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_dump_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_dump_open,
 183         .read = seq_read,
 184         .llseek = seq_lseek,
 185         .release = single_release,
 186         .write = xenvif_write_io_ring,
 187 };
 188
 189 static void xenvif_debugfs_addif(struct xenvif *vif)
 190 {
 191         struct dentry *pfile;
 192         int i;
 193
 194         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 195                 return;
 196
 197         vif->xenvif_dbg_root = debugfs_create_dir(vif->dev->name,
 198                                                   xen_netback_dbg_root);
 199         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) {
 200                 for (i = 0; i < vif->num_queues; ++i) {
 201                         char filename[sizeof("io_ring_q") + 4];
 202
 203                         snprintf(filename, sizeof(filename), "io_ring_q%d", i);
 204                         pfile = debugfs_create_file(filename,
 205                                                     S_IRUSR | S_IWUSR,
 206                                                     vif->xenvif_dbg_root,
 207                                                     &vif->queues[i],
 208                                                     &xenvif_dbg_io_ring_ops_fops);
 209                         if (IS_ERR_OR_NULL(pfile))
 210                                 pr_warn("Creation of io_ring file returned %ld!\n",
 211                                         PTR_ERR(pfile));
 212                 }
 213         } else
 214                 netdev_warn(vif->dev,
 215                             "Creation of vif debugfs dir returned %ld!\n",
 216                             PTR_ERR(vif->xenvif_dbg_root));
 217 }
 218
 219 static void xenvif_debugfs_delif(struct xenvif *vif)
 220 {
 221         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 222                 return;
 223
 224         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root))
 225                 debugfs_remove_recursive(vif->xenvif_dbg_root);
 226         vif->xenvif_dbg_root = NULL;
 227 }
 228 #endif /* CONFIG_DEBUG_FS */
 229
 230 static int netback_remove(struct xenbus_device *dev)
 231 {
 232         struct backend_info *be = dev_get_drvdata(&dev->dev);
 233
 234         set_backend_state(be, XenbusStateClosed);
 235
 236         unregister_hotplug_status_watch(be);
 237         if (be->vif) {
 238                 kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE);
 239                 xen_unregister_watchers(be->vif);
 240                 xenbus_rm(XBT_NIL, dev->nodename, "hotplug-status");
 241                 xenvif_free(be->vif);
 242                 be->vif = NULL;
 243         }
 244         kfree(be->hotplug_script);
 245         kfree(be);
 246         dev_set_drvdata(&dev->dev, NULL);
 247         return 0;
 248 }
 249
 250
 251 /**
 252  * Entry point to this code when a new device is created.  Allocate the basic
 253  * structures and switch to InitWait.
 254  */
 255 static int netback_probe(struct xenbus_device *dev,
 256                          const struct xenbus_device_id *id)
 257 {
 258         const char *message;
 259         struct xenbus_transaction xbt;
 260         int err;
 261         int sg;
 262         const char *script;
 263         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 264                                           GFP_KERNEL);
 265         if (!be) {
 266                 xenbus_dev_fatal(dev, -ENOMEM,
 267                                  "allocating backend structure");
 268                 return -ENOMEM;
 269         }
 270
 271         be->dev = dev;
 272         dev_set_drvdata(&dev->dev, be);
 273
 274         sg = 1;
 275
 276         do {
 277                 err = xenbus_transaction_start(&xbt);
 278                 if (err) {
 279                         xenbus_dev_fatal(dev, err, "starting transaction");
 280                         goto fail;
 281                 }
 282
 283                 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg);
 284                 if (err) {
 285                         message = "writing feature-sg";
 286                         goto abort_transaction;
 287                 }
 288
 289                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4",
 290                                     "%d", sg);
 291                 if (err) {
 292                         message = "writing feature-gso-tcpv4";
 293                         goto abort_transaction;
 294                 }
 295
 296                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6",
 297                                     "%d", sg);
 298                 if (err) {
 299                         message = "writing feature-gso-tcpv6";
 300                         goto abort_transaction;
 301                 }
 302
 303                 /* We support partial checksum setup for IPv6 packets */
 304                 err = xenbus_printf(xbt, dev->nodename,
 305                                     "feature-ipv6-csum-offload",
 306                                     "%d", 1);
 307                 if (err) {
 308                         message = "writing feature-ipv6-csum-offload";
 309                         goto abort_transaction;
 310                 }
 311
 312                 /* We support rx-copy path. */
 313                 err = xenbus_printf(xbt, dev->nodename,
 314                                     "feature-rx-copy", "%d", 1);
 315                 if (err) {
 316                         message = "writing feature-rx-copy";
 317                         goto abort_transaction;
 318                 }
 319
 320                 /*
 321                  * We don't support rx-flip path (except old guests who don't
 322                  * grok this feature flag).
