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 /* Enlarge the array size in order to fully show blkback name. */
  27 #define BLKBACK_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                             unsigned int);
  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, BLKBACK_NAME_LEN, "blkback.%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[BLKBACK_NAME_LEN];
  88
  89         /* Not ready to connect? */
  90         if (!blkif->irq || !blkif->vbd.bdev)
  91                 return;
  92
  93         /* Already connected? */
  94         if (blkif->be->dev->state == XenbusStateConnected)
  95                 return;
  96
  97         /* Attempt to connect: exit if we fail to. */
  98         connect(blkif->be);
  99         if (blkif->be->dev->state != XenbusStateConnected)
 100                 return;
 101
 102         err = blkback_name(blkif, name);
 103         if (err) {
 104                 xenbus_dev_error(blkif->be->dev, err, "get blkback dev name");
 105                 return;
 106         }
 107
 108         err = filemap_write_and_wait(blkif->vbd.bdev->bd_inode->i_mapping);
 109         if (err) {
 110                 xenbus_dev_error(blkif->be->dev, err, "block flush");
 111                 return;
 112         }
 113         invalidate_inode_pages2(blkif->vbd.bdev->bd_inode->i_mapping);
 114
 115         blkif->xenblkd = kthread_run(xen_blkif_schedule, blkif, "%s", name);
 116         if (IS_ERR(blkif->xenblkd)) {
 117                 err = PTR_ERR(blkif->xenblkd);
 118                 blkif->xenblkd = NULL;
 119                 xenbus_dev_error(blkif->be->dev, err, "start xenblkd");
 120                 return;
 121         }
 122 }
 123
 124 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
 125 {
 126         struct xen_blkif *blkif;
 127         struct pending_req *req, *n;
 128         int i, j;
 129
 130         BUILD_BUG_ON(MAX_INDIRECT_PAGES > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
 131
 132         blkif = kmem_cache_zalloc(xen_blkif_cachep, GFP_KERNEL);
 133         if (!blkif)
 134                 return ERR_PTR(-ENOMEM);
 135
 136         blkif->domid = domid;
 137         spin_lock_init(&blkif->blk_ring_lock);
 138         atomic_set(&blkif->refcnt, 1);
 139         init_waitqueue_head(&blkif->wq);
 140         init_completion(&blkif->drain_complete);
 141         atomic_set(&blkif->drain, 0);
 142         blkif->st_print = jiffies;
 143         blkif->persistent_gnts.rb_node = NULL;
 144         spin_lock_init(&blkif->free_pages_lock);
 145         INIT_LIST_HEAD(&blkif->free_pages);
 146         INIT_LIST_HEAD(&blkif->persistent_purge_list);
 147         blkif->free_pages_num = 0;
 148         atomic_set(&blkif->persistent_gnt_in_use, 0);
 149         atomic_set(&blkif->inflight, 0);
 150         INIT_WORK(&blkif->persistent_purge_work, xen_blkbk_unmap_purged_grants);
 151
 152         INIT_LIST_HEAD(&blkif->pending_free);
 153         INIT_WORK(&blkif->free_work, xen_blkif_deferred_free);
 154
 155         for (i = 0; i < XEN_BLKIF_REQS; i++) {
 156                 req = kzalloc(sizeof(*req), GFP_KERNEL);
 157                 if (!req)
 158                         goto fail;
 159                 list_add_tail(&req->free_list,
 160                               &blkif->pending_free);
 161                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
 162                         req->segments[j] = kzalloc(sizeof(*req->segments[0]),
 163                                                    GFP_KERNEL);
 164                         if (!req->segments[j])
 165                                 goto fail;
 166                 }
 167                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
 168                         req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
 169                                                          GFP_KERNEL);
 170                         if (!req->indirect_pages[j])
 171                                 goto fail;
 172                 }
 173         }
 174         spin_lock_init(&blkif->pending_free_lock);
 175         init_waitqueue_head(&blkif->pending_free_wq);
 176         init_waitqueue_head(&blkif->shutdown_wq);
 177
 178         return blkif;
 179
 180 fail:
 181         list_for_each_entry_safe(req, n, &blkif->pending_free, free_list) {
 182                 list_del(&req->free_list);
 183                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
 184                         if (!req->segments[j])
 185                                 break;
 186                         kfree(req->segments[j]);
 187                 }
 188                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
