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         bool busy = false;
 248
 249         for (r = 0; r < blkif->nr_rings; r++) {
 250                 struct xen_blkif_ring *ring = &blkif->rings[r];
 251                 unsigned int i = 0;
 252
 253                 if (!ring->active)
 254                         continue;
 255
 256                 if (ring->xenblkd) {
 257                         kthread_stop(ring->xenblkd);
 258                         wake_up(&ring->shutdown_wq);
 259                 }
 260
 261                 /* The above kthread_stop() guarantees that at this point we
 262                  * don't have any discard_io or other_io requests. So, checking
 263                  * for inflight IO is enough.
 264                  */
 265                 if (atomic_read(&ring->inflight) > 0) {
 266                         busy = true;
 267                         continue;
 268                 }
 269
 270                 if (ring->irq) {
 271                         unbind_from_irqhandler(ring->irq, ring);
 272                         ring->irq = 0;
 273                 }
 274
 275                 if (ring->blk_rings.common.sring) {
 276                         xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
 277                         ring->blk_rings.common.sring = NULL;
 278                 }
 279
 280                 /* Remove all persistent grants and the cache of ballooned pages. */
 281                 xen_blkbk_free_caches(ring);
 282
 283                 /* Check that there is no request in use */
 284                 list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
 285                         list_del(&req->free_list);
 286
 287                         for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
 288                                 kfree(req->segments[j]);
 289
 290                         for (j = 0; j < MAX_INDIRECT_PAGES; j++)
 291                                 kfree(req->indirect_pages[j]);
 292
 293                         kfree(req);
 294                         i++;
 295                 }
 296
 297                 BUG_ON(atomic_read(&ring->persistent_gnt_in_use) != 0);
 298                 BUG_ON(!list_empty(&ring->persistent_purge_list));
 299                 BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts));
 300                 BUG_ON(!list_empty(&ring->free_pages));
 301                 BUG_ON(ring->free_pages_num != 0);
 302                 BUG_ON(ring->persistent_gnt_c != 0);
 303                 WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
 304                 ring->active = false;
 305         }
 306         if (busy)
 307                 return -EBUSY;
 308
 309         blkif->nr_ring_pages = 0;
 310         /*
 311          * blkif->rings was allocated in connect_ring, so we should free it in
 312          * here.
 313          */
 314         kfree(blkif->rings);
 315         blkif->rings = NULL;
 316         blkif->nr_rings = 0;
 317
 318         return 0;
 319 }
 320
 321 static void xen_blkif_free(struct xen_blkif *blkif)
 322 {
 323         WARN_ON(xen_blkif_disconnect(blkif));
 324         xen_vbd_free(&blkif->vbd);
 325         kfree(blkif->be->mode);
 326         kfree(blkif->be);
 327
 328         /* Make sure everything is drained before shutting down */
 329         kmem_cache_free(xen_blkif_cachep, blkif);
 330 }
 331
 332 int __init xen_blkif_interface_init(void)
 333 {
 334         xen_blkif_cachep = kmem_cache_create("blkif_cache",
 335                                              sizeof(struct xen_blkif),
 336                                              0, 0, NULL);
 337         if (!xen_blkif_cachep)
 338                 return -ENOMEM;
 339
 340         return 0;
 341 }
 342
 343 /*
 344  *  sysfs interface for VBD I/O requests
 345  */
 346
 347 #define VBD_SHOW_ALLRING(name, format)                                  \
 348         static ssize_t show_##name(struct device *_dev,                 \
 349                                    struct device_attribute *attr,       \
 350                                    char *buf)                           \
 351         {                                                               \
 352                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 353                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 354                 struct xen_blkif *blkif = be->blkif;                    \
 355                 unsigned int i;                                         \
 356                 unsigned long long result = 0;                          \
 357                                                                         \
 358                 if (!blkif->rings)                              \
 359                         goto out;                                       \
