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 #include <stdarg.h>
  18 #include <linux/module.h>
  19 #include <linux/kthread.h>
  20 #include <xen/events.h>
  21 #include <xen/grant_table.h>
  22 #include "common.h"
  23
  24 struct backend_info {
  25         struct xenbus_device    *dev;
  26         struct xen_blkif        *blkif;
  27         struct xenbus_watch     backend_watch;
  28         unsigned                major;
  29         unsigned                minor;
  30         char                    *mode;
  31 };
  32
  33 static struct kmem_cache *xen_blkif_cachep;
  34 static void connect(struct backend_info *);
  35 static int connect_ring(struct backend_info *);
  36 static void backend_changed(struct xenbus_watch *, const char **,
  37                             unsigned int);
  38
  39 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be)
  40 {
  41         return be->dev;
  42 }
  43
  44 static int blkback_name(struct xen_blkif *blkif, char *buf)
  45 {
  46         char *devpath, *devname;
  47         struct xenbus_device *dev = blkif->be->dev;
  48
  49         devpath = xenbus_read(XBT_NIL, dev->nodename, "dev", NULL);
  50         if (IS_ERR(devpath))
  51                 return PTR_ERR(devpath);
  52
  53         devname = strstr(devpath, "/dev/");
  54         if (devname != NULL)
  55                 devname += strlen("/dev/");
  56         else
  57                 devname  = devpath;
  58
  59         snprintf(buf, TASK_COMM_LEN, "blkback.%d.%s", blkif->domid, devname);
  60         kfree(devpath);
  61
  62         return 0;
  63 }
  64
  65 static void xen_update_blkif_status(struct xen_blkif *blkif)
  66 {
  67         int err;
  68         char name[TASK_COMM_LEN];
  69
  70         /* Not ready to connect? */
  71         if (!blkif->irq || !blkif->vbd.bdev)
  72                 return;
  73
  74         /* Already connected? */
  75         if (blkif->be->dev->state == XenbusStateConnected)
  76                 return;
  77
  78         /* Attempt to connect: exit if we fail to. */
  79         connect(blkif->be);
  80         if (blkif->be->dev->state != XenbusStateConnected)
  81                 return;
  82
  83         err = blkback_name(blkif, name);
  84         if (err) {
  85                 xenbus_dev_error(blkif->be->dev, err, "get blkback dev name");
  86                 return;
  87         }
  88
  89         err = filemap_write_and_wait(blkif->vbd.bdev->bd_inode->i_mapping);
  90         if (err) {
  91                 xenbus_dev_error(blkif->be->dev, err, "block flush");
  92                 return;
  93         }
  94         invalidate_inode_pages2(blkif->vbd.bdev->bd_inode->i_mapping);
  95
  96         blkif->xenblkd = kthread_run(xen_blkif_schedule, blkif, "%s", name);
  97         if (IS_ERR(blkif->xenblkd)) {
  98                 err = PTR_ERR(blkif->xenblkd);
  99                 blkif->xenblkd = NULL;
 100                 xenbus_dev_error(blkif->be->dev, err, "start xenblkd");
 101                 return;
 102         }
 103 }
 104
 105 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
 106 {
 107         struct xen_blkif *blkif;
 108         struct pending_req *req, *n;
 109         int i, j;
 110
 111         BUILD_BUG_ON(MAX_INDIRECT_PAGES > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
 112
 113         blkif = kmem_cache_zalloc(xen_blkif_cachep, GFP_KERNEL);
 114         if (!blkif)
 115                 return ERR_PTR(-ENOMEM);
 116
 117         blkif->domid = domid;
 118         spin_lock_init(&blkif->blk_ring_lock);
 119         atomic_set(&blkif->refcnt, 1);
 120         init_waitqueue_head(&blkif->wq);
 121         init_completion(&blkif->drain_complete);
 122         atomic_set(&blkif->drain, 0);
 123         blkif->st_print = jiffies;
 124         init_waitqueue_head(&blkif->waiting_to_free);
 125         blkif->persistent_gnts.rb_node = NULL;
 126         spin_lock_init(&blkif->free_pages_lock);