 189                         if (!req->indirect_pages[j])
 190                                 break;
 191                         kfree(req->indirect_pages[j]);
 192                 }
 193                 kfree(req);
 194         }
 195
 196         kmem_cache_free(xen_blkif_cachep, blkif);
 197
 198         return ERR_PTR(-ENOMEM);
 199 }
 200
 201 static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t gref,
 202                          unsigned int evtchn)
 203 {
 204         int err;
 205
 206         /* Already connected through? */
 207         if (blkif->irq)
 208                 return 0;
 209
 210         err = xenbus_map_ring_valloc(blkif->be->dev, &gref, 1,
 211                                      &blkif->blk_ring);
 212         if (err < 0)
 213                 return err;
 214
 215         switch (blkif->blk_protocol) {
 216         case BLKIF_PROTOCOL_NATIVE:
 217         {
 218                 struct blkif_sring *sring;
 219                 sring = (struct blkif_sring *)blkif->blk_ring;
 220                 BACK_RING_INIT(&blkif->blk_rings.native, sring, PAGE_SIZE);
 221                 break;
 222         }
 223         case BLKIF_PROTOCOL_X86_32:
 224         {
 225                 struct blkif_x86_32_sring *sring_x86_32;
 226                 sring_x86_32 = (struct blkif_x86_32_sring *)blkif->blk_ring;
 227                 BACK_RING_INIT(&blkif->blk_rings.x86_32, sring_x86_32, PAGE_SIZE);
 228                 break;
 229         }
 230         case BLKIF_PROTOCOL_X86_64:
 231         {
 232                 struct blkif_x86_64_sring *sring_x86_64;
 233                 sring_x86_64 = (struct blkif_x86_64_sring *)blkif->blk_ring;
 234                 BACK_RING_INIT(&blkif->blk_rings.x86_64, sring_x86_64, PAGE_SIZE);
 235                 break;
 236         }
 237         default:
 238                 BUG();
 239         }
 240
 241         err = bind_interdomain_evtchn_to_irqhandler(blkif->domid, evtchn,
 242                                                     xen_blkif_be_int, 0,
 243                                                     "blkif-backend", blkif);
 244         if (err < 0) {
 245                 xenbus_unmap_ring_vfree(blkif->be->dev, blkif->blk_ring);
 246                 blkif->blk_rings.common.sring = NULL;
 247                 return err;
 248         }
 249         blkif->irq = err;
 250
 251         return 0;
 252 }
 253
 254 static int xen_blkif_disconnect(struct xen_blkif *blkif)
 255 {
 256         if (blkif->xenblkd) {
 257                 kthread_stop(blkif->xenblkd);
 258                 wake_up(&blkif->shutdown_wq);
 259                 blkif->xenblkd = NULL;
 260         }
 261
 262         /* The above kthread_stop() guarantees that at this point we
 263          * don't have any discard_io or other_io requests. So, checking
 264          * for inflight IO is enough.
 265          */
 266         if (atomic_read(&blkif->inflight) > 0)
 267                 return -EBUSY;
 268
 269         if (blkif->irq) {
 270                 unbind_from_irqhandler(blkif->irq, blkif);
 271                 blkif->irq = 0;
 272         }
 273
 274         if (blkif->blk_rings.common.sring) {
 275                 xenbus_unmap_ring_vfree(blkif->be->dev, blkif->blk_ring);
 276                 blkif->blk_rings.common.sring = NULL;
 277         }
 278
 279         /* Remove all persistent grants and the cache of ballooned pages. */
 280         xen_blkbk_free_caches(blkif);
 281
 282         return 0;
 283 }
 284
 285 static void xen_blkif_free(struct xen_blkif *blkif)
 286 {
 287         struct pending_req *req, *n;
 288         int i = 0, j;
 289
 290         xen_blkif_disconnect(blkif);
 291         xen_vbd_free(&blkif->vbd);
 292
 293         /* Make sure everything is drained before shutting down */
 294         BUG_ON(blkif->persistent_gnt_c != 0);
 295         BUG_ON(atomic_read(&blkif->persistent_gnt_in_use) != 0);
 296         BUG_ON(blkif->free_pages_num != 0);
 297         BUG_ON(!list_empty(&blkif->persistent_purge_list));
 298         BUG_ON(!list_empty(&blkif->free_pages));
 299         BUG_ON(!RB_EMPTY_ROOT(&blkif->persistent_gnts));
 300
 301         /* Check that there is no request in use */
 302         list_for_each_entry_safe(req, n, &blkif->pending_free, free_list) {
 303                 list_del(&req->free_list);
 304
 305                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
 306                         kfree(req->segments[j]);
 307
 308                 for (j = 0; j < MAX_INDIRECT_PAGES; j++)
 309                         kfree(req->indirect_pages[j]);
 310
 311                 kfree(req);
 312                 i++;
 313         }
 314
 315         WARN_ON(i != XEN_BLKIF_REQS);
 316
