drivers/net/xen-netback/xenbus.c

   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*
   3  * Xenbus code for netif backend
   4  *
   5  * Copyright (C) 2005 Rusty Russell <[email protected]>
   6  * Copyright (C) 2005 XenSource Ltd
   7 */
   8
   9 #include "common.h"
  10 #include <linux/vmalloc.h>
  11 #include <linux/rtnetlink.h>
  12
  13 static int connect_data_rings(struct backend_info *be,
  14                               struct xenvif_queue *queue);
  15 static void connect(struct backend_info *be);
  16 static int read_xenbus_vif_flags(struct backend_info *be);
  17 static int backend_create_xenvif(struct backend_info *be);
  18 static void unregister_hotplug_status_watch(struct backend_info *be);
  19 static void xen_unregister_watchers(struct xenvif *vif);
  20 static void set_backend_state(struct backend_info *be,
  21                               enum xenbus_state state);
  22
  23 #ifdef CONFIG_DEBUG_FS
  24 struct dentry *xen_netback_dbg_root = NULL;
  25
  26 static int xenvif_read_io_ring(struct seq_file *m, void *v)
  27 {
  28         struct xenvif_queue *queue = m->private;
  29         struct xen_netif_tx_back_ring *tx_ring = &queue->tx;
  30         struct xen_netif_rx_back_ring *rx_ring = &queue->rx;
  31         struct netdev_queue *dev_queue;
  32
  33         if (tx_ring->sring) {
  34                 struct xen_netif_tx_sring *sring = tx_ring->sring;
  35
  36                 seq_printf(m, "Queue %d\nTX: nr_ents %u\n", queue->id,
  37                            tx_ring->nr_ents);
  38                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  39                            sring->req_prod,
  40                            sring->req_prod - sring->rsp_prod,
  41                            tx_ring->req_cons,
  42                            tx_ring->req_cons - sring->rsp_prod,
  43                            sring->req_event,
  44                            sring->req_event - sring->rsp_prod);
  45                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n",
  46                            sring->rsp_prod,
  47                            tx_ring->rsp_prod_pvt,
  48                            tx_ring->rsp_prod_pvt - sring->rsp_prod,
  49                            sring->rsp_event,
  50                            sring->rsp_event - sring->rsp_prod);
  51                 seq_printf(m, "pending prod %u pending cons %u nr_pending_reqs %u\n",
  52                            queue->pending_prod,
  53                            queue->pending_cons,
  54                            nr_pending_reqs(queue));
  55                 seq_printf(m, "dealloc prod %u dealloc cons %u dealloc_queue %u\n\n",
  56                            queue->dealloc_prod,
  57                            queue->dealloc_cons,
  58                            queue->dealloc_prod - queue->dealloc_cons);
  59         }
  60
  61         if (rx_ring->sring) {
  62                 struct xen_netif_rx_sring *sring = rx_ring->sring;
  63
  64                 seq_printf(m, "RX: nr_ents %u\n", rx_ring->nr_ents);
  65                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  66                            sring->req_prod,
  67                            sring->req_prod - sring->rsp_prod,
  68                            rx_ring->req_cons,
  69                            rx_ring->req_cons - sring->rsp_prod,
  70                            sring->req_event,
  71                            sring->req_event - sring->rsp_prod);
  72                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n\n",
  73                            sring->rsp_prod,
  74                            rx_ring->rsp_prod_pvt,
  75                            rx_ring->rsp_prod_pvt - sring->rsp_prod,
  76                            sring->rsp_event,
  77                            sring->rsp_event - sring->rsp_prod);
  78         }
  79
  80         seq_printf(m, "NAPI state: %lx NAPI weight: %d TX queue len %u\n"
  81                    "Credit timer_pending: %d, credit: %lu, usec: %lu\n"
  82                    "remaining: %lu, expires: %lu, now: %lu\n",
  83                    queue->napi.state, queue->napi.weight,
  84                    skb_queue_len(&queue->tx_queue),
  85                    timer_pending(&queue->credit_timeout),
  86                    queue->credit_bytes,
  87                    queue->credit_usec,
  88                    queue->remaining_credit,
  89                    queue->credit_timeout.expires,
  90                    jiffies);
  91
  92         dev_queue = netdev_get_tx_queue(queue->vif->dev, queue->id);
  93
