drivers/net/xen-netback/xenbus.c

   1 /*
   2  * Xenbus code for netif backend
   3  *
   4  * Copyright (C) 2005 Rusty Russell <rusty@rustcorp.com.au>
   5  * Copyright (C) 2005 XenSource Ltd
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation; either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, see <http://www.gnu.org/licenses/>.
  19 */
  20
  21 #include "common.h"
  22 #include <linux/vmalloc.h>
  23 #include <linux/rtnetlink.h>
  24
  25 struct backend_info {
  26         struct xenbus_device *dev;
  27         struct xenvif *vif;
  28
  29         /* This is the state that will be reflected in xenstore when any
  30          * active hotplug script completes.
  31          */
  32         enum xenbus_state state;
  33
  34         enum xenbus_state frontend_state;
  35         struct xenbus_watch hotplug_status_watch;
  36         u8 have_hotplug_status_watch:1;
  37
  38         const char *hotplug_script;
  39 };
  40
  41 static int connect_data_rings(struct backend_info *be,
  42                               struct xenvif_queue *queue);
  43 static void connect(struct backend_info *be);
  44 static int read_xenbus_vif_flags(struct backend_info *be);
  45 static int backend_create_xenvif(struct backend_info *be);
  46 static void unregister_hotplug_status_watch(struct backend_info *be);
  47 static void xen_unregister_watchers(struct xenvif *vif);
  48 static void set_backend_state(struct backend_info *be,
  49                               enum xenbus_state state);
  50
  51 #ifdef CONFIG_DEBUG_FS
  52 struct dentry *xen_netback_dbg_root = NULL;
  53
  54 static int xenvif_read_io_ring(struct seq_file *m, void *v)
  55 {
  56         struct xenvif_queue *queue = m->private;
  57         struct xen_netif_tx_back_ring *tx_ring = &queue->tx;
  58         struct xen_netif_rx_back_ring *rx_ring = &queue->rx;
  59         struct netdev_queue *dev_queue;
  60
  61         if (tx_ring->sring) {
  62                 struct xen_netif_tx_sring *sring = tx_ring->sring;
  63
  64                 seq_printf(m, "Queue %d\nTX: nr_ents %u\n", queue->id,
  65                            tx_ring->nr_ents);
  66                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  67                            sring->req_prod,
  68                            sring->req_prod - sring->rsp_prod,
  69                            tx_ring->req_cons,
  70                            tx_ring->req_cons - sring->rsp_prod,
  71                            sring->req_event,
  72                            sring->req_event - sring->rsp_prod);
  73                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n",
  74                            sring->rsp_prod,
  75                            tx_ring->rsp_prod_pvt,
  76                            tx_ring->rsp_prod_pvt - sring->rsp_prod,
  77                            sring->rsp_event,
  78                            sring->rsp_event - sring->rsp_prod);
  79                 seq_printf(m, "pending prod %u pending cons %u nr_pending_reqs %u\n",
  80                            queue->pending_prod,
  81                            queue->pending_cons,
  82                            nr_pending_reqs(queue));
  83                 seq_printf(m, "dealloc prod %u dealloc cons %u dealloc_queue %u\n\n",
  84                            queue->dealloc_prod,
  85                            queue->dealloc_cons,
  86                            queue->dealloc_prod - queue->dealloc_cons);
  87         }
  88
  89         if (rx_ring->sring) {
  90                 struct xen_netif_rx_sring *sring = rx_ring->sring;
  91
  92                 seq_printf(m, "RX: nr_ents %u\n", rx_ring->nr_ents);
  93                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  94                            sring->req_prod,
  95                            sring->req_prod - sring->rsp_prod,
  96                            rx_ring->req_cons,
  97                            rx_ring->req_cons - sring->rsp_prod,
  98                            sring->req_event,
  99                            sring->req_event - sring->rsp_prod);
 100                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n\n",
 101                            sring->rsp_prod,
 102                            rx_ring->rsp_prod_pvt,
 103                            rx_ring->rsp_prod_pvt - sring->rsp_prod,
 104                            sring->rsp_event,
 105                            sring->rsp_event - sring->rsp_prod);
 106         }
 107
 108         seq_printf(m, "NAPI state: %lx NAPI weight: %d TX queue len %u\n"
 109                    "Credit timer_pending: %d, credit: %lu, usec: %lu\n"
 110                    "remaining: %lu, expires: %lu, now: %lu\n",
 111                    queue->napi.state, queue->napi.weight,
 112                    skb_queue_len(&queue->tx_queue),
 113                    timer_pending(&queue->credit_timeout),
 114                    queue->credit_bytes,
 115                    queue->credit_usec,
 116                    queue->remaining_credit,
 117                    queue->credit_timeout.expires,
 118                    jiffies);
 119
 120         dev_queue = netdev_get_tx_queue(queue->vif->dev, queue->id);
 121
 122         seq_printf(m, "\nRx internal queue: len %u max %u pkts %u %s\n",
 123                    queue->rx_queue_len, queue->rx_queue_max,
 124                    skb_queue_len(&queue->rx_queue),
 125                    netif_tx_queue_stopped(dev_queue) ? "stopped" : "running");
 126
 127         return 0;
 128 }
 129
 130 #define XENVIF_KICK_STR "kick"
 131 #define BUFFER_SIZE     32
 132
 133 static ssize_t
 134 xenvif_write_io_ring(struct file *filp, const char __user *buf, size_t count,
 135                      loff_t *ppos)
 136 {
 137         struct xenvif_queue *queue =
