }
1067
1068                 /* we can adjust a headroom for netfront XDP processing */
1069                 err = xenbus_printf(xbt, dev->nodename,
1070                                     "feature-xdp-headroom", "%d",
1071                                     provides_xdp_headroom);
1072                 if (err) {
1073                         message = "writing feature-xdp-headroom";
1074                         goto abort_transaction;
1075                 }
1076
1077                 /* We don't support rx-flip path (except old guests who
1078                  * don't grok this feature flag).
1079                  */
1080                 err = xenbus_printf(xbt, dev->nodename,
1081                                     "feature-rx-flip", "%d", 0);
1082                 if (err) {
1083                         message = "writing feature-rx-flip";
1084                         goto abort_transaction;
1085                 }
1086
1087                 /* We support dynamic multicast-control. */
1088                 err = xenbus_printf(xbt, dev->nodename,
1089                                     "feature-multicast-control", "%d", 1);
1090                 if (err) {
1091                         message = "writing feature-multicast-control";
1092                         goto abort_transaction;
1093                 }
1094
1095                 err = xenbus_printf(xbt, dev->nodename,
1096                                     "feature-dynamic-multicast-control",
1097                                     "%d", 1);
1098                 if (err) {
1099                         message = "writing feature-dynamic-multicast-control";
1100                         goto abort_transaction;
1101                 }
1102
1103                 err = xenbus_transaction_end(xbt, 0);
1104         } while (err == -EAGAIN);
1105
1106         if (err) {
1107                 xenbus_dev_fatal(dev, err, "completing transaction");
1108                 goto fail;
1109         }
1110
1111         /* Split event channels support, this is optional so it is not
1112          * put inside the above loop.
1113          */
1114         err = xenbus_printf(XBT_NIL, dev->nodename,
1115                             "feature-split-event-channels",
1116                             "%u", separate_tx_rx_irq);
1117         if (err)
1118                 pr_debug("Error writing feature-split-event-channels\n");
1119
1120         /* Multi-queue support: This is an optional feature. */
1121         err = xenbus_printf(XBT_NIL, dev->nodename,
1122                             "multi-queue-max-queues", "%u", xenvif_max_queues);
1123         if (err)
1124                 pr_debug("Error writing multi-queue-max-queues\n");
1125
1126         err = xenbus_printf(XBT_NIL, dev->nodename,
1127                             "feature-ctrl-ring",
1128                             "%u", true);
1129         if (err)
1130                 pr_debug("Error writing feature-ctrl-ring\n");
1131
1132         backend_switch_state(be, XenbusStateInitWait);
1133
1134         script = xenbus_read(XBT_NIL, dev->nodename, "script", NULL);
1135         if (IS_ERR(script)) {
1136                 err = PTR_ERR(script);
1137                 xenbus_dev_fatal(dev, err, "reading script");
1138                 goto fail;
1139         }
1140
1141         be->hotplug_script = script;
1142
1143         /* This kicks hotplug scripts, so do it immediately. */
1144         err = backend_create_xenvif(be);
1145         if (err)
1146                 goto fail;
1147
1148         return 0;
1149
1150 abort_transaction:
1151         xenbus_transaction_end(xbt, 1);
1152         xenbus_dev_fatal(dev, err, "%s", message);
1153 fail:
1154         pr_debug("failed\n");
1155         netback_remove(dev);
1156         return err;
1157 }
1158
1159 static const struct xenbus_device_id netback_ids[] = {
1160         { "vif" },
1161         { "" }
1162 };
1163
1164 static struct xenbus_driver netback_driver = {
1165         .ids = netback_ids,
1166         .probe = netback_probe,
1167         .remove = netback_remove,
1168         .uevent = netback_uevent,
1169         .otherend_changed = frontend_changed,
1170         .allow_rebind = true,
1171 };
1172
1173 int xenvif_xenbus_init(void)
1174 {
1175         return xenbus_register_backend(&netback_driver);
1176 }
1177
1178 void xenvif_xenbus_fini(void)
1179 {
1180         return xenbus_unregister_driver(&netback_driver);
1181 }