drivers/block/xen-blkback/xenbus.c

   1 /*  Xenbus code for blkif backend
   2     Copyright (C) 2005 Rusty Russell <rusty@rustcorp.com.au>
   3     Copyright (C) 2005 XenSource Ltd
   4
   5     This program is free software; you can redistribute it and/or modify
   6     it under the terms of the GNU General Public License as published by
   7     the Free Software Foundation; either version 2 of the License, or
   8     (at your option) any later version.
   9
  10     This program is distributed in the hope that it will be useful,
  11     but WITHOUT ANY WARRANTY; without even the implied warranty of
  12     MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13     GNU General Public License for more details.
  14
  15 */
  16
  17 #define pr_fmt(fmt) "xen-blkback: " fmt
  18
  19 #include <stdarg.h>
  20 #include <linux/module.h>
  21 #include <linux/kthread.h>
  22 #include <xen/events.h>
  23 #include <xen/grant_table.h>
  24 #include "common.h"
  25
  26 /* On the XenBus the max length of 'ring-ref%u'. */
  27 #define RINGREF_NAME_LEN (20)
  28
  29 struct backend_info {
  30         struct xenbus_device    *dev;
  31         struct xen_blkif        *blkif;
  32         struct xenbus_watch     backend_watch;
  33         unsigned                major;
  34         unsigned                minor;
  35         char                    *mode;
  36 };
  37
  38 static struct kmem_cache *xen_blkif_cachep;
  39 static void connect(struct backend_info *);
  40 static int connect_ring(struct backend_info *);
  41 static void backend_changed(struct xenbus_watch *, const char *,
  42                             const char *);
  43 static void xen_blkif_free(struct xen_blkif *blkif);
  44 static void xen_vbd_free(struct xen_vbd *vbd);
  45
  46 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be)
  47 {
  48         return be->dev;
  49 }
  50
  51 /*
  52  * The last request could free the device from softirq context and
  53  * xen_blkif_free() can sleep.
  54  */
  55 static void xen_blkif_deferred_free(struct work_struct *work)
  56 {
  57         struct xen_blkif *blkif;
  58
  59         blkif = container_of(work, struct xen_blkif, free_work);
  60         xen_blkif_free(blkif);
  61 }
  62
  63 static int blkback_name(struct xen_blkif *blkif, char *buf)
  64 {
  65         char *devpath, *devname;
  66         struct xenbus_device *dev = blkif->be->dev;
  67
  68         devpath = xenbus_read(XBT_NIL, dev->nodename, "dev", NULL);
  69         if (IS_ERR(devpath))
  70                 return PTR_ERR(devpath);
  71
  72         devname = strstr(devpath, "/dev/");
  73         if (devname != NULL)
  74                 devname += strlen("/dev/");
  75         else
  76                 devname  = devpath;
  77
  78         snprintf(buf, TASK_COMM_LEN, "%d.%s", blkif->domid, devname);
  79         kfree(devpath);
  80
  81         return 0;
  82 }
  83
  84 static void xen_update_blkif_status(struct xen_blkif *blkif)
  85 {
  86         int err;
  87         char name[TASK_COMM_LEN];
  88         struct xen_blkif_ring *ring;
  89         int i;
  90
  91         /* Not ready to connect? */
  92         if (!blkif->rings || !blkif->rings[0].irq || !blkif->vbd.bdev)
  93                 return;
  94
  95         /* Already connected? */
  96         if (blkif->be->dev->state == XenbusStateConnected)
  97                 return;
  98
  99         /* Attempt to connect: exit if we fail to. */
 100         connect(blkif->be);
 101         if (blkif->be->dev->state != XenbusStateConnected)
 102                 return;
 103
 104         err = blkback_name(blkif, name);
 105         if (err) {
 106                 xenbus_dev_error(blkif->be->dev, err, "get blkback dev name");
 107                 return;
 108         }
 109
 110         err = filemap_write_and_wait(blkif->vbd.bdev->bd_inode->i_mapping);
 111         if (err) {
 112                 xenbus_dev_error(blkif->be->dev, err, "block flush");
 113                 return;
 114         }
 115         invalidate_inode_pages2(blkif->vbd.bdev->bd_inode->i_mapping);
 116
 117         for (i = 0; i < blkif->nr_rings; i++) {
 118                 ring = &blkif->rings[i];
 119                 ring->xenblkd = kthread_run(xen_blkif_schedule, ring, "%s-%d", name, i);
 120                 if (IS_ERR(ring->xenblkd)) {
 121                         err = PTR_ERR(ring->xenblkd);
 122                         ring->xenblkd = NULL;
 123                         xenbus_dev_fatal(blkif->be->dev, err,
 124                                         "start %s-%d xenblkd", name, i);
 125                         goto out;
 126                 }
 127         }
 128         return;
 129
 130 out:
 131         while (--i >= 0) {
 132                 ring = &blkif->rings[i];
 133                 kthread_stop(ring->xenblkd);
 134         }
 135         return;
 136 }
 137
 138 static int xen_blkif_alloc_rings(struct xen_blkif *blkif)
 139 {
 140         unsigned int r;
 141
 142         blkif->rings = kcalloc(blkif->nr_rings, sizeof(struct xen_blkif_ring),
 143                                GFP_KERNEL);
 144         if (!blkif->rings)
