drivers/net/xen-netback/xenbus.c

   1 /*
   2  * Xenbus code for netif backend
   3  *
   4  * Copyright (C) 2005 Rusty Russell <rusty@rustcorp.com.au>
   5  * Copyright (C) 2005 XenSource Ltd
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation; either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program; if not, see <http://www.gnu.org/licenses/>.
  19 */
  20
  21 #include "common.h"
  22 #include <linux/vmalloc.h>
  23 #include <linux/rtnetlink.h>
  24
  25 struct backend_info {
  26         struct xenbus_device *dev;
  27         struct xenvif *vif;
  28
  29         /* This is the state that will be reflected in xenstore when any
  30          * active hotplug script completes.
  31          */
  32         enum xenbus_state state;
  33
  34         enum xenbus_state frontend_state;
  35         struct xenbus_watch hotplug_status_watch;
  36         u8 have_hotplug_status_watch:1;
  37
  38         const char *hotplug_script;
  39 };
  40
  41 static int connect_data_rings(struct backend_info *be,
  42                               struct xenvif_queue *queue);
  43 static void connect(struct backend_info *be);
  44 static int read_xenbus_vif_flags(struct backend_info *be);
  45 static int backend_create_xenvif(struct backend_info *be);
  46 static void unregister_hotplug_status_watch(struct backend_info *be);
  47 static void xen_unregister_watchers(struct xenvif *vif);
  48 static void set_backend_state(struct backend_info *be,
  49                               enum xenbus_state state);
  50
  51 #ifdef CONFIG_DEBUG_FS
  52 struct dentry *xen_netback_dbg_root = NULL;
  53
  54 static int xenvif_read_io_ring(struct seq_file *m, void *v)
  55 {
  56         struct xenvif_queue *queue = m->private;
  57         struct xen_netif_tx_back_ring *tx_ring = &queue->tx;
  58         struct xen_netif_rx_back_ring *rx_ring = &queue->rx;
  59         struct netdev_queue *dev_queue;
  60
  61         if (tx_ring->sring) {
  62                 struct xen_netif_tx_sring *sring = tx_ring->sring;
  63
  64                 seq_printf(m, "Queue %d\nTX: nr_ents %u\n", queue->id,
  65                            tx_ring->nr_ents);
  66                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  67                            sring->req_prod,
  68                            sring->req_prod - sring->rsp_prod,
  69                            tx_ring->req_cons,
  70                            tx_ring->req_cons - sring->rsp_prod,
  71                            sring->req_event,
  72                            sring->req_event - sring->rsp_prod);
  73                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n",
  74                            sring->rsp_prod,
  75                            tx_ring->rsp_prod_pvt,
  76                            tx_ring->rsp_prod_pvt - sring->rsp_prod,
  77                            sring->rsp_event,
  78                            sring->rsp_event - sring->rsp_prod);
  79                 seq_printf(m, "pending prod %u pending cons %u nr_pending_reqs %u\n",
  80                            queue->pending_prod,
  81                            queue->pending_cons,
  82                            nr_pending_reqs(queue));
  83                 seq_printf(m, "dealloc prod %u dealloc cons %u dealloc_queue %u\n\n",
  84                            queue->dealloc_prod,
  85                            queue->dealloc_cons,
  86                            queue->dealloc_prod - queue->dealloc_cons);
  87         }
  88
  89         if (rx_ring->sring) {
  90                 struct xen_netif_rx_sring *sring = rx_ring->sring;
  91
  92                 seq_printf(m, "RX: nr_ents %u\n", rx_ring->nr_ents);
  93                 seq_printf(m, "req prod %u (%d) cons %u (%d) event %u (%d)\n",
  94                            sring->req_prod,
  95                            sring->req_prod - sring->rsp_prod,
  96                            rx_ring->req_cons,
  97                            rx_ring->req_cons - sring->rsp_prod,
  98                            sring->req_event,
  99                            sring->req_event - sring->rsp_prod);
 100                 seq_printf(m, "rsp prod %u (base) pvt %u (%d) event %u (%d)\n\n",
 101                            sring->rsp_prod,
 102                            rx_ring->rsp_prod_pvt,
 103                            rx_ring->rsp_prod_pvt - sring->rsp_prod,
 104                            sring->rsp_event,
 105                            sring->rsp_event - sring->rsp_prod);
 106         }
 107
 108         seq_printf(m, "NAPI state: %lx NAPI weight: %d TX queue len %u\n"
 109                    "Credit timer_pending: %d, credit: %lu, usec: %lu\n"
 110                    "remaining: %lu, expires: %lu, now: %lu\n",
 111                    queue->napi.state, queue->napi.weight,
 112                    skb_queue_len(&queue->tx_queue),
 113                    timer_pending(&queue->credit_timeout),
 114                    queue->credit_bytes,
 115                    queue->credit_usec,
 116                    queue->remaining_credit,
 117                    queue->credit_timeout.expires,
 118                    jiffies);
 119
 120         dev_queue = netdev_get_tx_queue(queue->vif->dev, queue->id);
 121
 122         seq_printf(m, "\nRx internal queue: len %u max %u pkts %u %s\n",
 123                    queue->rx_queue_len, queue->rx_queue_max,
 124                    skb_queue_len(&queue->rx_queue),
 125                    netif_tx_queue_stopped(dev_queue) ? "stopped" : "running");
 126
 127         return 0;
 128 }
 129
 130 #define XENVIF_KICK_STR "kick"
 131 #define BUFFER_SIZE     32
 132
 133 static ssize_t
 134 xenvif_write_io_ring(struct file *filp, const char __user *buf, size_t count,
 135                      loff_t *ppos)