  94         seq_printf(m, "\nRx internal queue: len %u max %u pkts %u %s\n",
  95                    queue->rx_queue_len, queue->rx_queue_max,
  96                    skb_queue_len(&queue->rx_queue),
  97                    netif_tx_queue_stopped(dev_queue) ? "stopped" : "running");
  98
  99         return 0;
 100 }
 101
 102 #define XENVIF_KICK_STR "kick"
 103 #define BUFFER_SIZE     32
 104
 105 static ssize_t
 106 xenvif_write_io_ring(struct file *filp, const char __user *buf, size_t count,
 107                      loff_t *ppos)
 108 {
 109         struct xenvif_queue *queue =
 110                 ((struct seq_file *)filp->private_data)->private;
 111         int len;
 112         char write[BUFFER_SIZE];
 113
 114         /* don't allow partial writes and check the length */
 115         if (*ppos != 0)
 116                 return 0;
 117         if (count >= sizeof(write))
 118                 return -ENOSPC;
 119
 120         len = simple_write_to_buffer(write,
 121                                      sizeof(write) - 1,
 122                                      ppos,
 123                                      buf,
 124                                      count);
 125         if (len < 0)
 126                 return len;
 127
 128         write[len] = '\0';
 129
 130         if (!strncmp(write, XENVIF_KICK_STR, sizeof(XENVIF_KICK_STR) - 1))
 131                 xenvif_interrupt(0, (void *)queue);
 132         else {
 133                 pr_warn("Unknown command to io_ring_q%d. Available: kick\n",
 134                         queue->id);
 135                 count = -EINVAL;
 136         }
 137         return count;
 138 }
 139
 140 static int xenvif_io_ring_open(struct inode *inode, struct file *filp)
 141 {
 142         int ret;
 143         void *queue = NULL;
 144
 145         if (inode->i_private)
 146                 queue = inode->i_private;
 147         ret = single_open(filp, xenvif_read_io_ring, queue);
 148         filp->f_mode |= FMODE_PWRITE;
 149         return ret;
 150 }
 151
 152 static const struct file_operations xenvif_dbg_io_ring_ops_fops = {
 153         .owner = THIS_MODULE,
 154         .open = xenvif_io_ring_open,
 155         .read = seq_read,
 156         .llseek = seq_lseek,
 157         .release = single_release,
 158         .write = xenvif_write_io_ring,
 159 };
 160
 161 static int xenvif_ctrl_show(struct seq_file *m, void *v)
 162 {
 163         struct xenvif *vif = m->private;
 164
 165         xenvif_dump_hash_info(vif, m);
 166
 167         return 0;
 168 }
 169 DEFINE_SHOW_ATTRIBUTE(xenvif_ctrl);
 170
 171 static void xenvif_debugfs_addif(struct xenvif *vif)
 172 {
 173         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                 /* fall through - 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.callback = xen_net_rate_changed;
 561         err = register_xenbus_watch(&vif->credit_watch);
 562         if (err) {
 563                 pr_err("Failed to set watcher %s\n", vif->credit_watch.node);
 564                 kfree(node);
 565                 vif->credit_watch.node = NULL;
 566                 vif->credit_watch.callback = NULL;
 567         }
 568         return err;
 569 }
 570
 571 static void xen_unregister_credit_watch(struct xenvif *vif)
 572 {
 573         if (vif->credit_watch.node) {
 574                 unregister_xenbus_watch(&vif->credit_watch);
 575                 kfree(vif->credit_watch.node);
 576                 vif->credit_watch.node = NULL;
 577         }
 578 }
 579
 580 static void xen_mcast_ctrl_changed(struct xenbus_watch *watch,
 581                                    const char *path, const char *token)
 582 {
 583         struct xenvif *vif = container_of(watch, struct xenvif,
 584                                           mcast_ctrl_watch);
 585         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 586
 587         vif->multicast_control = !!xenbus_read_unsigned(dev->otherend,
 588                                         "request-multicast-control", 0);
 589 }
 590
 591 static int xen_register_mcast_ctrl_watch(struct xenbus_device *dev,
 592                                          struct xenvif *vif)
 593 {
 594         int err = 0;
 595         char *node;
 596         unsigned maxlen = strlen(dev->otherend) +
 597                 sizeof("/request-multicast-control");
 598
 599         if (vif->mcast_ctrl_watch.node) {
 600                 pr_err_ratelimited("Watch is already registered\n");
 601                 return -EADDRINUSE;
 602         }
 603
 604         node = kmalloc(maxlen, GFP_KERNEL);
 605         if (!node) {
 606                 pr_err("Failed to allocate memory for watch\n");