 145                 return -ENOMEM;
 146
 147         for (r = 0; r < blkif->nr_rings; r++) {
 148                 struct xen_blkif_ring *ring = &blkif->rings[r];
 149
 150                 spin_lock_init(&ring->blk_ring_lock);
 151                 init_waitqueue_head(&ring->wq);
 152                 INIT_LIST_HEAD(&ring->pending_free);
 153                 INIT_LIST_HEAD(&ring->persistent_purge_list);
 154                 INIT_WORK(&ring->persistent_purge_work, xen_blkbk_unmap_purged_grants);
 155                 spin_lock_init(&ring->free_pages_lock);
 156                 INIT_LIST_HEAD(&ring->free_pages);
 157
 158                 spin_lock_init(&ring->pending_free_lock);
 159                 init_waitqueue_head(&ring->pending_free_wq);
 160                 init_waitqueue_head(&ring->shutdown_wq);
 161                 ring->blkif = blkif;
 162                 ring->st_print = jiffies;
 163                 ring->active = true;
 164         }
 165
 166         return 0;
 167 }
 168
 169 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
 170 {
 171         struct xen_blkif *blkif;
 172
 173         BUILD_BUG_ON(MAX_INDIRECT_PAGES > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
 174
 175         blkif = kmem_cache_zalloc(xen_blkif_cachep, GFP_KERNEL);
 176         if (!blkif)
 177                 return ERR_PTR(-ENOMEM);
 178
 179         blkif->domid = domid;
 180         atomic_set(&blkif->refcnt, 1);
 181         init_completion(&blkif->drain_complete);
 182
 183         /*
 184          * Because freeing back to the cache may be deferred, it is not
 185          * safe to unload the module (and hence destroy the cache) until
 186          * this has completed. To prevent premature unloading, take an
 187          * extra module reference here and release only when the object
 188          * has been freed back to the cache.
 189          */
 190         __module_get(THIS_MODULE);
 191         INIT_WORK(&blkif->free_work, xen_blkif_deferred_free);
 192
 193         return blkif;
 194 }
 195
 196 static int xen_blkif_map(struct xen_blkif_ring *ring, grant_ref_t *gref,
 197                          unsigned int nr_grefs, unsigned int evtchn)
 198 {
 199         int err;
 200         struct xen_blkif *blkif = ring->blkif;
 201
 202         /* Already connected through? */
 203         if (ring->irq)
 204                 return 0;
 205
 206         err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
 207                                      &ring->blk_ring);
 208         if (err < 0)
 209                 return err;
 210
 211         switch (blkif->blk_protocol) {
 212         case BLKIF_PROTOCOL_NATIVE:
 213         {
 214                 struct blkif_sring *sring;
 215                 sring = (struct blkif_sring *)ring->blk_ring;
 216                 BACK_RING_INIT(&ring->blk_rings.native, sring,
 217                                XEN_PAGE_SIZE * nr_grefs);
 218                 break;
 219         }
 220         case BLKIF_PROTOCOL_X86_32:
 221         {
 222                 struct blkif_x86_32_sring *sring_x86_32;
 223                 sring_x86_32 = (struct blkif_x86_32_sring *)ring->blk_ring;
 224                 BACK_RING_INIT(&ring->blk_rings.x86_32, sring_x86_32,
 225                                XEN_PAGE_SIZE * nr_grefs);
 226                 break;
 227         }
 228         case BLKIF_PROTOCOL_X86_64:
 229         {
 230                 struct blkif_x86_64_sring *sring_x86_64;
 231                 sring_x86_64 = (struct blkif_x86_64_sring *)ring->blk_ring;
 232                 BACK_RING_INIT(&ring->blk_rings.x86_64, sring_x86_64,
 233                                XEN_PAGE_SIZE * nr_grefs);
 234                 break;
 235         }
 236         default:
 237                 BUG();
 238         }
 239
 240         err = bind_interdomain_evtchn_to_irqhandler(blkif->domid, evtchn,
 241                                                     xen_blkif_be_int, 0,
 242                                                     "blkif-backend", ring);
 243         if (err < 0) {
 244                 xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
 245                 ring->blk_rings.common.sring = NULL;
 246                 return err;
 247         }
 248         ring->irq = err;
 249
 250         return 0;
 251 }
 252
 253 static int xen_blkif_disconnect(struct xen_blkif *blkif)
 254 {
 255         struct pending_req *req, *n;
 256         unsigned int j, r;
 257         bool busy = false;
 258
 259         for (r = 0; r < blkif->nr_rings; r++) {
 260                 struct xen_blkif_ring *ring = &blkif->rings[r];
 261                 unsigned int i = 0;
 262
 263                 if (!ring->active)
 264                         continue;
 265
 266                 if (ring->xenblkd) {
 267                         kthread_stop(ring->xenblkd);
 268                         wake_up(&ring->shutdown_wq);
 269                 }
 270
 271                 /* The above kthread_stop() guarantees that at this point we
 272                  * don't have any discard_io or other_io requests. So, checking
 273                  * for inflight IO is enough.
