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                                                     0600,
 228                                                     vif->xenvif_dbg_root,
 229                                                     &vif->queues[i],
 230                                                     &xenvif_dbg_io_ring_ops_fops);
 231                         if (IS_ERR_OR_NULL(pfile))
 232                                 pr_warn("Creation of io_ring file returned %ld!\n",
 233                                         PTR_ERR(pfile));
 234                 }
 235
 236                 if (vif->ctrl_irq) {
 237                         pfile = debugfs_create_file("ctrl",
 238                                                     0400,
 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         debugfs_remove_recursive(vif->xenvif_dbg_root);
 258         vif->xenvif_dbg_root = NULL;
 259 }
 260 #endif /* CONFIG_DEBUG_FS */
 261
 262 static int netback_remove(struct xenbus_device *dev)
 263 {
 264         struct backend_info *be = dev_get_drvdata(&dev->dev);
 265
 266         set_backend_state(be, XenbusStateClosed);
 267
 268         unregister_hotplug_status_watch(be);
 269         if (be->vif) {
 270                 kobject_uevent(&dev->dev.kobj, KOBJ_OFFLINE);
 271                 xen_unregister_watchers(be->vif);
 272                 xenbus_rm(XBT_NIL, dev->nodename, "hotplug-status");
 273                 xenvif_free(be->vif);
 274                 be->vif = NULL;
 275         }
 276         kfree(be->hotplug_script);
 277         kfree(be);
 278         dev_set_drvdata(&dev->dev, NULL);
 279         return 0;
 280 }
 281
 282
 283 /**
 284  * Entry point to this code when a new device is created.  Allocate the basic
 285  * structures and switch to InitWait.
 286  */
 287 static int netback_probe(struct xenbus_device *dev,
 288                          const struct xenbus_device_id *id)
 289 {
 290         const char *message;
 291         struct xenbus_transaction xbt;
 292         int err;
 293         int sg;
 294         const char *script;
 295         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 296                                           GFP_KERNEL);
 297         if (!be) {
 298                 xenbus_dev_fatal(dev, -ENOMEM,
 299                                  "allocating backend structure");
 300                 return -ENOMEM;
 301         }
 302
 303         be->dev = dev;
 304         dev_set_drvdata(&dev->dev, be);
 305
 306         be->state = XenbusStateInitialising;
 307         err = xenbus_switch_state(dev, XenbusStateInitialising);
 308         if (err)
 309                 goto fail;
 310
 311         sg = 1;
 312
 313         do {
 314                 err = xenbus_transaction_start(&xbt);
 315                 if (err) {
 316                         xenbus_dev_fatal(dev, err, "starting transaction");
 317                         goto fail;
 318                 }
 319
 320                 err = xenbus_printf(xbt, dev->nodename, "feature-sg", "%d", sg);
 321                 if (err) {
 322                         message = "writing feature-sg";
 323                         goto abort_transaction;
 324                 }
 325
 326                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv4",
 327                                     "%d", sg);
 328                 if (err) {
 329                         message = "writing feature-gso-tcpv4";
 330                         goto abort_transaction;
 331                 }
 332
 333                 err = xenbus_printf(xbt, dev->nodename, "feature-gso-tcpv6",
 334                                     "%d", sg);
 335                 if (err) {
 336                         message = "writing feature-gso-tcpv6";
 337                         goto abort_transaction;
 338                 }
 339
 340                 /* We support partial checksum setup for IPv6 packets */
 341                 err = xenbus_printf(xbt, dev->nodename,
 342                                     "feature-ipv6-csum-offload",
 343                                     "%d", 1);
 344                 if (err) {
 345                         message = "writing feature-ipv6-csum-offload";
 346                         goto abort_transaction;
 347                 }
 348
 349                 /* We support rx-copy path. */
 350                 err = xenbus_printf(xbt, dev->nodename,
 351                                     "feature-rx-copy", "%d", 1);
 352                 if (err) {
 353                         message = "writing feature-rx-copy";
 354                         goto abort_transaction;
 355                 }
 356
 357                 /*
 358                  * We don't support rx-flip path (except old guests who don't
 359                  * grok this feature flag).