 127         INIT_LIST_HEAD(&blkif->free_pages);
 128         INIT_LIST_HEAD(&blkif->persistent_purge_list);
 129         blkif->free_pages_num = 0;
 130         atomic_set(&blkif->persistent_gnt_in_use, 0);
 131         atomic_set(&blkif->inflight, 0);
 132
 133         INIT_LIST_HEAD(&blkif->pending_free);
 134
 135         for (i = 0; i < XEN_BLKIF_REQS; i++) {
 136                 req = kzalloc(sizeof(*req), GFP_KERNEL);
 137                 if (!req)
 138                         goto fail;
 139                 list_add_tail(&req->free_list,
 140                               &blkif->pending_free);
 141                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
 142                         req->segments[j] = kzalloc(sizeof(*req->segments[0]),
 143                                                    GFP_KERNEL);
 144                         if (!req->segments[j])
 145                                 goto fail;
 146                 }
 147                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
 148                         req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
 149                                                          GFP_KERNEL);
 150                         if (!req->indirect_pages[j])
 151                                 goto fail;
 152                 }
 153         }
 154         spin_lock_init(&blkif->pending_free_lock);
 155         init_waitqueue_head(&blkif->pending_free_wq);
 156         init_waitqueue_head(&blkif->shutdown_wq);
 157
 158         return blkif;
 159
 160 fail:
 161         list_for_each_entry_safe(req, n, &blkif->pending_free, free_list) {
 162                 list_del(&req->free_list);
 163                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
 164                         if (!req->segments[j])
 165                                 break;
 166                         kfree(req->segments[j]);
 167                 }
 168                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
 169                         if (!req->indirect_pages[j])
 170                                 break;
 171                         kfree(req->indirect_pages[j]);
 172                 }
 173                 kfree(req);
 174         }
 175
 176         kmem_cache_free(xen_blkif_cachep, blkif);
 177
 178         return ERR_PTR(-ENOMEM);
 179 }
 180
 181 static int xen_blkif_map(struct xen_blkif *blkif, unsigned long shared_page,
 182                          unsigned int evtchn)
 183 {
 184         int err;
 185
 186         /* Already connected through? */
 187         if (blkif->irq)
 188                 return 0;
 189
 190         err = xenbus_map_ring_valloc(blkif->be->dev, shared_page, &blkif->blk_ring);
 191         if (err < 0)
 192                 return err;
 193
 194         switch (blkif->blk_protocol) {
 195         case BLKIF_PROTOCOL_NATIVE:
 196         {
 197                 struct blkif_sring *sring;
 198                 sring = (struct blkif_sring *)blkif->blk_ring;
 199                 BACK_RING_INIT(&blkif->blk_rings.native, sring, PAGE_SIZE);
 200                 break;
 201         }
 202         case BLKIF_PROTOCOL_X86_32:
 203         {
 204                 struct blkif_x86_32_sring *sring_x86_32;
 205                 sring_x86_32 = (struct blkif_x86_32_sring *)blkif->blk_ring;
 206                 BACK_RING_INIT(&blkif->blk_rings.x86_32, sring_x86_32, PAGE_SIZE);
 207                 break;
 208         }
 209         case BLKIF_PROTOCOL_X86_64:
 210         {
 211                 struct blkif_x86_64_sring *sring_x86_64;
 212                 sring_x86_64 = (struct blkif_x86_64_sring *)blkif->blk_ring;
 213                 BACK_RING_INIT(&blkif->blk_rings.x86_64, sring_x86_64, PAGE_SIZE);
 214                 break;
 215         }
 216         default:
 217                 BUG();
 218         }
 219
 220         err = bind_interdomain_evtchn_to_irqhandler(blkif->domid, evtchn,