 136 {
 137         struct xenvif_queue *queue =
 138                 ((struct seq_file *)filp->private_data)->private;
 139         int len;
 140         char write[BUFFER_SIZE];
 141
 142         /* don't allow partial writes and check the length */
 143         if (*ppos != 0)
 144                 return 0;
 145         if (count >= sizeof(write))
 146                 return -ENOSPC;
 147
 148         len = simple_write_to_buffer(write,
 149                                      sizeof(write) - 1,
 150                                      ppos,
 151                                      buf,
 152                                      count);
 153         if (len < 0)
 154                 return len;
 155
 156         write[len] = '\0';
 157
 158         if (!strncmp(write, XENVIF_KICK_STR, sizeof(XENVIF_KICK_STR) - 1))
 159                 xenvif_interrupt(0, (void *)queue);
 160         else {
 161                 pr_warn("Unknown command to io_ring_q%d. Available: kick\n",
 162                         queue->id);
 163                 count = -EINVAL;
 164         }
 165         return count;
 166 }
 167
 168 static int xenvif_io_ring_open(struct inode *inode, struct file *filp)
 169 {
 170         int ret;
 171         void *queue = NULL;
 172
 173         if (inode->i_private)
 174                 queue = inode->i_private;
 175         ret = single_open(filp, xenvif_read_io_ring, queue);
 176         filp->f_mode |= FMODE_PWRITE;
 177         return ret;
 178 }
 179
 180 static const struct file_operations xenvif_dbg_io_ring_ops_fops = {
 181         .owner = THIS_MODULE,
 182         .open = xenvif_io_ring_open,
 183         .read = seq_read,
 184         .llseek = seq_lseek,
 185         .release = single_release,
 186         .write = xenvif_write_io_ring,
 187 };
 188
 189 static int xenvif_read_ctrl(struct seq_file *m, void *v)
 190 {
 191         struct xenvif *vif = m->private;
 192
 193         xenvif_dump_hash_info(vif, m);
 194
 195         return 0;
 196 }
 197
 198 static int xenvif_ctrl_open(struct inode *inode, struct file *filp)
 199 {
 200         return single_open(filp, xenvif_read_ctrl, inode->i_private);
 201 }
 202
 203 static const struct file_operations xenvif_dbg_ctrl_ops_fops = {
 204         .owner = THIS_MODULE,
 205         .open = xenvif_ctrl_open,
 206         .read = seq_read,
 207         .llseek = seq_lseek,
 208         .release = single_release,
 209 };
 210
 211 static void xenvif_debugfs_addif(struct xenvif *vif)
 212 {
 213         struct dentry *pfile;
 214         int i;
 215
 216         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 217                 return;
 218
 219         vif->xenvif_dbg_root = debugfs_create_dir(vif->dev->name,
 220                                                   xen_netback_dbg_root);
 221         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root)) {
 222                 for (i = 0; i < vif->num_queues; ++i) {
 223                         char filename[sizeof("io_ring_q") + 4];
 224
 225                         snprintf(filename, sizeof(filename), "io_ring_q%d", i);
 226                         pfile = debugfs_create_file(filename,
 227                                                     S_IRUSR | S_IWUSR,
 228                                                     vif->xenvif_dbg_root,
 229                                                     &vif->queues[i],
 230                                                     &xenvif_dbg_io_ring_ops_fops);
 231                         if (IS_ERR_OR_NULL(pfile))
 232                                 pr_warn("Creation of io_ring file returned %ld!\n",
 233                                         PTR_ERR(pfile));
 234                 }
 235
 236                 if (vif->ctrl_task) {
 237                         pfile = debugfs_create_file("ctrl",
 238                                                     S_IRUSR,
 239                                                     vif->xenvif_dbg_root,
 240                                                     vif,
 241                                                     &xenvif_dbg_ctrl_ops_fops);
 242                         if (IS_ERR_OR_NULL(pfile))
 243                                 pr_warn("Creation of ctrl file returned %ld!\n",
 244                                         PTR_ERR(pfile));
 245                 }
 246         } else
 247                 netdev_warn(vif->dev,
 248                             "Creation of vif debugfs dir returned %ld!\n",
 249                             PTR_ERR(vif->xenvif_dbg_root));
 250 }
 251
 252 static void xenvif_debugfs_delif(struct xenvif *vif)
 253 {
 254         if (IS_ERR_OR_NULL(xen_netback_dbg_root))
 255                 return;
 256
 257         if (!IS_ERR_OR_NULL(vif->xenvif_dbg_root))
 258                 debugfs_remove_recursive(vif->xenvif_dbg_root);
 259         vif->xenvif_dbg_root = NULL;
 260 }
 261 #endif /* CONFIG_DEBUG_FS */
 262
 263 static int netback_remove(struct xenbus_device *dev)
 264 {
 265         struct backend_info *be = dev_get_drvdata(&dev->dev);
 266
 267         set_backend_state(be, XenbusStateClosed);
 268
 269         unregister_hotplug_status_watch(be);
 270         if (be->vif) {
 271                 kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE);
 272                 xen_unregister_watchers(be->vif);
 273                 xenbus_rm(XBT_NIL, dev->nodename, "hotplug-status");
 274                 xenvif_free(be->vif);
 275                 be->vif = NULL;
 276         }
 277         kfree(be->hotplug_script);
 278         kfree(be);
 279         dev_set_drvdata(&dev->dev, NULL);
 280         return 0;
 281 }
 282
 283
 284 /**
 285  * Entry point to this code when a new device is created.  Allocate the basic
 286  * structures and switch to InitWait.