 323                  */
 324                 err = xenbus_printf(xbt, dev->nodename,
 325                                     "feature-rx-flip", "%d", 0);
 326                 if (err) {
 327                         message = "writing feature-rx-flip";
 328                         goto abort_transaction;
 329                 }
 330
 331                 /* We support dynamic multicast-control. */
 332                 err = xenbus_printf(xbt, dev->nodename,
 333                                     "feature-multicast-control", "%d", 1);
 334                 if (err) {
 335                         message = "writing feature-multicast-control";
 336                         goto abort_transaction;
 337                 }
 338
 339                 err = xenbus_printf(xbt, dev->nodename,
 340                                     "feature-dynamic-multicast-control",
 341                                     "%d", 1);
 342                 if (err) {
 343                         message = "writing feature-dynamic-multicast-control";
 344                         goto abort_transaction;
 345                 }
 346
 347                 err = xenbus_transaction_end(xbt, 0);
 348         } while (err == -EAGAIN);
 349
 350         if (err) {
 351                 xenbus_dev_fatal(dev, err, "completing transaction");
 352                 goto fail;
 353         }
 354
 355         /*
 356          * Split event channels support, this is optional so it is not
 357          * put inside the above loop.
 358          */
 359         err = xenbus_printf(XBT_NIL, dev->nodename,
 360                             "feature-split-event-channels",
 361                             "%u", separate_tx_rx_irq);
 362         if (err)
 363                 pr_debug("Error writing feature-split-event-channels\n");
 364
 365         /* Multi-queue support: This is an optional feature. */
 366         err = xenbus_printf(XBT_NIL, dev->nodename,
 367                             "multi-queue-max-queues", "%u", xenvif_max_queues);
 368         if (err)
 369                 pr_debug("Error writing multi-queue-max-queues\n");
 370
 371         err = xenbus_printf(XBT_NIL, dev->nodename,
 372                             "feature-ctrl-ring",
 373                             "%u", true);
 374         if (err)
 375                 pr_debug("Error writing feature-ctrl-ring\n");
 376
 377         script = xenbus_read(XBT_NIL, dev->nodename, "script", NULL);
 378         if (IS_ERR(script)) {
 379                 err = PTR_ERR(script);
 380                 xenbus_dev_fatal(dev, err, "reading script");
 381                 goto fail;
 382         }
 383
 384         be->hotplug_script = script;
 385
 386         err = xenbus_switch_state(dev, XenbusStateInitWait);
 387         if (err)
 388                 goto fail;
 389
 390         be->state = XenbusStateInitWait;
 391
 392         /* This kicks hotplug scripts, so do it immediately. */
 393         err = backend_create_xenvif(be);
 394         if (err)
 395                 goto fail;
 396
 397         return 0;
 398
 399 abort_transaction:
 400         xenbus_transaction_end(xbt, 1);
 401         xenbus_dev_fatal(dev, err, "%s", message);
 402 fail:
 403         pr_debug("failed\n");
 404         netback_remove(dev);
 405         return err;
 406 }
 407
 408
 409 /*
 410  * Handle the creation of the hotplug script environment.  We add the script
 411  * and vif variables to the environment, for the benefit of the vif-* hotplug
 412  * scripts.