 360                  */
 361                 err = xenbus_printf(xbt, dev->nodename,
 362                                     "feature-rx-flip", "%d", 0);
 363                 if (err) {
 364                         message = "writing feature-rx-flip";
 365                         goto abort_transaction;
 366                 }
 367
 368                 /* We support dynamic multicast-control. */
 369                 err = xenbus_printf(xbt, dev->nodename,
 370                                     "feature-multicast-control", "%d", 1);
 371                 if (err) {
 372                         message = "writing feature-multicast-control";
 373                         goto abort_transaction;
 374                 }
 375
 376                 err = xenbus_printf(xbt, dev->nodename,
 377                                     "feature-dynamic-multicast-control",
 378                                     "%d", 1);
 379                 if (err) {
 380                         message = "writing feature-dynamic-multicast-control";
 381                         goto abort_transaction;
 382                 }
 383
 384                 err = xenbus_transaction_end(xbt, 0);
 385         } while (err == -EAGAIN);
 386
 387         if (err) {
 388                 xenbus_dev_fatal(dev, err, "completing transaction");
 389                 goto fail;
 390         }
 391
 392         /*
 393          * Split event channels support, this is optional so it is not
 394          * put inside the above loop.
 395          */
 396         err = xenbus_printf(XBT_NIL, dev->nodename,
 397                             "feature-split-event-channels",
 398                             "%u", separate_tx_rx_irq);
 399         if (err)
 400                 pr_debug("Error writing feature-split-event-channels\n");
 401
 402         /* Multi-queue support: This is an optional feature. */
 403         err = xenbus_printf(XBT_NIL, dev->nodename,
 404                             "multi-queue-max-queues", "%u", xenvif_max_queues);
 405         if (err)
 406                 pr_debug("Error writing multi-queue-max-queues\n");
 407
 408         err = xenbus_printf(XBT_NIL, dev->nodename,
 409                             "feature-ctrl-ring",
 410                             "%u", true);
 411         if (err)
 412                 pr_debug("Error writing feature-ctrl-ring\n");
 413
 414         script = xenbus_read(XBT_NIL, dev->nodename, "script", NULL);
 415         if (IS_ERR(script)) {
 416                 err = PTR_ERR(script);
 417                 xenbus_dev_fatal(dev, err, "reading script");
 418                 goto fail;
 419         }
 420
 421         be->hotplug_script = script;
 422
 423
 424         /* This kicks hotplug scripts, so do it immediately. */
 425         err = backend_create_xenvif(be);
 426         if (err)
 427                 goto fail;
 428
 429         return 0;
 430
 431 abort_transaction:
 432         xenbus_transaction_end(xbt, 1);
 433         xenbus_dev_fatal(dev, err, "%s", message);
 434 fail:
 435         pr_debug("failed\n");
 436         netback_remove(dev);
 437         return err;
 438 }
 439
 440
 441 /*
 442  * Handle the creation of the hotplug script environment.  We add the script
 443  * and vif variables to the environment, for the benefit of the vif-* hotplug
 444  * scripts.
 445  */
 446 static int netback_uevent(struct xenbus_device *xdev,
 447                           struct kobj_uevent_env *env)
 448 {
 449         struct backend_info *be = dev_get_drvdata(&xdev->dev);
 450
 451         if (!be)
 452                 return 0;
 453
 454         if (add_uevent_var(env, "script=%s", be->hotplug_script))
 455                 return -ENOMEM;
 456
 457         if (!be->vif)
 458                 return 0;
 459
 460         return add_uevent_var(env, "vif=%s", be->vif->dev->name);
 461 }
 462
 463
 464 static int backend_create_xenvif(struct backend_info *be)
 465 {
 466         int err;
 467         long handle;
 468         struct xenbus_device *dev = be->dev;
 469         struct xenvif *vif;
 470
 471         if (be->vif != NULL)
 472                 return 0;
 473
 474         err = xenbus_scanf(XBT_NIL, dev->nodename, "handle", "%li", &handle);
 475         if (err != 1) {
 476                 xenbus_dev_fatal(dev, err, "reading handle");
 477                 return (err < 0) ? err : -EINVAL;
 478         }
 479
 480         vif = xenvif_alloc(&dev->dev, dev->otherend_id, handle);
 481         if (IS_ERR(vif)) {
 482                 err = PTR_ERR(vif);
 483                 xenbus_dev_fatal(dev, err, "creating interface");
 484                 return err;
 485         }
 486         be->vif = vif;
 487
 488         kobject_uevent(&dev->dev.kobj, KOBJ_ONLINE);
 489         return 0;
 490 }
 491
 492 static void backend_disconnect(struct backend_info *be)
 493 {
 494         struct xenvif *vif = be->vif;
 495
 496         if (vif) {
 497                 unsigned int num_queues = vif->num_queues;
 498                 unsigned int queue_index;
 499
 500                 xen_unregister_watchers(vif);
 501 #ifdef CONFIG_DEBUG_FS
 502                 xenvif_debugfs_delif(vif);
 503 #endif /* CONFIG_DEBUG_FS */
 504                 xenvif_disconnect_data(vif);
 505
 506                 /* At this point some of the handlers may still be active
 507                  * so we need to have additional synchronization here.