 138                 ((struct seq_file *)filp->private_data)->private;
 139         int len;
 140         char write[BUFFER_SIZE];
 141
 142         /* don't allow partial writes and check the length */
 143         if (*ppos != 0)
 144                 return 0;
 145         if (count >= sizeof(write))
 146                 return -ENOSPC;
 147
 148         len = simple_write_to_buffer(write,
 149                                      sizeof(write) - 1,
 150                                      ppos,
 151                                      buf,
 152                                      count);
 153         if (len < 0)
 154                 return len;
 155
 156         write[len] = '\0';
 157
 158         if (!strncmp(write, XENVIF_KICK_STR, sizeof(XENVIF_KICK_STR) - 1))
 159                 xenvif_interrupt(0, (void *)queue);
 160         else {
 161                 pr_warn("Unknown command to io_ring_q%d. Available: kick\n",
 162                         queue->id);
 163                 count = -EINVAL;
 164         }
 165         return count;
 166 }
 167
 168 static int xenvif_io_ring_open(struct inode *inode, struct file *filp)
 169 {
 170         int ret;
 171         void *queue = NULL;
 172
 173         if (inode->i_private)
 174                 queue = inode->i_private;
 175         ret = single_open(filp, xenvif_read_io_ring, queue);
 176         filp->f_mode |= FMODE_PWRITE;
 177         return ret;
 178 }
 179
 180 static const struct file_operations xenvif_dbg_io_ring_ops_fops = {
 181         .owner = THIS_MODULE,
 182         .open = xenvif_io_ring_open,
 183         .read = seq_read,
 184         .llseek = seq_lseek,
 185         .release = single_release,
 186         .write = xenvif_write_io_ring,
 187 };
 188
 189 static int xenvif_read_ctrl(struct seq_file *m, void *v)
 190 {
 191         struct xenvif *vif = m->private;
 192
 193         xenvif_dump_hash_info(vif, m);
 194
 195         return 0;
 196 }
 197
 198 static int xenvif_ctrl_open(struct inode *inode, struct file *filp)
 199 {
 200         return single_open(filp, xenvif_read_ctrl, inode->i_private);
 201 }
 202
 203 static const struct file_operations xenvif_dbg_ctrl_ops_fops = {
 204         .owner = THIS_MODULE,
 205         .open = xenvif_ctrl_open,
 206         .read = seq_read,
 207         .llseek = seq_lseek,
 208         .release = single_release,
 209 };
 210
 211 static void xenvif_debugfs_addif(struct xenvif *vif)
 212 {
 213         struct dentry *pfile;
 214         int i;
 215
 216         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 217                 return;
 218
 219         vif->xenvif_dbg_root = debugfs_create_dir(vif->dev->name,
 220                                                   xen_netback_dbg_root);
 221         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) {
 222                 for (i = 0; i < vif->num_queues; ++i) {
 223                         char filename[sizeof("io_ring_q") + 4];
 224
 225                         snprintf(filename, sizeof(filename), "io_ring_q%d", i);
 226                         pfile = debugfs_create_file(filename,
 227                                                     S_IRUSR | S_IWUSR,
 228                                                     vif->xenvif_dbg_root,
 229                                                     &vif->queues[i],
 230                                                     &xenvif_dbg_io_ring_ops_fops);
 231                         if (IS_ERR_OR_NULL(pfile))
 232                                 pr_warn("Creation of io_ring file returned %ld!\n",
 233                                         PTR_ERR(pfile));
 234                 }
 235
 236                 if (vif->ctrl_irq) {
 237                         pfile = debugfs_create_file("ctrl",
 238                                                     S_IRUSR,
 239                                                     vif->xenvif_dbg_root,
 240                                                     vif,
 241                                                     &xenvif_dbg_ctrl_ops_fops);
 242                         if (IS_ERR_OR_NULL(pfile))
 243                                 pr_warn("Creation of ctrl file returned %ld!\n",
 244                                         PTR_ERR(pfile));
 245                 }
 246         } else
 247                 netdev_warn(vif->dev,
 248                             "Creation of vif debugfs dir returned %ld!\n",
 249                             PTR_ERR(vif->xenvif_dbg_root));
 250 }
 251
 252 static void xenvif_debugfs_delif(struct xenvif *vif)
 253 {
 254         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 255                 return;
 256
 257         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root))
 258                 debugfs_remove_recursive(vif->xenvif_dbg_root);
 259         vif->xenvif_dbg_root = NULL;
 260 }
 261 #endif /* CONFIG_DEBUG_FS */
 262
 263 static int netback_remove(struct xenbus_device *dev)
 264 {
 265         struct backend_info *be = dev_get_drvdata(&dev->dev);
 266
 267         set_backend_state(be, XenbusStateClosed);
 268
 269         unregister_hotplug_status_watch(be);
 270         if (be->vif) {
 271                 kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE);
 272                 xen_unregister_watchers(be->vif);
 273                 xenbus_rm(XBT_NIL, dev->nodename, "hotplug-status");
 274                 xenvif_free(be->vif);
 275                 be->vif = NULL;
 276         }
 277         kfree(be->hotplug_script);
 278         kfree(be);
 279         dev_set_drvdata(&dev->dev, NULL);
 280         return 0;
 281 }
 282
 283
 284 /**
 285  * Entry point to this code when a new device is created.  Allocate the basic
 286  * structures and switch to InitWait.