 607                 return -ENOMEM;
 608         }
 609         snprintf(node, maxlen, "%s/request-multicast-control",
 610                  dev->otherend);
 611         vif->mcast_ctrl_watch.node = node;
 612         vif->mcast_ctrl_watch.callback = xen_mcast_ctrl_changed;
 613         err = register_xenbus_watch(&vif->mcast_ctrl_watch);
 614         if (err) {
 615                 pr_err("Failed to set watcher %s\n",
 616                        vif->mcast_ctrl_watch.node);
 617                 kfree(node);
 618                 vif->mcast_ctrl_watch.node = NULL;
 619                 vif->mcast_ctrl_watch.callback = NULL;
 620         }
 621         return err;
 622 }
 623
 624 static void xen_unregister_mcast_ctrl_watch(struct xenvif *vif)
 625 {
 626         if (vif->mcast_ctrl_watch.node) {
 627                 unregister_xenbus_watch(&vif->mcast_ctrl_watch);
 628                 kfree(vif->mcast_ctrl_watch.node);
 629                 vif->mcast_ctrl_watch.node = NULL;
 630         }
 631 }
 632
 633 static void xen_register_watchers(struct xenbus_device *dev,
 634                                   struct xenvif *vif)
 635 {
 636         xen_register_credit_watch(dev, vif);
 637         xen_register_mcast_ctrl_watch(dev, vif);
 638 }
 639
 640 static void xen_unregister_watchers(struct xenvif *vif)
 641 {
 642         xen_unregister_mcast_ctrl_watch(vif);
 643         xen_unregister_credit_watch(vif);
 644 }
 645
 646 static void unregister_hotplug_status_watch(struct backend_info *be)
 647 {
 648         if (be->have_hotplug_status_watch) {
 649                 unregister_xenbus_watch(&be->hotplug_status_watch);
 650                 kfree(be->hotplug_status_watch.node);
 651         }
 652         be->have_hotplug_status_watch = 0;
 653 }
 654
 655 static void hotplug_status_changed(struct xenbus_watch *watch,
 656                                    const char *path,
 657                                    const char *token)
 658 {
 659         struct backend_info *be = container_of(watch,
 660                                                struct backend_info,
 661                                                hotplug_status_watch);
 662         char *str;
 663         unsigned int len;
 664
 665         str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len);
 666         if (IS_ERR(str))
 667                 return;
 668         if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) {
 669                 /* Complete any pending state change */
 670                 xenbus_switch_state(be->dev, be->state);
 671
 672                 /* Not interested in this watch anymore. */
 673                 unregister_hotplug_status_watch(be);
 674                 xenbus_rm(XBT_NIL, be->dev->nodename, "hotplug-status");
 675         }
 676         kfree(str);
 677 }
 678
 679 static int connect_ctrl_ring(struct backend_info *be)
 680 {
 681         struct xenbus_device *dev = be->dev;
 682         struct xenvif *vif = be->vif;
 683         unsigned int val;
 684         grant_ref_t ring_ref;
 685         unsigned int evtchn;
 686         int err;
 687
 688         err = xenbus_scanf(XBT_NIL, dev->otherend,
 689                            "ctrl-ring-ref", "%u", &val);
 690         if (err < 0)
 691                 goto done; /* The frontend does not have a control ring */
 692
 693         ring_ref = val;
 694
 695         err = xenbus_scanf(XBT_NIL, dev->otherend,
 696                            "event-channel-ctrl", "%u", &val);
 697         if (err < 0) {
 698                 xenbus_dev_fatal(dev, err,
 699                                  "reading %s/event-channel-ctrl",
 700                                  dev->otherend);
 701                 goto fail;
 702         }
 703
 704         evtchn = val;
 705
 706         err = xenvif_connect_ctrl(vif, ring_ref, evtchn);
 707         if (err) {
 708                 xenbus_dev_fatal(dev, err,
 709                                  "mapping shared-frame %u port %u",
 710                                  ring_ref, evtchn);
 711                 goto fail;
 712         }
 713
 714 done:
 715         return 0;
 716
 717 fail:
 718         return err;
 719 }
 720
 721 static void connect(struct backend_info *be)
 722 {
 723         int err;
 724         struct xenbus_device *dev = be->dev;
 725         unsigned long credit_bytes, credit_usec;
 726         unsigned int queue_index;
 727         unsigned int requested_num_queues;
 728         struct xenvif_queue *queue;
 729
 730         /* Check whether the frontend requested multiple queues
 731          * and read the number requested.