 317         kmem_cache_free(xen_blkif_cachep, blkif);
 318 }
 319
 320 int __init xen_blkif_interface_init(void)
 321 {
 322         xen_blkif_cachep = kmem_cache_create("blkif_cache",
 323                                              sizeof(struct xen_blkif),
 324                                              0, 0, NULL);
 325         if (!xen_blkif_cachep)
 326                 return -ENOMEM;
 327
 328         return 0;
 329 }
 330
 331 /*
 332  *  sysfs interface for VBD I/O requests
 333  */
 334
 335 #define VBD_SHOW(name, format, args...)                                 \
 336         static ssize_t show_##name(struct device *_dev,                 \
 337                                    struct device_attribute *attr,       \
 338                                    char *buf)                           \
 339         {                                                               \
 340                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 341                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 342                                                                         \
 343                 return sprintf(buf, format, ##args);                    \
 344         }                                                               \
 345         static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 346
 347 VBD_SHOW(oo_req,  "%llu\n", be->blkif->st_oo_req);
 348 VBD_SHOW(rd_req,  "%llu\n", be->blkif->st_rd_req);
 349 VBD_SHOW(wr_req,  "%llu\n", be->blkif->st_wr_req);
 350 VBD_SHOW(f_req,  "%llu\n", be->blkif->st_f_req);
 351 VBD_SHOW(ds_req,  "%llu\n", be->blkif->st_ds_req);
 352 VBD_SHOW(rd_sect, "%llu\n", be->blkif->st_rd_sect);
 353 VBD_SHOW(wr_sect, "%llu\n", be->blkif->st_wr_sect);
 354
 355 static struct attribute *xen_vbdstat_attrs[] = {
 356         &dev_attr_oo_req.attr,
 357         &dev_attr_rd_req.attr,
 358         &dev_attr_wr_req.attr,
 359         &dev_attr_f_req.attr,
 360         &dev_attr_ds_req.attr,
 361         &dev_attr_rd_sect.attr,
 362         &dev_attr_wr_sect.attr,
 363         NULL
 364 };
 365
 366 static struct attribute_group xen_vbdstat_group = {
 367         .name = "statistics",
 368         .attrs = xen_vbdstat_attrs,
 369 };
 370
 371 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
 372 VBD_SHOW(mode, "%s\n", be->mode);
 373
 374 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
 375 {
 376         int error;
 377
 378         error = device_create_file(&dev->dev, &dev_attr_physical_device);
 379         if (error)
 380                 goto fail1;
 381
 382         error = device_create_file(&dev->dev, &dev_attr_mode);
 383         if (error)
 384                 goto fail2;
 385
 386         error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
 387         if (error)
 388                 goto fail3;
 389
 390         return 0;
 391
 392 fail3:  sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 393 fail2:  device_remove_file(&dev->dev, &dev_attr_mode);
 394 fail1:  device_remove_file(&dev->dev, &dev_attr_physical_device);
 395         return error;
 396 }
 397
 398 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
 399 {
 400         sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 401         device_remove_file(&dev->dev, &dev_attr_mode);
 402         device_remove_file(&dev->dev, &dev_attr_physical_device);
 403 }
 404
 405
 406 static void xen_vbd_free(struct xen_vbd *vbd)
 407 {
 408         if (vbd->bdev)
 409                 blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
 410         vbd->bdev = NULL;
 411 }
 412
 413 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
 414                           unsigned major, unsigned minor, int readonly,
 415                           int cdrom)
 416 {
 417         struct xen_vbd *vbd;
 418         struct block_device *bdev;
 419         struct request_queue *q;
 420
 421         vbd = &blkif->vbd;
 422         vbd->handle   = handle;
 423         vbd->readonly = readonly;
 424         vbd->type     = 0;
 425
 426         vbd->pdevice  = MKDEV(major, minor);
 427
 428         bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
 429                                  FMODE_READ : FMODE_WRITE, NULL);
 430
 431         if (IS_ERR(bdev)) {
 432                 pr_warn("xen_vbd_create: device %08x could not be opened\n",
 433                         vbd->pdevice);
 434                 return -ENOENT;
 435         }
 436
 437         vbd->bdev = bdev;
 438         if (vbd->bdev->bd_disk == NULL) {
 439                 pr_warn("xen_vbd_create: device %08x doesn't exist\n",
 440                         vbd->pdevice);
 441                 xen_vbd_free(vbd);
 442                 return -ENOENT;