 287  */
 288 static int netback_probe(struct xenbus_device *dev,
 289                          const struct xenbus_device_id *id)
 290 {
 291         const char *message;
 292         struct xenbus_transaction xbt;
 293         int err;
 294         int sg;
 295         const char *script;
 296         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 297                                           GFP_KERNEL);
 298         if (!be) {
 299                 xenbus_dev_fatal(dev, -ENOMEM,
 300                                  "allocating backend structure");
 301                 return -ENOMEM;
 302         }
 303
 304         be->dev = dev;
 305         dev_set_drvdata(&dev->dev, be);
 306
 307         be->state = XenbusStateInitialising;
 308         err = xenbus_switch_state(dev, XenbusStateInitialising);
 309         if (err)
 310                 goto fail;
 311
 312         sg = 1;
 313
 314         do {
 315                 err = xenbus_transaction_start(&xbt);
 316                 if (err) {
 317                         xenbus_dev_fatal(dev, err, "starting transaction");
 318                         goto fail;
 319                 }
 320
 321                 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg);
 322                 if (err) {
 323                         message = "writing feature-sg";
 324                         goto abort_transaction;
 325                 }
 326
 327                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4",
 328                                     "%d", sg);
 329                 if (err) {
 330                         message = "writing feature-gso-tcpv4";
 331                         goto abort_transaction;
 332                 }
 333
 334                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6",
 335                                     "%d", sg);
 336                 if (err) {
 337                         message = "writing feature-gso-tcpv6";
 338                         goto abort_transaction;
 339                 }
 340
 341                 /* We support partial checksum setup for IPv6 packets */
 342                 err = xenbus_printf(xbt, dev->nodename,
 343                                     "feature-ipv6-csum-offload",
 344                                     "%d", 1);
 345                 if (err) {
 346                         message = "writing feature-ipv6-csum-offload";
 347                         goto abort_transaction;
 348                 }
 349
 350                 /* We support rx-copy path. */
 351                 err = xenbus_printf(xbt, dev->nodename,
 352                                     "feature-rx-copy", "%d", 1);
 353                 if (err) {
 354                         message = "writing feature-rx-copy";
 355                         goto abort_transaction;
 356                 }
 357
 358                 /*
 359                  * We don't support rx-flip path (except old guests who don't
 360                  * grok this feature flag).
 361                  */
 362                 err = xenbus_printf(xbt, dev->nodename,
 363                                     "feature-rx-flip", "%d", 0);
 364                 if (err) {
 365                         message = "writing feature-rx-flip";
 366                         goto abort_transaction;
 367                 }
 368
 369                 /* We support dynamic multicast-control. */
 370                 err = xenbus_printf(xbt, dev->nodename,
 371                                     "feature-multicast-control", "%d", 1);
 372                 if (err) {
 373                         message = "writing feature-multicast-control";
 374                         goto abort_transaction;
 375                 }
 376
 377                 err = xenbus_printf(xbt, dev->nodename,
 378                                     "feature-dynamic-multicast-control",
 379                                     "%d", 1);
 380                 if (err) {
 381                         message = "writing feature-dynamic-multicast-control";
 382                         goto abort_transaction;
 383                 }
 384
 385                 err = xenbus_transaction_end(xbt, 0);
 386         } while (err == -EAGAIN);
 387
 388         if (err) {
 389                 xenbus_dev_fatal(dev, err, "completing transaction");
 390                 goto fail;
 391         }
 392
 393         /*
 394          * Split event channels support, this is optional so it is not
 395          * put inside the above loop.
 396          */
 397         err = xenbus_printf(XBT_NIL, dev->nodename,
 398                             "feature-split-event-channels",
 399                             "%u", separate_tx_rx_irq);
 400         if (err)
 401                 pr_debug("Error writing feature-split-event-channels\n");
 402
 403         /* Multi-queue support: This is an optional feature. */
 404         err = xenbus_printf(XBT_NIL, dev->nodename,
 405                             "multi-queue-max-queues", "%u", xenvif_max_queues);
 406         if (err)
 407                 pr_debug("Error writing multi-queue-max-queues\n");
 408
 409         err = xenbus_printf(XBT_NIL, dev->nodename,
 410                             "feature-ctrl-ring",
 411                             "%u", true);
 412         if (err)
 413                 pr_debug("Error writing feature-ctrl-ring\n");
 414
 415         script = xenbus_read(XBT_NIL, dev->nodename, "script", NULL);
 416         if (IS_ERR(script)) {
 417                 err = PTR_ERR(script);
 418                 xenbus_dev_fatal(dev, err, "reading script");
 419                 goto fail;
 420         }
 421
 422         be->hotplug_script = script;
 423
 424
 425         /* This kicks hotplug scripts, so do it immediately. */
 426         err = backend_create_xenvif(be);
 427         if (err)
 428                 goto fail;
 429
 430         return 0;
 431
 432 abort_transaction:
 433         xenbus_transaction_end(xbt, 1);
 434         xenbus_dev_fatal(dev, err, "%s", message);
 435 fail:
 436         pr_debug("failed\n");
 437         netback_remove(dev);
 438         return err;
 439 }
 440
 441
 442 /*
 443  * Handle the creation of the hotplug script environment.  We add the script
 444  * and vif variables to the environment, for the benefit of the vif-* hotplug
 445  * scripts.
