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 #define pr_fmt(fmt) "xen-blkback: " fmt
  18
  19 #include <stdarg.h>
  20 #include <linux/module.h>
  21 #include <linux/kthread.h>
  22 #include <xen/events.h>
  23 #include <xen/grant_table.h>
  24 #include "common.h"
  25
  26 /* On the XenBus the max length of 'ring-ref%u'. */
  27 #define RINGREF_NAME_LEN (20)
  28
  29 struct backend_info {
  30         struct xenbus_device    *dev;
  31         struct xen_blkif        *blkif;
  32         struct xenbus_watch     backend_watch;
  33         unsigned                major;
  34         unsigned                minor;
  35         char                    *mode;
  36 };
  37
  38 static struct kmem_cache *xen_blkif_cachep;
  39 static void connect(struct backend_info *);
  40 static int connect_ring(struct backend_info *);
  41 static void backend_changed(struct xenbus_watch *, const char *,
  42                             const char *);
  43 static void xen_blkif_free(struct xen_blkif *blkif);
  44 static void xen_vbd_free(struct xen_vbd *vbd);
  45
  46 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be)
  47 {
  48         return be->dev;
  49 }
  50
  51 /*
  52  * The last request could free the device from softirq context and
  53  * xen_blkif_free() can sleep.
  54  */
  55 static void xen_blkif_deferred_free(struct work_struct *work)
  56 {
  57         struct xen_blkif *blkif;
  58
  59         blkif = container_of(work, struct xen_blkif, free_work);
  60         xen_blkif_free(blkif);
  61 }
  62
  63 static int blkback_name(struct xen_blkif *blkif, char *buf)
  64 {
  65         char *devpath, *devname;
  66         struct xenbus_device *dev = blkif->be->dev;
  67
  68         devpath = xenbus_read(XBT_NIL, dev->nodename, "dev", NULL);
  69         if (IS_ERR(devpath))
  70                 return PTR_ERR(devpath);
  71
  72         devname = strstr(devpath, "/dev/");
  73         if (devname != NULL)
  74                 devname += strlen("/dev/");
  75         else
  76                 devname  = devpath;
  77
  78         snprintf(buf, TASK_COMM_LEN, "%d.%s", blkif->domid, devname);
  79         kfree(devpath);
  80
  81         return 0;
  82 }
  83
  84 static void xen_update_blkif_status(struct xen_blkif *blkif)
  85 {
  86         int err;
  87         char name[TASK_COMM_LEN];
  88         struct xen_blkif_ring *ring;
  89         int i;
  90
  91         /* Not ready to connect? */
  92         if (!blkif->rings || !blkif->rings[0].irq || !blkif->vbd.bdev)
  93                 return;
  94
  95         /* Already connected? */
  96         if (blkif->be->dev->state == XenbusStateConnected)
  97                 return;
  98
  99         /* Attempt to connect: exit if we fail to. */
 100         connect(blkif->be);
 101         if (blkif->be->dev->state != XenbusStateConnected)
 102                 return;
 103
 104         err = blkback_name(blkif, name);
 105         if (err) {
 106                 xenbus_dev_error(blkif->be->dev, err, "get blkback dev name");
 107                 return;
 108         }
 109
 110         err = filemap_write_and_wait(blkif->vbd.bdev->bd_inode->i_mapping);
 111         if (err) {
 112                 xenbus_dev_error(blkif->be->dev, err, "block flush");
 113                 return;
 114         }
 115         invalidate_inode_pages2(blkif->vbd.bdev->bd_inode->i_mapping);
 116
 117         for (i = 0; i < blkif->nr_rings; i++) {
 118                 ring = &blkif->rings[i];
 119                 ring->xenblkd = kthread_run(xen_blkif_schedule, ring, "%s-%d", name, i);
 120                 if (IS_ERR(ring->xenblkd)) {
 121                         err = PTR_ERR(ring->xenblkd);
 122                         ring->xenblkd = NULL;
 123                         xenbus_dev_fatal(blkif->be->dev, err,
 124                                         "start %s-%d xenblkd", name, i);
 125                         goto out;
 126                 }
 127         }
 128         return;
 129
 130 out:
 131         while (--i >= 0) {
 132                 ring = &blkif->rings[i];
 133                 kthread_stop(ring->xenblkd);
 134         }
 135         return;
 136 }
 137
 138 static int xen_blkif_alloc_rings(struct xen_blkif *blkif)
 139 {
 140         unsigned int r;
 141
 142         blkif->rings = kzalloc(blkif->nr_rings * sizeof(struct xen_blkif_ring), GFP_KERNEL);
 143         if (!blkif->rings)
 144                 return -ENOMEM;
 145
 146         for (r = 0; r < blkif->nr_rings; r++) {
 147                 struct xen_blkif_ring *ring = &blkif->rings[r];