 360                                                                         \
 361                 for (i = 0; i < blkif->nr_rings; i++) {         \
 362                         struct xen_blkif_ring *ring = &blkif->rings[i]; \
 363                                                                         \
 364                         result += ring->st_##name;                      \
 365                 }                                                       \
 366                                                                         \
 367 out:                                                                    \
 368                 return sprintf(buf, format, result);                    \
 369         }                                                               \
 370         static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 371
 372 VBD_SHOW_ALLRING(oo_req,  "%llu\n");
 373 VBD_SHOW_ALLRING(rd_req,  "%llu\n");
 374 VBD_SHOW_ALLRING(wr_req,  "%llu\n");
 375 VBD_SHOW_ALLRING(f_req,  "%llu\n");
 376 VBD_SHOW_ALLRING(ds_req,  "%llu\n");
 377 VBD_SHOW_ALLRING(rd_sect, "%llu\n");
 378 VBD_SHOW_ALLRING(wr_sect, "%llu\n");
 379
 380 static struct attribute *xen_vbdstat_attrs[] = {
 381         &dev_attr_oo_req.attr,
 382         &dev_attr_rd_req.attr,
 383         &dev_attr_wr_req.attr,
 384         &dev_attr_f_req.attr,
 385         &dev_attr_ds_req.attr,
 386         &dev_attr_rd_sect.attr,
 387         &dev_attr_wr_sect.attr,
 388         NULL
 389 };
 390
 391 static const struct attribute_group xen_vbdstat_group = {
 392         .name = "statistics",
 393         .attrs = xen_vbdstat_attrs,
 394 };
 395
 396 #define VBD_SHOW(name, format, args...)                                 \
 397         static ssize_t show_##name(struct device *_dev,                 \
 398                                    struct device_attribute *attr,       \
 399                                    char *buf)                           \
 400         {                                                               \
 401                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 402                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 403                                                                         \
 404                 return sprintf(buf, format, ##args);                    \
 405         }                                                               \
 406         static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 407
 408 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
 409 VBD_SHOW(mode, "%s\n", be->mode);
 410
 411 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
 412 {
 413         int error;
 414
 415         error = device_create_file(&dev->dev, &dev_attr_physical_device);
 416         if (error)
 417                 goto fail1;
 418
 419         error = device_create_file(&dev->dev, &dev_attr_mode);
 420         if (error)
 421                 goto fail2;
 422
 423         error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
 424         if (error)
 425                 goto fail3;
 426
 427         return 0;
 428
 429 fail3:  sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 430 fail2:  device_remove_file(&dev->dev, &dev_attr_mode);
 431 fail1:  device_remove_file(&dev->dev, &dev_attr_physical_device);
 432         return error;
 433 }
 434
 435 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
 436 {
 437         sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 438         device_remove_file(&dev->dev, &dev_attr_mode);
 439         device_remove_file(&dev->dev, &dev_attr_physical_device);
 440 }
 441
 442
 443 static void xen_vbd_free(struct xen_vbd *vbd)
 444 {
 445         if (vbd->bdev)
 446                 blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
 447         vbd->bdev = NULL;
 448 }
 449
 450 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
 451                           unsigned major, unsigned minor, int readonly,
 452                           int cdrom)
 453 {
 454         struct xen_vbd *vbd;
 455         struct block_device *bdev;
 456         struct request_queue *q;
 457
 458         vbd = &blkif->vbd;
 459         vbd->handle   = handle;
 460         vbd->readonly = readonly;
 461         vbd->type     = 0;
 462
 463         vbd->pdevice  = MKDEV(major, minor);
 464
 465         bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
 466                                  FMODE_READ : FMODE_WRITE, NULL);
 467
 468         if (IS_ERR(bdev)) {
 469                 pr_warn("xen_vbd_create: device %08x could not be opened\n",
 470                         vbd->pdevice);
 471                 return -ENOENT;
 472         }
 473
 474         vbd->bdev = bdev;
 475         if (vbd->bdev->bd_disk == NULL) {