 446  */
 447 static int netback_uevent(struct xenbus_device *xdev,
 448                           struct kobj_uevent_env *env)
 449 {
 450         struct backend_info *be = dev_get_drvdata(&xdev->dev);
 451
 452         if (!be)
 453                 return 0;
 454
 455         if (add_uevent_var(env, "script=%s", be->hotplug_script))
 456                 return -ENOMEM;
 457
 458         if (!be->vif)
 459                 return 0;
 460
 461         return add_uevent_var(env, "vif=%s", be->vif->dev->name);
 462 }
 463
 464
 465 static int backend_create_xenvif(struct backend_info *be)
 466 {
 467         int err;
 468         long handle;
 469         struct xenbus_device *dev = be->dev;
 470         struct xenvif *vif;
 471
 472         if (be->vif != NULL)
 473                 return 0;
 474
 475         err = xenbus_scanf(XBT_NIL, dev->nodename, "handle", "%li", &handle);
 476         if (err != 1) {
 477                 xenbus_dev_fatal(dev, err, "reading handle");
 478                 return (err < 0) ? err : -EINVAL;
 479         }
 480
 481         vif = xenvif_alloc(&dev->dev, dev->otherend_id, handle);
 482         if (IS_ERR(vif)) {
 483                 err = PTR_ERR(vif);
 484                 xenbus_dev_fatal(dev, err, "creating interface");
 485                 return err;
 486         }
 487         be->vif = vif;
 488
 489         kobject_uevent(&dev->dev.kobj, KOBJ_ONLINE);
 490         return 0;
 491 }
 492
 493 static void backend_disconnect(struct backend_info *be)
 494 {
 495         struct xenvif *vif = be->vif;
 496
 497         if (vif) {
 498                 unsigned int queue_index;
 499                 struct xenvif_queue *queues;
 500
 501                 xen_unregister_watchers(vif);
 502 #ifdef CONFIG_DEBUG_FS
 503                 xenvif_debugfs_delif(vif);
 504 #endif /* CONFIG_DEBUG_FS */
 505                 xenvif_disconnect_data(vif);
 506                 for (queue_index = 0;
 507                      queue_index < vif->num_queues;
 508                      ++queue_index)
 509                         xenvif_deinit_queue(&vif->queues[queue_index]);
 510
 511                 spin_lock(&vif->lock);
 512                 queues = vif->queues;
 513                 vif->num_queues = 0;
 514                 vif->queues = NULL;
 515                 spin_unlock(&vif->lock);
 516
 517                 vfree(queues);
 518
 519                 xenvif_disconnect_ctrl(vif);
 520         }
 521 }
 522
 523 static void backend_connect(struct backend_info *be)
 524 {
 525         if (be->vif)
 526                 connect(be);
 527 }
 528
 529 static inline void backend_switch_state(struct backend_info *be,
 530                                         enum xenbus_state state)
 531 {
 532         struct xenbus_device *dev = be->dev;
 533
 534         pr_debug("%s -> %s\n", dev->nodename, xenbus_strstate(state));
 535         be->state = state;
 536
 537         /* If we are waiting for a hotplug script then defer the
 538          * actual xenbus state change.
 539          */
 540         if (!be->have_hotplug_status_watch)
 541                 xenbus_switch_state(dev, state);
 542 }
 543
 544 /* Handle backend state transitions:
 545  *
 546  * The backend state starts in Initialising and the following transitions are
 547  * allowed.
 548  *
 549  * Initialising -> InitWait -> Connected
 550  *          \
 551  *           \        ^    \         |
 552  *            \       |     \        |
 553  *             \      |      \       |
 554  *              \     |       \      |
 555  *               \    |        \     |
 556  *                \   |         \    |
 557  *                 V  |          V   V
 558  *
 559  *                  Closed  <-> Closing
 560  *
 561  * The state argument specifies the eventual state of the backend and the
 562  * function transitions to that state via the shortest path.
 563  */
 564 static void set_backend_state(struct backend_info *be,
 565                               enum xenbus_state state)
 566 {
 567         while (be->state != state) {
 568                 switch (be->state) {
 569                 case XenbusStateInitialising:
 570                         switch (state) {
 571                         case XenbusStateInitWait:
 572                         case XenbusStateConnected:
 573                         case XenbusStateClosing:
 574                                 backend_switch_state(be, XenbusStateInitWait);
 575                                 break;
 576                         case XenbusStateClosed:
 577                                 backend_switch_state(be, XenbusStateClosed);
 578                                 break;
 579                         default:
 580                                 BUG();
 581                         }
 582                         break;
 583                 case XenbusStateClosed:
 584                         switch (state) {
 585                         case XenbusStateInitWait:
 586                         case XenbusStateConnected:
 587                                 backend_switch_state(be, XenbusStateInitWait);
 588                                 break;
 589                         case XenbusStateClosing:
 590                                 backend_switch_state(be, XenbusStateClosing);
 591                                 break;
 592                         default:
 593                                 BUG();
 594                         }
 595                         break;
 596                 case XenbusStateInitWait:
 597                         switch (state) {
 598                         case XenbusStateConnected:
 599                                 backend_connect(be);
 600                                 backend_switch_state(be, XenbusStateConnected);
 601                                 break;
 602                         case XenbusStateClosing:
 603                         case XenbusStateClosed:
 604                                 backend_switch_state(be, XenbusStateClosing);
 605                                 break;
 606                         default:
 607                                 BUG();
 608                         }
 609                         break;
 610                 case XenbusStateConnected:
 611                         switch (state) {
 612                         case XenbusStateInitWait:
 613                         case XenbusStateClosing:
 614                         case XenbusStateClosed:
 615                                 backend_disconnect(be);
 616                                 backend_switch_state(be, XenbusStateClosing);
 617                                 break;
 618                         default:
 619                                 BUG();
 620                         }
 621                         break;
 622                 case XenbusStateClosing:
 623                         switch (state) {
 624                         case XenbusStateInitWait:
 625                         case XenbusStateConnected:
 626                         case XenbusStateClosed:
 627                                 backend_switch_state(be, XenbusStateClosed);
 628                                 break;
 629                         default:
 630                                 BUG();
 631                         }
 632                         break;
 633                 default:
 634                         BUG();
 635                 }
 636         }
 637 }
 638
 639 /**
 640  * Callback received when the frontend's state changes.