 508                  */
 509                 vif->num_queues = 0;
 510                 synchronize_net();
 511
 512                 for (queue_index = 0; queue_index < num_queues; ++queue_index)
 513                         xenvif_deinit_queue(&vif->queues[queue_index]);
 514
 515                 vfree(vif->queues);
 516                 vif->queues = NULL;
 517
 518                 xenvif_disconnect_ctrl(vif);
 519         }
 520 }
 521
 522 static void backend_connect(struct backend_info *be)
 523 {
 524         if (be->vif)
 525                 connect(be);
 526 }
 527
 528 static inline void backend_switch_state(struct backend_info *be,
 529                                         enum xenbus_state state)
 530 {
 531         struct xenbus_device *dev = be->dev;
 532
 533         pr_debug("%s -> %s\n", dev->nodename, xenbus_strstate(state));
 534         be->state = state;
 535
 536         /* If we are waiting for a hotplug script then defer the
 537          * actual xenbus state change.
 538          */
 539         if (!be->have_hotplug_status_watch)
 540                 xenbus_switch_state(dev, state);
 541 }
 542
 543 /* Handle backend state transitions:
 544  *
 545  * The backend state starts in Initialising and the following transitions are
 546  * allowed.
 547  *
 548  * Initialising -> InitWait -> Connected
 549  *          \
 550  *           \        ^    \         |
 551  *            \       |     \        |
 552  *             \      |      \       |
 553  *              \     |       \      |
 554  *               \    |        \     |
 555  *                \   |         \    |
 556  *                 V  |          V   V
 557  *
 558  *                  Closed  <-> Closing
 559  *
 560  * The state argument specifies the eventual state of the backend and the
 561  * function transitions to that state via the shortest path.
 562  */
 563 static void set_backend_state(struct backend_info *be,
 564                               enum xenbus_state state)
 565 {
 566         while (be->state != state) {
 567                 switch (be->state) {
 568                 case XenbusStateInitialising:
 569                         switch (state) {
 570                         case XenbusStateInitWait:
 571                         case XenbusStateConnected:
 572                         case XenbusStateClosing:
 573                                 backend_switch_state(be, XenbusStateInitWait);
 574                                 break;
 575                         case XenbusStateClosed:
 576                                 backend_switch_state(be, XenbusStateClosed);
 577                                 break;
 578                         default:
 579                                 BUG();
 580                         }
 581                         break;
 582                 case XenbusStateClosed:
 583                         switch (state) {
 584                         case XenbusStateInitWait:
 585                         case XenbusStateConnected:
 586                                 backend_switch_state(be, XenbusStateInitWait);
 587                                 break;
 588                         case XenbusStateClosing:
 589                                 backend_switch_state(be, XenbusStateClosing);
 590                                 break;
 591                         default:
 592                                 BUG();
 593                         }
 594                         break;
 595                 case XenbusStateInitWait:
 596                         switch (state) {
 597                         case XenbusStateConnected:
 598                                 backend_connect(be);
 599                                 backend_switch_state(be, XenbusStateConnected);
 600                                 break;
 601                         case XenbusStateClosing:
 602                         case XenbusStateClosed:
 603                                 backend_switch_state(be, XenbusStateClosing);
 604                                 break;
 605                         default:
 606                                 BUG();
 607                         }
 608                         break;
 609                 case XenbusStateConnected:
 610                         switch (state) {
 611                         case XenbusStateInitWait:
 612                         case XenbusStateClosing:
 613                         case XenbusStateClosed:
 614                                 backend_disconnect(be);
 615                                 backend_switch_state(be, XenbusStateClosing);
 616                                 break;
 617                         default:
 618                                 BUG();
 619                         }
 620                         break;
 621                 case XenbusStateClosing:
 622                         switch (state) {
 623                         case XenbusStateInitWait:
 624                         case XenbusStateConnected:
 625                         case XenbusStateClosed:
 626                                 backend_switch_state(be, XenbusStateClosed);
 627                                 break;
 628                         default:
 629                                 BUG();
 630                         }
 631                         break;
 632                 default:
 633                         BUG();
 634                 }
 635         }
 636 }
 637
 638 /**
 639  * Callback received when the frontend's state changes.