 476                 pr_warn("xen_vbd_create: device %08x doesn't exist\n",
 477                         vbd->pdevice);
 478                 xen_vbd_free(vbd);
 479                 return -ENOENT;
 480         }
 481         vbd->size = vbd_sz(vbd);
 482
 483         if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
 484                 vbd->type |= VDISK_CDROM;
 485         if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
 486                 vbd->type |= VDISK_REMOVABLE;
 487
 488         q = bdev_get_queue(bdev);
 489         if (q && test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 490                 vbd->flush_support = true;
 491
 492         if (q && blk_queue_secure_erase(q))
 493                 vbd->discard_secure = true;
 494
 495         pr_debug("Successful creation of handle=%04x (dom=%u)\n",
 496                 handle, blkif->domid);
 497         return 0;
 498 }
 499 static int xen_blkbk_remove(struct xenbus_device *dev)
 500 {
 501         struct backend_info *be = dev_get_drvdata(&dev->dev);
 502
 503         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 504
 505         if (be->major || be->minor)
 506                 xenvbd_sysfs_delif(dev);
 507
 508         if (be->backend_watch.node) {
 509                 unregister_xenbus_watch(&be->backend_watch);
 510                 kfree(be->backend_watch.node);
 511                 be->backend_watch.node = NULL;
 512         }
 513
 514         dev_set_drvdata(&dev->dev, NULL);
 515
 516         if (be->blkif) {
 517                 xen_blkif_disconnect(be->blkif);
 518
 519                 /* Put the reference we set in xen_blkif_alloc(). */
 520                 xen_blkif_put(be->blkif);
 521         }
 522
 523         return 0;
 524 }
 525
 526 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
 527                               struct backend_info *be, int state)
 528 {
 529         struct xenbus_device *dev = be->dev;
 530         int err;
 531
 532         err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
 533                             "%d", state);
 534         if (err)
 535                 dev_warn(&dev->dev, "writing feature-flush-cache (%d)", err);
 536
 537         return err;
 538 }
 539
 540 static void xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
 541 {
 542         struct xenbus_device *dev = be->dev;
 543         struct xen_blkif *blkif = be->blkif;
 544         int err;
 545         int state = 0;
 546         struct block_device *bdev = be->blkif->vbd.bdev;
 547         struct request_queue *q = bdev_get_queue(bdev);
 548
 549         if (!xenbus_read_unsigned(dev->nodename, "discard-enable", 1))
 550                 return;
 551
 552         if (blk_queue_discard(q)) {
 553                 err = xenbus_printf(xbt, dev->nodename,
 554                         "discard-granularity", "%u",
 555                         q->limits.discard_granularity);
 556                 if (err) {
 557                         dev_warn(&dev->dev, "writing discard-granularity (%d)", err);
 558                         return;
 559                 }
 560                 err = xenbus_printf(xbt, dev->nodename,
 561                         "discard-alignment", "%u",
 562                         q->limits.discard_alignment);
 563                 if (err) {
 564                         dev_warn(&dev->dev, "writing discard-alignment (%d)", err);
 565                         return;
 566                 }
 567                 state = 1;
 568                 /* Optional. */
 569                 err = xenbus_printf(xbt, dev->nodename,
 570                                     "discard-secure", "%d",
 571                                     blkif->vbd.discard_secure);
 572                 if (err) {
 573                         dev_warn(&dev->dev, "writing discard-secure (%d)", err);
 574                         return;
 575                 }
 576         }
 577         err = xenbus_printf(xbt, dev->nodename, "feature-discard",
 578                             "%d", state);
 579         if (err)
 580                 dev_warn(&dev->dev, "writing feature-discard (%d)", err);
 581 }
 582 int xen_blkbk_barrier(struct xenbus_transaction xbt,
 583                       struct backend_info *be, int state)
 584 {
 585         struct xenbus_device *dev = be->dev;
 586         int err;
 587
 588         err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
 589                             "%d", state);
 590         if (err)
 591                 dev_warn(&dev->dev, "writing feature-barrier (%d)", err);
 592
 593         return err;
 594 }
 595
 596 /*
 597  * Entry point to this code when a new device is created.  Allocate the basic
 598  * structures, and watch the store waiting for the hotplug scripts to tell us
 599  * the device's physical major and minor numbers.  Switch to InitWait.