 274                  */
 275                 if (atomic_read(&ring->inflight) > 0) {
 276                         busy = true;
 277                         continue;
 278                 }
 279
 280                 if (ring->irq) {
 281                         unbind_from_irqhandler(ring->irq, ring);
 282                         ring->irq = 0;
 283                 }
 284
 285                 if (ring->blk_rings.common.sring) {
 286                         xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
 287                         ring->blk_rings.common.sring = NULL;
 288                 }
 289
 290                 /* Remove all persistent grants and the cache of ballooned pages. */
 291                 xen_blkbk_free_caches(ring);
 292
 293                 /* Check that there is no request in use */
 294                 list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
 295                         list_del(&req->free_list);
 296
 297                         for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
 298                                 kfree(req->segments[j]);
 299
 300                         for (j = 0; j < MAX_INDIRECT_PAGES; j++)
 301                                 kfree(req->indirect_pages[j]);
 302
 303                         kfree(req);
 304                         i++;
 305                 }
 306
 307                 BUG_ON(atomic_read(&ring->persistent_gnt_in_use) != 0);
 308                 BUG_ON(!list_empty(&ring->persistent_purge_list));
 309                 BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts));
 310                 BUG_ON(!list_empty(&ring->free_pages));
 311                 BUG_ON(ring->free_pages_num != 0);
 312                 BUG_ON(ring->persistent_gnt_c != 0);
 313                 WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
 314                 ring->active = false;
 315         }
 316         if (busy)
 317                 return -EBUSY;
 318
 319         blkif->nr_ring_pages = 0;
 320         /*
 321          * blkif->rings was allocated in connect_ring, so we should free it in
 322          * here.
 323          */
 324         kfree(blkif->rings);
 325         blkif->rings = NULL;
 326         blkif->nr_rings = 0;
 327
 328         return 0;
 329 }
 330
 331 static void xen_blkif_free(struct xen_blkif *blkif)
 332 {
 333         WARN_ON(xen_blkif_disconnect(blkif));
 334         xen_vbd_free(&blkif->vbd);
 335         kfree(blkif->be->mode);
 336         kfree(blkif->be);
 337
 338         /* Make sure everything is drained before shutting down */
 339         kmem_cache_free(xen_blkif_cachep, blkif);
 340         module_put(THIS_MODULE);
 341 }
 342
 343 int __init xen_blkif_interface_init(void)
 344 {
 345         xen_blkif_cachep = kmem_cache_create("blkif_cache",
 346                                              sizeof(struct xen_blkif),
 347                                              0, 0, NULL);
 348         if (!xen_blkif_cachep)
 349                 return -ENOMEM;
 350
 351         return 0;
 352 }
 353
 354 /*
 355  *  sysfs interface for VBD I/O requests
 356  */
 357
 358 #define VBD_SHOW_ALLRING(name, format)                                  \
 359         static ssize_t show_##name(struct device *_dev,                 \
 360                                    struct device_attribute *attr,       \
 361                                    char *buf)                           \
 362         {                                                               \
 363                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 364                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 365                 struct xen_blkif *blkif = be->blkif;                    \
 366                 unsigned int i;                                         \
 367                 unsigned long long result = 0;                          \
 368                                                                         \
 369                 if (!blkif->rings)                              \
 370                         goto out;                                       \
 371                                                                         \
 372                 for (i = 0; i < blkif->nr_rings; i++) {         \
 373                         struct xen_blkif_ring *ring = &blkif->rings[i]; \
 374                                                                         \
 375                         result += ring->st_##name;                      \
 376                 }                                                       \
 377                                                                         \
 378 out:                                                                    \
 379                 return sprintf(buf, format, result);                    \
 380         }                                                               \
 381         static DEVICE_ATTR(name, 0444, show_##name, NULL)
 382
 383 VBD_SHOW_ALLRING(oo_req,  "%llu\n");
 384 VBD_SHOW_ALLRING(rd_req,  "%llu\n");
 385 VBD_SHOW_ALLRING(wr_req,  "%llu\n");
 386 VBD_SHOW_ALLRING(f_req,  "%llu\n");
 387 VBD_SHOW_ALLRING(ds_req,  "%llu\n");
 388 VBD_SHOW_ALLRING(rd_sect, "%llu\n");
 389 VBD_SHOW_ALLRING(wr_sect, "%llu\n");
 390
 391 static struct attribute *xen_vbdstat_attrs[] = {
 392         &dev_attr_oo_req.attr,
 393         &dev_attr_rd_req.attr,
 394         &dev_attr_wr_req.attr,
 395         &dev_attr_f_req.attr,
 396         &dev_attr_ds_req.attr,