 221                                                     xen_blkif_be_int, 0,
 222                                                     "blkif-backend", blkif);
 223         if (err < 0) {
 224                 xenbus_unmap_ring_vfree(blkif->be->dev, blkif->blk_ring);
 225                 blkif->blk_rings.common.sring = NULL;
 226                 return err;
 227         }
 228         blkif->irq = err;
 229
 230         return 0;
 231 }
 232
 233 static void xen_blkif_disconnect(struct xen_blkif *blkif)
 234 {
 235         if (blkif->xenblkd) {
 236                 kthread_stop(blkif->xenblkd);
 237                 wake_up(&blkif->shutdown_wq);
 238                 blkif->xenblkd = NULL;
 239         }
 240
 241         atomic_dec(&blkif->refcnt);
 242         wait_event(blkif->waiting_to_free, atomic_read(&blkif->refcnt) == 0);
 243         atomic_inc(&blkif->refcnt);
 244
 245         if (blkif->irq) {
 246                 unbind_from_irqhandler(blkif->irq, blkif);
 247                 blkif->irq = 0;
 248         }
 249
 250         if (blkif->blk_rings.common.sring) {
 251                 xenbus_unmap_ring_vfree(blkif->be->dev, blkif->blk_ring);
 252                 blkif->blk_rings.common.sring = NULL;
 253         }
 254 }
 255
 256 static void xen_blkif_free(struct xen_blkif *blkif)
 257 {
 258         struct pending_req *req, *n;
 259         int i = 0, j;
 260
 261         if (!atomic_dec_and_test(&blkif->refcnt))
 262                 BUG();
 263
 264         /* Remove all persistent grants and the cache of ballooned pages. */
 265         xen_blkbk_free_caches(blkif);
 266
 267         /* Make sure everything is drained before shutting down */
 268         BUG_ON(blkif->persistent_gnt_c != 0);
 269         BUG_ON(atomic_read(&blkif->persistent_gnt_in_use) != 0);
 270         BUG_ON(blkif->free_pages_num != 0);
 271         BUG_ON(!list_empty(&blkif->persistent_purge_list));
 272         BUG_ON(!list_empty(&blkif->free_pages));
 273         BUG_ON(!RB_EMPTY_ROOT(&blkif->persistent_gnts));
 274
 275         /* Check that there is no request in use */
 276         list_for_each_entry_safe(req, n, &blkif->pending_free, free_list) {
 277                 list_del(&req->free_list);
 278
 279                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
 280                         kfree(req->segments[j]);
 281
 282                 for (j = 0; j < MAX_INDIRECT_PAGES; j++)
 283                         kfree(req->indirect_pages[j]);
 284
 285                 kfree(req);
 286                 i++;
 287         }
 288
 289         WARN_ON(i != XEN_BLKIF_REQS);
 290
 291         kmem_cache_free(xen_blkif_cachep, blkif);
 292 }
 293
 294 int __init xen_blkif_interface_init(void)
 295 {
 296         xen_blkif_cachep = kmem_cache_create("blkif_cache",
 297                                              sizeof(struct xen_blkif),
 298                                              0, 0, NULL);
 299         if (!xen_blkif_cachep)
 300                 return -ENOMEM;
 301
 302         return 0;
 303 }
 304
 305 /*
 306  *  sysfs interface for VBD I/O requests
 307  */
 308
 309 #define VBD_SHOW(name, format, args...)                                 \
 310         static ssize_t show_##name(struct device *_dev,                 \
 311                                    struct device_attribute *attr,       \
 312                                    char *buf)                           \
 313         {                                                               \
 314                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 315                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 316                                                                         \
 317                 return sprintf(buf, format, ##args);                    \
 318         }                                                               \
 319         static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 320