 148
 149                 spin_lock_init(&ring->blk_ring_lock);
 150                 init_waitqueue_head(&ring->wq);
 151                 INIT_LIST_HEAD(&ring->pending_free);
 152                 INIT_LIST_HEAD(&ring->persistent_purge_list);
 153                 INIT_WORK(&ring->persistent_purge_work, xen_blkbk_unmap_purged_grants);
 154                 spin_lock_init(&ring->free_pages_lock);
 155                 INIT_LIST_HEAD(&ring->free_pages);
 156
 157                 spin_lock_init(&ring->pending_free_lock);
 158                 init_waitqueue_head(&ring->pending_free_wq);
 159                 init_waitqueue_head(&ring->shutdown_wq);
 160                 ring->blkif = blkif;
 161                 ring->st_print = jiffies;
 162                 ring->active = true;
 163         }
 164
 165         return 0;
 166 }
 167
 168 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
 169 {
 170         struct xen_blkif *blkif;
 171
 172         BUILD_BUG_ON(MAX_INDIRECT_PAGES > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
 173
 174         blkif = kmem_cache_zalloc(xen_blkif_cachep, GFP_KERNEL);
 175         if (!blkif)
 176                 return ERR_PTR(-ENOMEM);
 177
 178         blkif->domid = domid;
 179         atomic_set(&blkif->refcnt, 1);
 180         init_completion(&blkif->drain_complete);
 181         INIT_WORK(&blkif->free_work, xen_blkif_deferred_free);
 182
 183         return blkif;
 184 }
 185
 186 static int xen_blkif_map(struct xen_blkif_ring *ring, grant_ref_t *gref,
 187                          unsigned int nr_grefs, unsigned int evtchn)
 188 {
 189         int err;
 190         struct xen_blkif *blkif = ring->blkif;
 191
 192         /* Already connected through? */
 193         if (ring->irq)
 194                 return 0;
 195
 196         err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
 197                                      &ring->blk_ring);
 198         if (err < 0)
 199                 return err;
 200
 201         switch (blkif->blk_protocol) {
 202         case BLKIF_PROTOCOL_NATIVE:
 203         {
 204                 struct blkif_sring *sring;
 205                 sring = (struct blkif_sring *)ring->blk_ring;
 206                 BACK_RING_INIT(&ring->blk_rings.native, sring,
 207                                XEN_PAGE_SIZE * nr_grefs);
 208                 break;
 209         }
 210         case BLKIF_PROTOCOL_X86_32:
 211         {
 212                 struct blkif_x86_32_sring *sring_x86_32;
 213                 sring_x86_32 = (struct blkif_x86_32_sring *)ring->blk_ring;
 214                 BACK_RING_INIT(&ring->blk_rings.x86_32, sring_x86_32,
 215                                XEN_PAGE_SIZE * nr_grefs);
 216                 break;
 217         }
 218         case BLKIF_PROTOCOL_X86_64:
 219         {
 220                 struct blkif_x86_64_sring *sring_x86_64;
 221                 sring_x86_64 = (struct blkif_x86_64_sring *)ring->blk_ring;
 222                 BACK_RING_INIT(&ring->blk_rings.x86_64, sring_x86_64,
 223                                XEN_PAGE_SIZE * nr_grefs);
 224                 break;
 225         }
 226         default:
 227                 BUG();
 228         }
 229
 230         err = bind_interdomain_evtchn_to_irqhandler(blkif->domid, evtchn,
 231                                                     xen_blkif_be_int, 0,
 232                                                     "blkif-backend", ring);
 233         if (err < 0) {
 234                 xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
 235                 ring->blk_rings.common.sring = NULL;
 236                 return err;
 237         }
 238         ring->irq = err;
 239
 240         return 0;
 241 }
 242
 243 static int xen_blkif_disconnect(struct xen_blkif *blkif)
 244 {
 245         struct pending_req *req, *n;
 246         unsigned int j, r;
 247
 248         for (r = 0; r < blkif->nr_rings; r++) {
 249                 struct xen_blkif_ring *ring = &blkif->rings[r];
 250                 unsigned int i = 0;
 251
 252                 if (!ring->active)
 253                         continue;
 254
 255                 if (ring->xenblkd) {
 256                         kthread_stop(ring->xenblkd);
 257                         wake_up(&ring->shutdown_wq);
 258                 }
 259
 260                 /* The above kthread_stop() guarantees that at this point we
 261                  * don't have any discard_io or other_io requests. So, checking
 262                  * for inflight IO is enough.