 397         &dev_attr_rd_sect.attr,
 398         &dev_attr_wr_sect.attr,
 399         NULL
 400 };
 401
 402 static const struct attribute_group xen_vbdstat_group = {
 403         .name = "statistics",
 404         .attrs = xen_vbdstat_attrs,
 405 };
 406
 407 #define VBD_SHOW(name, format, args...)                                 \
 408         static ssize_t show_##name(struct device *_dev,                 \
 409                                    struct device_attribute *attr,       \
 410                                    char *buf)                           \
 411         {                                                               \
 412                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 413                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 414                                                                         \
 415                 return sprintf(buf, format, ##args);                    \
 416         }                                                               \
 417         static DEVICE_ATTR(name, 0444, show_##name, NULL)
 418
 419 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
 420 VBD_SHOW(mode, "%s\n", be->mode);
 421
 422 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
 423 {
 424         int error;
 425
 426         error = device_create_file(&dev->dev, &dev_attr_physical_device);
 427         if (error)
 428                 goto fail1;
 429
 430         error = device_create_file(&dev->dev, &dev_attr_mode);
 431         if (error)
 432                 goto fail2;
 433
 434         error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
 435         if (error)
 436                 goto fail3;
 437
 438         return 0;
 439
 440 fail3:  sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 441 fail2:  device_remove_file(&dev->dev, &dev_attr_mode);
 442 fail1:  device_remove_file(&dev->dev, &dev_attr_physical_device);
 443         return error;
 444 }
 445
 446 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
 447 {
 448         sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 449         device_remove_file(&dev->dev, &dev_attr_mode);
 450         device_remove_file(&dev->dev, &dev_attr_physical_device);
 451 }
 452
 453
 454 static void xen_vbd_free(struct xen_vbd *vbd)
 455 {
 456         if (vbd->bdev)
 457                 blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
 458         vbd->bdev = NULL;
 459 }
 460
 461 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
 462                           unsigned major, unsigned minor, int readonly,
 463                           int cdrom)
 464 {
 465         struct xen_vbd *vbd;
 466         struct block_device *bdev;
 467         struct request_queue *q;
 468
 469         vbd = &blkif->vbd;
 470         vbd->handle   = handle;
 471         vbd->readonly = readonly;
 472         vbd->type     = 0;
 473
 474         vbd->pdevice  = MKDEV(major, minor);
 475
 476         bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
 477                                  FMODE_READ : FMODE_WRITE, NULL);
 478
 479         if (IS_ERR(bdev)) {
 480                 pr_warn("xen_vbd_create: device %08x could not be opened\n",
 481                         vbd->pdevice);
 482                 return -ENOENT;
 483         }
 484
 485         vbd->bdev = bdev;
 486         if (vbd->bdev->bd_disk == NULL) {
 487                 pr_warn("xen_vbd_create: device %08x doesn't exist\n",
 488                         vbd->pdevice);
 489                 xen_vbd_free(vbd);
 490                 return -ENOENT;
 491         }
 492         vbd->size = vbd_sz(vbd);
 493
 494         if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
 495                 vbd->type |= VDISK_CDROM;
 496         if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
 497                 vbd->type |= VDISK_REMOVABLE;
 498
 499         q = bdev_get_queue(bdev);
 500         if (q && test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 501                 vbd->flush_support = true;
 502
 503         if (q && blk_queue_secure_erase(q))
 504                 vbd->discard_secure = true;
 505
 506         pr_debug("Successful creation of handle=%04x (dom=%u)\n",
 507                 handle, blkif->domid);
 508         return 0;
 509 }
 510 static int xen_blkbk_remove(struct xenbus_device *dev)
 511 {
 512         struct backend_info *be = dev_get_drvdata(&dev->dev);
 513
 514         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 515
 516         if (be->major || be->minor)
 517                 xenvbd_sysfs_delif(dev);
 518
 519         if (be->backend_watch.node) {
 520                 unregister_xenbus_watch(&be->backend_watch);
 521                 kfree(be->backend_watch.node);
 522                 be->backend_watch.node = NULL;
 523         }
 524
 525         dev_set_drvdata(&dev->dev, NULL);
 526
 527         if (be->blkif) {
 528                 xen_blkif_disconnect(be->blkif);
 529
 530                 /* Put the reference we set in xen_blkif_alloc(). */
 531                 xen_blkif_put(be->blkif);
 532         }
 533
 534         return 0;
 535 }
 536
 537 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
 538                               struct backend_info *be, int state)
 539 {
 540         struct xenbus_device *dev = be->dev;
 541         int err;
 542
 543         err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
 544                             "%d", state);
 545         if (err)