 641  */
 642 static void frontend_changed(struct xenbus_device *dev,
 643                              enum xenbus_state frontend_state)
 644 {
 645         struct backend_info *be = dev_get_drvdata(&dev->dev);
 646
 647         pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state));
 648
 649         be->frontend_state = frontend_state;
 650
 651         switch (frontend_state) {
 652         case XenbusStateInitialising:
 653                 set_backend_state(be, XenbusStateInitWait);
 654                 break;
 655
 656         case XenbusStateInitialised:
 657                 break;
 658
 659         case XenbusStateConnected:
 660                 set_backend_state(be, XenbusStateConnected);
 661                 break;
 662
 663         case XenbusStateClosing:
 664                 set_backend_state(be, XenbusStateClosing);
 665                 break;
 666
 667         case XenbusStateClosed:
 668                 set_backend_state(be, XenbusStateClosed);
 669                 if (xenbus_dev_is_online(dev))
 670                         break;
 671                 /* fall through if not online */
 672         case XenbusStateUnknown:
 673                 set_backend_state(be, XenbusStateClosed);
 674                 device_unregister(&dev->dev);
 675                 break;
 676
 677         default:
 678                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 679                                  frontend_state);
 680                 break;
 681         }
 682 }
 683
 684
 685 static void xen_net_read_rate(struct xenbus_device *dev,
 686                               unsigned long *bytes, unsigned long *usec)
 687 {
 688         char *s, *e;
 689         unsigned long b, u;
 690         char *ratestr;
 691
 692         /* Default to unlimited bandwidth. */
 693         *bytes = ~0UL;
 694         *usec = 0;
 695
 696         ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL);
 697         if (IS_ERR(ratestr))
 698                 return;
 699
 700         s = ratestr;
 701         b = simple_strtoul(s, &e, 10);
 702         if ((s == e) || (*e != ','))
 703                 goto fail;
 704
 705         s = e + 1;
 706         u = simple_strtoul(s, &e, 10);
 707         if ((s == e) || (*e != '\0'))
 708                 goto fail;
 709
 710         *bytes = b;
 711         *usec = u;
 712
 713         kfree(ratestr);
 714         return;
 715
 716  fail:
 717         pr_warn("Failed to parse network rate limit. Traffic unlimited.\n");
 718         kfree(ratestr);
 719 }
 720
 721 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
 722 {
 723         char *s, *e, *macstr;
 724         int i;
 725
 726         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
 727         if (IS_ERR(macstr))
 728                 return PTR_ERR(macstr);
 729
 730         for (i = 0; i < ETH_ALEN; i++) {
 731                 mac[i] = simple_strtoul(s, &e, 16);
 732                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
 733                         kfree(macstr);
 734                         return -ENOENT;
 735                 }
 736                 s = e+1;
 737         }
 738
 739         kfree(macstr);
 740         return 0;
 741 }
 742
 743 static void xen_net_rate_changed(struct xenbus_watch *watch,
 744                                  const char *path, const char *token)
 745 {
 746         struct xenvif *vif = container_of(watch, struct xenvif, credit_watch);
 747         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 748         unsigned long   credit_bytes;
 749         unsigned long   credit_usec;
 750         unsigned int queue_index;
 751
 752         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 753         for (queue_index = 0; queue_index < vif->num_queues; queue_index++) {
 754                 struct xenvif_queue *queue = &vif->queues[queue_index];
 755
 756                 queue->credit_bytes = credit_bytes;
 757                 queue->credit_usec = credit_usec;
 758                 if (!mod_timer_pending(&queue->credit_timeout, jiffies) &&
 759                         queue->remaining_credit > queue->credit_bytes) {
 760                         queue->remaining_credit = queue->credit_bytes;
 761                 }
 762         }
 763 }
 764
 765 static int xen_register_credit_watch(struct xenbus_device *dev,
 766                                      struct xenvif *vif)
 767 {
 768         int err = 0;
 769         char *node;
 770         unsigned maxlen = strlen(dev->nodename) + sizeof("/rate");
 771
 772         if (vif->credit_watch.node)
 773                 return -EADDRINUSE;
 774
 775         node = kmalloc(maxlen, GFP_KERNEL);
 776         if (!node)
 777                 return -ENOMEM;
 778         snprintf(node, maxlen, "%s/rate", dev->nodename);
 779         vif->credit_watch.node = node;
 780         vif->credit_watch.callback = xen_net_rate_changed;
 781         err = register_xenbus_watch(&vif->credit_watch);
 782         if (err) {
 783                 pr_err("Failed to set watcher %s\n", vif->credit_watch.node);
 784                 kfree(node);
 785                 vif->credit_watch.node = NULL;
 786                 vif->credit_watch.callback = NULL;
 787         }
 788         return err;
 789 }
 790
 791 static void xen_unregister_credit_watch(struct xenvif *vif)
 792 {
 793         if (vif->credit_watch.node) {
 794                 unregister_xenbus_watch(&vif->credit_watch);
 795                 kfree(vif->credit_watch.node);
 796                 vif->credit_watch.node = NULL;
 797         }
 798 }
 799
 800 static void xen_mcast_ctrl_changed(struct xenbus_watch *watch,