 413  */
 414 static int netback_uevent(struct xenbus_device *xdev,
 415                           struct kobj_uevent_env *env)
 416 {
 417         struct backend_info *be = dev_get_drvdata(&xdev->dev);
 418
 419         if (!be)
 420                 return 0;
 421
 422         if (add_uevent_var(env, "script=%s", be->hotplug_script))
 423                 return -ENOMEM;
 424
 425         if (!be->vif)
 426                 return 0;
 427
 428         return add_uevent_var(env, "vif=%s", be->vif->dev->name);
 429 }
 430
 431
 432 static int backend_create_xenvif(struct backend_info *be)
 433 {
 434         int err;
 435         long handle;
 436         struct xenbus_device *dev = be->dev;
 437         struct xenvif *vif;
 438
 439         if (be->vif != NULL)
 440                 return 0;
 441
 442         err = xenbus_scanf(XBT_NIL, dev->nodename, "handle", "%li", &handle);
 443         if (err != 1) {
 444                 xenbus_dev_fatal(dev, err, "reading handle");
 445                 return (err < 0) ? err : -EINVAL;
 446         }
 447
 448         vif = xenvif_alloc(&dev->dev, dev->otherend_id, handle);
 449         if (IS_ERR(vif)) {
 450                 err = PTR_ERR(vif);
 451                 xenbus_dev_fatal(dev, err, "creating interface");
 452                 return err;
 453         }
 454         be->vif = vif;
 455
 456         kobject_uevent(&dev->dev.kobj, KOBJ_ONLINE);
 457         return 0;
 458 }
 459
 460 static void backend_disconnect(struct backend_info *be)
 461 {
 462         if (be->vif) {
 463                 xen_unregister_watchers(be->vif);
 464 #ifdef CONFIG_DEBUG_FS
 465                 xenvif_debugfs_delif(be->vif);
 466 #endif /* CONFIG_DEBUG_FS */
 467                 xenvif_disconnect_data(be->vif);
 468                 xenvif_disconnect_ctrl(be->vif);
 469         }
 470 }
 471
 472 static void backend_connect(struct backend_info *be)
 473 {
 474         if (be->vif)
 475                 connect(be);
 476 }
 477
 478 static inline void backend_switch_state(struct backend_info *be,
 479                                         enum xenbus_state state)
 480 {
 481         struct xenbus_device *dev = be->dev;
 482
 483         pr_debug("%s -> %s\n", dev->nodename, xenbus_strstate(state));
 484         be->state = state;
 485
 486         /* If we are waiting for a hotplug script then defer the
 487          * actual xenbus state change.
 488          */
 489         if (!be->have_hotplug_status_watch)
 490                 xenbus_switch_state(dev, state);
 491 }
 492
 493 /* Handle backend state transitions:
 494  *
 495  * The backend state starts in InitWait and the following transitions are
 496  * allowed.
 497  *
 498  * InitWait -> Connected
 499  *
 500  *    ^    \         |
 501  *    |     \        |
 502  *    |      \       |
 503  *    |       \      |
 504  *    |        \     |
 505  *    |         \    |
 506  *    |          V   V
 507  *
 508  *  Closed  <-> Closing
 509  *
 510  * The state argument specifies the eventual state of the backend and the
 511  * function transitions to that state via the shortest path.
 512  */
 513 static void set_backend_state(struct backend_info *be,
 514                               enum xenbus_state state)
 515 {
 516         while (be->state != state) {
 517                 switch (be->state) {
 518                 case XenbusStateClosed:
 519                         switch (state) {
 520                         case XenbusStateInitWait:
 521                         case XenbusStateConnected:
 522                                 backend_switch_state(be, XenbusStateInitWait);
 523                                 break;
 524                         case XenbusStateClosing:
 525                                 backend_switch_state(be, XenbusStateClosing);
 526                                 break;
 527                         default:
 528                                 BUG();
 529                         }
 530                         break;
 531                 case XenbusStateInitWait:
 532                         switch (state) {
 533                         case XenbusStateConnected:
 534                                 backend_connect(be);
 535                                 backend_switch_state(be, XenbusStateConnected);
 536                                 break;
 537                         case XenbusStateClosing:
 538                         case XenbusStateClosed:
 539                                 backend_switch_state(be, XenbusStateClosing);
 540                                 break;
 541                         default:
 542                                 BUG();
 543                         }
 544                         break;
 545                 case XenbusStateConnected:
 546                         switch (state) {
 547                         case XenbusStateInitWait:
 548                         case XenbusStateClosing:
 549                         case XenbusStateClosed:
 550                                 backend_disconnect(be);
 551                                 backend_switch_state(be, XenbusStateClosing);
 552                                 break;
 553                         default:
 554                                 BUG();
 555                         }
 556                         break;
 557                 case XenbusStateClosing:
 558                         switch (state) {
 559                         case XenbusStateInitWait:
 560                         case XenbusStateConnected:
 561                         case XenbusStateClosed:
 562                                 backend_switch_state(be, XenbusStateClosed);
 563                                 break;
 564                         default:
 565                                 BUG();
 566                         }
 567                         break;
 568                 default:
 569                         BUG();
 570                 }
 571         }
 572 }
 573
 574 /**
 575  * Callback received when the frontend's state changes.