 263                  */
 264                 if (atomic_read(&ring->inflight) > 0)
 265                         return -EBUSY;
 266
 267                 if (ring->irq) {
 268                         unbind_from_irqhandler(ring->irq, ring);
 269                         ring->irq = 0;
 270                 }
 271
 272                 if (ring->blk_rings.common.sring) {
 273                         xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
 274                         ring->blk_rings.common.sring = NULL;
 275                 }
 276
 277                 /* Remove all persistent grants and the cache of ballooned pages. */
 278                 xen_blkbk_free_caches(ring);
 279
 280                 /* Check that there is no request in use */
 281                 list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
 282                         list_del(&req->free_list);
 283
 284                         for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
 285                                 kfree(req->segments[j]);
 286
 287                         for (j = 0; j < MAX_INDIRECT_PAGES; j++)
 288                                 kfree(req->indirect_pages[j]);
 289
 290                         kfree(req);
 291                         i++;
 292                 }
 293
 294                 BUG_ON(atomic_read(&ring->persistent_gnt_in_use) != 0);
 295                 BUG_ON(!list_empty(&ring->persistent_purge_list));
 296                 BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts));
 297                 BUG_ON(!list_empty(&ring->free_pages));
 298                 BUG_ON(ring->free_pages_num != 0);
 299                 BUG_ON(ring->persistent_gnt_c != 0);
 300                 WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
 301                 ring->active = false;
 302         }
 303         blkif->nr_ring_pages = 0;
 304         /*
 305          * blkif->rings was allocated in connect_ring, so we should free it in
 306          * here.
 307          */
 308         kfree(blkif->rings);
 309         blkif->rings = NULL;
 310         blkif->nr_rings = 0;
 311
 312         return 0;
 313 }
 314
 315 static void xen_blkif_free(struct xen_blkif *blkif)
 316 {
 317         WARN_ON(xen_blkif_disconnect(blkif));
 318         xen_vbd_free(&blkif->vbd);
 319         kfree(blkif->be->mode);
 320         kfree(blkif->be);
 321
 322         /* Make sure everything is drained before shutting down */
 323         kmem_cache_free(xen_blkif_cachep, blkif);
 324 }
 325
 326 int __init xen_blkif_interface_init(void)
 327 {
 328         xen_blkif_cachep = kmem_cache_create("blkif_cache",
 329                                              sizeof(struct xen_blkif),
 330                                              0, 0, NULL);
 331         if (!xen_blkif_cachep)
 332                 return -ENOMEM;
 333
 334         return 0;
 335 }
 336
 337 /*
 338  *  sysfs interface for VBD I/O requests
 339  */
 340
 341 #define VBD_SHOW_ALLRING(name, format)                                  \
 342         static ssize_t show_##name(struct device *_dev,                 \
 343                                    struct device_attribute *attr,       \
 344                                    char *buf)                           \
 345         {                                                               \
 346                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 347                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 348                 struct xen_blkif *blkif = be->blkif;                    \
 349                 unsigned int i;                                         \
 350                 unsigned long long result = 0;                          \
 351                                                                         \
 352                 if (!blkif->rings)                              \
 353                         goto out;                                       \
 354                                                                         \
 355                 for (i = 0; i < blkif->nr_rings; i++) {         \
 356                         struct xen_blkif_ring *ring = &blkif->rings[i]; \
 357                                                                         \
 358                         result += ring->st_##name;                      \
 359                 }                                                       \
 360                                                                         \
 361 out:                                                                    \
 362                 return sprintf(buf, format, result);                    \
 363         }                                                               \
 364         static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 365
 366 VBD_SHOW_ALLRING(oo_req,  "%llu\n");
 367 VBD_SHOW_ALLRING(rd_req,  "%llu\n");
 368 VBD_SHOW_ALLRING(wr_req,  "%llu\n");
 369 VBD_SHOW_ALLRING(f_req,  "%llu\n");
 370 VBD_SHOW_ALLRING(ds_req,  "%llu\n");
 371 VBD_SHOW_ALLRING(rd_sect, "%llu\n");
 372 VBD_SHOW_ALLRING(wr_sect, "%llu\n");
 373
 374 static struct attribute *xen_vbdstat_attrs[] = {
 375         &dev_attr_oo_req.attr,
 376         &dev_attr_rd_req.attr,
 377         &dev_attr_wr_req.attr,
 378         &dev_attr_f_req.attr,
 379         &dev_attr_ds_req.attr,
 380         &dev_attr_rd_sect.attr,
 381         &dev_attr_wr_sect.attr,
 382         NULL
 383 };
 384
 385 static const struct attribute_group xen_vbdstat_group = {
 386         .name = "statistics",
 387         .attrs = xen_vbdstat_attrs,
 388 };
 389
 390 #define VBD_SHOW(name, format, args...)                                 \
 391         static ssize_t show_##name(struct device *_dev,                 \