 732          */
 733         requested_num_queues = xenbus_read_unsigned(dev->otherend,
 734                                         "multi-queue-num-queues", 1);
 735         if (requested_num_queues > xenvif_max_queues) {
 736                 /* buggy or malicious guest */
 737                 xenbus_dev_fatal(dev, -EINVAL,
 738                                  "guest requested %u queues, exceeding the maximum of %u.",
 739                                  requested_num_queues, xenvif_max_queues);
 740                 return;
 741         }
 742
 743         err = xen_net_read_mac(dev, be->vif->fe_dev_addr);
 744         if (err) {
 745                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
 746                 return;
 747         }
 748
 749         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 750         xen_unregister_watchers(be->vif);
 751         xen_register_watchers(dev, be->vif);
 752         read_xenbus_vif_flags(be);
 753
 754         err = connect_ctrl_ring(be);
 755         if (err) {
 756                 xenbus_dev_fatal(dev, err, "connecting control ring");
 757                 return;
 758         }
 759
 760         /* Use the number of queues requested by the frontend */
 761         be->vif->queues = vzalloc(array_size(requested_num_queues,
 762                                              sizeof(struct xenvif_queue)));
 763         if (!be->vif->queues) {
 764                 xenbus_dev_fatal(dev, -ENOMEM,
 765                                  "allocating queues");
 766                 return;
 767         }
 768
 769         be->vif->num_queues = requested_num_queues;
 770         be->vif->stalled_queues = requested_num_queues;
 771
 772         for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) {
 773                 queue = &be->vif->queues[queue_index];
 774                 queue->vif = be->vif;
 775                 queue->id = queue_index;
 776                 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
 777                                 be->vif->dev->name, queue->id);
 778
 779                 err = xenvif_init_queue(queue);
 780                 if (err) {
 781                         /* xenvif_init_queue() cleans up after itself on
 782                          * failure, but we need to clean up any previously
 783                          * initialised queues. Set num_queues to i so that
 784                          * earlier queues can be destroyed using the regular
 785                          * disconnect logic.
 786                          */
 787                         be->vif->num_queues = queue_index;
 788                         goto err;
 789                 }
 790
 791                 queue->credit_bytes = credit_bytes;
 792                 queue->remaining_credit = credit_bytes;
 793                 queue->credit_usec = credit_usec;
 794
 795                 err = connect_data_rings(be, queue);
 796                 if (err) {
 797                         /* connect_data_rings() cleans up after itself on
 798                          * failure, but we need to clean up after
 799                          * xenvif_init_queue() here, and also clean up any
 800                          * previously initialised queues.
 801                          */
 802                         xenvif_deinit_queue(queue);
 803                         be->vif->num_queues = queue_index;
 804                         goto err;
 805                 }
 806         }
 807
 808 #ifdef CONFIG_DEBUG_FS
 809         xenvif_debugfs_addif(be->vif);
 810 #endif /* CONFIG_DEBUG_FS */
 811
 812         /* Initialisation completed, tell core driver the number of
 813          * active queues.