 801                                    const char *path, const char *token)
 802 {
 803         struct xenvif *vif = container_of(watch, struct xenvif,
 804                                           mcast_ctrl_watch);
 805         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 806
 807         vif->multicast_control = !!xenbus_read_unsigned(dev->otherend,
 808                                         "request-multicast-control", 0);
 809 }
 810
 811 static int xen_register_mcast_ctrl_watch(struct xenbus_device *dev,
 812                                          struct xenvif *vif)
 813 {
 814         int err = 0;
 815         char *node;
 816         unsigned maxlen = strlen(dev->otherend) +
 817                 sizeof("/request-multicast-control");
 818
 819         if (vif->mcast_ctrl_watch.node) {
 820                 pr_err_ratelimited("Watch is already registered\n");
 821                 return -EADDRINUSE;
 822         }
 823
 824         node = kmalloc(maxlen, GFP_KERNEL);
 825         if (!node) {
 826                 pr_err("Failed to allocate memory for watch\n");
 827                 return -ENOMEM;
 828         }
 829         snprintf(node, maxlen, "%s/request-multicast-control",
 830                  dev->otherend);
 831         vif->mcast_ctrl_watch.node = node;
 832         vif->mcast_ctrl_watch.callback = xen_mcast_ctrl_changed;
 833         err = register_xenbus_watch(&vif->mcast_ctrl_watch);
 834         if (err) {
 835                 pr_err("Failed to set watcher %s\n",
 836                        vif->mcast_ctrl_watch.node);
 837                 kfree(node);
 838                 vif->mcast_ctrl_watch.node = NULL;
 839                 vif->mcast_ctrl_watch.callback = NULL;
 840         }
 841         return err;
 842 }
 843
 844 static void xen_unregister_mcast_ctrl_watch(struct xenvif *vif)
 845 {
 846         if (vif->mcast_ctrl_watch.node) {
 847                 unregister_xenbus_watch(&vif->mcast_ctrl_watch);
 848                 kfree(vif->mcast_ctrl_watch.node);
 849                 vif->mcast_ctrl_watch.node = NULL;
 850         }
 851 }
 852
 853 static void xen_register_watchers(struct xenbus_device *dev,
 854                                   struct xenvif *vif)
 855 {
 856         xen_register_credit_watch(dev, vif);
 857         xen_register_mcast_ctrl_watch(dev, vif);
 858 }
 859
 860 static void xen_unregister_watchers(struct xenvif *vif)
 861 {
 862         xen_unregister_mcast_ctrl_watch(vif);
 863         xen_unregister_credit_watch(vif);
 864 }
 865
 866 static void unregister_hotplug_status_watch(struct backend_info *be)
 867 {
 868         if (be->have_hotplug_status_watch) {
 869                 unregister_xenbus_watch(&be->hotplug_status_watch);
 870                 kfree(be->hotplug_status_watch.node);
 871         }
 872         be->have_hotplug_status_watch = 0;
 873 }
 874
 875 static void hotplug_status_changed(struct xenbus_watch *watch,
 876                                    const char *path,
 877                                    const char *token)
 878 {
 879         struct backend_info *be = container_of(watch,
 880                                                struct backend_info,
 881                                                hotplug_status_watch);
 882         char *str;
 883         unsigned int len;
 884
 885         str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len);
 886         if (IS_ERR(str))
 887                 return;
 888         if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) {
 889                 /* Complete any pending state change */
 890                 xenbus_switch_state(be->dev, be->state);
 891
 892                 /* Not interested in this watch anymore. */
 893                 unregister_hotplug_status_watch(be);
 894         }
 895         kfree(str);
 896 }
 897
 898 static int connect_ctrl_ring(struct backend_info *be)
 899 {
 900         struct xenbus_device *dev = be->dev;
 901         struct xenvif *vif = be->vif;
 902         unsigned int val;
 903         grant_ref_t ring_ref;
 904         unsigned int evtchn;
 905         int err;
 906
 907         err = xenbus_scanf(XBT_NIL, dev->otherend,
 908                            "ctrl-ring-ref", "%u", &val);
 909         if (err < 0)
 910                 goto done; /* The frontend does not have a control ring */
 911
 912         ring_ref = val;
 913
 914         err = xenbus_scanf(XBT_NIL, dev->otherend,
 915                            "event-channel-ctrl", "%u", &val);
 916         if (err < 0) {
 917                 xenbus_dev_fatal(dev, err,
 918                                  "reading %s/event-channel-ctrl",
 919                                  dev->otherend);
 920                 goto fail;
 921         }
 922
 923         evtchn = val;
 924
 925         err = xenvif_connect_ctrl(vif, ring_ref, evtchn);
 926         if (err) {
 927                 xenbus_dev_fatal(dev, err,
 928                                  "mapping shared-frame %u port %u",
 929                                  ring_ref, evtchn);
 930                 goto fail;
 931         }
 932
 933 done:
 934         return 0;
 935
 936 fail:
 937         return err;
 938 }
 939
 940 static void connect(struct backend_info *be)
 941 {
 942         int err;
 943         struct xenbus_device *dev = be->dev;
 944         unsigned long credit_bytes, credit_usec;
 945         unsigned int queue_index;
 946         unsigned int requested_num_queues;
 947         struct xenvif_queue *queue;
 948
 949         /* Check whether the frontend requested multiple queues
 950          * and read the number requested.