 392                                    struct device_attribute *attr,       \
 393                                    char *buf)                           \
 394         {                                                               \
 395                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 396                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 397                                                                         \
 398                 return sprintf(buf, format, ##args);                    \
 399         }                                                               \
 400         static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 401
 402 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
 403 VBD_SHOW(mode, "%s\n", be->mode);
 404
 405 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
 406 {
 407         int error;
 408
 409         error = device_create_file(&dev->dev, &dev_attr_physical_device);
 410         if (error)
 411                 goto fail1;
 412
 413         error = device_create_file(&dev->dev, &dev_attr_mode);
 414         if (error)
 415                 goto fail2;
 416
 417         error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
 418         if (error)
 419                 goto fail3;
 420
 421         return 0;
 422
 423 fail3:  sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 424 fail2:  device_remove_file(&dev->dev, &dev_attr_mode);
 425 fail1:  device_remove_file(&dev->dev, &dev_attr_physical_device);
 426         return error;
 427 }
 428
 429 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
 430 {
 431         sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 432         device_remove_file(&dev->dev, &dev_attr_mode);
 433         device_remove_file(&dev->dev, &dev_attr_physical_device);
 434 }
 435
 436
 437 static void xen_vbd_free(struct xen_vbd *vbd)
 438 {
 439         if (vbd->bdev)
 440                 blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
 441         vbd->bdev = NULL;
 442 }
 443
 444 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
 445                           unsigned major, unsigned minor, int readonly,
 446                           int cdrom)
 447 {
 448         struct xen_vbd *vbd;
 449         struct block_device *bdev;
 450         struct request_queue *q;
 451
 452         vbd = &blkif->vbd;
 453         vbd->handle   = handle;
 454         vbd->readonly = readonly;
 455         vbd->type     = 0;
 456
 457         vbd->pdevice  = MKDEV(major, minor);
 458
 459         bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
 460                                  FMODE_READ : FMODE_WRITE, NULL);
 461
 462         if (IS_ERR(bdev)) {
 463                 pr_warn("xen_vbd_create: device %08x could not be opened\n",
 464                         vbd->pdevice);
 465                 return -ENOENT;
 466         }
 467
 468         vbd->bdev = bdev;
 469         if (vbd->bdev->bd_disk == NULL) {
 470                 pr_warn("xen_vbd_create: device %08x doesn't exist\n",
 471                         vbd->pdevice);
 472                 xen_vbd_free(vbd);
 473                 return -ENOENT;
 474         }
 475         vbd->size = vbd_sz(vbd);
 476
 477         if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
 478                 vbd->type |= VDISK_CDROM;
 479         if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
 480                 vbd->type |= VDISK_REMOVABLE;
 481
 482         q = bdev_get_queue(bdev);
 483         if (q && test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 484                 vbd->flush_support = true;
 485
 486         if (q && blk_queue_secure_erase(q))
 487                 vbd->discard_secure = true;
 488
 489         pr_debug("Successful creation of handle=%04x (dom=%u)\n",
 490                 handle, blkif->domid);
 491         return 0;
 492 }
 493 static int xen_blkbk_remove(struct xenbus_device *dev)
 494 {
 495         struct backend_info *be = dev_get_drvdata(&dev->dev);
 496
 497         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 498
 499         if (be->major || be->minor)
 500                 xenvbd_sysfs_delif(dev);
 501
 502         if (be->backend_watch.node) {
 503                 unregister_xenbus_watch(&be->backend_watch);
 504                 kfree(be->backend_watch.node);
 505                 be->backend_watch.node = NULL;
 506         }
 507
 508         dev_set_drvdata(&dev->dev, NULL);
 509
 510         if (be->blkif) {
 511                 xen_blkif_disconnect(be->blkif);
 512
 513                 /* Put the reference we set in xen_blkif_alloc(). */
 514                 xen_blkif_put(be->blkif);
 515         }
 516
 517         return 0;
 518 }
 519
 520 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
 521                               struct backend_info *be, int state)
 522 {
 523         struct xenbus_device *dev = be->dev;
 524         int err;
 525
 526         err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
 527                             "%d", state);
 528         if (err)
 529                 dev_warn(&dev->dev, "writing feature-flush-cache (%d)", err);
 530
 531         return err;
 532 }
 533
 534 static void xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
 535 {
 536         struct xenbus_device *dev = be->dev;
 537         struct xen_blkif *blkif = be->blkif;
 538         int err;
 539         int state = 0;
 540         struct block_device *bdev = be->blkif->vbd.bdev;
 541         struct request_queue *q = bdev_get_queue(bdev);
 542
 543         if (!xenbus_read_unsigned(dev->nodename, "discard-enable", 1))
 544                 return;
 545
 546         if (blk_queue_discard(q)) {
 547                 err = xenbus_printf(xbt, dev->nodename,