 321 VBD_SHOW(oo_req,  "%llu\n", be->blkif->st_oo_req);
 322 VBD_SHOW(rd_req,  "%llu\n", be->blkif->st_rd_req);
 323 VBD_SHOW(wr_req,  "%llu\n", be->blkif->st_wr_req);
 324 VBD_SHOW(f_req,  "%llu\n", be->blkif->st_f_req);
 325 VBD_SHOW(ds_req,  "%llu\n", be->blkif->st_ds_req);
 326 VBD_SHOW(rd_sect, "%llu\n", be->blkif->st_rd_sect);
 327 VBD_SHOW(wr_sect, "%llu\n", be->blkif->st_wr_sect);
 328
 329 static struct attribute *xen_vbdstat_attrs[] = {
 330         &dev_attr_oo_req.attr,
 331         &dev_attr_rd_req.attr,
 332         &dev_attr_wr_req.attr,
 333         &dev_attr_f_req.attr,
 334         &dev_attr_ds_req.attr,
 335         &dev_attr_rd_sect.attr,
 336         &dev_attr_wr_sect.attr,
 337         NULL
 338 };
 339
 340 static struct attribute_group xen_vbdstat_group = {
 341         .name = "statistics",
 342         .attrs = xen_vbdstat_attrs,
 343 };
 344
 345 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
 346 VBD_SHOW(mode, "%s\n", be->mode);
 347
 348 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
 349 {
 350         int error;
 351
 352         error = device_create_file(&dev->dev, &dev_attr_physical_device);
 353         if (error)
 354                 goto fail1;
 355
 356         error = device_create_file(&dev->dev, &dev_attr_mode);
 357         if (error)
 358                 goto fail2;
 359
 360         error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
 361         if (error)
 362                 goto fail3;
 363
 364         return 0;
 365
 366 fail3:  sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 367 fail2:  device_remove_file(&dev->dev, &dev_attr_mode);
 368 fail1:  device_remove_file(&dev->dev, &dev_attr_physical_device);
 369         return error;
 370 }
 371
 372 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
 373 {
 374         sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 375         device_remove_file(&dev->dev, &dev_attr_mode);
 376         device_remove_file(&dev->dev, &dev_attr_physical_device);
 377 }
 378
 379
 380 static void xen_vbd_free(struct xen_vbd *vbd)
 381 {
 382         if (vbd->bdev)
 383                 blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
 384         vbd->bdev = NULL;
 385 }
 386
 387 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
 388                           unsigned major, unsigned minor, int readonly,
 389                           int cdrom)
 390 {
 391         struct xen_vbd *vbd;
 392         struct block_device *bdev;
 393         struct request_queue *q;
 394
 395         vbd = &blkif->vbd;
 396         vbd->handle   = handle;
 397         vbd->readonly = readonly;
 398         vbd->type     = 0;
 399
 400         vbd->pdevice  = MKDEV(major, minor);
 401
 402         bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
 403                                  FMODE_READ : FMODE_WRITE, NULL);
 404
 405         if (IS_ERR(bdev)) {
 406                 DPRINTK("xen_vbd_create: device %08x could not be opened.\n",
 407                         vbd->pdevice);
 408                 return -ENOENT;
 409         }
 410
 411         vbd->bdev = bdev;
 412         if (vbd->bdev->bd_disk == NULL) {
 413                 DPRINTK("xen_vbd_create: device %08x doesn't exist.\n",
 414                         vbd->pdevice);
 415                 xen_vbd_free(vbd);
 416                 return -ENOENT;
 417         }
 418         vbd->size = vbd_sz(vbd);
 419
 420         if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
 421                 vbd->type |= VDISK_CDROM;
 422         if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
 423                 vbd->type |= VDISK_REMOVABLE;
 424
 425         q = bdev_get_queue(bdev);
 426         if (q && q->flush_flags)
 427                 vbd->flush_support = true;
 428
 429         if (q && blk_queue_secdiscard(q))
 430                 vbd->discard_secure = true;
 431
 432         DPRINTK("Successful creation of handle=%04x (dom=%u)\n",