 546                 dev_warn(&dev->dev, "writing feature-flush-cache (%d)", err);
 547
 548         return err;
 549 }
 550
 551 static void xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
 552 {
 553         struct xenbus_device *dev = be->dev;
 554         struct xen_blkif *blkif = be->blkif;
 555         int err;
 556         int state = 0;
 557         struct block_device *bdev = be->blkif->vbd.bdev;
 558         struct request_queue *q = bdev_get_queue(bdev);
 559
 560         if (!xenbus_read_unsigned(dev->nodename, "discard-enable", 1))
 561                 return;
 562
 563         if (blk_queue_discard(q)) {
 564                 err = xenbus_printf(xbt, dev->nodename,
 565                         "discard-granularity", "%u",
 566                         q->limits.discard_granularity);
 567                 if (err) {
 568                         dev_warn(&dev->dev, "writing discard-granularity (%d)", err);
 569                         return;
 570                 }
 571                 err = xenbus_printf(xbt, dev->nodename,
 572                         "discard-alignment", "%u",
 573                         q->limits.discard_alignment);
 574                 if (err) {
 575                         dev_warn(&dev->dev, "writing discard-alignment (%d)", err);
 576                         return;
 577                 }
 578                 state = 1;
 579                 /* Optional. */
 580                 err = xenbus_printf(xbt, dev->nodename,
 581                                     "discard-secure", "%d",
 582                                     blkif->vbd.discard_secure);
 583                 if (err) {
 584                         dev_warn(&dev->dev, "writing discard-secure (%d)", err);
 585                         return;
 586                 }
 587         }
 588         err = xenbus_printf(xbt, dev->nodename, "feature-discard",
 589                             "%d", state);
 590         if (err)
 591                 dev_warn(&dev->dev, "writing feature-discard (%d)", err);
 592 }
 593 int xen_blkbk_barrier(struct xenbus_transaction xbt,
 594                       struct backend_info *be, int state)
 595 {
 596         struct xenbus_device *dev = be->dev;
 597         int err;
 598
 599         err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
 600                             "%d", state);
 601         if (err)
 602                 dev_warn(&dev->dev, "writing feature-barrier (%d)", err);
 603
 604         return err;
 605 }
 606
 607 /*
 608  * Entry point to this code when a new device is created.  Allocate the basic
 609  * structures, and watch the store waiting for the hotplug scripts to tell us
 610  * the device's physical major and minor numbers.  Switch to InitWait.
 611  */
 612 static int xen_blkbk_probe(struct xenbus_device *dev,
 613                            const struct xenbus_device_id *id)
 614 {
 615         int err;
 616         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 617                                           GFP_KERNEL);
 618
 619         /* match the pr_debug in xen_blkbk_remove */
 620         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 621
 622         if (!be) {
 623                 xenbus_dev_fatal(dev, -ENOMEM,
 624                                  "allocating backend structure");
 625                 return -ENOMEM;
 626         }
 627         be->dev = dev;
 628         dev_set_drvdata(&dev->dev, be);
 629
 630         be->blkif = xen_blkif_alloc(dev->otherend_id);
 631         if (IS_ERR(be->blkif)) {
 632                 err = PTR_ERR(be->blkif);
 633                 be->blkif = NULL;
 634                 xenbus_dev_fatal(dev, err, "creating block interface");
 635                 goto fail;
 636         }
 637
 638         err = xenbus_printf(XBT_NIL, dev->nodename,
 639                             "feature-max-indirect-segments", "%u",
 640                             MAX_INDIRECT_SEGMENTS);
 641         if (err)
 642                 dev_warn(&dev->dev,
 643                          "writing %s/feature-max-indirect-segments (%d)",
 644                          dev->nodename, err);
 645
 646         /* Multi-queue: advertise how many queues are supported by us.*/
 647         err = xenbus_printf(XBT_NIL, dev->nodename,
 648                             "multi-queue-max-queues", "%u", xenblk_max_queues);
 649         if (err)
 650                 pr_warn("Error writing multi-queue-max-queues\n");
 651
 652         /* setup back pointer */
 653         be->blkif->be = be;
 654
 655         err = xenbus_watch_pathfmt(dev, &be->backend_watch, backend_changed,
 656                                    "%s/%s", dev->nodename, "physical-device");
 657         if (err)
 658                 goto fail;
 659
 660         err = xenbus_printf(XBT_NIL, dev->nodename, "max-ring-page-order", "%u",
 661                             xen_blkif_max_ring_order);
 662         if (err)
 663                 pr_warn("%s write out 'max-ring-page-order' failed\n", __func__);
 664
 665         err = xenbus_switch_state(dev, XenbusStateInitWait);
 666         if (err)
 667                 goto fail;
 668
 669         return 0;
 670
 671 fail:
 672         pr_warn("%s failed\n", __func__);
 673         xen_blkbk_remove(dev);
 674         return err;
 675 }
 676
 677
 678 /*
 679  * Callback received when the hotplug scripts have placed the physical-device
 680  * node.  Read it and the mode node, and create a vbd.  If the frontend is
 681  * ready, connect.