 548                         "discard-granularity", "%u",
 549                         q->limits.discard_granularity);
 550                 if (err) {
 551                         dev_warn(&dev->dev, "writing discard-granularity (%d)", err);
 552                         return;
 553                 }
 554                 err = xenbus_printf(xbt, dev->nodename,
 555                         "discard-alignment", "%u",
 556                         q->limits.discard_alignment);
 557                 if (err) {
 558                         dev_warn(&dev->dev, "writing discard-alignment (%d)", err);
 559                         return;
 560                 }
 561                 state = 1;
 562                 /* Optional. */
 563                 err = xenbus_printf(xbt, dev->nodename,
 564                                     "discard-secure", "%d",
 565                                     blkif->vbd.discard_secure);
 566                 if (err) {
 567                         dev_warn(&dev->dev, "writing discard-secure (%d)", err);
 568                         return;
 569                 }
 570         }
 571         err = xenbus_printf(xbt, dev->nodename, "feature-discard",
 572                             "%d", state);
 573         if (err)
 574                 dev_warn(&dev->dev, "writing feature-discard (%d)", err);
 575 }
 576 int xen_blkbk_barrier(struct xenbus_transaction xbt,
 577                       struct backend_info *be, int state)
 578 {
 579         struct xenbus_device *dev = be->dev;
 580         int err;
 581
 582         err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
 583                             "%d", state);
 584         if (err)
 585                 dev_warn(&dev->dev, "writing feature-barrier (%d)", err);
 586
 587         return err;
 588 }
 589
 590 /*
 591  * Entry point to this code when a new device is created.  Allocate the basic
 592  * structures, and watch the store waiting for the hotplug scripts to tell us
 593  * the device's physical major and minor numbers.  Switch to InitWait.
 594  */
 595 static int xen_blkbk_probe(struct xenbus_device *dev,
 596                            const struct xenbus_device_id *id)
 597 {
 598         int err;
 599         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 600                                           GFP_KERNEL);
 601
 602         /* match the pr_debug in xen_blkbk_remove */
 603         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 604
 605         if (!be) {
 606                 xenbus_dev_fatal(dev, -ENOMEM,
 607                                  "allocating backend structure");
 608                 return -ENOMEM;
 609         }
 610         be->dev = dev;
 611         dev_set_drvdata(&dev->dev, be);
 612
 613         be->blkif = xen_blkif_alloc(dev->otherend_id);
 614         if (IS_ERR(be->blkif)) {
 615                 err = PTR_ERR(be->blkif);
 616                 be->blkif = NULL;
 617                 xenbus_dev_fatal(dev, err, "creating block interface");
 618                 goto fail;
 619         }
 620
 621         err = xenbus_printf(XBT_NIL, dev->nodename,
 622                             "feature-max-indirect-segments", "%u",
 623                             MAX_INDIRECT_SEGMENTS);
 624         if (err)
 625                 dev_warn(&dev->dev,
 626                          "writing %s/feature-max-indirect-segments (%d)",
 627                          dev->nodename, err);
 628
 629         /* Multi-queue: advertise how many queues are supported by us.*/
 630         err = xenbus_printf(XBT_NIL, dev->nodename,
 631                             "multi-queue-max-queues", "%u", xenblk_max_queues);
 632         if (err)
 633                 pr_warn("Error writing multi-queue-max-queues\n");
 634
 635         /* setup back pointer */
 636         be->blkif->be = be;
 637
 638         err = xenbus_watch_pathfmt(dev, &be->backend_watch, backend_changed,
 639                                    "%s/%s", dev->nodename, "physical-device");
 640         if (err)
 641                 goto fail;
 642
 643         err = xenbus_printf(XBT_NIL, dev->nodename, "max-ring-page-order", "%u",
 644                             xen_blkif_max_ring_order);
 645         if (err)
 646                 pr_warn("%s write out 'max-ring-page-order' failed\n", __func__);
 647
 648         err = xenbus_switch_state(dev, XenbusStateInitWait);
 649         if (err)
 650                 goto fail;
 651
 652         return 0;
 653
 654 fail:
 655         pr_warn("%s failed\n", __func__);
 656         xen_blkbk_remove(dev);
 657         return err;
 658 }
 659
 660
 661 /*
 662  * Callback received when the hotplug scripts have placed the physical-device
 663  * node.  Read it and the mode node, and create a vbd.  If the frontend is
 664  * ready, connect.
 665  */
 666 static void backend_changed(struct xenbus_watch *watch,
 667                             const char *path, const char *token)
 668 {
 669         int err;
 670         unsigned major;
 671         unsigned minor;
 672         struct backend_info *be
 673                 = container_of(watch, struct backend_info, backend_watch);
 674         struct xenbus_device *dev = be->dev;
 675         int cdrom = 0;
 676         unsigned long handle;
 677         char *device_type;
 678
 679         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 680
 681         err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
 682                            &major, &minor);
 683         if (XENBUS_EXIST_ERR(err)) {
 684                 /*
 685                  * Since this watch will fire once immediately after it is
 686                  * registered, we expect this.  Ignore it, and wait for the
 687                  * hotplug scripts.
