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