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