 576  */
 577 static void frontend_changed(struct xenbus_device *dev,
 578                              enum xenbus_state frontend_state)
 579 {
 580         struct backend_info *be = dev_get_drvdata(&dev->dev);
 581
 582         pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state));
 583
 584         be->frontend_state = frontend_state;
 585
 586         switch (frontend_state) {
 587         case XenbusStateInitialising:
 588                 set_backend_state(be, XenbusStateInitWait);
 589                 break;
 590
 591         case XenbusStateInitialised:
 592                 break;
 593
 594         case XenbusStateConnected:
 595                 set_backend_state(be, XenbusStateConnected);
 596                 break;
 597
 598         case XenbusStateClosing:
 599                 set_backend_state(be, XenbusStateClosing);
 600                 break;
 601
 602         case XenbusStateClosed:
 603                 set_backend_state(be, XenbusStateClosed);
 604                 if (xenbus_dev_is_online(dev))
 605                         break;
 606                 /* fall through if not online */
 607         case XenbusStateUnknown:
 608                 set_backend_state(be, XenbusStateClosed);
 609                 device_unregister(&dev->dev);
 610                 break;
 611
 612         default:
 613                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 614                                  frontend_state);
 615                 break;
 616         }
 617 }
 618
 619
 620 static void xen_net_read_rate(struct xenbus_device *dev,
 621                               unsigned long *bytes, unsigned long *usec)
 622 {
 623         char *s, *e;
 624         unsigned long b, u;
 625         char *ratestr;
 626
 627         /* Default to unlimited bandwidth. */
 628         *bytes = ~0UL;
 629         *usec = 0;
 630
 631         ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL);
 632         if (IS_ERR(ratestr))
 633                 return;
 634
 635         s = ratestr;
 636         b = simple_strtoul(s, &e, 10);
 637         if ((s == e) || (*e != ','))
 638                 goto fail;
 639
 640         s = e + 1;
 641         u = simple_strtoul(s, &e, 10);
 642         if ((s == e) || (*e != '\0'))
 643                 goto fail;
 644
 645         *bytes = b;
 646         *usec = u;
 647
 648         kfree(ratestr);
 649         return;
 650
 651  fail:
 652         pr_warn("Failed to parse network rate limit. Traffic unlimited.\n");
 653         kfree(ratestr);
 654 }
 655
 656 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
 657 {
 658         char *s, *e, *macstr;
 659         int i;
 660
 661         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
 662         if (IS_ERR(macstr))
 663                 return PTR_ERR(macstr);
 664
 665         for (i = 0; i < ETH_ALEN; i++) {
 666                 mac[i] = simple_strtoul(s, &e, 16);
 667                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
 668                         kfree(macstr);
 669                         return -ENOENT;
 670                 }
 671                 s = e+1;
 672         }
 673
 674         kfree(macstr);
 675         return 0;
 676 }
 677
 678 static void xen_net_rate_changed(struct xenbus_watch *watch,
 679                                 const char **vec, unsigned int len)
 680 {
 681         struct xenvif *vif = container_of(watch, struct xenvif, credit_watch);
 682         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 683         unsigned long   credit_bytes;
 684         unsigned long   credit_usec;
 685         unsigned int queue_index;
 686
 687         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 688         for (queue_index = 0; queue_index < vif->num_queues; queue_index++) {
 689                 struct xenvif_queue *queue = &vif->queues[queue_index];
 690
 691                 queue->credit_bytes = credit_bytes;
 692                 queue->credit_usec = credit_usec;
 693                 if (!mod_timer_pending(&queue->credit_timeout, jiffies) &&
 694                         queue->remaining_credit > queue->credit_bytes) {
 695                         queue->remaining_credit = queue->credit_bytes;
 696                 }
 697         }
 698 }
 699
 700 static int xen_register_credit_watch(struct xenbus_device *dev,
 701                                      struct xenvif *vif)
 702 {
 703         int err = 0;
 704         char *node;
 705         unsigned maxlen = strlen(dev->nodename) + sizeof("/rate");
 706
 707         if (vif->credit_watch.node)
 708                 return -EADDRINUSE;
 709
 710         node = kmalloc(maxlen, GFP_KERNEL);