 443         }
 444         vbd->size = vbd_sz(vbd);
 445
 446         if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
 447                 vbd->type |= VDISK_CDROM;
 448         if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
 449                 vbd->type |= VDISK_REMOVABLE;
 450
 451         q = bdev_get_queue(bdev);
 452         if (q && q->flush_flags)
 453                 vbd->flush_support = true;
 454
 455         if (q && blk_queue_secdiscard(q))
 456                 vbd->discard_secure = true;
 457
 458         pr_debug("Successful creation of handle=%04x (dom=%u)\n",
 459                 handle, blkif->domid);
 460         return 0;
 461 }
 462 static int xen_blkbk_remove(struct xenbus_device *dev)
 463 {
 464         struct backend_info *be = dev_get_drvdata(&dev->dev);
 465
 466         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 467
 468         if (be->major || be->minor)
 469                 xenvbd_sysfs_delif(dev);
 470
 471         if (be->backend_watch.node) {
 472                 unregister_xenbus_watch(&be->backend_watch);
 473                 kfree(be->backend_watch.node);
 474                 be->backend_watch.node = NULL;
 475         }
 476
 477         dev_set_drvdata(&dev->dev, NULL);
 478
 479         if (be->blkif) {
 480                 xen_blkif_disconnect(be->blkif);
 481                 xen_blkif_put(be->blkif);
 482         }
 483
 484         kfree(be->mode);
 485         kfree(be);
 486         return 0;
 487 }
 488
 489 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
 490                               struct backend_info *be, int state)
 491 {
 492         struct xenbus_device *dev = be->dev;
 493         int err;
 494
 495         err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
 496                             "%d", state);
 497         if (err)
 498                 dev_warn(&dev->dev, "writing feature-flush-cache (%d)", err);
 499
 500         return err;
 501 }
 502
 503 static void xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
 504 {
 505         struct xenbus_device *dev = be->dev;
 506         struct xen_blkif *blkif = be->blkif;
 507         int err;
 508         int state = 0, discard_enable;
 509         struct block_device *bdev = be->blkif->vbd.bdev;
 510         struct request_queue *q = bdev_get_queue(bdev);
 511
 512         err = xenbus_scanf(XBT_NIL, dev->nodename, "discard-enable", "%d",
 513                            &discard_enable);
 514         if (err == 1 && !discard_enable)
 515                 return;
 516
 517         if (blk_queue_discard(q)) {
 518                 err = xenbus_printf(xbt, dev->nodename,
 519                         "discard-granularity", "%u",
 520                         q->limits.discard_granularity);
 521                 if (err) {
 522                         dev_warn(&dev->dev, "writing discard-granularity (%d)", err);
 523                         return;
 524                 }
 525                 err = xenbus_printf(xbt, dev->nodename,
 526                         "discard-alignment", "%u",
 527                         q->limits.discard_alignment);
 528                 if (err) {
 529                         dev_warn(&dev->dev, "writing discard-alignment (%d)", err);
 530                         return;
 531                 }
 532                 state = 1;
 533                 /* Optional. */
 534                 err = xenbus_printf(xbt, dev->nodename,
 535                                     "discard-secure", "%d",
 536                                     blkif->vbd.discard_secure);
 537                 if (err) {
 538                         dev_warn(&dev->dev, "writing discard-secure (%d)", err);
 539                         return;
 540                 }
 541         }
 542         err = xenbus_printf(xbt, dev->nodename, "feature-discard",
 543                             "%d", state);
 544         if (err)
 545                 dev_warn(&dev->dev, "writing feature-discard (%d)", err);
 546 }
 547 int xen_blkbk_barrier(struct xenbus_transaction xbt,
 548                       struct backend_info *be, int state)
 549 {
 550         struct xenbus_device *dev = be->dev;
 551         int err;
 552
 553         err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
 554                             "%d", state);
 555         if (err)
 556                 dev_warn(&dev->dev, "writing feature-barrier (%d)", err);
 557
 558         return err;
 559 }
 560
 561 /*
 562  * Entry point to this code when a new device is created.  Allocate the basic
 563  * structures, and watch the store waiting for the hotplug scripts to tell us
 564  * the device's physical major and minor numbers.  Switch to InitWait.