 600  */
 601 static int xen_blkbk_probe(struct xenbus_device *dev,
 602                            const struct xenbus_device_id *id)
 603 {
 604         int err;
 605         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 606                                           GFP_KERNEL);
 607
 608         /* match the pr_debug in xen_blkbk_remove */
 609         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 610
 611         if (!be) {
 612                 xenbus_dev_fatal(dev, -ENOMEM,
 613                                  "allocating backend structure");
 614                 return -ENOMEM;
 615         }
 616         be->dev = dev;
 617         dev_set_drvdata(&dev->dev, be);
 618
 619         be->blkif = xen_blkif_alloc(dev->otherend_id);
 620         if (IS_ERR(be->blkif)) {
 621                 err = PTR_ERR(be->blkif);
 622                 be->blkif = NULL;
 623                 xenbus_dev_fatal(dev, err, "creating block interface");
 624                 goto fail;
 625         }
 626
 627         err = xenbus_printf(XBT_NIL, dev->nodename,
 628                             "feature-max-indirect-segments", "%u",
 629                             MAX_INDIRECT_SEGMENTS);
 630         if (err)
 631                 dev_warn(&dev->dev,
 632                          "writing %s/feature-max-indirect-segments (%d)",
 633                          dev->nodename, err);
 634
 635         /* Multi-queue: advertise how many queues are supported by us.*/
 636         err = xenbus_printf(XBT_NIL, dev->nodename,
 637                             "multi-queue-max-queues", "%u", xenblk_max_queues);
 638         if (err)
 639                 pr_warn("Error writing multi-queue-max-queues\n");
 640
 641         /* setup back pointer */
 642         be->blkif->be = be;
 643
 644         err = xenbus_watch_pathfmt(dev, &be->backend_watch, backend_changed,
 645                                    "%s/%s", dev->nodename, "physical-device");
 646         if (err)
 647                 goto fail;
 648
 649         err = xenbus_printf(XBT_NIL, dev->nodename, "max-ring-page-order", "%u",
 650                             xen_blkif_max_ring_order);
 651         if (err)
 652                 pr_warn("%s write out 'max-ring-page-order' failed\n", __func__);
 653
 654         err = xenbus_switch_state(dev, XenbusStateInitWait);
 655         if (err)
 656                 goto fail;
 657
 658         return 0;
 659
 660 fail:
 661         pr_warn("%s failed\n", __func__);
 662         xen_blkbk_remove(dev);
 663         return err;
 664 }
 665
 666
 667 /*
 668  * Callback received when the hotplug scripts have placed the physical-device
 669  * node.  Read it and the mode node, and create a vbd.  If the frontend is
 670  * ready, connect.
 671  */
 672 static void backend_changed(struct xenbus_watch *watch,
 673                             const char *path, const char *token)
 674 {
 675         int err;
 676         unsigned major;
 677         unsigned minor;
 678         struct backend_info *be
 679                 = container_of(watch, struct backend_info, backend_watch);
 680         struct xenbus_device *dev = be->dev;
 681         int cdrom = 0;
 682         unsigned long handle;
 683         char *device_type;
 684
 685         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 686
 687         err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
 688                            &major, &minor);
 689         if (XENBUS_EXIST_ERR(err)) {
 690                 /*
 691                  * Since this watch will fire once immediately after it is
 692                  * registered, we expect this.  Ignore it, and wait for the
 693                  * hotplug scripts.
 694                  */
 695                 return;
 696         }
 697         if (err != 2) {
 698                 xenbus_dev_fatal(dev, err, "reading physical-device");
 699                 return;
 700         }
 701
 702         if (be->major | be->minor) {
 703                 if (be->major != major || be->minor != minor)
 704                         pr_warn("changing physical device (from %x:%x to %x:%x) not supported.\n",
 705                                 be->major, be->minor, major, minor);
 706                 return;
 707         }
 708
 709         be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
 710         if (IS_ERR(be->mode)) {
 711                 err = PTR_ERR(be->mode);
 712                 be->mode = NULL;
 713                 xenbus_dev_fatal(dev, err, "reading mode");
 714                 return;
 715         }
 716
 717         device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
 718         if (!IS_ERR(device_type)) {
 719                 cdrom = strcmp(device_type, "cdrom") == 0;
 720                 kfree(device_type);
 721         }
 722
 723         /* Front end dir is a number, which is used as the handle. */
 724         err = kstrtoul(strrchr(dev->otherend, '/') + 1, 0, &handle);
 725         if (err) {
 726                 kfree(be->mode);
 727                 be->mode = NULL;
 728                 return;
 729         }
 730
 731         be->major = major;
 732         be->minor = minor;
 733
 734         err = xen_vbd_create(be->blkif, handle, major, minor,
 735                              !strchr(be->mode, 'w'), cdrom);
 736
 737         if (err)
 738                 xenbus_dev_fatal(dev, err, "creating vbd structure");
 739         else {
 740                 err = xenvbd_sysfs_addif(dev);
 741                 if (err) {
 742                         xen_vbd_free(&be->blkif->vbd);
 743                         xenbus_dev_fatal(dev, err, "creating sysfs entries");
 744                 }
 745         }
 746
 747         if (err) {
 748                 kfree(be->mode);
 749                 be->mode = NULL;
 750                 be->major = 0;
 751                 be->minor = 0;
 752         } else {
 753                 /* We're potentially connected now */
 754                 xen_update_blkif_status(be->blkif);
 755         }
 756 }
 757
 758
 759 /*
 760  * Callback received when the frontend's state changes.