 682  */
 683 static void backend_changed(struct xenbus_watch *watch,
 684                             const char *path, const char *token)
 685 {
 686         int err;
 687         unsigned major;
 688         unsigned minor;
 689         struct backend_info *be
 690                 = container_of(watch, struct backend_info, backend_watch);
 691         struct xenbus_device *dev = be->dev;
 692         int cdrom = 0;
 693         unsigned long handle;
 694         char *device_type;
 695
 696         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 697
 698         err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
 699                            &major, &minor);
 700         if (XENBUS_EXIST_ERR(err)) {
 701                 /*
 702                  * Since this watch will fire once immediately after it is
 703                  * registered, we expect this.  Ignore it, and wait for the
 704                  * hotplug scripts.
 705                  */
 706                 return;
 707         }
 708         if (err != 2) {
 709                 xenbus_dev_fatal(dev, err, "reading physical-device");
 710                 return;
 711         }
 712
 713         if (be->major | be->minor) {
 714                 if (be->major != major || be->minor != minor)
 715                         pr_warn("changing physical device (from %x:%x to %x:%x) not supported.\n",
 716                                 be->major, be->minor, major, minor);
 717                 return;
 718         }
 719
 720         be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
 721         if (IS_ERR(be->mode)) {
 722                 err = PTR_ERR(be->mode);
 723                 be->mode = NULL;
 724                 xenbus_dev_fatal(dev, err, "reading mode");
 725                 return;
 726         }
 727
 728         device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
 729         if (!IS_ERR(device_type)) {
 730                 cdrom = strcmp(device_type, "cdrom") == 0;
 731                 kfree(device_type);
 732         }
 733
 734         /* Front end dir is a number, which is used as the handle. */
 735         err = kstrtoul(strrchr(dev->otherend, '/') + 1, 0, &handle);
 736         if (err) {
 737                 kfree(be->mode);
 738                 be->mode = NULL;
 739                 return;
 740         }
 741
 742         be->major = major;
 743         be->minor = minor;
 744
 745         err = xen_vbd_create(be->blkif, handle, major, minor,
 746                              !strchr(be->mode, 'w'), cdrom);
 747
 748         if (err)
 749                 xenbus_dev_fatal(dev, err, "creating vbd structure");
 750         else {
 751                 err = xenvbd_sysfs_addif(dev);
 752                 if (err) {
 753                         xen_vbd_free(&be->blkif->vbd);
 754                         xenbus_dev_fatal(dev, err, "creating sysfs entries");
 755                 }
 756         }
 757
 758         if (err) {
 759                 kfree(be->mode);
 760                 be->mode = NULL;
 761                 be->major = 0;
 762                 be->minor = 0;
 763         } else {
 764                 /* We're potentially connected now */
 765                 xen_update_blkif_status(be->blkif);
 766         }
 767 }
 768
 769
 770 /*
 771  * Callback received when the frontend's state changes.
 772  */
 773 static void frontend_changed(struct xenbus_device *dev,
 774                              enum xenbus_state frontend_state)
 775 {
 776         struct backend_info *be = dev_get_drvdata(&dev->dev);
 777         int err;
 778
 779         pr_debug("%s %p %s\n", __func__, dev, xenbus_strstate(frontend_state));
 780
 781         switch (frontend_state) {
 782         case XenbusStateInitialising:
 783                 if (dev->state == XenbusStateClosed) {
 784                         pr_info("%s: prepare for reconnect\n", dev->nodename);
 785                         xenbus_switch_state(dev, XenbusStateInitWait);
 786                 }
 787                 break;
 788
 789         case XenbusStateInitialised:
 790         case XenbusStateConnected:
 791                 /*
 792                  * Ensure we connect even when two watches fire in
 793                  * close succession and we miss the intermediate value
 794                  * of frontend_state.
 795                  */
 796                 if (dev->state == XenbusStateConnected)
 797                         break;
 798
 799                 /*
 800                  * Enforce precondition before potential leak point.
 801                  * xen_blkif_disconnect() is idempotent.
 802                  */
 803                 err = xen_blkif_disconnect(be->blkif);
 804                 if (err) {
 805                         xenbus_dev_fatal(dev, err, "pending I/O");
 806                         break;
 807                 }
 808
 809                 err = connect_ring(be);
 810                 if (err) {
 811                         /*
 812                          * Clean up so that memory resources can be used by
 813                          * other devices. connect_ring reported already error.
 814                          */
 815                         xen_blkif_disconnect(be->blkif);
 816                         break;
 817                 }
 818                 xen_update_blkif_status(be->blkif);
 819                 break;
 820
 821         case XenbusStateClosing:
 822                 xenbus_switch_state(dev, XenbusStateClosing);
 823                 break;
 824
 825         case XenbusStateClosed:
 826                 xen_blkif_disconnect(be->blkif);
 827                 xenbus_switch_state(dev, XenbusStateClosed);
 828                 if (xenbus_dev_is_online(dev))
 829                         break;
 830                 /* fall through */
 831                 /* if not online */
 832         case XenbusStateUnknown:
 833                 /* implies xen_blkif_disconnect() via xen_blkbk_remove() */
 834                 device_unregister(&dev->dev);
 835                 break;
 836
 837         default:
 838                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 839                                  frontend_state);
 840                 break;
 841         }
 842 }
 843
 844
 845 /* ** Connection ** */
 846
 847
 848 /*
 849  * Write the physical details regarding the block device to the store, and
 850  * switch to Connected state.