 951          */
 952         requested_num_queues = xenbus_read_unsigned(dev->otherend,
 953                                         "multi-queue-num-queues", 1);
 954         if (requested_num_queues > xenvif_max_queues) {
 955                 /* buggy or malicious guest */
 956                 xenbus_dev_fatal(dev, -EINVAL,
 957                                  "guest requested %u queues, exceeding the maximum of %u.",
 958                                  requested_num_queues, xenvif_max_queues);
 959                 return;
 960         }
 961
 962         err = xen_net_read_mac(dev, be->vif->fe_dev_addr);
 963         if (err) {
 964                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
 965                 return;
 966         }
 967
 968         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 969         xen_unregister_watchers(be->vif);
 970         xen_register_watchers(dev, be->vif);
 971         read_xenbus_vif_flags(be);
 972
 973         err = connect_ctrl_ring(be);
 974         if (err) {
 975                 xenbus_dev_fatal(dev, err, "connecting control ring");
 976                 return;
 977         }
 978
 979         /* Use the number of queues requested by the frontend */
 980         be->vif->queues = vzalloc(requested_num_queues *
 981                                   sizeof(struct xenvif_queue));
 982         if (!be->vif->queues) {
 983                 xenbus_dev_fatal(dev, -ENOMEM,
 984                                  "allocating queues");
 985                 return;
 986         }
 987
 988         be->vif->num_queues = requested_num_queues;
 989         be->vif->stalled_queues = requested_num_queues;
 990
 991         for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) {
 992                 queue = &be->vif->queues[queue_index];
 993                 queue->vif = be->vif;
 994                 queue->id = queue_index;
 995                 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
 996                                 be->vif->dev->name, queue->id);
 997
 998                 err = xenvif_init_queue(queue);
 999                 if (err) {
1000                         /* xenvif_init_queue() cleans up after itself on
1001                          * failure, but we need to clean up any previously
1002                          * initialised queues. Set num_queues to i so that
1003                          * earlier queues can be destroyed using the regular
1004                          * disconnect logic.
1005                          */
1006                         be->vif->num_queues = queue_index;
1007                         goto err;
1008                 }
1009
1010                 queue->credit_bytes = credit_bytes;
1011                 queue->remaining_credit = credit_bytes;
1012                 queue->credit_usec = credit_usec;
1013
1014                 err = connect_data_rings(be, queue);
1015                 if (err) {
1016                         /* connect_data_rings() cleans up after itself on
1017                          * failure, but we need to clean up after
1018                          * xenvif_init_queue() here, and also clean up any
1019                          * previously initialised queues.
1020                          */
1021                         xenvif_deinit_queue(queue);
1022                         be->vif->num_queues = queue_index;
1023                         goto err;
1024                 }
1025         }
1026
1027 #ifdef CONFIG_DEBUG_FS
1028         xenvif_debugfs_addif(be->vif);
1029 #endif /* CONFIG_DEBUG_FS */
1030
1031         /* Initialisation completed, tell core driver the number of
1032          * active queues.
1033          */
1034         rtnl_lock();
1035         netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues);
1036         netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues);
1037         rtnl_unlock();
1038
1039         xenvif_carrier_on(be->vif);
1040
1041         unregister_hotplug_status_watch(be);
1042         err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch,
1043                                    hotplug_status_changed,
1044                                    "%s/%s", dev->nodename, "hotplug-status");
1045         if (!err)
1046                 be->have_hotplug_status_watch = 1;
1047
1048         netif_tx_wake_all_queues(be->vif->dev);
1049
1050         return;
1051
1052 err:
1053         if (be->vif->num_queues > 0)
1054                 xenvif_disconnect_data(be->vif); /* Clean up existing queues */
1055         for (queue_index = 0; queue_index < be->vif->num_queues; ++queue_index)
1056                 xenvif_deinit_queue(&be->vif->queues[queue_index]);
1057         vfree(be->vif->queues);
1058         be->vif->queues = NULL;
1059         be->vif->num_queues = 0;
1060         xenvif_disconnect_ctrl(be->vif);
1061         return;
1062 }
1063
1064
1065 static int connect_data_rings(struct backend_info *be,
1066                               struct xenvif_queue *queue)
1067 {
1068         struct xenbus_device *dev = be->dev;
1069         unsigned int num_queues = queue->vif->num_queues;
1070         unsigned long tx_ring_ref, rx_ring_ref;
1071         unsigned int tx_evtchn, rx_evtchn;
1072         int err;
1073         char *xspath;
1074         size_t xspathsize;
1075         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1076
1077         /* If the frontend requested 1 queue, or we have fallen back
1078          * to single queue due to lack of frontend support for multi-
1079          * queue, expect the remaining XenStore keys in the toplevel
1080          * directory. Otherwise, expect them in a subdirectory called
1081          * queue-N.
