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