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                                 backend_switch_state(be, XenbusStateInitWait);
 515                                 break;
 516                         case XenbusStateClosing:
 517                                 backend_switch_state(be, XenbusStateClosing);
 518                                 break;
 519                         default:
 520                                 BUG();
 521                         }
 522                         break;
 523                 case XenbusStateInitWait:
 524                         switch (state) {
 525                         case XenbusStateConnected:
 526                                 backend_connect(be);
 527                                 backend_switch_state(be, XenbusStateConnected);
 528                                 break;
 529                         case XenbusStateClosing:
 530                         case XenbusStateClosed:
 531                                 backend_switch_state(be, XenbusStateClosing);
 532                                 break;
 533                         default:
 534                                 BUG();
 535                         }
 536                         break;
 537                 case XenbusStateConnected:
 538                         switch (state) {
 539                         case XenbusStateInitWait:
 540                         case XenbusStateClosing:
 541                         case XenbusStateClosed:
 542                                 backend_disconnect(be);
 543                                 backend_switch_state(be, XenbusStateClosing);
 544                                 break;
 545                         default:
 546                                 BUG();
 547                         }
 548                         break;
 549                 case XenbusStateClosing:
 550                         switch (state) {
 551                         case XenbusStateInitWait:
 552                         case XenbusStateConnected:
 553                         case XenbusStateClosed:
 554                                 backend_switch_state(be, XenbusStateClosed);
 555                                 break;
 556                         default:
 557                                 BUG();
 558                         }
 559                         break;
 560                 default:
 561                         BUG();
 562                 }
 563         }
 564 }
 565
 566 /**
 567  * Callback received when the frontend's state changes.
 568  */
 569 static void frontend_changed(struct xenbus_device *dev,
 570                              enum xenbus_state frontend_state)
 571 {
 572         struct backend_info *be = dev_get_drvdata(&dev->dev);
 573
 574         pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state));
 575
 576         be->frontend_state = frontend_state;
 577
 578         switch (frontend_state) {
 579         case XenbusStateInitialising:
 580                 set_backend_state(be, XenbusStateInitWait);
 581                 break;
 582
 583         case XenbusStateInitialised:
 584                 break;
 585
 586         case XenbusStateConnected:
 587                 set_backend_state(be, XenbusStateConnected);
 588                 break;
 589
 590         case XenbusStateClosing:
 591                 set_backend_state(be, XenbusStateClosing);
 592                 break;
 593
 594         case XenbusStateClosed:
 595                 set_backend_state(be, XenbusStateClosed);
 596                 if (xenbus_dev_is_online(dev))
 597                         break;
 598                 /* fall through if not online */
 599         case XenbusStateUnknown:
 600                 set_backend_state(be, XenbusStateClosed);
 601                 device_unregister(&dev->dev);
 602                 break;
 603
 604         default:
 605                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 606                                  frontend_state);
 607                 break;
 608         }
 609 }
 610
 611
 612 static void xen_net_read_rate(struct xenbus_device *dev,
 613                               unsigned long *bytes, unsigned long *usec)
 614 {
 615         char *s, *e;
 616         unsigned long b, u;
 617         char *ratestr;
 618
 619         /* Default to unlimited bandwidth. */
 620         *bytes = ~0UL;
 621         *usec = 0;
 622
 623         ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL);
 624         if (IS_ERR(ratestr))
 625                 return;
 626
 627         s = ratestr;
 628         b = simple_strtoul(s, &e, 10);
 629         if ((s == e) || (*e != ','))
 630                 goto fail;
 631
 632         s = e + 1;
 633         u = simple_strtoul(s, &e, 10);
 634         if ((s == e) || (*e != '\0'))
 635                 goto fail;
 636
 637         *bytes = b;
 638         *usec = u;
 639
 640         kfree(ratestr);
 641         return;
 642
 643  fail:
 644         pr_warn("Failed to parse network rate limit. Traffic unlimited.\n");