 711         if (!node)
 712                 return -ENOMEM;
 713         snprintf(node, maxlen, "%s/rate", dev->nodename);
 714         vif->credit_watch.node = node;
 715         vif->credit_watch.callback = xen_net_rate_changed;
 716         err = register_xenbus_watch(&vif->credit_watch);
 717         if (err) {
 718                 pr_err("Failed to set watcher %s\n", vif->credit_watch.node);
 719                 kfree(node);
 720                 vif->credit_watch.node = NULL;
 721                 vif->credit_watch.callback = NULL;
 722         }
 723         return err;
 724 }
 725
 726 static void xen_unregister_credit_watch(struct xenvif *vif)
 727 {
 728         if (vif->credit_watch.node) {
 729                 unregister_xenbus_watch(&vif->credit_watch);
 730                 kfree(vif->credit_watch.node);
 731                 vif->credit_watch.node = NULL;
 732         }
 733 }
 734
 735 static void xen_mcast_ctrl_changed(struct xenbus_watch *watch,
 736                                    const char **vec, unsigned int len)
 737 {
 738         struct xenvif *vif = container_of(watch, struct xenvif,
 739                                           mcast_ctrl_watch);
 740         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 741         int val;
 742
 743         if (xenbus_scanf(XBT_NIL, dev->otherend,
 744                          "request-multicast-control", "%d", &val) < 0)
 745                 val = 0;
 746         vif->multicast_control = !!val;
 747 }
 748
 749 static int xen_register_mcast_ctrl_watch(struct xenbus_device *dev,
 750                                          struct xenvif *vif)
 751 {
 752         int err = 0;
 753         char *node;
 754         unsigned maxlen = strlen(dev->otherend) +
 755                 sizeof("/request-multicast-control");
 756
 757         if (vif->mcast_ctrl_watch.node) {
 758                 pr_err_ratelimited("Watch is already registered\n");
 759                 return -EADDRINUSE;
 760         }
 761
 762         node = kmalloc(maxlen, GFP_KERNEL);
 763         if (!node) {
 764                 pr_err("Failed to allocate memory for watch\n");
 765                 return -ENOMEM;
 766         }
 767         snprintf(node, maxlen, "%s/request-multicast-control",
 768                  dev->otherend);
 769         vif->mcast_ctrl_watch.node = node;
 770         vif->mcast_ctrl_watch.callback = xen_mcast_ctrl_changed;
 771         err = register_xenbus_watch(&vif->mcast_ctrl_watch);
 772         if (err) {
 773                 pr_err("Failed to set watcher %s\n",
 774                        vif->mcast_ctrl_watch.node);
 775                 kfree(node);
 776                 vif->mcast_ctrl_watch.node = NULL;
 777                 vif->mcast_ctrl_watch.callback = NULL;
 778         }
 779         return err;
 780 }
 781
 782 static void xen_unregister_mcast_ctrl_watch(struct xenvif *vif)
 783 {
 784         if (vif->mcast_ctrl_watch.node) {
 785                 unregister_xenbus_watch(&vif->mcast_ctrl_watch);
 786                 kfree(vif->mcast_ctrl_watch.node);
 787                 vif->mcast_ctrl_watch.node = NULL;
 788         }
 789 }
 790
 791 static void xen_register_watchers(struct xenbus_device *dev,
 792                                   struct xenvif *vif)
 793 {
 794         xen_register_credit_watch(dev, vif);
 795         xen_register_mcast_ctrl_watch(dev, vif);
 796 }
 797
 798 static void xen_unregister_watchers(struct xenvif *vif)
 799 {
 800         xen_unregister_mcast_ctrl_watch(vif);
 801         xen_unregister_credit_watch(vif);
 802 }
 803
 804 static void unregister_hotplug_status_watch(struct backend_info *be)
 805 {
 806         if (be->have_hotplug_status_watch) {
 807                 unregister_xenbus_watch(&be->hotplug_status_watch);
 808                 kfree(be->hotplug_status_watch.node);
 809         }
 810         be->have_hotplug_status_watch = 0;
 811 }
 812
 813 static void hotplug_status_changed(struct xenbus_watch *watch,
 814                                    const char **vec,
 815                                    unsigned int vec_size)
 816 {
 817         struct backend_info *be = container_of(watch,
 818                                                struct backend_info,
 819                                                hotplug_status_watch);
 820         char *str;
 821         unsigned int len;
 822
 823         str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len);
 824         if (IS_ERR(str))
 825                 return;
 826         if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) {
 827                 /* Complete any pending state change */