 287  */
 288 static int netback_probe(struct xenbus_device *dev,
 289                          const struct xenbus_device_id *id)
 290 {
 291         const char *message;
 292         struct xenbus_transaction xbt;
 293         int err;
 294         int sg;
 295         const char *script;
 296         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 297                                           GFP_KERNEL);
 298         if (!be) {
 299                 xenbus_dev_fatal(dev, -ENOMEM,
 300                                  "allocating backend structure");
 301                 return -ENOMEM;
 302         }
 303
 304         be->dev = dev;
 305         dev_set_drvdata(&dev->dev, be);
 306
 307         sg = 1;
 308
 309         do {
 310                 err = xenbus_transaction_start(&xbt);
 311                 if (err) {
 312                         xenbus_dev_fatal(dev, err, "starting transaction");
 313                         goto fail;
 314                 }
 315
 316                 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg);
 317                 if (err) {
 318                         message = "writing feature-sg";
 319                         goto abort_transaction;
 320                 }
 321
 322                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4",
 323                                     "%d", sg);
 324                 if (err) {
 325                         message = "writing feature-gso-tcpv4";
 326                         goto abort_transaction;
 327                 }
 328
 329                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6",
 330                                     "%d", sg);
 331                 if (err) {
 332                         message = "writing feature-gso-tcpv6";
 333                         goto abort_transaction;
 334                 }
 335
 336                 /* We support partial checksum setup for IPv6 packets */
 337                 err = xenbus_printf(xbt, dev->nodename,
 338                                     "feature-ipv6-csum-offload",
 339                                     "%d", 1);
 340                 if (err) {
 341                         message = "writing feature-ipv6-csum-offload";
 342                         goto abort_transaction;
 343                 }
 344
 345                 /* We support rx-copy path. */
 346                 err = xenbus_printf(xbt, dev->nodename,
 347                                     "feature-rx-copy", "%d", 1);
 348                 if (err) {
 349                         message = "writing feature-rx-copy";
 350                         goto abort_transaction;
 351                 }
 352
 353                 /*
 354                  * We don't support rx-flip path (except old guests who don't
 355                  * grok this feature flag).
 356                  */
 357                 err = xenbus_printf(xbt, dev->nodename,
 358                                     "feature-rx-flip", "%d", 0);
 359                 if (err) {
 360                         message = "writing feature-rx-flip";
 361                         goto abort_transaction;
 362                 }
 363
 364                 /* We support dynamic multicast-control. */
 365                 err = xenbus_printf(xbt, dev->nodename,
 366                                     "feature-multicast-control", "%d", 1);
 367                 if (err) {
 368                         message = "writing feature-multicast-control";
 369                         goto abort_transaction;
 370                 }
 371
 372                 err = xenbus_printf(xbt, dev->nodename,
 373                                     "feature-dynamic-multicast-control",
 374                                     "%d", 1);
 375                 if (err) {
 376                         message = "writing feature-dynamic-multicast-control";
 377                         goto abort_transaction;
 378                 }
 379
 380                 err = xenbus_transaction_end(xbt, 0);
 381         } while (err == -EAGAIN);
 382
 383         if (err) {
 384                 xenbus_dev_fatal(dev, err, "completing transaction");
 385                 goto fail;
 386         }
 387
 388         /*
 389          * Split event channels support, this is optional so it is not
 390          * put inside the above loop.
 391          */
 392         err = xenbus_printf(XBT_NIL, dev->nodename,
 393                             "feature-split-event-channels",
 394                             "%u", separate_tx_rx_irq);
 395         if (err)
 396                 pr_debug("Error writing feature-split-event-channels\n");
 397
 398         /* Multi-queue support: This is an optional feature. */
 399         err = xenbus_printf(XBT_NIL, dev->nodename,
 400                             "multi-queue-max-queues", "%u", xenvif_max_queues);
 401         if (err)
 402                 pr_debug("Error writing multi-queue-max-queues\n");
 403
 404         err = xenbus_printf(XBT_NIL, dev->nodename,
 405                             "feature-ctrl-ring",
 406                             "%u", true);
 407         if (err)
 408                 pr_debug("Error writing feature-ctrl-ring\n");
 409
 410         script = xenbus_read(XBT_NIL, dev->nodename, "script", NULL);
 411         if (IS_ERR(script)) {
 412                 err = PTR_ERR(script);
 413                 xenbus_dev_fatal(dev, err, "reading script");
 414                 goto fail;
 415         }
 416
 417         be->hotplug_script = script;
 418
 419         err = xenbus_switch_state(dev, XenbusStateInitWait);
 420         if (err)
 421                 goto fail;
 422
 423         be->state = XenbusStateInitWait;
 424
 425         /* This kicks hotplug scripts, so do it immediately. */
 426         err = backend_create_xenvif(be);
 427         if (err)
 428                 goto fail;
 429
 430         return 0;
 431
 432 abort_transaction:
 433         xenbus_transaction_end(xbt, 1);
 434         xenbus_dev_fatal(dev, err, "%s", message);
 435 fail:
 436         pr_debug("failed\n");
 437         netback_remove(dev);
 438         return err;
 439 }
 440
 441
 442 /*
 443  * Handle the creation of the hotplug script environment.  We add the script
 444  * and vif variables to the environment, for the benefit of the vif-* hotplug
 445  * scripts.