 640  */
 641 static void frontend_changed(struct xenbus_device *dev,
 642                              enum xenbus_state frontend_state)
 643 {
 644         struct backend_info *be = dev_get_drvdata(&dev->dev);
 645
 646         pr_debug("%s -> %s\n", dev->otherend, xenbus_strstate(frontend_state));
 647
 648         be->frontend_state = frontend_state;
 649
 650         switch (frontend_state) {
 651         case XenbusStateInitialising:
 652                 set_backend_state(be, XenbusStateInitWait);
 653                 break;
 654
 655         case XenbusStateInitialised:
 656                 break;
 657
 658         case XenbusStateConnected:
 659                 set_backend_state(be, XenbusStateConnected);
 660                 break;
 661
 662         case XenbusStateClosing:
 663                 set_backend_state(be, XenbusStateClosing);
 664                 break;
 665
 666         case XenbusStateClosed:
 667                 set_backend_state(be, XenbusStateClosed);
 668                 if (xenbus_dev_is_online(dev))
 669                         break;
 670                 /* fall through if not online */
 671         case XenbusStateUnknown:
 672                 set_backend_state(be, XenbusStateClosed);
 673                 device_unregister(&dev->dev);
 674                 break;
 675
 676         default:
 677                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 678                                  frontend_state);
 679                 break;
 680         }
 681 }
 682
 683
 684 static void xen_net_read_rate(struct xenbus_device *dev,
 685                               unsigned long *bytes, unsigned long *usec)
 686 {
 687         char *s, *e;
 688         unsigned long b, u;
 689         char *ratestr;
 690
 691         /* Default to unlimited bandwidth. */
 692         *bytes = ~0UL;
 693         *usec = 0;
 694
 695         ratestr = xenbus_read(XBT_NIL, dev->nodename, "rate", NULL);
 696         if (IS_ERR(ratestr))
 697                 return;
 698
 699         s = ratestr;
 700         b = simple_strtoul(s, &e, 10);
 701         if ((s == e) || (*e != ','))
 702                 goto fail;
 703
 704         s = e + 1;
 705         u = simple_strtoul(s, &e, 10);
 706         if ((s == e) || (*e != '\0'))
 707                 goto fail;
 708
 709         *bytes = b;
 710         *usec = u;
 711
 712         kfree(ratestr);
 713         return;
 714
 715  fail:
 716         pr_warn("Failed to parse network rate limit. Traffic unlimited.\n");
 717         kfree(ratestr);
 718 }
 719
 720 static int xen_net_read_mac(struct xenbus_device *dev, u8 mac[])
 721 {
 722         char *s, *e, *macstr;
 723         int i;
 724
 725         macstr = s = xenbus_read(XBT_NIL, dev->nodename, "mac", NULL);
 726         if (IS_ERR(macstr))
 727                 return PTR_ERR(macstr);
 728
 729         for (i = 0; i < ETH_ALEN; i++) {
 730                 mac[i] = simple_strtoul(s, &e, 16);
 731                 if ((s == e) || (*e != ((i == ETH_ALEN-1) ? '\0' : ':'))) {
 732                         kfree(macstr);
 733                         return -ENOENT;
 734                 }
 735                 s = e+1;
 736         }
 737
 738         kfree(macstr);
 739         return 0;
 740 }
 741
 742 static void xen_net_rate_changed(struct xenbus_watch *watch,
 743                                  const char *path, const char *token)
 744 {
 745         struct xenvif *vif = container_of(watch, struct xenvif, credit_watch);
 746         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 747         unsigned long   credit_bytes;
 748         unsigned long   credit_usec;
 749         unsigned int queue_index;
 750
 751         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 752         for (queue_index = 0; queue_index < vif->num_queues; queue_index++) {
 753                 struct xenvif_queue *queue = &vif->queues[queue_index];
 754
 755                 queue->credit_bytes = credit_bytes;
 756                 queue->credit_usec = credit_usec;
 757                 if (!mod_timer_pending(&queue->credit_timeout, jiffies) &&
 758                         queue->remaining_credit > queue->credit_bytes) {
 759                         queue->remaining_credit = queue->credit_bytes;
 760                 }
 761         }
 762 }
 763
 764 static int xen_register_credit_watch(struct xenbus_device *dev,
 765                                      struct xenvif *vif)
 766 {
 767         int err = 0;
 768         char *node;
 769         unsigned maxlen = strlen(dev->nodename) + sizeof("/rate");
 770
 771         if (vif->credit_watch.node)