 433                 handle, blkif->domid);
 434         return 0;
 435 }
 436 static int xen_blkbk_remove(struct xenbus_device *dev)
 437 {
 438         struct backend_info *be = dev_get_drvdata(&dev->dev);
 439
 440         DPRINTK("");
 441
 442         if (be->major || be->minor)
 443                 xenvbd_sysfs_delif(dev);
 444
 445         if (be->backend_watch.node) {
 446                 unregister_xenbus_watch(&be->backend_watch);
 447                 kfree(be->backend_watch.node);
 448                 be->backend_watch.node = NULL;
 449         }
 450
 451         if (be->blkif) {
 452                 xen_blkif_disconnect(be->blkif);
 453                 xen_vbd_free(&be->blkif->vbd);
 454                 xen_blkif_free(be->blkif);
 455                 be->blkif = NULL;
 456         }
 457
 458         kfree(be->mode);
 459         kfree(be);
 460         dev_set_drvdata(&dev->dev, NULL);
 461         return 0;
 462 }
 463
 464 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
 465                               struct backend_info *be, int state)
 466 {
 467         struct xenbus_device *dev = be->dev;
 468         int err;
 469
 470         err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
 471                             "%d", state);
 472         if (err)
 473                 dev_warn(&dev->dev, "writing feature-flush-cache (%d)", err);
 474
 475         return err;
 476 }
 477
 478 static void xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
 479 {
 480         struct xenbus_device *dev = be->dev;
 481         struct xen_blkif *blkif = be->blkif;
 482         int err;
 483         int state = 0;
 484         struct block_device *bdev = be->blkif->vbd.bdev;
 485         struct request_queue *q = bdev_get_queue(bdev);
 486
 487         if (blk_queue_discard(q)) {
 488                 err = xenbus_printf(xbt, dev->nodename,
 489                         "discard-granularity", "%u",
 490                         q->limits.discard_granularity);
 491                 if (err) {
 492                         dev_warn(&dev->dev, "writing discard-granularity (%d)", err);
 493                         return;
 494                 }
 495                 err = xenbus_printf(xbt, dev->nodename,
 496                         "discard-alignment", "%u",
 497                         q->limits.discard_alignment);
 498                 if (err) {
 499                         dev_warn(&dev->dev, "writing discard-alignment (%d)", err);
 500                         return;
 501                 }
 502                 state = 1;
 503                 /* Optional. */
 504                 err = xenbus_printf(xbt, dev->nodename,
 505                                     "discard-secure", "%d",
 506                                     blkif->vbd.discard_secure);
 507                 if (err) {
 508                         dev_warn(&dev->dev, "writing discard-secure (%d)", err);
 509                         return;
 510                 }
 511         }
 512         err = xenbus_printf(xbt, dev->nodename, "feature-discard",
 513                             "%d", state);
 514         if (err)
 515                 dev_warn(&dev->dev, "writing feature-discard (%d)", err);
 516 }
 517 int xen_blkbk_barrier(struct xenbus_transaction xbt,
 518                       struct backend_info *be, int state)
 519 {
 520         struct xenbus_device *dev = be->dev;
 521         int err;
 522
 523         err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
 524                             "%d", state);
 525         if (err)
 526                 dev_warn(&dev->dev, "writing feature-barrier (%d)", err);
 527
 528         return err;
 529 }
 530
 531 /*
 532  * Entry point to this code when a new device is created.  Allocate the basic
 533  * structures, and watch the store waiting for the hotplug scripts to tell us
 534  * the device's physical major and minor numbers.  Switch to InitWait.