 446  */
 447 static int netback_uevent(struct xenbus_device *xdev,
 448                           struct kobj_uevent_env *env)
 449 {
 450         struct backend_info *be = dev_get_drvdata(&xdev->dev);
 451
 452         if (!be)
 453                 return 0;
 454
 455         if (add_uevent_var(env, "script=%s", be->hotplug_script))
 456                 return -ENOMEM;
 457
 458         if (!be->vif)
 459                 return 0;
 460
 461         return add_uevent_var(env, "vif=%s", be->vif->dev->name);
 462 }
 463
 464
 465 static int backend_create_xenvif(struct backend_info *be)
 466 {
 467         int err;
 468         long handle;
 469         struct xenbus_device *dev = be->dev;
 470         struct xenvif *vif;
 471
 472         if (be->vif != NULL)
 473                 return 0;
 474
 475         err = xenbus_scanf(XBT_NIL, dev->nodename, "handle", "%li", &handle);
 476         if (err != 1) {
 477                 xenbus_dev_fatal(dev, err, "reading handle");
 478                 return (err < 0) ? err : -EINVAL;
 479         }
 480
 481         vif = xenvif_alloc(&dev->dev, dev->otherend_id, handle);
 482         if (IS_ERR(vif)) {
 483                 err = PTR_ERR(vif);
 484                 xenbus_dev_fatal(dev, err, "creating interface");
 485                 return err;
 486         }
 487         be->vif = vif;
 488
 489         kobject_uevent(&dev->dev.kobj, KOBJ_ONLINE);
 490         return 0;
 491 }
 492
 493 static void backend_disconnect(struct backend_info *be)
 494 {
 495         if (be->vif) {
 496                 xen_unregister_watchers(be->vif);
 497 #ifdef CONFIG_DEBUG_FS
 498                 xenvif_debugfs_delif(be->vif);
 499 #endif /* CONFIG_DEBUG_FS */
 500                 xenvif_disconnect_data(be->vif);
 501                 xenvif_disconnect_ctrl(be->vif);
 502         }
 503 }
 504
 505 static void backend_connect(struct backend_info *be)
 506 {
 507         if (be->vif)
 508                 connect(be);
 509 }
 510
 511 static inline void backend_switch_state(struct backend_info *be,
 512                                         enum xenbus_state state)
 513 {
 514         struct xenbus_device *dev = be->dev;
 515
 516         pr_debug("%s -> %s\n", dev->nodename, xenbus_strstate(state));
 517         be->state = state;
 518
 519         /* If we are waiting for a hotplug script then defer the
 520          * actual xenbus state change.
 521          */
 522         if (!be->have_hotplug_status_watch)
 523                 xenbus_switch_state(dev, state);
 524 }
 525
 526 /* Handle backend state transitions:
 527  *
 528  * The backend state starts in InitWait and the following transitions are
 529  * allowed.
 530  *
 531  * InitWait -> Connected
 532  *
 533  *    ^    \         |
 534  *    |     \        |
 535  *    |      \       |
 536  *    |       \      |
 537  *    |        \     |
 538  *    |         \    |
 539  *    |          V   V
 540  *
 541  *  Closed  <-> Closing
 542  *
 543  * The state argument specifies the eventual state of the backend and the
 544  * function transitions to that state via the shortest path.
 545  */
 546 static void set_backend_state(struct backend_info *be,
 547                               enum xenbus_state state)
 548 {
 549         while (be->state != state) {
 550                 switch (be->state) {
 551                 case XenbusStateClosed:
 552                         switch (state) {
 553                         case XenbusStateInitWait:
 554                         case XenbusStateConnected:
 555                                 backend_switch_state(be, XenbusStateInitWait);
 556                                 break;
 557                         case XenbusStateClosing:
 558                                 backend_switch_state(be, XenbusStateClosing);
 559                                 break;
 560                         default:
 561                                 BUG();
 562                         }
 563                         break;
 564                 case XenbusStateInitWait:
 565                         switch (state) {
 566                         case XenbusStateConnected:
 567                                 backend_connect(be);
 568                                 backend_switch_state(be, XenbusStateConnected);
 569                                 break;
 570                         case XenbusStateClosing:
 571                         case XenbusStateClosed:
 572                                 backend_switch_state(be, XenbusStateClosing);
 573                                 break;
 574                         default:
 575                                 BUG();
 576                         }
 577                         break;
 578                 case XenbusStateConnected:
 579                         switch (state) {
 580                         case XenbusStateInitWait:
 581                         case XenbusStateClosing:
 582                         case XenbusStateClosed:
 583                                 backend_disconnect(be);
 584                                 backend_switch_state(be, XenbusStateClosing);
 585                                 break;
 586                         default:
 587                                 BUG();
 588                         }
 589                         break;
 590                 case XenbusStateClosing:
 591                         switch (state) {
 592                         case XenbusStateInitWait:
 593                         case XenbusStateConnected:
 594                         case XenbusStateClosed:
 595                                 backend_switch_state(be, XenbusStateClosed);
 596                                 break;
 597                         default:
 598                                 BUG();
 599                         }
 600                         break;
 601                 default:
 602                         BUG();
 603                 }
 604         }
 605 }
 606
 607 /**
 608  * Callback received when the frontend's state changes.