 828                 xenbus_switch_state(be->dev, be->state);
 829
 830                 /* Not interested in this watch anymore. */
 831                 unregister_hotplug_status_watch(be);
 832         }
 833         kfree(str);
 834 }
 835
 836 static int connect_ctrl_ring(struct backend_info *be)
 837 {
 838         struct xenbus_device *dev = be->dev;
 839         struct xenvif *vif = be->vif;
 840         unsigned int val;
 841         grant_ref_t ring_ref;
 842         unsigned int evtchn;
 843         int err;
 844
 845         err = xenbus_gather(XBT_NIL, dev->otherend,
 846                             "ctrl-ring-ref", "%u", &val, NULL);
 847         if (err)
 848                 goto done; /* The frontend does not have a control ring */
 849
 850         ring_ref = val;
 851
 852         err = xenbus_gather(XBT_NIL, dev->otherend,
 853                             "event-channel-ctrl", "%u", &val, NULL);
 854         if (err) {
 855                 xenbus_dev_fatal(dev, err,
 856                                  "reading %s/event-channel-ctrl",
 857                                  dev->otherend);
 858                 goto fail;
 859         }
 860
 861         evtchn = val;
 862
 863         err = xenvif_connect_ctrl(vif, ring_ref, evtchn);
 864         if (err) {
 865                 xenbus_dev_fatal(dev, err,
 866                                  "mapping shared-frame %u port %u",
 867                                  ring_ref, evtchn);
 868                 goto fail;
 869         }
 870
 871 done:
 872         return 0;
 873
 874 fail:
 875         return err;
 876 }
 877
 878 static void connect(struct backend_info *be)
 879 {
 880         int err;
 881         struct xenbus_device *dev = be->dev;
 882         unsigned long credit_bytes, credit_usec;
 883         unsigned int queue_index;
 884         unsigned int requested_num_queues;
 885         struct xenvif_queue *queue;
 886
 887         /* Check whether the frontend requested multiple queues
 888          * and read the number requested.
 889          */
 890         err = xenbus_scanf(XBT_NIL, dev->otherend,
 891                            "multi-queue-num-queues",
 892                            "%u", &requested_num_queues);
 893         if (err < 0) {
 894                 requested_num_queues = 1; /* Fall back to single queue */
 895         } else if (requested_num_queues > xenvif_max_queues) {
 896                 /* buggy or malicious guest */
 897                 xenbus_dev_fatal(dev, err,
 898                                  "guest requested %u queues, exceeding the maximum of %u.",
 899                                  requested_num_queues, xenvif_max_queues);
 900                 return;
 901         }
 902
 903         err = xen_net_read_mac(dev, be->vif->fe_dev_addr);
 904         if (err) {
 905                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
 906                 return;
 907         }
 908
 909         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 910         xen_unregister_watchers(be->vif);
 911         xen_register_watchers(dev, be->vif);
 912         read_xenbus_vif_flags(be);
 913
 914         err = connect_ctrl_ring(be);
 915         if (err) {
 916                 xenbus_dev_fatal(dev, err, "connecting control ring");
 917                 return;
 918         }
 919
 920         /* Use the number of queues requested by the frontend */
 921         be->vif->queues = vzalloc(requested_num_queues *
 922                                   sizeof(struct xenvif_queue));
 923         if (!be->vif->queues) {
 924                 xenbus_dev_fatal(dev, -ENOMEM,
 925                                  "allocating queues");
 926                 return;
 927         }
 928
 929         be->vif->num_queues = requested_num_queues;
 930         be->vif->stalled_queues = requested_num_queues;
 931
 932         for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) {
 933                 queue = &be->vif->queues[queue_index];
 934                 queue->vif = be->vif;
 935                 queue->id = queue_index;
 936                 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
 937                                 be->vif->dev->name, queue->id);
 938
 939                 err = xenvif_init_queue(queue);
 940                 if (err) {
 941                         /* xenvif_init_queue() cleans up after itself on
 942                          * failure, but we need to clean up any previously
 943                          * initialised queues. Set num_queues to i so that
 944                          * earlier queues can be destroyed using the regular
 945                          * disconnect logic.