 814          */
 815         rtnl_lock();
 816         netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues);
 817         netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues);
 818         rtnl_unlock();
 819
 820         xenvif_carrier_on(be->vif);
 821
 822         unregister_hotplug_status_watch(be);
 823         err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch,
 824                                    hotplug_status_changed,
 825                                    "%s/%s", dev->nodename, "hotplug-status");
 826         if (!err)
 827                 be->have_hotplug_status_watch = 1;
 828
 829         netif_tx_wake_all_queues(be->vif->dev);
 830
 831         return;
 832
 833 err:
 834         if (be->vif->num_queues > 0)
 835                 xenvif_disconnect_data(be->vif); /* Clean up existing queues */
 836         for (queue_index = 0; queue_index < be->vif->num_queues; ++queue_index)
 837                 xenvif_deinit_queue(&be->vif->queues[queue_index]);
 838         vfree(be->vif->queues);
 839         be->vif->queues = NULL;
 840         be->vif->num_queues = 0;
 841         xenvif_disconnect_ctrl(be->vif);
 842         return;
 843 }
 844
 845
 846 static int connect_data_rings(struct backend_info *be,
 847                               struct xenvif_queue *queue)
 848 {
 849         struct xenbus_device *dev = be->dev;
 850         unsigned int num_queues = queue->vif->num_queues;
 851         unsigned long tx_ring_ref, rx_ring_ref;
 852         unsigned int tx_evtchn, rx_evtchn;
 853         int err;
 854         char *xspath;
 855         size_t xspathsize;
 856         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
 857
 858         /* If the frontend requested 1 queue, or we have fallen back
 859          * to single queue due to lack of frontend support for multi-
 860          * queue, expect the remaining XenStore keys in the toplevel
 861          * directory. Otherwise, expect them in a subdirectory called
 862          * queue-N.
 863          */
 864         if (num_queues == 1) {
 865                 xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL);
 866                 if (!xspath) {
 867                         xenbus_dev_fatal(dev, -ENOMEM,
 868                                          "reading ring references");
 869                         return -ENOMEM;
 870                 }
 871                 strcpy(xspath, dev->otherend);
 872         } else {
 873                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
 874                 xspath = kzalloc(xspathsize, GFP_KERNEL);
 875                 if (!xspath) {
 876                         xenbus_dev_fatal(dev, -ENOMEM,
 877                                          "reading ring references");
 878                         return -ENOMEM;
 879                 }
 880                 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend,
 881                          queue->id);
 882         }
 883
 884         err = xenbus_gather(XBT_NIL, xspath,
 885                             "tx-ring-ref", "%lu", &tx_ring_ref,
 886                             "rx-ring-ref", "%lu", &rx_ring_ref, NULL);
 887         if (err) {
 888                 xenbus_dev_fatal(dev, err,
 889                                  "reading %s/ring-ref",
 890                                  xspath);
 891                 goto err;
 892         }
 893
 894         /* Try split event channels first, then single event channel. */
 895         err = xenbus_gather(XBT_NIL, xspath,
 896                             "event-channel-tx", "%u", &tx_evtchn,
 897                             "event-channel-rx", "%u", &rx_evtchn, NULL);
 898         if (err < 0) {
 899                 err = xenbus_scanf(XBT_NIL, xspath,
 900                                    "event-channel", "%u", &tx_evtchn);
 901                 if (err < 0) {
 902                         xenbus_dev_fatal(dev, err,
 903                                          "reading %s/event-channel(-tx/rx)",
 904                                          xspath);
 905                         goto err;
 906                 }
 907                 rx_evtchn = tx_evtchn;
 908         }
 909
 910         /* Map the shared frame, irq etc. */
 911         err = xenvif_connect_data(queue, tx_ring_ref, rx_ring_ref,
 912                                   tx_evtchn, rx_evtchn);
 913         if (err) {
 914                 xenbus_dev_fatal(dev, err,
 915                                  "mapping shared-frames %lu/%lu port tx %u rx %u",
 916                                  tx_ring_ref, rx_ring_ref,
 917                                  tx_evtchn, rx_evtchn);
 918                 goto err;
 919         }
 920
 921         err = 0;
 922 err: /* Regular return falls through with err == 0 */
 923         kfree(xspath);
 924         return err;
 925 }
 926
 927 static int read_xenbus_vif_flags(struct backend_info *be)
 928 {
 929         struct xenvif *vif = be->vif;
 930         struct xenbus_device *dev = be->dev;
 931         unsigned int rx_copy;
 932         int err;
 933
 934         err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u",
 935                            &rx_copy);
 936         if (err == -ENOENT) {
 937                 err = 0;
 938                 rx_copy = 0;
 939         }
 940         if (err < 0) {
 941                 xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy",
 942                                  dev->otherend);
 943                 return err;
 944         }
 945         if (!rx_copy)
 946                 return -EOPNOTSUPP;
 947
 948         if (!xenbus_read_unsigned(dev->otherend, "feature-rx-notify", 0)) {
 949                 /* - Reduce drain timeout to poll more frequently for
 950                  *   Rx requests.