 609  */
 610 static void frontend_changed(struct xenbus_device *dev,
 611                              enum xenbus_state frontend_state)
 612 {
 613         struct backend_info *be = dev_get_drvdata(&dev->dev);
 614
 615         pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state));
 616
 617         be->frontend_state = frontend_state;
 618
 619         switch (frontend_state) {
 620         case XenbusStateInitialising:
 621                 set_backend_state(be, XenbusStateInitWait);
 622                 break;
 623
 624         case XenbusStateInitialised:
 625                 break;
 626
 627         case XenbusStateConnected:
 628                 set_backend_state(be, XenbusStateConnected);
 629                 break;
 630
 631         case XenbusStateClosing:
 632                 set_backend_state(be, XenbusStateClosing);
 633                 break;
 634
 635         case XenbusStateClosed:
 636                 set_backend_state(be, XenbusStateClosed);
 637                 if (xenbus_dev_is_online(dev))
 638                         break;
 639                 /* fall through if not online */
 640         case XenbusStateUnknown:
 641                 set_backend_state(be, XenbusStateClosed);
 642                 device_unregister(&dev->dev);
 643                 break;
 644
 645         default:
 646                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 647                                  frontend_state);
 648                 break;
 649         }
 650 }
 651
 652
 653 static void xen_net_read_rate(struct xenbus_device *dev,
 654                               unsigned long *bytes, unsigned long *usec)
 655 {
 656         char *s, *e;
 657         unsigned long b, u;
 658         char *ratestr;
 659
 660         /* Default to unlimited bandwidth. */
 661         *bytes = ~0UL;
 662         *usec = 0;
 663
 664         ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL);
 665         if (IS_ERR(ratestr))
 666                 return;
 667
 668         s = ratestr;
 669         b = simple_strtoul(s, &e, 10);
 670         if ((s == e) || (*e != ','))
 671                 goto fail;
 672
 673         s = e + 1;
 674         u = simple_strtoul(s, &e, 10);
 675         if ((s == e) || (*e != '\0'))
 676                 goto fail;
 677
 678         *bytes = b;
 679         *usec = u;
 680
 681         kfree(ratestr);
 682         return;
 683
 684  fail:
 685         pr_warn("Failed to parse network rate limit. Traffic unlimited.\n");
 686         kfree(ratestr);
 687 }
 688
 689 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
 690 {
 691         char *s, *e, *macstr;
 692         int i;
 693
 694         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
 695         if (IS_ERR(macstr))
 696                 return PTR_ERR(macstr);
 697
 698         for (i = 0; i < ETH_ALEN; i++) {
 699                 mac[i] = simple_strtoul(s, &e, 16);
 700                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
 701                         kfree(macstr);
 702                         return -ENOENT;
 703                 }
 704                 s = e+1;
 705         }
 706
 707         kfree(macstr);
 708         return 0;
 709 }
 710
 711 static void xen_net_rate_changed(struct xenbus_watch *watch,
 712                                 const char **vec, unsigned int len)
 713 {
 714         struct xenvif *vif = container_of(watch, struct xenvif, credit_watch);
 715         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 716         unsigned long   credit_bytes;
 717         unsigned long   credit_usec;
 718         unsigned int queue_index;
 719
 720         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 721         for (queue_index = 0; queue_index < vif->num_queues; queue_index++) {
 722                 struct xenvif_queue *queue = &vif->queues[queue_index];
 723
 724                 queue->credit_bytes = credit_bytes;
 725                 queue->credit_usec = credit_usec;
 726                 if (!mod_timer_pending(&queue->credit_timeout, jiffies) &&
 727                         queue->remaining_credit > queue->credit_bytes) {
 728                         queue->remaining_credit = queue->credit_bytes;
 729                 }
 730         }
 731 }
 732
 733 static int xen_register_credit_watch(struct xenbus_device *dev,
 734                                      struct xenvif *vif)
 735 {
 736         int err = 0;
 737         char *node;
 738         unsigned maxlen = strlen(dev->nodename) + sizeof("/rate");
 739
 740         if (vif->credit_watch.node)
 741                 return -EADDRINUSE;
 742
 743         node = kmalloc(maxlen, GFP_KERNEL);
 744         if (!node)
 745                 return -ENOMEM;
 746         snprintf(node, maxlen, "%s/rate", dev->nodename);
 747         vif->credit_watch.node = node;
 748         vif->credit_watch.callback = xen_net_rate_changed;
 749         err = register_xenbus_watch(&vif->credit_watch);
 750         if (err) {
 751                 pr_err("Failed to set watcher %s\n", vif->credit_watch.node);
 752                 kfree(node);
 753                 vif->credit_watch.node = NULL;
 754                 vif->credit_watch.callback = NULL;
 755         }
 756         return err;
 757 }
 758
 759 static void xen_unregister_credit_watch(struct xenvif *vif)
 760 {
 761         if (vif->credit_watch.node) {
 762                 unregister_xenbus_watch(&vif->credit_watch);
 763                 kfree(vif->credit_watch.node);
 764                 vif->credit_watch.node = NULL;
 765         }
 766 }
 767
 768 static void xen_mcast_ctrl_changed(struct xenbus_watch *watch,