 535  */
 536 static int xen_blkbk_probe(struct xenbus_device *dev,
 537                            const struct xenbus_device_id *id)
 538 {
 539         int err;
 540         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 541                                           GFP_KERNEL);
 542         if (!be) {
 543                 xenbus_dev_fatal(dev, -ENOMEM,
 544                                  "allocating backend structure");
 545                 return -ENOMEM;
 546         }
 547         be->dev = dev;
 548         dev_set_drvdata(&dev->dev, be);
 549
 550         be->blkif = xen_blkif_alloc(dev->otherend_id);
 551         if (IS_ERR(be->blkif)) {
 552                 err = PTR_ERR(be->blkif);
 553                 be->blkif = NULL;
 554                 xenbus_dev_fatal(dev, err, "creating block interface");
 555                 goto fail;
 556         }
 557
 558         /* setup back pointer */
 559         be->blkif->be = be;
 560
 561         err = xenbus_watch_pathfmt(dev, &be->backend_watch, backend_changed,
 562                                    "%s/%s", dev->nodename, "physical-device");
 563         if (err)
 564                 goto fail;
 565
 566         err = xenbus_switch_state(dev, XenbusStateInitWait);
 567         if (err)
 568                 goto fail;
 569
 570         return 0;
 571
 572 fail:
 573         DPRINTK("failed");
 574         xen_blkbk_remove(dev);
 575         return err;
 576 }
 577
 578
 579 /*
 580  * Callback received when the hotplug scripts have placed the physical-device
 581  * node.  Read it and the mode node, and create a vbd.  If the frontend is
 582  * ready, connect.
 583  */
 584 static void backend_changed(struct xenbus_watch *watch,
 585                             const char **vec, unsigned int len)
 586 {
 587         int err;
 588         unsigned major;
 589         unsigned minor;
 590         struct backend_info *be
 591                 = container_of(watch, struct backend_info, backend_watch);
 592         struct xenbus_device *dev = be->dev;
 593         int cdrom = 0;
 594         unsigned long handle;
 595         char *device_type;
 596
 597         DPRINTK("");
 598
 599         err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
 600                            &major, &minor);
 601         if (XENBUS_EXIST_ERR(err)) {
 602                 /*
 603                  * Since this watch will fire once immediately after it is
 604                  * registered, we expect this.  Ignore it, and wait for the
 605                  * hotplug scripts.
 606                  */
 607                 return;
 608         }
 609         if (err != 2) {
 610                 xenbus_dev_fatal(dev, err, "reading physical-device");
 611                 return;
 612         }
 613
 614         if (be->major | be->minor) {
 615                 if (be->major != major || be->minor != minor)
 616                         pr_warn(DRV_PFX "changing physical device (from %x:%x to %x:%x) not supported.\n",
 617                                 be->major, be->minor, major, minor);
 618                 return;
 619         }
 620
 621         be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
 622         if (IS_ERR(be->mode)) {
 623                 err = PTR_ERR(be->mode);
 624                 be->mode = NULL;
 625                 xenbus_dev_fatal(dev, err, "reading mode");
 626                 return;
 627         }
 628
 629         device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
 630         if (!IS_ERR(device_type)) {
 631                 cdrom = strcmp(device_type, "cdrom") == 0;
 632                 kfree(device_type);
 633         }
 634
 635         /* Front end dir is a number, which is used as the handle. */
 636         err = kstrtoul(strrchr(dev->otherend, '/') + 1, 0, &handle);
 637         if (err)
 638                 return;
 639
 640         be->major = major;
 641         be->minor = minor;
 642
 643         err = xen_vbd_create(be->blkif, handle, major, minor,
 644                              !strchr(be->mode, 'w'), cdrom);
 645
 646         if (err)
 647                 xenbus_dev_fatal(dev, err, "creating vbd structure");
 648         else {
 649                 err = xenvbd_sysfs_addif(dev);
 650                 if (err) {
 651                         xen_vbd_free(&be->blkif->vbd);
 652                         xenbus_dev_fatal(dev, err, "creating sysfs entries");
 653                 }
 654         }
 655
 656         if (err) {
 657                 kfree(be->mode);
 658                 be->mode = NULL;
 659                 be->major = 0;
 660                 be->minor = 0;
 661         } else {
 662                 /* We're potentially connected now */
 663                 xen_update_blkif_status(be->blkif);
 664         }
 665 }
 666
 667
 668 /*
 669  * Callback received when the frontend's state changes.