 688                  */
 689                 return;
 690         }
 691         if (err != 2) {
 692                 xenbus_dev_fatal(dev, err, "reading physical-device");
 693                 return;
 694         }
 695
 696         if (be->major | be->minor) {
 697                 if (be->major != major || be->minor != minor)
 698                         pr_warn("changing physical device (from %x:%x to %x:%x) not supported.\n",
 699                                 be->major, be->minor, major, minor);
 700                 return;
 701         }
 702
 703         be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
 704         if (IS_ERR(be->mode)) {
 705                 err = PTR_ERR(be->mode);
 706                 be->mode = NULL;
 707                 xenbus_dev_fatal(dev, err, "reading mode");
 708                 return;
 709         }
 710
 711         device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
 712         if (!IS_ERR(device_type)) {
 713                 cdrom = strcmp(device_type, "cdrom") == 0;
 714                 kfree(device_type);
 715         }
 716
 717         /* Front end dir is a number, which is used as the handle. */
 718         err = kstrtoul(strrchr(dev->otherend, '/') + 1, 0, &handle);
 719         if (err) {
 720                 kfree(be->mode);
 721                 be->mode = NULL;
 722                 return;
 723         }
 724
 725         be->major = major;
 726         be->minor = minor;
 727
 728         err = xen_vbd_create(be->blkif, handle, major, minor,
 729                              !strchr(be->mode, 'w'), cdrom);
 730
 731         if (err)
 732                 xenbus_dev_fatal(dev, err, "creating vbd structure");
 733         else {
 734                 err = xenvbd_sysfs_addif(dev);
 735                 if (err) {
 736                         xen_vbd_free(&be->blkif->vbd);
 737                         xenbus_dev_fatal(dev, err, "creating sysfs entries");
 738                 }
 739         }
 740
 741         if (err) {
 742                 kfree(be->mode);
 743                 be->mode = NULL;
 744                 be->major = 0;
 745                 be->minor = 0;
 746         } else {
 747                 /* We're potentially connected now */
 748                 xen_update_blkif_status(be->blkif);
 749         }
 750 }
 751
 752
 753 /*
 754  * Callback received when the frontend's state changes.
 755  */
 756 static void frontend_changed(struct xenbus_device *dev,
 757                              enum xenbus_state frontend_state)
 758 {
 759         struct backend_info *be = dev_get_drvdata(&dev->dev);
 760         int err;
 761
 762         pr_debug("%s %p %s\n", __func__, dev, xenbus_strstate(frontend_state));
 763
 764         switch (frontend_state) {
 765         case XenbusStateInitialising:
 766                 if (dev->state == XenbusStateClosed) {
 767                         pr_info("%s: prepare for reconnect\n", dev->nodename);
 768                         xenbus_switch_state(dev, XenbusStateInitWait);
 769                 }
 770                 break;
 771
 772         case XenbusStateInitialised:
 773         case XenbusStateConnected:
 774                 /*
 775                  * Ensure we connect even when two watches fire in
 776                  * close succession and we miss the intermediate value
 777                  * of frontend_state.
 778                  */
 779                 if (dev->state == XenbusStateConnected)
 780                         break;
 781
 782                 /*
 783                  * Enforce precondition before potential leak point.
 784                  * xen_blkif_disconnect() is idempotent.
 785                  */
 786                 err = xen_blkif_disconnect(be->blkif);
 787                 if (err) {
 788                         xenbus_dev_fatal(dev, err, "pending I/O");
 789                         break;
 790                 }
 791
 792                 err = connect_ring(be);
 793                 if (err) {
 794                         /*
 795                          * Clean up so that memory resources can be used by
 796                          * other devices. connect_ring reported already error.
 797                          */
 798                         xen_blkif_disconnect(be->blkif);
 799                         break;
 800                 }
 801                 xen_update_blkif_status(be->blkif);
 802                 break;
 803
 804         case XenbusStateClosing:
 805                 xenbus_switch_state(dev, XenbusStateClosing);
 806                 break;
 807
 808         case XenbusStateClosed:
 809                 xen_blkif_disconnect(be->blkif);
 810                 xenbus_switch_state(dev, XenbusStateClosed);
 811                 if (xenbus_dev_is_online(dev))
 812                         break;
 813                 /* fall through if not online */
 814         case XenbusStateUnknown:
 815                 /* implies xen_blkif_disconnect() via xen_blkbk_remove() */
 816                 device_unregister(&dev->dev);
 817                 break;
 818
 819         default:
 820                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 821                                  frontend_state);
 822                 break;
 823         }
 824 }
 825
 826
 827 /* ** Connection ** */
 828
 829
 830 /*
 831  * Write the physical details regarding the block device to the store, and
 832  * switch to Connected state.