 769                                    const char **vec, unsigned int len)
 770 {
 771         struct xenvif *vif = container_of(watch, struct xenvif,
 772                                           mcast_ctrl_watch);
 773         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 774         int val;
 775
 776         if (xenbus_scanf(XBT_NIL, dev->otherend,
 777                          "request-multicast-control", "%d", &val) < 0)
 778                 val = 0;
 779         vif->multicast_control = !!val;
 780 }
 781
 782 static int xen_register_mcast_ctrl_watch(struct xenbus_device *dev,
 783                                          struct xenvif *vif)
 784 {
 785         int err = 0;
 786         char *node;
 787         unsigned maxlen = strlen(dev->otherend) +
 788                 sizeof("/request-multicast-control");
 789
 790         if (vif->mcast_ctrl_watch.node) {
 791                 pr_err_ratelimited("Watch is already registered\n");
 792                 return -EADDRINUSE;
 793         }
 794
 795         node = kmalloc(maxlen, GFP_KERNEL);
 796         if (!node) {
 797                 pr_err("Failed to allocate memory for watch\n");
 798                 return -ENOMEM;
 799         }
 800         snprintf(node, maxlen, "%s/request-multicast-control",
 801                  dev->otherend);
 802         vif->mcast_ctrl_watch.node = node;
 803         vif->mcast_ctrl_watch.callback = xen_mcast_ctrl_changed;
 804         err = register_xenbus_watch(&vif->mcast_ctrl_watch);
 805         if (err) {
 806                 pr_err("Failed to set watcher %s\n",
 807                        vif->mcast_ctrl_watch.node);
 808                 kfree(node);
 809                 vif->mcast_ctrl_watch.node = NULL;
 810                 vif->mcast_ctrl_watch.callback = NULL;
 811         }
 812         return err;
 813 }
 814
 815 static void xen_unregister_mcast_ctrl_watch(struct xenvif *vif)
 816 {
 817         if (vif->mcast_ctrl_watch.node) {
 818                 unregister_xenbus_watch(&vif->mcast_ctrl_watch);
 819                 kfree(vif->mcast_ctrl_watch.node);
 820                 vif->mcast_ctrl_watch.node = NULL;
 821         }
 822 }
 823
 824 static void xen_register_watchers(struct xenbus_device *dev,
 825                                   struct xenvif *vif)
 826 {
 827         xen_register_credit_watch(dev, vif);
 828         xen_register_mcast_ctrl_watch(dev, vif);
 829 }
 830
 831 static void xen_unregister_watchers(struct xenvif *vif)
 832 {
 833         xen_unregister_mcast_ctrl_watch(vif);
 834         xen_unregister_credit_watch(vif);
 835 }
 836
 837 static void unregister_hotplug_status_watch(struct backend_info *be)
 838 {
 839         if (be->have_hotplug_status_watch) {
 840                 unregister_xenbus_watch(&be->hotplug_status_watch);
 841                 kfree(be->hotplug_status_watch.node);
 842         }
 843         be->have_hotplug_status_watch = 0;
 844 }
 845
 846 static void hotplug_status_changed(struct xenbus_watch *watch,
 847                                    const char **vec,
 848                                    unsigned int vec_size)
 849 {
 850         struct backend_info *be = container_of(watch,
 851                                                struct backend_info,
 852                                                hotplug_status_watch);
 853         char *str;
 854         unsigned int len;
 855
 856         str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len);
 857         if (IS_ERR(str))
 858                 return;
 859         if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) {
 860                 /* Complete any pending state change */
 861                 xenbus_switch_state(be->dev, be->state);
 862
 863                 /* Not interested in this watch anymore. */
 864                 unregister_hotplug_status_watch(be);
 865         }
 866         kfree(str);
 867 }
 868
 869 static int connect_ctrl_ring(struct backend_info *be)
 870 {
 871         struct xenbus_device *dev = be->dev;
 872         struct xenvif *vif = be->vif;
 873         unsigned int val;
 874         grant_ref_t ring_ref;
 875         unsigned int evtchn;
 876         int err;
 877
 878         err = xenbus_gather(XBT_NIL, dev->otherend,
 879                             "ctrl-ring-ref", "%u", &val, NULL);
 880         if (err)
 881                 goto done; /* The frontend does not have a control ring */
 882
 883         ring_ref = val;
 884
 885         err = xenbus_gather(XBT_NIL, dev->otherend,
 886                             "event-channel-ctrl", "%u", &val, NULL);
 887         if (err) {
 888                 xenbus_dev_fatal(dev, err,
 889                                  "reading %s/event-channel-ctrl",
 890                                  dev->otherend);
 891                 goto fail;
 892         }
 893
 894         evtchn = val;
 895
 896         err = xenvif_connect_ctrl(vif, ring_ref, evtchn);
 897         if (err) {
 898                 xenbus_dev_fatal(dev, err,
 899                                  "mapping shared-frame %u port %u",
 900                                  ring_ref, evtchn);
 901                 goto fail;
 902         }
 903
 904 done:
 905         return 0;
 906
 907 fail:
 908         return err;
 909 }
 910
 911 static void connect(struct backend_info *be)
 912 {
 913         int err;
 914         struct xenbus_device *dev = be->dev;
 915         unsigned long credit_bytes, credit_usec;
 916         unsigned int queue_index;
 917         unsigned int requested_num_queues;
 918         struct xenvif_queue *queue;
 919
 920         /* Check whether the frontend requested multiple queues
 921          * and read the number requested.