 670  */
 671 static void frontend_changed(struct xenbus_device *dev,
 672                              enum xenbus_state frontend_state)
 673 {
 674         struct backend_info *be = dev_get_drvdata(&dev->dev);
 675         int err;
 676
 677         DPRINTK("%s", xenbus_strstate(frontend_state));
 678
 679         switch (frontend_state) {
 680         case XenbusStateInitialising:
 681                 if (dev->state == XenbusStateClosed) {
 682                         pr_info(DRV_PFX "%s: prepare for reconnect\n",
 683                                 dev->nodename);
 684                         xenbus_switch_state(dev, XenbusStateInitWait);
 685                 }
 686                 break;
 687
 688         case XenbusStateInitialised:
 689         case XenbusStateConnected:
 690                 /*
 691                  * Ensure we connect even when two watches fire in
 692                  * close succession and we miss the intermediate value
 693                  * of frontend_state.
 694                  */
 695                 if (dev->state == XenbusStateConnected)
 696                         break;
 697
 698                 /*
 699                  * Enforce precondition before potential leak point.
 700                  * xen_blkif_disconnect() is idempotent.
 701                  */
 702                 xen_blkif_disconnect(be->blkif);
 703
 704                 err = connect_ring(be);
 705                 if (err)
 706                         break;
 707                 xen_update_blkif_status(be->blkif);
 708                 break;
 709
 710         case XenbusStateClosing:
 711                 xenbus_switch_state(dev, XenbusStateClosing);
 712                 break;
 713
 714         case XenbusStateClosed:
 715                 xen_blkif_disconnect(be->blkif);
 716                 xenbus_switch_state(dev, XenbusStateClosed);
 717                 if (xenbus_dev_is_online(dev))
 718                         break;
 719                 /* fall through if not online */
 720         case XenbusStateUnknown:
 721                 /* implies xen_blkif_disconnect() via xen_blkbk_remove() */
 722                 device_unregister(&dev->dev);
 723                 break;
 724
 725         default:
 726                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 727                                  frontend_state);
 728                 break;
 729         }
 730 }
 731
 732
 733 /* ** Connection ** */
 734
 735
 736 /*
 737  * Write the physical details regarding the block device to the store, and
 738  * switch to Connected state.
 739  */
 740 static void connect(struct backend_info *be)
 741 {
 742         struct xenbus_transaction xbt;
 743         int err;
 744         struct xenbus_device *dev = be->dev;
 745
 746         DPRINTK("%s", dev->otherend);
 747
 748         /* Supply the information about the device the frontend needs */
 749 again:
 750         err = xenbus_transaction_start(&xbt);
 751         if (err) {
 752                 xenbus_dev_fatal(dev, err, "starting transaction");
 753                 return;
 754         }
 755
 756         /* If we can't advertise it is OK. */
 757         xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
 758
 759         xen_blkbk_discard(xbt, be);
 760
 761         xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
 762
 763         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u", 1);
 764         if (err) {
 765                 xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
 766                                  dev->nodename);
 767                 goto abort;
 768         }
 769         err = xenbus_printf(xbt, dev->nodename, "feature-max-indirect-segments", "%u",
 770                             MAX_INDIRECT_SEGMENTS);
 771         if (err)
 772                 dev_warn(&dev->dev, "writing %s/feature-max-indirect-segments (%d)",
 773                          dev->nodename, err);
 774
 775         err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
 776                             (unsigned long long)vbd_sz(&be->blkif->vbd));
 777         if (err) {
 778                 xenbus_dev_fatal(dev, err, "writing %s/sectors",
 779                                  dev->nodename);
 780                 goto abort;
 781         }
 782
 783         /* FIXME: use a typename instead */
 784         err = xenbus_printf(xbt, dev->nodename, "info", "%u",
 785                             be->blkif->vbd.type |
 786                             (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
