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