1082          */
1083         if (num_queues == 1) {
1084                 xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL);
1085                 if (!xspath) {
1086                         xenbus_dev_fatal(dev, -ENOMEM,
1087                                          "reading ring references");
1088                         return -ENOMEM;
1089                 }
1090                 strcpy(xspath, dev->otherend);
1091         } else {
1092                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1093                 xspath = kzalloc(xspathsize, GFP_KERNEL);
1094                 if (!xspath) {
1095                         xenbus_dev_fatal(dev, -ENOMEM,
1096                                          "reading ring references");
1097                         return -ENOMEM;
1098                 }
1099                 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend,
1100                          queue->id);
1101         }
1102
1103         err = xenbus_gather(XBT_NIL, xspath,
1104                             "tx-ring-ref", "%lu", &tx_ring_ref,
1105                             "rx-ring-ref", "%lu", &rx_ring_ref, NULL);
1106         if (err) {
1107                 xenbus_dev_fatal(dev, err,
1108                                  "reading %s/ring-ref",
1109                                  xspath);
1110                 goto err;
1111         }
1112
1113         /* Try split event channels first, then single event channel. */
1114         err = xenbus_gather(XBT_NIL, xspath,
1115                             "event-channel-tx", "%u", &tx_evtchn,
1116                             "event-channel-rx", "%u", &rx_evtchn, NULL);
1117         if (err < 0) {
1118                 err = xenbus_scanf(XBT_NIL, xspath,
1119                                    "event-channel", "%u", &tx_evtchn);
1120                 if (err < 0) {
1121                         xenbus_dev_fatal(dev, err,
1122                                          "reading %s/event-channel(-tx/rx)",
1123                                          xspath);
1124                         goto err;
1125                 }
1126                 rx_evtchn = tx_evtchn;
1127         }
1128
1129         /* Map the shared frame, irq etc. */
1130         err = xenvif_connect_data(queue, tx_ring_ref, rx_ring_ref,
1131                                   tx_evtchn, rx_evtchn);
1132         if (err) {
1133                 xenbus_dev_fatal(dev, err,
1134                                  "mapping shared-frames %lu/%lu port tx %u rx %u",
1135                                  tx_ring_ref, rx_ring_ref,
1136                                  tx_evtchn, rx_evtchn);
1137                 goto err;
1138         }
1139
1140         err = 0;
1141 err: /* Regular return falls through with err == 0 */
1142         kfree(xspath);
1143         return err;
1144 }
1145
1146 static int read_xenbus_vif_flags(struct backend_info *be)
1147 {
1148         struct xenvif *vif = be->vif;
1149         struct xenbus_device *dev = be->dev;
1150         unsigned int rx_copy;
1151         int err;
1152
1153         err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u",
1154                            &rx_copy);
1155         if (err == -ENOENT) {
1156                 err = 0;
1157                 rx_copy = 0;
1158         }
1159         if (err < 0) {
1160                 xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy",
1161                                  dev->otherend);
1162                 return err;
1163         }
1164         if (!rx_copy)
1165                 return -EOPNOTSUPP;
1166
1167         if (!xenbus_read_unsigned(dev->otherend, "feature-rx-notify", 0)) {
1168                 /* - Reduce drain timeout to poll more frequently for
1169                  *   Rx requests.
1170                  * - Disable Rx stall detection.
1171                  */
1172                 be->vif->drain_timeout = msecs_to_jiffies(30);
1173                 be->vif->stall_timeout = 0;
1174         }
1175
1176         vif->can_sg = !!xenbus_read_unsigned(dev->otherend, "feature-sg", 0);
1177
1178         vif->gso_mask = 0;
1179
1180         if (xenbus_read_unsigned(dev->otherend, "feature-gso-tcpv4", 0))
1181                 vif->gso_mask |= GSO_BIT(TCPV4);
1182
1183         if (xenbus_read_unsigned(dev->otherend, "feature-gso-tcpv6", 0))
1184                 vif->gso_mask |= GSO_BIT(TCPV6);
1185
1186         vif->ip_csum = !xenbus_read_unsigned(dev->otherend,
1187                                              "feature-no-csum-offload", 0);
1188
1189         vif->ipv6_csum = !!xenbus_read_unsigned(dev->otherend,
1190                                                 "feature-ipv6-csum-offload", 0);
1191
1192         return 0;
1193 }
1194
1195 static const struct xenbus_device_id netback_ids[] = {
1196         { "vif" },
1197         { "" }
1198 };
1199
1200 static struct xenbus_driver netback_driver = {
1201         .ids = netback_ids,
1202         .probe = netback_probe,
1203         .remove = netback_remove,
1204         .uevent = netback_uevent,
1205         .otherend_changed = frontend_changed,
1206 };
1207
1208 int xenvif_xenbus_init(void)
1209 {
1210         return xenbus_register_backend(&netback_driver);
1211 }
1212
1213 void xenvif_xenbus_fini(void)
1214 {
1215         return xenbus_unregister_driver(&netback_driver);
1216 }