 772                 return -EADDRINUSE;
 773
 774         node = kmalloc(maxlen, GFP_KERNEL);
 775         if (!node)
 776                 return -ENOMEM;
 777         snprintf(node, maxlen, "%s/rate", dev->nodename);
 778         vif->credit_watch.node = node;
 779         vif->credit_watch.callback = xen_net_rate_changed;
 780         err = register_xenbus_watch(&vif->credit_watch);
 781         if (err) {
 782                 pr_err("Failed to set watcher %s\n", vif->credit_watch.node);
 783                 kfree(node);
 784                 vif->credit_watch.node = NULL;
 785                 vif->credit_watch.callback = NULL;
 786         }
 787         return err;
 788 }
 789
 790 static void xen_unregister_credit_watch(struct xenvif *vif)
 791 {
 792         if (vif->credit_watch.node) {
 793                 unregister_xenbus_watch(&vif->credit_watch);
 794                 kfree(vif->credit_watch.node);
 795                 vif->credit_watch.node = NULL;
 796         }
 797 }
 798
 799 static void xen_mcast_ctrl_changed(struct xenbus_watch *watch,
 800                                    const char *path, const char *token)
 801 {
 802         struct xenvif *vif = container_of(watch, struct xenvif,
 803                                           mcast_ctrl_watch);
 804         struct xenbus_device *dev = xenvif_to_xenbus_device(vif);
 805
 806         vif->multicast_control = !!xenbus_read_unsigned(dev->otherend,
 807                                         "request-multicast-control", 0);
 808 }
 809
 810 static int xen_register_mcast_ctrl_watch(struct xenbus_device *dev,
 811                                          struct xenvif *vif)
 812 {
 813         int err = 0;
 814         char *node;
 815         unsigned maxlen = strlen(dev->otherend) +
 816                 sizeof("/request-multicast-control");
 817
 818         if (vif->mcast_ctrl_watch.node) {
 819                 pr_err_ratelimited("Watch is already registered\n");
 820                 return -EADDRINUSE;
 821         }
 822
 823         node = kmalloc(maxlen, GFP_KERNEL);
 824         if (!node) {
 825                 pr_err("Failed to allocate memory for watch\n");
 826                 return -ENOMEM;
 827         }
 828         snprintf(node, maxlen, "%s/request-multicast-control",
 829                  dev->otherend);
 830         vif->mcast_ctrl_watch.node = node;
 831         vif->mcast_ctrl_watch.callback = xen_mcast_ctrl_changed;
 832         err = register_xenbus_watch(&vif->mcast_ctrl_watch);
 833         if (err) {
 834                 pr_err("Failed to set watcher %s\n",
 835                        vif->mcast_ctrl_watch.node);
 836                 kfree(node);
 837                 vif->mcast_ctrl_watch.node = NULL;
 838                 vif->mcast_ctrl_watch.callback = NULL;
 839         }
 840         return err;
 841 }
 842
 843 static void xen_unregister_mcast_ctrl_watch(struct xenvif *vif)
 844 {
 845         if (vif->mcast_ctrl_watch.node) {
 846                 unregister_xenbus_watch(&vif->mcast_ctrl_watch);
 847                 kfree(vif->mcast_ctrl_watch.node);
 848                 vif->mcast_ctrl_watch.node = NULL;
 849         }
 850 }
 851
 852 static void xen_register_watchers(struct xenbus_device *dev,
 853                                   struct xenvif *vif)
 854 {
 855         xen_register_credit_watch(dev, vif);
 856         xen_register_mcast_ctrl_watch(dev, vif);
 857 }
 858
 859 static void xen_unregister_watchers(struct xenvif *vif)
 860 {
 861         xen_unregister_mcast_ctrl_watch(vif);
 862         xen_unregister_credit_watch(vif);
 863 }
 864
 865 static void unregister_hotplug_status_watch(struct backend_info *be)
 866 {
 867         if (be->have_hotplug_status_watch) {
 868                 unregister_xenbus_watch(&be->hotplug_status_watch);
 869                 kfree(be->hotplug_status_watch.node);
 870         }
 871         be->have_hotplug_status_watch = 0;
 872 }
 873
 874 static void hotplug_status_changed(struct xenbus_watch *watch,
 875                                    const char *path,
 876                                    const char *token)
 877 {
 878         struct backend_info *be = container_of(watch,
 879                                                struct backend_info,
 880                                                hotplug_status_watch);
 881         char *str;
 882         unsigned int len;
 883
 884         str = xenbus_read(XBT_NIL, be->dev->nodename, "hotplug-status", &len);
 885         if (IS_ERR(str))
 886                 return;
 887         if (len == sizeof("connected")-1 && !memcmp(str, "connected", len)) {