 851  */
 852 static void connect(struct backend_info *be)
 853 {
 854         struct xenbus_transaction xbt;
 855         int err;
 856         struct xenbus_device *dev = be->dev;
 857
 858         pr_debug("%s %s\n", __func__, dev->otherend);
 859
 860         /* Supply the information about the device the frontend needs */
 861 again:
 862         err = xenbus_transaction_start(&xbt);
 863         if (err) {
 864                 xenbus_dev_fatal(dev, err, "starting transaction");
 865                 return;
 866         }
 867
 868         /* If we can't advertise it is OK. */
 869         xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
 870
 871         xen_blkbk_discard(xbt, be);
 872
 873         xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
 874
 875         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u", 1);
 876         if (err) {
 877                 xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
 878                                  dev->nodename);
 879                 goto abort;
 880         }
 881
 882         err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
 883                             (unsigned long long)vbd_sz(&be->blkif->vbd));
 884         if (err) {
 885                 xenbus_dev_fatal(dev, err, "writing %s/sectors",
 886                                  dev->nodename);
 887                 goto abort;
 888         }
 889
 890         /* FIXME: use a typename instead */
 891         err = xenbus_printf(xbt, dev->nodename, "info", "%u",
 892                             be->blkif->vbd.type |
 893                             (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
 894         if (err) {
 895                 xenbus_dev_fatal(dev, err, "writing %s/info",
 896                                  dev->nodename);
 897                 goto abort;
 898         }
 899         err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
 900                             (unsigned long)
 901                             bdev_logical_block_size(be->blkif->vbd.bdev));
 902         if (err) {
 903                 xenbus_dev_fatal(dev, err, "writing %s/sector-size",
 904                                  dev->nodename);
 905                 goto abort;
 906         }
 907         err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
 908                             bdev_physical_block_size(be->blkif->vbd.bdev));
 909         if (err)
 910                 xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
 911                                  dev->nodename);
 912
 913         err = xenbus_transaction_end(xbt, 0);
 914         if (err == -EAGAIN)
 915                 goto again;
 916         if (err)
 917                 xenbus_dev_fatal(dev, err, "ending transaction");
 918
 919         err = xenbus_switch_state(dev, XenbusStateConnected);
 920         if (err)
 921                 xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
 922                                  dev->nodename);
 923
 924         return;
 925  abort:
 926         xenbus_transaction_end(xbt, 1);
 927 }
 928
 929 /*
 930  * Each ring may have multi pages, depends on "ring-page-order".
 931  */
 932 static int read_per_ring_refs(struct xen_blkif_ring *ring, const char *dir)
 933 {
 934         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 935         struct pending_req *req, *n;
 936         int err, i, j;
 937         struct xen_blkif *blkif = ring->blkif;
 938         struct xenbus_device *dev = blkif->be->dev;
 939         unsigned int ring_page_order, nr_grefs, evtchn;
 940
 941         err = xenbus_scanf(XBT_NIL, dir, "event-channel", "%u",
 942                           &evtchn);
 943         if (err != 1) {
 944                 err = -EINVAL;
 945                 xenbus_dev_fatal(dev, err, "reading %s/event-channel", dir);
 946                 return err;
 947         }
 948
 949         err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-page-order", "%u",
 950                           &ring_page_order);
 951         if (err != 1) {
 952                 err = xenbus_scanf(XBT_NIL, dir, "ring-ref", "%u", &ring_ref[0]);
 953                 if (err != 1) {
 954                         err = -EINVAL;
 955                         xenbus_dev_fatal(dev, err, "reading %s/ring-ref", dir);
 956                         return err;
 957                 }
 958                 nr_grefs = 1;
 959         } else {
 960                 unsigned int i;
 961
 962                 if (ring_page_order > xen_blkif_max_ring_order) {
 963                         err = -EINVAL;
 964                         xenbus_dev_fatal(dev, err, "%s/request %d ring page order exceed max:%d",
 965                                          dir, ring_page_order,
 966                                          xen_blkif_max_ring_order);
 967                         return err;
 968                 }
 969
 970                 nr_grefs = 1 << ring_page_order;
 971                 for (i = 0; i < nr_grefs; i++) {
 972                         char ring_ref_name[RINGREF_NAME_LEN];
 973
 974                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
 975                         err = xenbus_scanf(XBT_NIL, dir, ring_ref_name,
 976                                            "%u", &ring_ref[i]);
 977                         if (err != 1) {
 978                                 err = -EINVAL;
 979                                 xenbus_dev_fatal(dev, err, "reading %s/%s",
 980                                                  dir, ring_ref_name);
 981                                 return err;
 982                         }
 983                 }
 984         }
 985         blkif->nr_ring_pages = nr_grefs;
 986
 987         err = -ENOMEM;
 988         for (i = 0; i < nr_grefs * XEN_BLKIF_REQS_PER_PAGE; i++) {
 989                 req = kzalloc(sizeof(*req), GFP_KERNEL);
 990                 if (!req)
 991                         goto fail;