 787         if (err) {
 788                 xenbus_dev_fatal(dev, err, "writing %s/info",
 789                                  dev->nodename);
 790                 goto abort;
 791         }
 792         err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
 793                             (unsigned long)
 794                             bdev_logical_block_size(be->blkif->vbd.bdev));
 795         if (err) {
 796                 xenbus_dev_fatal(dev, err, "writing %s/sector-size",
 797                                  dev->nodename);
 798                 goto abort;
 799         }
 800         err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
 801                             bdev_physical_block_size(be->blkif->vbd.bdev));
 802         if (err)
 803                 xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
 804                                  dev->nodename);
 805
 806         err = xenbus_transaction_end(xbt, 0);
 807         if (err == -EAGAIN)
 808                 goto again;
 809         if (err)
 810                 xenbus_dev_fatal(dev, err, "ending transaction");
 811
 812         err = xenbus_switch_state(dev, XenbusStateConnected);
 813         if (err)
 814                 xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
 815                                  dev->nodename);
 816
 817         return;
 818  abort:
 819         xenbus_transaction_end(xbt, 1);
 820 }
 821
 822
 823 static int connect_ring(struct backend_info *be)
 824 {
 825         struct xenbus_device *dev = be->dev;
 826         unsigned long ring_ref;
 827         unsigned int evtchn;
 828         unsigned int pers_grants;
 829         char protocol[64] = "";
 830         int err;
 831
 832         DPRINTK("%s", dev->otherend);
 833
 834         err = xenbus_gather(XBT_NIL, dev->otherend, "ring-ref", "%lu",
 835                             &ring_ref, "event-channel", "%u", &evtchn, NULL);
 836         if (err) {
 837                 xenbus_dev_fatal(dev, err,
 838                                  "reading %s/ring-ref and event-channel",
 839                                  dev->otherend);
 840                 return err;
 841         }
 842
 843         be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
 844         err = xenbus_gather(XBT_NIL, dev->otherend, "protocol",
 845                             "%63s", protocol, NULL);
 846         if (err)
 847                 strcpy(protocol, "unspecified, assuming native");
 848         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
 849                 be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
 850         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
 851                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
 852         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
 853                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
 854         else {
 855                 xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
 856                 return -1;
 857         }
 858         err = xenbus_gather(XBT_NIL, dev->otherend,
 859                             "feature-persistent", "%u",
 860                             &pers_grants, NULL);
 861         if (err)
 862                 pers_grants = 0;
 863
 864         be->blkif->vbd.feature_gnt_persistent = pers_grants;
 865         be->blkif->vbd.overflow_max_grants = 0;
 866
 867         pr_info(DRV_PFX "ring-ref %ld, event-channel %d, protocol %d (%s) %s\n",
 868                 ring_ref, evtchn, be->blkif->blk_protocol, protocol,
 869                 pers_grants ? "persistent grants" : "");
 870
 871         /* Map the shared frame, irq etc. */
 872         err = xen_blkif_map(be->blkif, ring_ref, evtchn);
 873         if (err) {
 874                 xenbus_dev_fatal(dev, err, "mapping ring-ref %lu port %u",
 875                                  ring_ref, evtchn);
 876                 return err;
 877         }
 878
 879         return 0;
 880 }
 881
 882
 883 /* ** Driver Registration ** */
 884
 885
 886 static const struct xenbus_device_id xen_blkbk_ids[] = {
 887         { "vbd" },
 888         { "" }
 889 };
 890
 891
 892 static DEFINE_XENBUS_DRIVER(xen_blkbk, ,
 893         .probe = xen_blkbk_probe,
 894         .remove = xen_blkbk_remove,
 895         .otherend_changed = frontend_changed
 896 );
 897
 898
 899 int xen_blkif_xenbus_init(void)
 900 {
 901         return xenbus_register_backend(&xen_blkbk_driver);
 902 }