 946                          */
 947                         be->vif->num_queues = queue_index;
 948                         goto err;
 949                 }
 950
 951                 queue->credit_bytes = credit_bytes;
 952                 queue->remaining_credit = credit_bytes;
 953                 queue->credit_usec = credit_usec;
 954
 955                 err = connect_data_rings(be, queue);
 956                 if (err) {
 957                         /* connect_data_rings() cleans up after itself on
 958                          * failure, but we need to clean up after
 959                          * xenvif_init_queue() here, and also clean up any
 960                          * previously initialised queues.
 961                          */
 962                         xenvif_deinit_queue(queue);
 963                         be->vif->num_queues = queue_index;
 964                         goto err;
 965                 }
 966         }
 967
 968 #ifdef CONFIG_DEBUG_FS
 969         xenvif_debugfs_addif(be->vif);
 970 #endif /* CONFIG_DEBUG_FS */
 971
 972         /* Initialisation completed, tell core driver the number of
 973          * active queues.
 974          */
 975         rtnl_lock();
 976         netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues);
 977         netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues);
 978         rtnl_unlock();
 979
 980         xenvif_carrier_on(be->vif);
 981
 982         unregister_hotplug_status_watch(be);
 983         err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch,
 984                                    hotplug_status_changed,
 985                                    "%s/%s", dev->nodename, "hotplug-status");
 986         if (!err)
 987                 be->have_hotplug_status_watch = 1;
 988
 989         netif_tx_wake_all_queues(be->vif->dev);
 990
 991         return;
 992
 993 err:
 994         if (be->vif->num_queues > 0)
 995                 xenvif_disconnect_data(be->vif); /* Clean up existing queues */
 996         vfree(be->vif->queues);
 997         be->vif->queues = NULL;
 998         be->vif->num_queues = 0;
 999         xenvif_disconnect_ctrl(be->vif);
1000         return;
1001 }
1002
1003
1004 static int connect_data_rings(struct backend_info *be,
1005                               struct xenvif_queue *queue)
1006 {
1007         struct xenbus_device *dev = be->dev;
1008         unsigned int num_queues = queue->vif->num_queues;
1009         unsigned long tx_ring_ref, rx_ring_ref;
1010         unsigned int tx_evtchn, rx_evtchn;
1011         int err;
1012         char *xspath;
1013         size_t xspathsize;
1014         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1015
1016         /* If the frontend requested 1 queue, or we have fallen back
1017          * to single queue due to lack of frontend support for multi-
1018          * queue, expect the remaining XenStore keys in the toplevel
1019          * directory. Otherwise, expect them in a subdirectory called
1020          * queue-N.
1021          */
1022         if (num_queues == 1) {
1023                 xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL);
1024                 if (!xspath) {
1025                         xenbus_dev_fatal(dev, -ENOMEM,
1026                                          "reading ring references");
1027                         return -ENOMEM;
1028                 }
1029                 strcpy(xspath, dev->otherend);
1030         } else {
1031                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1032                 xspath = kzalloc(xspathsize, GFP_KERNEL);
1033                 if (!xspath) {
1034                         xenbus_dev_fatal(dev, -ENOMEM,
1035                                          "reading ring references");
1036                         return -ENOMEM;
1037                 }
1038                 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend,
1039                          queue->id);
1040         }
1041
1042         err = xenbus_gather(XBT_NIL, xspath,
1043                             "tx-ring-ref", "%lu", &tx_ring_ref,
1044                             "rx-ring-ref", "%lu", &rx_ring_ref, NULL);
1045         if (err) {
1046                 xenbus_dev_fatal(dev, err,
1047                                  "reading %s/ring-ref",
1048                                  xspath);
1049                 goto err;
1050         }
1051
1052         /* Try split event channels first, then single event channel. */
1053         err = xenbus_gather(XBT_NIL, xspath,
1054                             "event-channel-tx", "%u", &tx_evtchn,
1055                             "event-channel-rx", "%u", &rx_evtchn, NULL);
1056         if (err < 0) {
1057                 err = xenbus_scanf(XBT_NIL, xspath,
1058                                    "event-channel", "%u", &tx_evtchn);