 922          */
 923         err = xenbus_scanf(XBT_NIL, dev->otherend,
 924                            "multi-queue-num-queues",
 925                            "%u", &requested_num_queues);
 926         if (err < 0) {
 927                 requested_num_queues = 1; /* Fall back to single queue */
 928         } else if (requested_num_queues > xenvif_max_queues) {
 929                 /* buggy or malicious guest */
 930                 xenbus_dev_fatal(dev, err,
 931                                  "guest requested %u queues, exceeding the maximum of %u.",
 932                                  requested_num_queues, xenvif_max_queues);
 933                 return;
 934         }
 935
 936         err = xen_net_read_mac(dev, be->vif->fe_dev_addr);
 937         if (err) {
 938                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
 939                 return;
 940         }
 941
 942         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 943         xen_unregister_watchers(be->vif);
 944         xen_register_watchers(dev, be->vif);
 945         read_xenbus_vif_flags(be);
 946
 947         err = connect_ctrl_ring(be);
 948         if (err) {
 949                 xenbus_dev_fatal(dev, err, "connecting control ring");
 950                 return;
 951         }
 952
 953         /* Use the number of queues requested by the frontend */
 954         be->vif->queues = vzalloc(requested_num_queues *
 955                                   sizeof(struct xenvif_queue));
 956         if (!be->vif->queues) {
 957                 xenbus_dev_fatal(dev, -ENOMEM,
 958                                  "allocating queues");
 959                 return;
 960         }
 961
 962         be->vif->num_queues = requested_num_queues;
 963         be->vif->stalled_queues = requested_num_queues;
 964
 965         for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) {
 966                 queue = &be->vif->queues[queue_index];
 967                 queue->vif = be->vif;
 968                 queue->id = queue_index;
 969                 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
 970                                 be->vif->dev->name, queue->id);
 971
 972                 err = xenvif_init_queue(queue);
 973                 if (err) {
 974                         /* xenvif_init_queue() cleans up after itself on
 975                          * failure, but we need to clean up any previously
 976                          * initialised queues. Set num_queues to i so that
 977                          * earlier queues can be destroyed using the regular
 978                          * disconnect logic.
 979                          */
 980                         be->vif->num_queues = queue_index;
 981                         goto err;
 982                 }
 983
 984                 queue->credit_bytes = credit_bytes;
 985                 queue->remaining_credit = credit_bytes;
 986                 queue->credit_usec = credit_usec;
 987
 988                 err = connect_data_rings(be, queue);
 989                 if (err) {
 990                         /* connect_data_rings() cleans up after itself on
 991                          * failure, but we need to clean up after
 992                          * xenvif_init_queue() here, and also clean up any
 993                          * previously initialised queues.
 994                          */
 995                         xenvif_deinit_queue(queue);
 996                         be->vif->num_queues = queue_index;
 997                         goto err;
 998                 }
 999         }
1000
1001 #ifdef CONFIG_DEBUG_FS
1002         xenvif_debugfs_addif(be->vif);
1003 #endif /* CONFIG_DEBUG_FS */
1004
1005         /* Initialisation completed, tell core driver the number of
1006          * active queues.
1007          */
1008         rtnl_lock();
1009         netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues);
1010         netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues);
1011         rtnl_unlock();
1012
1013         xenvif_carrier_on(be->vif);
1014
1015         unregister_hotplug_status_watch(be);
1016         err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch,
1017                                    hotplug_status_changed,
1018                                    "%s/%s", dev->nodename, "hotplug-status");
1019         if (!err)
1020                 be->have_hotplug_status_watch = 1;
1021
1022         netif_tx_wake_all_queues(be->vif->dev);
1023
1024         return;
1025
1026 err:
1027         if (be->vif->num_queues > 0)
1028                 xenvif_disconnect_data(be->vif); /* Clean up existing queues */
1029         vfree(be->vif->queues);
1030         be->vif->queues = NULL;
1031         be->vif->num_queues = 0;
1032         xenvif_disconnect_ctrl(be->vif);
1033         return;
1034 }
1035
1036
1037 static int connect_data_rings(struct backend_info *be,
1038                               struct xenvif_queue *queue)
1039 {
1040         struct xenbus_device *dev = be->dev;
1041         unsigned int num_queues = queue->vif->num_queues;
1042         unsigned long tx_ring_ref, rx_ring_ref;
1043         unsigned int tx_evtchn, rx_evtchn;
1044         int err;
1045         char *xspath;
1046         size_t xspathsize;
1047         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1048
1049         /* If the frontend requested 1 queue, or we have fallen back
1050          * to single queue due to lack of frontend support for multi-
1051          * queue, expect the remaining XenStore keys in the toplevel
1052          * directory. Otherwise, expect them in a subdirectory called
1053          * queue-N.