 833  */
 834 static void connect(struct backend_info *be)
 835 {
 836         struct xenbus_transaction xbt;
 837         int err;
 838         struct xenbus_device *dev = be->dev;
 839
 840         pr_debug("%s %s\n", __func__, dev->otherend);
 841
 842         /* Supply the information about the device the frontend needs */
 843 again:
 844         err = xenbus_transaction_start(&xbt);
 845         if (err) {
 846                 xenbus_dev_fatal(dev, err, "starting transaction");
 847                 return;
 848         }
 849
 850         /* If we can't advertise it is OK. */
 851         xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
 852
 853         xen_blkbk_discard(xbt, be);
 854
 855         xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
 856
 857         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u", 1);
 858         if (err) {
 859                 xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
 860                                  dev->nodename);
 861                 goto abort;
 862         }
 863
 864         err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
 865                             (unsigned long long)vbd_sz(&be->blkif->vbd));
 866         if (err) {
 867                 xenbus_dev_fatal(dev, err, "writing %s/sectors",
 868                                  dev->nodename);
 869                 goto abort;
 870         }
 871
 872         /* FIXME: use a typename instead */
 873         err = xenbus_printf(xbt, dev->nodename, "info", "%u",
 874                             be->blkif->vbd.type |
 875                             (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
 876         if (err) {
 877                 xenbus_dev_fatal(dev, err, "writing %s/info",
 878                                  dev->nodename);
 879                 goto abort;
 880         }
 881         err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
 882                             (unsigned long)
 883                             bdev_logical_block_size(be->blkif->vbd.bdev));
 884         if (err) {
 885                 xenbus_dev_fatal(dev, err, "writing %s/sector-size",
 886                                  dev->nodename);
 887                 goto abort;
 888         }
 889         err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
 890                             bdev_physical_block_size(be->blkif->vbd.bdev));
 891         if (err)
 892                 xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
 893                                  dev->nodename);
 894
 895         err = xenbus_transaction_end(xbt, 0);
 896         if (err == -EAGAIN)
 897                 goto again;
 898         if (err)
 899                 xenbus_dev_fatal(dev, err, "ending transaction");
 900
 901         err = xenbus_switch_state(dev, XenbusStateConnected);
 902         if (err)
 903                 xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
 904                                  dev->nodename);
 905
 906         return;
 907  abort:
 908         xenbus_transaction_end(xbt, 1);
 909 }
 910
 911 /*
 912  * Each ring may have multi pages, depends on "ring-page-order".
 913  */
 914 static int read_per_ring_refs(struct xen_blkif_ring *ring, const char *dir)
 915 {
 916         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 917         struct pending_req *req, *n;
 918         int err, i, j;
 919         struct xen_blkif *blkif = ring->blkif;
 920         struct xenbus_device *dev = blkif->be->dev;
 921         unsigned int ring_page_order, nr_grefs, evtchn;
 922
 923         err = xenbus_scanf(XBT_NIL, dir, "event-channel", "%u",
 924                           &evtchn);
 925         if (err != 1) {
 926                 err = -EINVAL;
 927                 xenbus_dev_fatal(dev, err, "reading %s/event-channel", dir);
 928                 return err;
 929         }
 930
 931         err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-page-order", "%u",
 932                           &ring_page_order);
 933         if (err != 1) {
 934                 err = xenbus_scanf(XBT_NIL, dir, "ring-ref", "%u", &ring_ref[0]);
 935                 if (err != 1) {
 936                         err = -EINVAL;
 937                         xenbus_dev_fatal(dev, err, "reading %s/ring-ref", dir);
 938                         return err;
 939                 }
 940                 nr_grefs = 1;
 941         } else {
 942                 unsigned int i;
 943
 944                 if (ring_page_order > xen_blkif_max_ring_order) {
 945                         err = -EINVAL;
 946                         xenbus_dev_fatal(dev, err, "%s/request %d ring page order exceed max:%d",
 947                                          dir, ring_page_order,
 948                                          xen_blkif_max_ring_order);
 949                         return err;
 950                 }
 951
 952                 nr_grefs = 1 << ring_page_order;
 953                 for (i = 0; i < nr_grefs; i++) {
 954                         char ring_ref_name[RINGREF_NAME_LEN];
 955
 956                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
 957                         err = xenbus_scanf(XBT_NIL, dir, ring_ref_name,
 958                                            "%u", &ring_ref[i]);
 959                         if (err != 1) {
 960                                 err = -EINVAL;
 961                                 xenbus_dev_fatal(dev, err, "reading %s/%s",
 962                                                  dir, ring_ref_name);
 963                                 return err;
 964                         }
 965                 }
 966         }
 967         blkif->nr_ring_pages = nr_grefs;
 968
 969         for (i = 0; i < nr_grefs * XEN_BLKIF_REQS_PER_PAGE; i++) {
 970                 req = kzalloc(sizeof(*req), GFP_KERNEL);
 971                 if (!req)
 972                         goto fail;
 973                 list_add_tail(&req->free_list, &ring->pending_free);