 565  */
 566 static int xen_blkbk_probe(struct xenbus_device *dev,
 567                            const struct xenbus_device_id *id)
 568 {
 569         int err;
 570         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 571                                           GFP_KERNEL);
 572
 573         /* match the pr_debug in xen_blkbk_remove */
 574         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 575
 576         if (!be) {
 577                 xenbus_dev_fatal(dev, -ENOMEM,
 578                                  "allocating backend structure");
 579                 return -ENOMEM;
 580         }
 581         be->dev = dev;
 582         dev_set_drvdata(&dev->dev, be);
 583
 584         be->blkif = xen_blkif_alloc(dev->otherend_id);
 585         if (IS_ERR(be->blkif)) {
 586                 err = PTR_ERR(be->blkif);
 587                 be->blkif = NULL;
 588                 xenbus_dev_fatal(dev, err, "creating block interface");
 589                 goto fail;
 590         }
 591
 592         /* setup back pointer */
 593         be->blkif->be = be;
 594
 595         err = xenbus_watch_pathfmt(dev, &be->backend_watch, backend_changed,
 596                                    "%s/%s", dev->nodename, "physical-device");
 597         if (err)
 598                 goto fail;
 599
 600         err = xenbus_switch_state(dev, XenbusStateInitWait);
 601         if (err)
 602                 goto fail;
 603
 604         return 0;
 605
 606 fail:
 607         pr_warn("%s failed\n", __func__);
 608         xen_blkbk_remove(dev);
 609         return err;
 610 }
 611
 612
 613 /*
 614  * Callback received when the hotplug scripts have placed the physical-device
 615  * node.  Read it and the mode node, and create a vbd.  If the frontend is
 616  * ready, connect.
 617  */
 618 static void backend_changed(struct xenbus_watch *watch,
 619                             const char **vec, unsigned int len)
 620 {
 621         int err;
 622         unsigned major;
 623         unsigned minor;
 624         struct backend_info *be
 625                 = container_of(watch, struct backend_info, backend_watch);
 626         struct xenbus_device *dev = be->dev;
 627         int cdrom = 0;
 628         unsigned long handle;
 629         char *device_type;
 630
 631         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 632
 633         err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
 634                            &major, &minor);
 635         if (XENBUS_EXIST_ERR(err)) {
 636                 /*
 637                  * Since this watch will fire once immediately after it is
 638                  * registered, we expect this.  Ignore it, and wait for the
 639                  * hotplug scripts.
 640                  */
 641                 return;
 642         }
 643         if (err != 2) {
 644                 xenbus_dev_fatal(dev, err, "reading physical-device");
 645                 return;
 646         }
 647
 648         if (be->major | be->minor) {
 649                 if (be->major != major || be->minor != minor)
 650                         pr_warn("changing physical device (from %x:%x to %x:%x) not supported.\n",
 651                                 be->major, be->minor, major, minor);
 652                 return;
 653         }
 654
 655         be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
 656         if (IS_ERR(be->mode)) {
 657                 err = PTR_ERR(be->mode);
 658                 be->mode = NULL;
 659                 xenbus_dev_fatal(dev, err, "reading mode");
 660                 return;
 661         }
 662
 663         device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
 664         if (!IS_ERR(device_type)) {
 665                 cdrom = strcmp(device_type, "cdrom") == 0;
 666                 kfree(device_type);
 667         }
 668
 669         /* Front end dir is a number, which is used as the handle. */
 670         err = kstrtoul(strrchr(dev->otherend, '/') + 1, 0, &handle);
 671         if (err)
 672                 return;
 673
 674         be->major = major;
 675         be->minor = minor;
 676
 677         err = xen_vbd_create(be->blkif, handle, major, minor,
 678                              !strchr(be->mode, 'w'), cdrom);
 679
 680         if (err)
 681                 xenbus_dev_fatal(dev, err, "creating vbd structure");
 682         else {
 683                 err = xenvbd_sysfs_addif(dev);
 684                 if (err) {
 685                         xen_vbd_free(&be->blkif->vbd);
 686                         xenbus_dev_fatal(dev, err, "creating sysfs entries");
 687                 }
 688         }
 689
 690         if (err) {
 691                 kfree(be->mode);
 692                 be->mode = NULL;
 693                 be->major = 0;
 694                 be->minor = 0;
 695         } else {
 696                 /* We're potentially connected now */
 697                 xen_update_blkif_status(be->blkif);
 698         }
 699 }
 700
 701
 702 /*
 703  * Callback received when the frontend's state changes.
