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