 761  */
 762 static void frontend_changed(struct xenbus_device *dev,
 763                              enum xenbus_state frontend_state)
 764 {
 765         struct backend_info *be = dev_get_drvdata(&dev->dev);
 766         int err;
 767
 768         pr_debug("%s %p %s\n", __func__, dev, xenbus_strstate(frontend_state));
 769
 770         switch (frontend_state) {
 771         case XenbusStateInitialising:
 772                 if (dev->state == XenbusStateClosed) {
 773                         pr_info("%s: prepare for reconnect\n", dev->nodename);
 774                         xenbus_switch_state(dev, XenbusStateInitWait);
 775                 }
 776                 break;
 777
 778         case XenbusStateInitialised:
 779         case XenbusStateConnected:
 780                 /*
 781                  * Ensure we connect even when two watches fire in
 782                  * close succession and we miss the intermediate value
 783                  * of frontend_state.
 784                  */
 785                 if (dev->state == XenbusStateConnected)
 786                         break;
 787
 788                 /*
 789                  * Enforce precondition before potential leak point.
 790                  * xen_blkif_disconnect() is idempotent.
 791                  */
 792                 err = xen_blkif_disconnect(be->blkif);
 793                 if (err) {
 794                         xenbus_dev_fatal(dev, err, "pending I/O");
 795                         break;
 796                 }
 797
 798                 err = connect_ring(be);
 799                 if (err) {
 800                         /*
 801                          * Clean up so that memory resources can be used by
 802                          * other devices. connect_ring reported already error.
 803                          */
 804                         xen_blkif_disconnect(be->blkif);
 805                         break;
 806                 }
 807                 xen_update_blkif_status(be->blkif);
 808                 break;
 809
 810         case XenbusStateClosing:
 811                 xenbus_switch_state(dev, XenbusStateClosing);
 812                 break;
 813
 814         case XenbusStateClosed:
 815                 xen_blkif_disconnect(be->blkif);
 816                 xenbus_switch_state(dev, XenbusStateClosed);
 817                 if (xenbus_dev_is_online(dev))
 818                         break;
 819                 /* fall through */
 820                 /* if not online */
 821         case XenbusStateUnknown:
 822                 /* implies xen_blkif_disconnect() via xen_blkbk_remove() */
 823                 device_unregister(&dev->dev);
 824                 break;
 825
 826         default:
 827                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 828                                  frontend_state);
 829                 break;
 830         }
 831 }
 832
 833
 834 /* ** Connection ** */
 835
 836
 837 /*
 838  * Write the physical details regarding the block device to the store, and
 839  * switch to Connected state.
 840  */
 841 static void connect(struct backend_info *be)
 842 {
 843         struct xenbus_transaction xbt;
 844         int err;
 845         struct xenbus_device *dev = be->dev;
 846
 847         pr_debug("%s %s\n", __func__, dev->otherend);
 848
 849         /* Supply the information about the device the frontend needs */
 850 again:
 851         err = xenbus_transaction_start(&xbt);
 852         if (err) {
 853                 xenbus_dev_fatal(dev, err, "starting transaction");
 854                 return;
 855         }
 856
 857         /* If we can't advertise it is OK. */
 858         xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
 859
 860         xen_blkbk_discard(xbt, be);
 861
 862         xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
 863
 864         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u", 1);
 865         if (err) {
 866                 xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
 867                                  dev->nodename);
 868                 goto abort;
 869         }
 870
 871         err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
 872                             (unsigned long long)vbd_sz(&be->blkif->vbd));
 873         if (err) {
 874                 xenbus_dev_fatal(dev, err, "writing %s/sectors",
 875                                  dev->nodename);
 876                 goto abort;
 877         }
 878
 879         /* FIXME: use a typename instead */
 880         err = xenbus_printf(xbt, dev->nodename, "info", "%u",
 881                             be->blkif->vbd.type |
 882                             (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
 883         if (err) {
 884                 xenbus_dev_fatal(dev, err, "writing %s/info",
 885                                  dev->nodename);
 886                 goto abort;
 887         }
 888         err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
 889                             (unsigned long)
 890                             bdev_logical_block_size(be->blkif->vbd.bdev));
 891         if (err) {
 892                 xenbus_dev_fatal(dev, err, "writing %s/sector-size",
 893                                  dev->nodename);
 894                 goto abort;
 895         }
 896         err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
 897                             bdev_physical_block_size(be->blkif->vbd.bdev));
 898         if (err)
 899                 xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
 900                                  dev->nodename);
 901
 902         err = xenbus_transaction_end(xbt, 0);
 903         if (err == -EAGAIN)
 904                 goto again;
 905         if (err)
 906                 xenbus_dev_fatal(dev, err, "ending transaction");
 907
 908         err = xenbus_switch_state(dev, XenbusStateConnected);
 909         if (err)
 910                 xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
 911                                  dev->nodename);
 912
 913         return;
 914  abort:
 915         xenbus_transaction_end(xbt, 1);
 916 }
 917
 918 /*
 919  * Each ring may have multi pages, depends on "ring-page-order".