 645         kfree(ratestr);
 646 }
 647
 648 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
 649 {
 650         char *s, *e, *macstr;
 651         int i;
 652
 653         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
 654         if (IS_ERR(macstr))
 655                 return PTR_ERR(macstr);
 656
 657         for (i = 0; i < ETH_ALEN; i++) {
 658                 mac[i] = simple_strtoul(s, &e, 16);
 659                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
 660                         kfree(macstr);
 661                         return -ENOENT;
 662                 }
 663                 s = e+1;
 664         }
 665
 666         kfree(macstr);
 667         return 0;
 668 }
 669
 670 static void xen_net_rate_changed(struct xenbus_watch *watch,
 671                                 const char **vec, unsigned int len)
 672 {
 673         struct xenvif *vif = container_of(watch, struct xenvif, credit_watch);
 674         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 675         unsigned long   credit_bytes;
 676         unsigned long   credit_usec;
 677         unsigned int queue_index;
 678
 679         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 680         for (queue_index = 0; queue_index < vif->num_queues; queue_index++) {
 681                 struct xenvif_queue *queue = &vif->queues[queue_index];
 682
 683                 queue->credit_bytes = credit_bytes;
 684                 queue->credit_usec = credit_usec;
 685                 if (!mod_timer_pending(&queue->credit_timeout, jiffies) &&
 686                         queue->remaining_credit > queue->credit_bytes) {
 687                         queue->remaining_credit = queue->credit_bytes;
 688                 }
 689         }
 690 }
 691
 692 static int xen_register_credit_watch(struct xenbus_device *dev,
 693                                      struct xenvif *vif)
 694 {
 695         int err = 0;
 696         char *node;
 697         unsigned maxlen = strlen(dev->nodename) + sizeof("/rate");
 698
 699         if (vif->credit_watch.node)
 700                 return -EADDRINUSE;
 701
 702         node = kmalloc(maxlen, GFP_KERNEL);
 703         if (!node)
 704                 return -ENOMEM;
 705         snprintf(node, maxlen, "%s/rate", dev->nodename);
 706         vif->credit_watch.node = node;
 707         vif->credit_watch.callback = xen_net_rate_changed;
 708         err = register_xenbus_watch(&vif->credit_watch);
 709         if (err) {
 710                 pr_err("Failed to set watcher %s\n", vif->credit_watch.node);
 711                 kfree(node);
 712                 vif->credit_watch.node = NULL;
 713                 vif->credit_watch.callback = NULL;
 714         }
 715         return err;
 716 }
 717
 718 static void xen_unregister_credit_watch(struct xenvif *vif)
 719 {
 720         if (vif->credit_watch.node) {
 721                 unregister_xenbus_watch(&vif->credit_watch);
 722                 kfree(vif->credit_watch.node);
 723                 vif->credit_watch.node = NULL;
 724         }
 725 }
 726
 727 static void xen_mcast_ctrl_changed(struct xenbus_watch *watch,
 728                                    const char **vec, unsigned int len)
 729 {
 730         struct xenvif *vif = container_of(watch, struct xenvif,
 731                                           mcast_ctrl_watch);
 732         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 733         int val;
 734
 735         if (xenbus_scanf(XBT_NIL, dev->otherend,
 736                          "request-multicast-control", "%d", &val) < 0)
 737                 val = 0;
 738         vif->multicast_control = !!val;
 739 }
 740
 741 static int xen_register_mcast_ctrl_watch(struct xenbus_device *dev,
 742                                          struct xenvif *vif)
 743 {
 744         int err = 0;
 745         char *node;
 746         unsigned maxlen = strlen(dev->otherend) +
 747                 sizeof("/request-multicast-control");
 748
 749         if (vif->mcast_ctrl_watch.node) {
 750                 pr_err_ratelimited("Watch is already registered\n");
 751                 return -EADDRINUSE;
 752         }
 753
 754         node = kmalloc(maxlen, GFP_KERNEL);
 755         if (!node) {
 756                 pr_err("Failed to allocate memory for watch\n");
 757                 return -ENOMEM;
 758         }
 759         snprintf(node, maxlen, "%s/request-multicast-control",
 760                  dev->otherend);
 761         vif->mcast_ctrl_watch.node = node;
 762         vif->mcast_ctrl_watch.callback = xen_mcast_ctrl_changed;
 763         err = register_xenbus_watch(&vif->mcast_ctrl_watch);
 764         if (err) {
 765                 pr_err("Failed to set watcher %s\n",
 766                        vif->mcast_ctrl_watch.node);
 767                 kfree(node);
 768                 vif->mcast_ctrl_watch.node = NULL;
 769                 vif->mcast_ctrl_watch.callback = NULL;
 770         }
