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