 951                  * - Disable Rx stall detection.
 952                  */
 953                 be->vif->drain_timeout = msecs_to_jiffies(30);
 954                 be->vif->stall_timeout = 0;
 955         }
 956
 957         vif->can_sg = !!xenbus_read_unsigned(dev->otherend, "feature-sg", 0);
 958
 959         vif->gso_mask = 0;
 960
 961         if (xenbus_read_unsigned(dev->otherend, "feature-gso-tcpv4", 0))
 962                 vif->gso_mask |= GSO_BIT(TCPV4);
 963
 964         if (xenbus_read_unsigned(dev->otherend, "feature-gso-tcpv6", 0))
 965                 vif->gso_mask |= GSO_BIT(TCPV6);
 966
 967         vif->ip_csum = !xenbus_read_unsigned(dev->otherend,
 968                                              "feature-no-csum-offload", 0);
 969
 970         vif->ipv6_csum = !!xenbus_read_unsigned(dev->otherend,
 971                                                 "feature-ipv6-csum-offload", 0);
 972
 973         read_xenbus_frontend_xdp(be, dev);
 974
 975         return 0;
 976 }
 977
 978 static int netback_remove(struct xenbus_device *dev)
 979 {
 980         struct backend_info *be = dev_get_drvdata(&dev->dev);
 981
 982         unregister_hotplug_status_watch(be);
 983         if (be->vif) {
 984                 kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE);
 985                 backend_disconnect(be);
 986                 xenvif_free(be->vif);
 987                 be->vif = NULL;
 988         }
 989         kfree(be->hotplug_script);
 990         kfree(be);
 991         dev_set_drvdata(&dev->dev, NULL);
 992         return 0;
 993 }
 994
 995 /**
 996  * Entry point to this code when a new device is created.  Allocate the basic
 997  * structures and switch to InitWait.
 998  */
 999 static int netback_probe(struct xenbus_device *dev,
1000                          const struct xenbus_device_id *id)
1001 {
1002         const char *message;
1003         struct xenbus_transaction xbt;
1004         int err;
1005         int sg;
1006         const char *script;
1007         struct backend_info *be = kzalloc(sizeof(*be), GFP_KERNEL);
1008
1009         if (!be) {
1010                 xenbus_dev_fatal(dev, -ENOMEM,
1011                                  "allocating backend structure");
1012                 return -ENOMEM;
1013         }
1014
1015         be->dev = dev;
1016         dev_set_drvdata(&dev->dev, be);
1017
1018         sg = 1;
1019
1020         do {
1021                 err = xenbus_transaction_start(&xbt);
1022                 if (err) {
1023                         xenbus_dev_fatal(dev, err, "starting transaction");
1024                         goto fail;
1025                 }
1026
1027                 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg);
1028                 if (err) {
1029                         message = "writing feature-sg";
1030                         goto abort_transaction;
1031                 }
1032
1033                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4",
1034                                     "%d", sg);
1035                 if (err) {
1036                         message = "writing feature-gso-tcpv4";
1037                         goto abort_transaction;
1038                 }
1039
1040                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6",
1041                                     "%d", sg);
1042                 if (err) {
1043                         message = "writing feature-gso-tcpv6";
1044                         goto abort_transaction;
1045                 }
1046
1047                 /* We support partial checksum setup for IPv6 packets */
1048                 err = xenbus_printf(xbt, dev->nodename,
1049                                     "feature-ipv6-csum-offload",
1050                                     "%d", 1);
1051                 if (err) {
1052                         message = "writing feature-ipv6-csum-offload";
1053                         goto abort_transaction;
1054                 }
1055
1056                 /* We support rx-copy path. */
1057                 err = xenbus_printf(xbt, dev->nodename,
1058                                     "feature-rx-copy", "%d", 1);
1059                 if (err) {
1060                         message = "writing feature-rx-copy";
1061                         goto abort_transaction;
1062                 }
1063
1064                 /* we can adjust a headroom for netfront XDP processing */
1065                 err = xenbus_printf(xbt, dev->nodename,
1066                                     "feature-xdp-headroom", "%d",
1067                                     provides_xdp_headroom);
1068                 if (err) {
1069                         message = "writing feature-xdp-headroom";
1070                         goto abort_transaction;
1071                 }
1072
1073                 /* We don't support rx-flip path (except old guests who
1074                  * don't grok this feature flag).