 771         return err;
 772 }
 773
 774 static void xen_unregister_mcast_ctrl_watch(struct xenvif *vif)
 775 {
 776         if (vif->mcast_ctrl_watch.node) {
 777                 unregister_xenbus_watch(&vif->mcast_ctrl_watch);
 778                 kfree(vif->mcast_ctrl_watch.node);
 779                 vif->mcast_ctrl_watch.node = NULL;
 780         }
 781 }
 782
 783 static void xen_register_watchers(struct xenbus_device *dev,
 784                                   struct xenvif *vif)
 785 {
 786         xen_register_credit_watch(dev, vif);
 787         xen_register_mcast_ctrl_watch(dev, vif);
 788 }
 789
 790 static void xen_unregister_watchers(struct xenvif *vif)
 791 {
 792         xen_unregister_mcast_ctrl_watch(vif);
 793         xen_unregister_credit_watch(vif);
 794 }
 795
 796 static void unregister_hotplug_status_watch(struct backend_info *be)
 797 {
 798         if (be->have_hotplug_status_watch) {
 799                 unregister_xenbus_watch(&be->hotplug_status_watch);
 800                 kfree(be->hotplug_status_watch.node);
 801         }
 802         be->have_hotplug_status_watch = 0;
 803 }
 804
 805 static void hotplug_status_changed(struct xenbus_watch *watch,
 806                                    const char **vec,
 807                                    unsigned int vec_size)
 808 {
 809         struct backend_info *be = container_of(watch,
 810                                                struct backend_info,
 811                                                hotplug_status_watch);
 812         char *str;
 813         unsigned int len;
 814
 815         str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len);
 816         if (IS_ERR(str))
 817                 return;
 818         if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) {
 819                 /* Complete any pending state change */
 820                 xenbus_switch_state(be->dev, be->state);
 821
 822                 /* Not interested in this watch anymore. */
 823                 unregister_hotplug_status_watch(be);
 824         }
 825         kfree(str);
 826 }
 827
 828 static void connect(struct backend_info *be)
 829 {
 830         int err;
 831         struct xenbus_device *dev = be->dev;
 832         unsigned long credit_bytes, credit_usec;
 833         unsigned int queue_index;
 834         unsigned int requested_num_queues;
 835         struct xenvif_queue *queue;
 836
 837         /* Check whether the frontend requested multiple queues
 838          * and read the number requested.
 839          */
 840         err = xenbus_scanf(XBT_NIL, dev->otherend,
 841                            "multi-queue-num-queues",
 842                            "%u", &requested_num_queues);
 843         if (err < 0) {
 844                 requested_num_queues = 1; /* Fall back to single queue */
 845         } else if (requested_num_queues > xenvif_max_queues) {
 846                 /* buggy or malicious guest */
 847                 xenbus_dev_fatal(dev, err,
 848                                  "guest requested %u queues, exceeding the maximum of %u.",
 849                                  requested_num_queues, xenvif_max_queues);
 850                 return;
 851         }
 852
 853         err = xen_net_read_mac(dev, be->vif->fe_dev_addr);
 854         if (err) {
 855                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
 856                 return;
 857         }
 858
 859         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 860         xen_unregister_watchers(be->vif);
 861         xen_register_watchers(dev, be->vif);
 862         read_xenbus_vif_flags(be);
 863
 864         /* Use the number of queues requested by the frontend */
 865         be->vif->queues = vzalloc(requested_num_queues *
 866                                   sizeof(struct xenvif_queue));
 867         if (!be->vif->queues) {
 868                 xenbus_dev_fatal(dev, -ENOMEM,
 869                                  "allocating queues");
 870                 return;
 871         }
 872
 873         be->vif->num_queues = requested_num_queues;
 874         be->vif->stalled_queues = requested_num_queues;
 875
 876         for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) {
 877                 queue = &be->vif->queues[queue_index];
 878                 queue->vif = be->vif;
 879                 queue->id = queue_index;
 880                 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
 881                                 be->vif->dev->name, queue->id);
 882
 883                 err = xenvif_init_queue(queue);
 884                 if (err) {
 885                         /* xenvif_init_queue() cleans up after itself on
 886                          * failure, but we need to clean up any previously
 887                          * initialised queues. Set num_queues to i so that
 888                          * earlier queues can be destroyed using the regular
 889                          * disconnect logic.