1054          */
1055         if (num_queues == 1) {
1056                 xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL);
1057                 if (!xspath) {
1058                         xenbus_dev_fatal(dev, -ENOMEM,
1059                                          "reading ring references");
1060                         return -ENOMEM;
1061                 }
1062                 strcpy(xspath, dev->otherend);
1063         } else {
1064                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1065                 xspath = kzalloc(xspathsize, GFP_KERNEL);
1066                 if (!xspath) {
1067                         xenbus_dev_fatal(dev, -ENOMEM,
1068                                          "reading ring references");
1069                         return -ENOMEM;
1070                 }
1071                 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend,
1072                          queue->id);
1073         }
1074
1075         err = xenbus_gather(XBT_NIL, xspath,
1076                             "tx-ring-ref", "%lu", &tx_ring_ref,
1077                             "rx-ring-ref", "%lu", &rx_ring_ref, NULL);
1078         if (err) {
1079                 xenbus_dev_fatal(dev, err,
1080                                  "reading %s/ring-ref",
1081                                  xspath);
1082                 goto err;
1083         }
1084
1085         /* Try split event channels first, then single event channel. */
1086         err = xenbus_gather(XBT_NIL, xspath,
1087                             "event-channel-tx", "%u", &tx_evtchn,
1088                             "event-channel-rx", "%u", &rx_evtchn, NULL);
1089         if (err < 0) {
1090                 err = xenbus_scanf(XBT_NIL, xspath,
1091                                    "event-channel", "%u", &tx_evtchn);
1092                 if (err < 0) {
1093                         xenbus_dev_fatal(dev, err,
1094                                          "reading %s/event-channel(-tx/rx)",
1095                                          xspath);
1096                         goto err;
1097                 }
1098                 rx_evtchn = tx_evtchn;
1099         }
1100
1101         /* Map the shared frame, irq etc. */
1102         err = xenvif_connect_data(queue, tx_ring_ref, rx_ring_ref,
1103                                   tx_evtchn, rx_evtchn);
1104         if (err) {
1105                 xenbus_dev_fatal(dev, err,
1106                                  "mapping shared-frames %lu/%lu port tx %u rx %u",
1107                                  tx_ring_ref, rx_ring_ref,
1108                                  tx_evtchn, rx_evtchn);
1109                 goto err;
1110         }
1111
1112         err = 0;
1113 err: /* Regular return falls through with err == 0 */
1114         kfree(xspath);
1115         return err;
1116 }
1117
1118 static int read_xenbus_vif_flags(struct backend_info *be)
1119 {
1120         struct xenvif *vif = be->vif;
1121         struct xenbus_device *dev = be->dev;
1122         unsigned int rx_copy;
1123         int err, val;
1124
1125         err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u",
1126                            &rx_copy);
1127         if (err == -ENOENT) {
1128                 err = 0;
1129                 rx_copy = 0;
1130         }
1131         if (err < 0) {
1132                 xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy",
1133                                  dev->otherend);
1134                 return err;
1135         }
1136         if (!rx_copy)
1137                 return -EOPNOTSUPP;
1138
1139         if (xenbus_scanf(XBT_NIL, dev->otherend,
1140                          "feature-rx-notify", "%d", &val) < 0)
1141                 val = 0;
1142         if (!val) {
1143                 /* - Reduce drain timeout to poll more frequently for
1144                  *   Rx requests.
1145                  * - Disable Rx stall detection.
1146                  */
1147                 be->vif->drain_timeout = msecs_to_jiffies(30);
1148                 be->vif->stall_timeout = 0;
1149         }
1150
1151         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-sg",
1152                          "%d", &val) < 0)
1153                 val = 0;
1154         vif->can_sg = !!val;
1155
1156         vif->gso_mask = 0;
1157         vif->gso_prefix_mask = 0;
1158
1159         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4",
1160                          "%d", &val) < 0)
1161                 val = 0;
1162         if (val)
1163                 vif->gso_mask |= GSO_BIT(TCPV4);
1164
1165         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv4-prefix",
1166                          "%d", &val) < 0)
1167                 val = 0;
1168         if (val)
1169                 vif->gso_prefix_mask |= GSO_BIT(TCPV4);
1170
1171         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6",
1172                          "%d", &val) < 0)
1173                 val = 0;
1174         if (val)
1175                 vif->gso_mask |= GSO_BIT(TCPV6);
1176
1177         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-gso-tcpv6-prefix",
1178                          "%d", &val) < 0)
1179                 val = 0;
1180         if (val)
1181                 vif->gso_prefix_mask |= GSO_BIT(TCPV6);
1182
1183         if (vif->gso_mask & vif->gso_prefix_mask) {
1184                 xenbus_dev_fatal(dev, err,
1185                                  "%s: gso and gso prefix flags are not "
1186                                  "mutually exclusive",
1187                                  dev->otherend);
1188                 return -EOPNOTSUPP;
1189         }
1190
1191         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-no-csum-offload",
1192                          "%d", &val) < 0)
1193                 val = 0;
1194         vif->ip_csum = !val;
1195
1196         if (xenbus_scanf(XBT_NIL, dev->otherend, "feature-ipv6-csum-offload",
1197                          "%d", &val) < 0)
1198                 val = 0;
1199         vif->ipv6_csum = !!val;
1200
1201         return 0;
1202 }
1203
1204 static const struct xenbus_device_id netback_ids[] = {
1205         { "vif" },
1206         { "" }
1207 };
1208
1209 static struct xenbus_driver netback_driver = {
1210         .ids = netback_ids,
1211         .probe = netback_probe,
1212         .remove = netback_remove,
1213         .uevent = netback_uevent,
1214         .otherend_changed = frontend_changed,
1215 };
1216
1217 int xenvif_xenbus_init(void)
1218 {
1219         return xenbus_register_backend(&netback_driver);
1220 }
1221
1222 void xenvif_xenbus_fini(void)
1223 {
1224         return xenbus_unregister_driver(&netback_driver);
1225 }