1075                  */
1076                 err = xenbus_printf(xbt, dev->nodename,
1077                                     "feature-rx-flip", "%d", 0);
1078                 if (err) {
1079                         message = "writing feature-rx-flip";
1080                         goto abort_transaction;
1081                 }
1082
1083                 /* We support dynamic multicast-control. */
1084                 err = xenbus_printf(xbt, dev->nodename,
1085                                     "feature-multicast-control", "%d", 1);
1086                 if (err) {
1087                         message = "writing feature-multicast-control";
1088                         goto abort_transaction;
1089                 }
1090
1091                 err = xenbus_printf(xbt, dev->nodename,
1092                                     "feature-dynamic-multicast-control",
1093                                     "%d", 1);
1094                 if (err) {
1095                         message = "writing feature-dynamic-multicast-control";
1096                         goto abort_transaction;
1097                 }
1098
1099                 err = xenbus_transaction_end(xbt, 0);
1100         } while (err == -EAGAIN);
1101
1102         if (err) {
1103                 xenbus_dev_fatal(dev, err, "completing transaction");
1104                 goto fail;
1105         }
1106
1107         /* Split event channels support, this is optional so it is not
1108          * put inside the above loop.
1109          */
1110         err = xenbus_printf(XBT_NIL, dev->nodename,
1111                             "feature-split-event-channels",
1112                             "%u", separate_tx_rx_irq);
1113         if (err)
1114                 pr_debug("Error writing feature-split-event-channels\n");
1115
1116         /* Multi-queue support: This is an optional feature. */
1117         err = xenbus_printf(XBT_NIL, dev->nodename,
1118                             "multi-queue-max-queues", "%u", xenvif_max_queues);
1119         if (err)
1120                 pr_debug("Error writing multi-queue-max-queues\n");
1121
1122         err = xenbus_printf(XBT_NIL, dev->nodename,
1123                             "feature-ctrl-ring",
1124                             "%u", true);
1125         if (err)
1126                 pr_debug("Error writing feature-ctrl-ring\n");
1127
1128         backend_switch_state(be, XenbusStateInitWait);
1129
1130         script = xenbus_read(XBT_NIL, dev->nodename, "script", NULL);
1131         if (IS_ERR(script)) {
1132                 err = PTR_ERR(script);
1133                 xenbus_dev_fatal(dev, err, "reading script");
1134                 goto fail;
1135         }
1136
1137         be->hotplug_script = script;
1138
1139         /* This kicks hotplug scripts, so do it immediately. */
1140         err = backend_create_xenvif(be);
1141         if (err)
1142                 goto fail;
1143
1144         return 0;
1145
1146 abort_transaction:
1147         xenbus_transaction_end(xbt, 1);
1148         xenbus_dev_fatal(dev, err, "%s", message);
1149 fail:
1150         pr_debug("failed\n");
1151         netback_remove(dev);
1152         return err;
1153 }
1154
1155 static const struct xenbus_device_id netback_ids[] = {
1156         { "vif" },
1157         { "" }
1158 };
1159
1160 static struct xenbus_driver netback_driver = {
1161         .ids = netback_ids,
1162         .probe = netback_probe,
1163         .remove = netback_remove,
1164         .uevent = netback_uevent,
1165         .otherend_changed = frontend_changed,
1166         .allow_rebind = true,
1167 };
1168
1169 int xenvif_xenbus_init(void)
1170 {
1171         return xenbus_register_backend(&netback_driver);
1172 }
1173
1174 void xenvif_xenbus_fini(void)
1175 {
1176         return xenbus_unregister_driver(&netback_driver);
1177 }