 888                 /* Complete any pending state change */
 889                 xenbus_switch_state(be->dev, be->state);
 890
 891                 /* Not interested in this watch anymore. */
 892                 unregister_hotplug_status_watch(be);
 893         }
 894         kfree(str);
 895 }
 896
 897 static int connect_ctrl_ring(struct backend_info *be)
 898 {
 899         struct xenbus_device *dev = be->dev;
 900         struct xenvif *vif = be->vif;
 901         unsigned int val;
 902         grant_ref_t ring_ref;
 903         unsigned int evtchn;
 904         int err;
 905
 906         err = xenbus_scanf(XBT_NIL, dev->otherend,
 907                            "ctrl-ring-ref", "%u", &val);
 908         if (err < 0)
 909                 goto done; /* The frontend does not have a control ring */
 910
 911         ring_ref = val;
 912
 913         err = xenbus_scanf(XBT_NIL, dev->otherend,
 914                            "event-channel-ctrl", "%u", &val);
 915         if (err < 0) {
 916                 xenbus_dev_fatal(dev, err,
 917                                  "reading %s/event-channel-ctrl",
 918                                  dev->otherend);
 919                 goto fail;
 920         }
 921
 922         evtchn = val;
 923
 924         err = xenvif_connect_ctrl(vif, ring_ref, evtchn);
 925         if (err) {
 926                 xenbus_dev_fatal(dev, err,
 927                                  "mapping shared-frame %u port %u",
 928                                  ring_ref, evtchn);
 929                 goto fail;
 930         }
 931
 932 done:
 933         return 0;
 934
 935 fail:
 936         return err;
 937 }
 938
 939 static void connect(struct backend_info *be)
 940 {
 941         int err;
 942         struct xenbus_device *dev = be->dev;
 943         unsigned long credit_bytes, credit_usec;
 944         unsigned int queue_index;
 945         unsigned int requested_num_queues;
 946         struct xenvif_queue *queue;
 947
 948         /* Check whether the frontend requested multiple queues
 949          * and read the number requested.
 950          */
 951         requested_num_queues = xenbus_read_unsigned(dev->otherend,
 952                                         "multi-queue-num-queues", 1);
 953         if (requested_num_queues > xenvif_max_queues) {
 954                 /* buggy or malicious guest */
 955                 xenbus_dev_fatal(dev, -EINVAL,
 956                                  "guest requested %u queues, exceeding the maximum of %u.",
 957                                  requested_num_queues, xenvif_max_queues);
 958                 return;
 959         }
 960
 961         err = xen_net_read_mac(dev, be->vif->fe_dev_addr);
 962         if (err) {
 963                 xenbus_dev_fatal(dev, err, "parsing %s/mac", dev->nodename);
 964                 return;
 965         }
 966
 967         xen_net_read_rate(dev, &credit_bytes, &credit_usec);
 968         xen_unregister_watchers(be->vif);
 969         xen_register_watchers(dev, be->vif);
 970         read_xenbus_vif_flags(be);
 971
 972         err = connect_ctrl_ring(be);
 973         if (err) {
 974                 xenbus_dev_fatal(dev, err, "connecting control ring");
 975                 return;
 976         }
 977
 978         /* Use the number of queues requested by the frontend */
 979         be->vif->queues = vzalloc(array_size(requested_num_queues,
 980                                              sizeof(struct xenvif_queue)));
 981         if (!be->vif->queues) {
 982                 xenbus_dev_fatal(dev, -ENOMEM,
 983                                  "allocating queues");
 984                 return;
 985         }
 986
 987         be->vif->num_queues = requested_num_queues;
 988         be->vif->stalled_queues = requested_num_queues;
 989
 990         for (queue_index = 0; queue_index < requested_num_queues; ++queue_index) {
 991                 queue = &be->vif->queues[queue_index];
 992                 queue->vif = be->vif;
 993                 queue->id = queue_index;
 994                 snprintf(queue->name, sizeof(queue->name), "%s-q%u",
 995                                 be->vif->dev->name, queue->id);
 996
 997                 err = xenvif_init_queue(queue);
 998                 if (err) {
 999                         /* xenvif_init_queue() cleans up after itself on
1000                          * failure, but we need to clean up any previously
1001                          * initialised queues. Set num_queues to i so that
1002                          * earlier queues can be destroyed using the regular
1003                          * disconnect logic.