 704  */
 705 static void frontend_changed(struct xenbus_device *dev,
 706                              enum xenbus_state frontend_state)
 707 {
 708         struct backend_info *be = dev_get_drvdata(&dev->dev);
 709         int err;
 710
 711         pr_debug("%s %p %s\n", __func__, dev, xenbus_strstate(frontend_state));
 712
 713         switch (frontend_state) {
 714         case XenbusStateInitialising:
 715                 if (dev->state == XenbusStateClosed) {
 716                         pr_info("%s: prepare for reconnect\n", dev->nodename);
 717                         xenbus_switch_state(dev, XenbusStateInitWait);
 718                 }
 719                 break;
 720
 721         case XenbusStateInitialised:
 722         case XenbusStateConnected:
 723                 /*
 724                  * Ensure we connect even when two watches fire in
 725                  * close succession and we miss the intermediate value
 726                  * of frontend_state.
 727                  */
 728                 if (dev->state == XenbusStateConnected)
 729                         break;
 730
 731                 /*
 732                  * Enforce precondition before potential leak point.
 733                  * xen_blkif_disconnect() is idempotent.
 734                  */
 735                 err = xen_blkif_disconnect(be->blkif);
 736                 if (err) {
 737                         xenbus_dev_fatal(dev, err, "pending I/O");
 738                         break;
 739                 }
 740
 741                 err = connect_ring(be);
 742                 if (err)
 743                         break;
 744                 xen_update_blkif_status(be->blkif);
 745                 break;
 746
 747         case XenbusStateClosing:
 748                 xenbus_switch_state(dev, XenbusStateClosing);
 749                 break;
 750
 751         case XenbusStateClosed:
 752                 xen_blkif_disconnect(be->blkif);
 753                 xenbus_switch_state(dev, XenbusStateClosed);
 754                 if (xenbus_dev_is_online(dev))
 755                         break;
 756                 /* fall through if not online */
 757         case XenbusStateUnknown:
 758                 /* implies xen_blkif_disconnect() via xen_blkbk_remove() */
 759                 device_unregister(&dev->dev);
 760                 break;
 761
 762         default:
 763                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 764                                  frontend_state);
 765                 break;
 766         }
 767 }
 768
 769
 770 /* ** Connection ** */
 771
 772
 773 /*
 774  * Write the physical details regarding the block device to the store, and
 775  * switch to Connected state.