 992                 list_add_tail(&req->free_list, &ring->pending_free);
 993                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
 994                         req->segments[j] = kzalloc(sizeof(*req->segments[0]), GFP_KERNEL);
 995                         if (!req->segments[j])
 996                                 goto fail;
 997                 }
 998                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
 999                         req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
1000                                                          GFP_KERNEL);
1001                         if (!req->indirect_pages[j])
1002                                 goto fail;
1003                 }
1004         }
1005
1006         /* Map the shared frame, irq etc. */
1007         err = xen_blkif_map(ring, ring_ref, nr_grefs, evtchn);
1008         if (err) {
1009                 xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
1010                 goto fail;
1011         }
1012
1013         return 0;
1014
1015 fail:
1016         list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
1017                 list_del(&req->free_list);
1018                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
1019                         if (!req->segments[j])
1020                                 break;
1021                         kfree(req->segments[j]);
1022                 }
1023                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1024                         if (!req->indirect_pages[j])
1025                                 break;
1026                         kfree(req->indirect_pages[j]);
1027                 }
1028                 kfree(req);
1029         }
1030         return err;
1031 }
1032
1033 static int connect_ring(struct backend_info *be)
1034 {
1035         struct xenbus_device *dev = be->dev;
1036         unsigned int pers_grants;
1037         char protocol[64] = "";
1038         int err, i;
1039         char *xspath;
1040         size_t xspathsize;
1041         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1042         unsigned int requested_num_queues = 0;
1043
1044         pr_debug("%s %s\n", __func__, dev->otherend);
1045
1046         be->blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
1047         err = xenbus_scanf(XBT_NIL, dev->otherend, "protocol",
1048                            "%63s", protocol);
1049         if (err <= 0)
1050                 strcpy(protocol, "unspecified, assuming default");
1051         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
1052                 be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
1053         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
1054                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
1055         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
1056                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
1057         else {
1058                 xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
1059                 return -ENOSYS;
1060         }
1061         pers_grants = xenbus_read_unsigned(dev->otherend, "feature-persistent",
1062                                            0);
1063         be->blkif->vbd.feature_gnt_persistent = pers_grants;
1064         be->blkif->vbd.overflow_max_grants = 0;
1065
1066         /*
1067          * Read the number of hardware queues from frontend.
1068          */
1069         requested_num_queues = xenbus_read_unsigned(dev->otherend,
1070                                                     "multi-queue-num-queues",
1071                                                     1);
1072         if (requested_num_queues > xenblk_max_queues
1073             || requested_num_queues == 0) {
1074                 /* Buggy or malicious guest. */
1075                 xenbus_dev_fatal(dev, err,
1076                                 "guest requested %u queues, exceeding the maximum of %u.",
1077                                 requested_num_queues, xenblk_max_queues);
1078                 return -ENOSYS;
1079         }
1080         be->blkif->nr_rings = requested_num_queues;
1081         if (xen_blkif_alloc_rings(be->blkif))
1082                 return -ENOMEM;
1083
1084         pr_info("%s: using %d queues, protocol %d (%s) %s\n", dev->nodename,
1085                  be->blkif->nr_rings, be->blkif->blk_protocol, protocol,
1086                  pers_grants ? "persistent grants" : "");
1087
1088         if (be->blkif->nr_rings == 1)
1089                 return read_per_ring_refs(&be->blkif->rings[0], dev->otherend);
1090         else {
1091                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1092                 xspath = kmalloc(xspathsize, GFP_KERNEL);
1093                 if (!xspath) {
1094                         xenbus_dev_fatal(dev, -ENOMEM, "reading ring references");
1095                         return -ENOMEM;
1096                 }
1097
1098                 for (i = 0; i < be->blkif->nr_rings; i++) {
1099                         memset(xspath, 0, xspathsize);
1100                         snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, i);
1101                         err = read_per_ring_refs(&be->blkif->rings[i], xspath);
1102                         if (err) {
1103                                 kfree(xspath);
1104                                 return err;
1105                         }
1106                 }
1107                 kfree(xspath);
1108         }
1109         return 0;
1110 }
1111
1112 static const struct xenbus_device_id xen_blkbk_ids[] = {
1113         { "vbd" },
1114         { "" }
1115 };
1116
1117 static struct xenbus_driver xen_blkbk_driver = {
1118         .ids  = xen_blkbk_ids,
1119         .probe = xen_blkbk_probe,
1120         .remove = xen_blkbk_remove,
1121         .otherend_changed = frontend_changed
1122 };
1123
1124 int xen_blkif_xenbus_init(void)
1125 {
1126         return xenbus_register_backend(&xen_blkbk_driver);
1127 }