1004                          */
1005                         be->vif->num_queues = queue_index;
1006                         goto err;
1007                 }
1008
1009                 queue->credit_bytes = credit_bytes;
1010                 queue->remaining_credit = credit_bytes;
1011                 queue->credit_usec = credit_usec;
1012
1013                 err = connect_data_rings(be, queue);
1014                 if (err) {
1015                         /* connect_data_rings() cleans up after itself on
1016                          * failure, but we need to clean up after
1017                          * xenvif_init_queue() here, and also clean up any
1018                          * previously initialised queues.
1019                          */
1020                         xenvif_deinit_queue(queue);
1021                         be->vif->num_queues = queue_index;
1022                         goto err;
1023                 }
1024         }
1025
1026 #ifdef CONFIG_DEBUG_FS
1027         xenvif_debugfs_addif(be->vif);
1028 #endif /* CONFIG_DEBUG_FS */
1029
1030         /* Initialisation completed, tell core driver the number of
1031          * active queues.
1032          */
1033         rtnl_lock();
1034         netif_set_real_num_tx_queues(be->vif->dev, requested_num_queues);
1035         netif_set_real_num_rx_queues(be->vif->dev, requested_num_queues);
1036         rtnl_unlock();
1037
1038         xenvif_carrier_on(be->vif);
1039
1040         unregister_hotplug_status_watch(be);
1041         err = xenbus_watch_pathfmt(dev, &be->hotplug_status_watch,
1042                                    hotplug_status_changed,
1043                                    "%s/%s", dev->nodename, "hotplug-status");
1044         if (!err)
1045                 be->have_hotplug_status_watch = 1;
1046
1047         netif_tx_wake_all_queues(be->vif->dev);
1048
1049         return;
1050
1051 err:
1052         if (be->vif->num_queues > 0)
1053                 xenvif_disconnect_data(be->vif); /* Clean up existing queues */
1054         for (queue_index = 0; queue_index < be->vif->num_queues; ++queue_index)
1055                 xenvif_deinit_queue(&be->vif->queues[queue_index]);
1056         vfree(be->vif->queues);
1057         be->vif->queues = NULL;
1058         be->vif->num_queues = 0;
1059         xenvif_disconnect_ctrl(be->vif);
1060         return;
1061 }
1062
1063
1064 static int connect_data_rings(struct backend_info *be,
1065                               struct xenvif_queue *queue)
1066 {
1067         struct xenbus_device *dev = be->dev;
1068         unsigned int num_queues = queue->vif->num_queues;
1069         unsigned long tx_ring_ref, rx_ring_ref;
1070         unsigned int tx_evtchn, rx_evtchn;
1071         int err;
1072         char *xspath;
1073         size_t xspathsize;
1074         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1075
1076         /* If the frontend requested 1 queue, or we have fallen back
1077          * to single queue due to lack of frontend support for multi-
1078          * queue, expect the remaining XenStore keys in the toplevel
1079          * directory. Otherwise, expect them in a subdirectory called
1080          * queue-N.