1059                 if (err < 0) {
1060                         xenbus_dev_fatal(dev, err,
1061                                          "reading %s/event-channel(-tx/rx)",
1062                                          xspath);
1063                         goto err;
1064                 }
1065                 rx_evtchn = tx_evtchn;
1066         }
1067
1068         /* Map the shared frame, irq etc. */
1069         err = xenvif_connect_data(queue, tx_ring_ref, rx_ring_ref,
1070                                   tx_evtchn, rx_evtchn);
1071         if (err) {
1072                 xenbus_dev_fatal(dev, err,
1073                                  "mapping shared-frames %lu/%lu port tx %u rx %u",
1074                                  tx_ring_ref, rx_ring_ref,
1075                                  tx_evtchn, rx_evtchn);
1076                 goto err;
1077         }
1078
1079         err = 0;
1080 err: /* Regular return falls through with err == 0 */
1081         kfree(xspath);
1082         return err;
1083 }
1084
1085 static int read_xenbus_vif_flags(struct backend_info *be)
1086 {
1087         struct xenvif *vif = be->vif;
1088         struct xenbus_device *dev = be->dev;
1089         unsigned int rx_copy;
1090         int err, val;
1091
1092         err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u",
1093                            &rx_copy);
1094         if (err == -ENOENT) {
1095                 err = 0;
1096                 rx_copy = 0;
1097         }
1098         if (err < 0) {
1099                 xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy",
1100                                  dev->otherend);
1101                 return err;
1102         }
1103         if (!rx_copy)
1104                 return -EOPNOTSUPP;
1105
1106         if (xenbus_scanf(XBT_NIL, dev->otherend,
1107                          "feature-rx-notify", "%d", &val) < 0)
1108                 val = 0;
1109         if (!val) {
1110                 /* - Reduce drain timeout to poll more frequently for
1111                  *   Rx requests.
1112                  * - Disable Rx stall detection.
1113                  */
1114                 be->vif->drain_timeout = msecs_to_jiffies(30);
1115                 be->vif->stall_timeout = 0;
1116         }
1117
1118         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-sg",
1119                          "%d", &val) < 0)
1120                 val = 0;
1121         vif->can_sg = !!val;
1122
1123         vif->gso_mask = 0;
1124         vif->gso_prefix_mask = 0;
1125
1126         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4",
1127                          "%d", &val) < 0)
1128                 val = 0;
1129         if (val)
1130                 vif->gso_mask |= GSO_BIT(TCPV4);
1131
1132         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4-prefix",
1133                          "%d", &val) < 0)
1134                 val = 0;
1135         if (val)
1136                 vif->gso_prefix_mask |= GSO_BIT(TCPV4);
1137
1138         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6",
1139                          "%d", &val) < 0)
1140                 val = 0;
1141         if (val)
1142                 vif->gso_mask |= GSO_BIT(TCPV6);
1143
1144         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6-prefix",
1145                          "%d", &val) < 0)
1146                 val = 0;
1147         if (val)
1148                 vif->gso_prefix_mask |= GSO_BIT(TCPV6);
1149
1150         if (vif->gso_mask & vif->gso_prefix_mask) {
1151                 xenbus_dev_fatal(dev, err,
1152                                  "%s: gso and gso prefix flags are not "
1153                                  "mutually exclusive",
1154                                  dev->otherend);
1155                 return -EOPNOTSUPP;
1156         }
1157
1158         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-no-csum-offload",
1159                          "%d", &val) < 0)
1160                 val = 0;
1161         vif->ip_csum = !val;
1162
1163         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-ipv6-csum-offload",
1164                          "%d", &val) < 0)
1165                 val = 0;
1166         vif->ipv6_csum = !!val;
1167
1168         return 0;
1169 }
1170
1171 static const struct xenbus_device_id netback_ids[] = {
1172         { "vif" },
1173         { "" }
1174 };
1175
1176 static struct xenbus_driver netback_driver = {
1177         .ids = netback_ids,
1178         .probe = netback_probe,
1179         .remove = netback_remove,
1180         .uevent = netback_uevent,
1181         .otherend_changed = frontend_changed,
1182 };
1183
1184 int xenvif_xenbus_init(void)
1185 {
1186         return xenbus_register_backend(&netback_driver);
1187 }
1188
1189 void xenvif_xenbus_fini(void)
1190 {
1191         return xenbus_unregister_driver(&netback_driver);
1192 }