 776  */
 777 static void connect(struct backend_info *be)
 778 {
 779         struct xenbus_transaction xbt;
 780         int err;
 781         struct xenbus_device *dev = be->dev;
 782
 783         pr_debug("%s %s\n", __func__, dev->otherend);
 784
 785         /* Supply the information about the device the frontend needs */
 786 again:
 787         err = xenbus_transaction_start(&xbt);
 788         if (err) {
 789                 xenbus_dev_fatal(dev, err, "starting transaction");
 790                 return;
 791         }
 792
 793         /* If we can't advertise it is OK. */
 794         xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
 795
 796         xen_blkbk_discard(xbt, be);
 797
 798         xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
 799
 800         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u", 1);
 801         if (err) {
 802                 xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
 803                                  dev->nodename);
 804                 goto abort;
 805         }
 806         err = xenbus_printf(xbt, dev->nodename, "feature-max-indirect-segments", "%u",
 807                             MAX_INDIRECT_SEGMENTS);
 808         if (err)
 809                 dev_warn(&dev->dev, "writing %s/feature-max-indirect-segments (%d)",
 810                          dev->nodename, err);
 811
 812         err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
 813                             (unsigned long long)vbd_sz(&be->blkif->vbd));
 814         if (err) {
 815                 xenbus_dev_fatal(dev, err, "writing %s/sectors",
 816                                  dev->nodename);
 817                 goto abort;
 818         }
 819
 820         /* FIXME: use a typename instead */
 821         err = xenbus_printf(xbt, dev->nodename, "info", "%u",
 822                             be->blkif->vbd.type |
 823                             (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
 824         if (err) {
 825                 xenbus_dev_fatal(dev, err, "writing %s/info",
 826                                  dev->nodename);
 827                 goto abort;
 828         }
 829         err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
 830                             (unsigned long)
 831                             bdev_logical_block_size(be->blkif->vbd.bdev));
 832         if (err) {
 833                 xenbus_dev_fatal(dev, err, "writing %s/sector-size",
 834                                  dev->nodename);
 835                 goto abort;
 836         }
 837         err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
 838                             bdev_physical_block_size(be->blkif->vbd.bdev));
 839         if (err)
 840                 xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
 841                                  dev->nodename);
 842
 843         err = xenbus_transaction_end(xbt, 0);
 844         if (err == -EAGAIN)
 845                 goto again;
 846         if (err)
 847                 xenbus_dev_fatal(dev, err, "ending transaction");
 848
 849         err = xenbus_switch_state(dev, XenbusStateConnected);
 850         if (err)
 851                 xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
 852                                  dev->nodename);
 853
 854         return;
 855  abort:
 856         xenbus_transaction_end(xbt, 1);
 857 }
 858
 859
 860 static int connect_ring(struct backend_info *be)
 861 {
 862         struct xenbus_device *dev = be->dev;
 863         unsigned long ring_ref;
 864         unsigned int evtchn;
 865         unsigned int pers_grants;
 866         char protocol[64] = "";
 867         int err;
 868
 869         pr_debug("%s %s\n", __func__, dev->otherend);
 870
 871         err = xenbus_gather(XBT_NIL, dev->otherend, "ring-ref", "%lu",
 872                             &ring_ref, "event-channel", "%u", &evtchn, NULL);
 873         if (err) {
 874                 xenbus_dev_fatal(dev, err,
 875                                  "reading %s/ring-ref and event-channel",
 876                                  dev->otherend);
 877                 return err;
 878         }
 879
 880         be->blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
 881         err = xenbus_gather(XBT_NIL, dev->otherend, "protocol",
 882                             "%63s", protocol, NULL);
 883         if (err)
 884                 strcpy(protocol, "unspecified, assuming default");
 885         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
 886                 be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
 887         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
 888                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
 889         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
 890                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
 891         else {
 892                 xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
 893                 return -1;
 894         }
 895         err = xenbus_gather(XBT_NIL, dev->otherend,
 896                             "feature-persistent", "%u",
 897                             &pers_grants, NULL);
 898         if (err)
 899                 pers_grants = 0;
 900
 901         be->blkif->vbd.feature_gnt_persistent = pers_grants;
 902         be->blkif->vbd.overflow_max_grants = 0;
 903
 904         pr_info("ring-ref %ld, event-channel %d, protocol %d (%s) %s\n",
 905                 ring_ref, evtchn, be->blkif->blk_protocol, protocol,
 906                 pers_grants ? "persistent grants" : "");
 907
 908         /* Map the shared frame, irq etc. */
 909         err = xen_blkif_map(be->blkif, ring_ref, evtchn);
 910         if (err) {
 911                 xenbus_dev_fatal(dev, err, "mapping ring-ref %lu port %u",
 912                                  ring_ref, evtchn);
 913                 return err;
 914         }
 915
 916         return 0;
 917 }
 918
 919 static const struct xenbus_device_id xen_blkbk_ids[] = {
 920         { "vbd" },
 921         { "" }
 922 };
 923
 924 static struct xenbus_driver xen_blkbk_driver = {
 925         .ids  = xen_blkbk_ids,
 926         .probe = xen_blkbk_probe,
 927         .remove = xen_blkbk_remove,
 928         .otherend_changed = frontend_changed
 929 };
 930
 931 int xen_blkif_xenbus_init(void)
 932 {
 933         return xenbus_register_backend(&xen_blkbk_driver);
 934 }