1081          */
1082         if (num_queues == 1) {
1083                 xspath = kzalloc(strlen(dev->otherend) + 1, GFP_KERNEL);
1084                 if (!xspath) {
1085                         xenbus_dev_fatal(dev, -ENOMEM,
1086                                          "reading ring references");
1087                         return -ENOMEM;
1088                 }
1089                 strcpy(xspath, dev->otherend);
1090         } else {
1091                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1092                 xspath = kzalloc(xspathsize, GFP_KERNEL);
1093                 if (!xspath) {
1094                         xenbus_dev_fatal(dev, -ENOMEM,
1095                                          "reading ring references");
1096                         return -ENOMEM;
1097                 }
1098                 snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend,
1099                          queue->id);
1100         }
1101
1102         err = xenbus_gather(XBT_NIL, xspath,
1103                             "tx-ring-ref", "%lu", &tx_ring_ref,
1104                             "rx-ring-ref", "%lu", &rx_ring_ref, NULL);
1105         if (err) {
1106                 xenbus_dev_fatal(dev, err,
1107                                  "reading %s/ring-ref",
1108                                  xspath);
1109                 goto err;
1110         }
1111
1112         /* Try split event channels first, then single event channel. */
1113         err = xenbus_gather(XBT_NIL, xspath,
1114                             "event-channel-tx", "%u", &tx_evtchn,
1115                             "event-channel-rx", "%u", &rx_evtchn, NULL);
1116         if (err < 0) {
1117                 err = xenbus_scanf(XBT_NIL, xspath,
1118                                    "event-channel", "%u", &tx_evtchn);
1119                 if (err < 0) {
1120                         xenbus_dev_fatal(dev, err,
1121                                          "reading %s/event-channel(-tx/rx)",
1122                                          xspath);
1123                         goto err;
1124                 }
1125                 rx_evtchn = tx_evtchn;
1126         }
1127
1128         /* Map the shared frame, irq etc. */
1129         err = xenvif_connect_data(queue, tx_ring_ref, rx_ring_ref,
1130                                   tx_evtchn, rx_evtchn);
1131         if (err) {
1132                 xenbus_dev_fatal(dev, err,
1133                                  "mapping shared-frames %lu/%lu port tx %u rx %u",
1134                                  tx_ring_ref, rx_ring_ref,
1135                                  tx_evtchn, rx_evtchn);
1136                 goto err;
1137         }
1138
1139         err = 0;
1140 err: /* Regular return falls through with err == 0 */
1141         kfree(xspath);
1142         return err;
1143 }
1144
1145 static int read_xenbus_vif_flags(struct backend_info *be)
1146 {
1147         struct xenvif *vif = be->vif;
1148         struct xenbus_device *dev = be->dev;
1149         unsigned int rx_copy;
1150         int err;
1151
1152         err = xenbus_scanf(XBT_NIL, dev->otherend, "request-rx-copy", "%u",
1153                            &rx_copy);
1154         if (err == -ENOENT) {
1155                 err = 0;
1156                 rx_copy = 0;
1157         }
1158         if (err < 0) {
1159                 xenbus_dev_fatal(dev, err, "reading %s/request-rx-copy",
1160                                  dev->otherend);
1161                 return err;
1162         }
1163         if (!rx_copy)
1164                 return -EOPNOTSUPP;
1165
1166         if (!xenbus_read_unsigned(dev->otherend, "feature-rx-notify", 0)) {
1167                 /* - Reduce drain timeout to poll more frequently for
1168                  *   Rx requests.
1169                  * - Disable Rx stall detection.
1170                  */
1171                 be->vif->drain_timeout = msecs_to_jiffies(30);
1172                 be->vif->stall_timeout = 0;
1173         }
1174
1175         vif->can_sg = !!xenbus_read_unsigned(dev->otherend, "feature-sg", 0);
1176
1177         vif->gso_mask = 0;
1178
1179         if (xenbus_read_unsigned(dev->otherend, "feature-gso-tcpv4", 0))
1180                 vif->gso_mask |= GSO_BIT(TCPV4);
1181
1182         if (xenbus_read_unsigned(dev->otherend, "feature-gso-tcpv6", 0))
1183                 vif->gso_mask |= GSO_BIT(TCPV6);
1184
1185         vif->ip_csum = !xenbus_read_unsigned(dev->otherend,
1186                                              "feature-no-csum-offload", 0);
1187
1188         vif->ipv6_csum = !!xenbus_read_unsigned(dev->otherend,
1189                                                 "feature-ipv6-csum-offload", 0);
1190
1191         return 0;
1192 }
1193
1194 static const struct xenbus_device_id netback_ids[] = {
1195         { "vif" },
1196         { "" }
1197 };
1198
1199 static struct xenbus_driver netback_driver = {
1200         .ids = netback_ids,
1201         .probe = netback_probe,
1202         .remove = netback_remove,
1203         .uevent = netback_uevent,
1204         .otherend_changed = frontend_changed,
1205 };
1206
1207 int xenvif_xenbus_init(void)
1208 {
1209         return xenbus_register_backend(&netback_driver);
1210 }
1211
1212 void xenvif_xenbus_fini(void)
1213 {
1214         return xenbus_unregister_driver(&netback_driver);
1215 }