 974                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
 975                         req->segments[j] = kzalloc(sizeof(*req->segments[0]), GFP_KERNEL);
 976                         if (!req->segments[j])
 977                                 goto fail;
 978                 }
 979                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
 980                         req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
 981                                                          GFP_KERNEL);
 982                         if (!req->indirect_pages[j])
 983                                 goto fail;
 984                 }
 985         }
 986
 987         /* Map the shared frame, irq etc. */
 988         err = xen_blkif_map(ring, ring_ref, nr_grefs, evtchn);
 989         if (err) {
 990                 xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
 991                 return err;
 992         }
 993
 994         return 0;
 995
 996 fail:
 997         list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
 998                 list_del(&req->free_list);
 999                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
1000                         if (!req->segments[j])
1001                                 break;
1002                         kfree(req->segments[j]);
1003                 }
1004                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1005                         if (!req->indirect_pages[j])
1006                                 break;
1007                         kfree(req->indirect_pages[j]);
1008                 }
1009                 kfree(req);
1010         }
1011         return -ENOMEM;
1012
1013 }
1014
1015 static int connect_ring(struct backend_info *be)
1016 {
1017         struct xenbus_device *dev = be->dev;
1018         unsigned int pers_grants;
1019         char protocol[64] = "";
1020         int err, i;
1021         char *xspath;
1022         size_t xspathsize;
1023         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1024         unsigned int requested_num_queues = 0;
1025
1026         pr_debug("%s %s\n", __func__, dev->otherend);
1027
1028         be->blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
1029         err = xenbus_scanf(XBT_NIL, dev->otherend, "protocol",
1030                            "%63s", protocol);
1031         if (err <= 0)
1032                 strcpy(protocol, "unspecified, assuming default");
1033         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
1034                 be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
1035         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
1036                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
1037         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
1038                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
1039         else {
1040                 xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
1041                 return -ENOSYS;
1042         }
1043         pers_grants = xenbus_read_unsigned(dev->otherend, "feature-persistent",
1044                                            0);
1045         be->blkif->vbd.feature_gnt_persistent = pers_grants;
1046         be->blkif->vbd.overflow_max_grants = 0;
1047
1048         /*
1049          * Read the number of hardware queues from frontend.
1050          */
1051         requested_num_queues = xenbus_read_unsigned(dev->otherend,
1052                                                     "multi-queue-num-queues",
1053                                                     1);
1054         if (requested_num_queues > xenblk_max_queues
1055             || requested_num_queues == 0) {
1056                 /* Buggy or malicious guest. */
1057                 xenbus_dev_fatal(dev, err,
1058                                 "guest requested %u queues, exceeding the maximum of %u.",
1059                                 requested_num_queues, xenblk_max_queues);
1060                 return -ENOSYS;
1061         }
1062         be->blkif->nr_rings = requested_num_queues;
1063         if (xen_blkif_alloc_rings(be->blkif))
1064                 return -ENOMEM;
1065
1066         pr_info("%s: using %d queues, protocol %d (%s) %s\n", dev->nodename,
1067                  be->blkif->nr_rings, be->blkif->blk_protocol, protocol,
1068                  pers_grants ? "persistent grants" : "");
1069
1070         if (be->blkif->nr_rings == 1)
1071                 return read_per_ring_refs(&be->blkif->rings[0], dev->otherend);
1072         else {
1073                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1074                 xspath = kmalloc(xspathsize, GFP_KERNEL);
1075                 if (!xspath) {
1076                         xenbus_dev_fatal(dev, -ENOMEM, "reading ring references");
1077                         return -ENOMEM;
1078                 }
1079
1080                 for (i = 0; i < be->blkif->nr_rings; i++) {
1081                         memset(xspath, 0, xspathsize);
1082                         snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, i);
1083                         err = read_per_ring_refs(&be->blkif->rings[i], xspath);
1084                         if (err) {
1085                                 kfree(xspath);
1086                                 return err;
1087                         }
1088                 }
1089                 kfree(xspath);
1090         }
1091         return 0;
1092 }
1093
1094 static const struct xenbus_device_id xen_blkbk_ids[] = {
1095         { "vbd" },
1096         { "" }
1097 };
1098
1099 static struct xenbus_driver xen_blkbk_driver = {
1100         .ids  = xen_blkbk_ids,
1101         .probe = xen_blkbk_probe,
1102         .remove = xen_blkbk_remove,
1103         .otherend_changed = frontend_changed
1104 };
1105
1106 int xen_blkif_xenbus_init(void)
1107 {
1108         return xenbus_register_backend(&xen_blkbk_driver);
1109 }