 890                          */
 891                         be->vif->num_queues = queue_index;
 892                         goto err;
 893                 }
 894
 895                 queue->credit_bytes = credit_bytes;
 896                 queue->remaining_credit = credit_bytes;
 897                 queue->credit_usec = credit_usec;
 898
 899                 err = connect_rings(be, queue);
 900                 if (err) {
 901                         /* connect_rings() cleans up after itself on failure,
 902                          * but we need to clean up after xenvif_init_queue() here,
 903                          * and also clean up any previously initialised queues.
 904                          */
 905                         xenvif_deinit_queue(queue);
 906                         be->vif->num_queues = queue_index;
 907                         goto err;
 908                 }
 909         }
 910
 911 #ifdef CONFIG_DEBUG_FS
 912         xenvif_debugfs_addif(be->vif);
 913 #endif /* CONFIG_DEBUG_FS */
 914
 915         /* Initialisation completed, tell core driver the number of
 916          * active queues.
 917          */
 918         rtnl_lock();
 919         netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues);
 920         netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues);
 921         rtnl_unlock();
 922
 923         xenvif_carrier_on(be->vif);
 924
 925         unregister_hotplug_status_watch(be);
 926         err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch,
 927                                    hotplug_status_changed,
 928                                    "%s/%s", dev->nodename, "hotplug-status");
 929         if (!err)
 930                 be->have_hotplug_status_watch = 1;
 931
 932         netif_tx_wake_all_queues(be->vif->dev);
 933
 934         return;
 935
 936 err:
 937         if (be->vif->num_queues > 0)
 938                 xenvif_disconnect(be->vif); /* Clean up existing queues */
 939         vfree(be->vif->queues);
 940         be->vif->queues = NULL;
 941         be->vif->num_queues = 0;
 942         return;
 943 }
 944
 945
 946 static int connect_rings(struct backend_info *be, struct xenvif_queue *queue)
 947 {
 948         struct xenbus_device *dev = be->dev;
 949         unsigned int num_queues = queue->vif->num_queues;
 950         unsigned long tx_ring_ref, rx_ring_ref;
 951         unsigned int tx_evtchn, rx_evtchn;
 952         int err;
 953         char *xspath;
 954         size_t xspathsize;
 955         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
 956
 957         /* If the frontend requested 1 queue, or we have fallen back
 958          * to single queue due to lack of frontend support for multi-
 959          * queue, expect the remaining XenStore keys in the toplevel
 960          * directory. Otherwise, expect them in a subdirectory called
 961          * queue-N.
 962          */
 963         if (num_queues == 1) {
 964                 xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL);
 965                 if (!xspath) {
 966                         xenbus_dev_fatal(dev, -ENOMEM,
 967                                          "reading ring references");
 968                         return -ENOMEM;
 969                 }
 970                 strcpy(xspath, dev->otherend);
 971         } else {
 972                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
 973                 xspath = kzalloc(xspathsize, GFP_KERNEL);
 974                 if (!xspath) {
 975                         xenbus_dev_fatal(dev, -ENOMEM,
 976                                          "reading ring references");
 977                         return -ENOMEM;
 978                 }
 979                 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend,
 980                          queue->id);
 981         }
 982
 983         err = xenbus_gather(XBT_NIL, xspath,
 984                             "tx-ring-ref", "%lu", &tx_ring_ref,
 985                             "rx-ring-ref", "%lu", &rx_ring_ref, NULL);
 986         if (err) {
 987                 xenbus_dev_fatal(dev, err,
 988                                  "reading %s/ring-ref",
 989                                  xspath);
 990                 goto err;
 991         }
 992
 993         /* Try split event channels first, then single event channel. */
 994         err = xenbus_gather(XBT_NIL, xspath,
 995                             "event-channel-tx", "%u", &tx_evtchn,
 996                             "event-channel-rx", "%u", &rx_evtchn, NULL);
 997         if (err < 0) {
 998                 err = xenbus_scanf(XBT_NIL, xspath,
 999                                    "event-channel", "%u", &tx_evtchn);
1000                 if (err < 0) {
1001                         xenbus_dev_fatal(dev, err,
1002                                          "reading %s/event-channel(-tx/rx)",
1003                                          xspath);
1004                         goto err;
1005                 }
1006                 rx_evtchn = tx_evtchn;
1007         }
1008
1009         /* Map the shared frame, irq etc. */
1010         err = xenvif_connect(queue, tx_ring_ref, rx_ring_ref,
1011                              tx_evtchn, rx_evtchn);
1012         if (err) {
1013                 xenbus_dev_fatal(dev, err,
1014                                  "mapping shared-frames %lu/%lu port tx %u rx %u",
1015                                  tx_ring_ref, rx_ring_ref,
1016                                  tx_evtchn, rx_evtchn);
1017                 goto err;
1018         }
1019
1020         err = 0;
1021 err: /* Regular return falls through with err == 0 */
1022         kfree(xspath);
1023         return err;
1024 }
1025
1026 static int read_xenbus_vif_flags(struct backend_info *be)
1027 {
1028         struct xenvif *vif = be->vif;
1029         struct xenbus_device *dev = be->dev;
1030         unsigned int rx_copy;
1031         int err, val;
1032
1033         err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u",
1034                            &rx_copy);
1035         if (err == -ENOENT) {
1036                 err = 0;
1037                 rx_copy = 0;
1038         }
1039         if (err < 0) {
1040                 xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy",
1041                                  dev->otherend);
1042                 return err;
1043         }
1044         if (!rx_copy)
1045                 return -EOPNOTSUPP;
1046
1047         if (xenbus_scanf(XBT_NIL, dev->otherend,
1048                          "feature-rx-notify", "%d", &val) < 0)
1049                 val = 0;
1050         if (!val) {
1051                 /* - Reduce drain timeout to poll more frequently for
1052                  *   Rx requests.
1053                  * - Disable Rx stall detection.
1054                  */
1055                 be->vif->drain_timeout = msecs_to_jiffies(30);
1056                 be->vif->stall_timeout = 0;
1057         }
1058
1059         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-sg",
1060                          "%d", &val) < 0)
1061                 val = 0;
1062         vif->can_sg = !!val;
1063
1064         vif->gso_mask = 0;
1065         vif->gso_prefix_mask = 0;
1066
1067         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4",
1068                          "%d", &val) < 0)
1069                 val = 0;
1070         if (val)
1071                 vif->gso_mask |= GSO_BIT(TCPV4);
1072
1073         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4-prefix",
1074                          "%d", &val) < 0)
1075                 val = 0;
1076         if (val)
1077                 vif->gso_prefix_mask |= GSO_BIT(TCPV4);
1078
1079         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6",
1080                          "%d", &val) < 0)
1081                 val = 0;
1082         if (val)
1083                 vif->gso_mask |= GSO_BIT(TCPV6);
1084
1085         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6-prefix",
1086                          "%d", &val) < 0)
1087                 val = 0;
1088         if (val)
1089                 vif->gso_prefix_mask |= GSO_BIT(TCPV6);
1090
1091         if (vif->gso_mask & vif->gso_prefix_mask) {
1092                 xenbus_dev_fatal(dev, err,
1093                                  "%s: gso and gso prefix flags are not "
1094                                  "mutually exclusive",
1095                                  dev->otherend);
1096                 return -EOPNOTSUPP;
1097         }
1098
1099         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-no-csum-offload",
1100                          "%d", &val) < 0)
1101                 val = 0;
1102         vif->ip_csum = !val;
1103
1104         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-ipv6-csum-offload",
1105                          "%d", &val) < 0)
1106                 val = 0;
1107         vif->ipv6_csum = !!val;
1108
1109         return 0;
1110 }
1111
1112 static const struct xenbus_device_id netback_ids[] = {
1113         { "vif" },
1114         { "" }
1115 };
1116
1117 static struct xenbus_driver netback_driver = {
1118         .ids = netback_ids,
1119         .probe = netback_probe,
1120         .remove = netback_remove,
1121         .uevent = netback_uevent,
1122         .otherend_changed = frontend_changed,
1123 };
1124
1125 int xenvif_xenbus_init(void)
1126 {
1127         return xenbus_register_backend(&netback_driver);
1128 }
1129
1130 void xenvif_xenbus_fini(void)
1131 {
1132         return xenbus_unregister_driver(&netback_driver);
1133 }