 920  */
 921 static int read_per_ring_refs(struct xen_blkif_ring *ring, const char *dir)
 922 {
 923         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 924         struct pending_req *req, *n;
 925         int err, i, j;
 926         struct xen_blkif *blkif = ring->blkif;
 927         struct xenbus_device *dev = blkif->be->dev;
 928         unsigned int ring_page_order, nr_grefs, evtchn;
 929
 930         err = xenbus_scanf(XBT_NIL, dir, "event-channel", "%u",
 931                           &evtchn);
 932         if (err != 1) {
 933                 err = -EINVAL;
 934                 xenbus_dev_fatal(dev, err, "reading %s/event-channel", dir);
 935                 return err;
 936         }
 937
 938         err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-page-order", "%u",
 939                           &ring_page_order);
 940         if (err != 1) {
 941                 err = xenbus_scanf(XBT_NIL, dir, "ring-ref", "%u", &ring_ref[0]);
 942                 if (err != 1) {
 943                         err = -EINVAL;
 944                         xenbus_dev_fatal(dev, err, "reading %s/ring-ref", dir);
 945                         return err;
 946                 }
 947                 nr_grefs = 1;
 948         } else {
 949                 unsigned int i;
 950
 951                 if (ring_page_order > xen_blkif_max_ring_order) {
 952                         err = -EINVAL;
 953                         xenbus_dev_fatal(dev, err, "%s/request %d ring page order exceed max:%d",
 954                                          dir, ring_page_order,
 955                                          xen_blkif_max_ring_order);
 956                         return err;
 957                 }
 958
 959                 nr_grefs = 1 << ring_page_order;
 960                 for (i = 0; i < nr_grefs; i++) {
 961                         char ring_ref_name[RINGREF_NAME_LEN];
 962
 963                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
 964                         err = xenbus_scanf(XBT_NIL, dir, ring_ref_name,
 965                                            "%u", &ring_ref[i]);
 966                         if (err != 1) {
 967                                 err = -EINVAL;
 968                                 xenbus_dev_fatal(dev, err, "reading %s/%s",
 969                                                  dir, ring_ref_name);
 970                                 return err;
 971                         }
 972                 }
 973         }
 974         blkif->nr_ring_pages = nr_grefs;
 975
 976         for (i = 0; i < nr_grefs * XEN_BLKIF_REQS_PER_PAGE; i++) {
 977                 req = kzalloc(sizeof(*req), GFP_KERNEL);
 978                 if (!req)
 979                         goto fail;
 980                 list_add_tail(&req->free_list, &ring->pending_free);
 981                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
 982                         req->segments[j] = kzalloc(sizeof(*req->segments[0]), GFP_KERNEL);
 983                         if (!req->segments[j])
 984                                 goto fail;
 985                 }
 986                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
 987                         req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
 988                                                          GFP_KERNEL);
 989                         if (!req->indirect_pages[j])
 990                                 goto fail;
 991                 }
 992         }
 993
 994         /* Map the shared frame, irq etc. */
 995         err = xen_blkif_map(ring, ring_ref, nr_grefs, evtchn);
 996         if (err) {
 997                 xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
 998                 return err;
 999         }
1000
1001         return 0;
1002
1003 fail:
1004         list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
1005                 list_del(&req->free_list);
1006                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
1007                         if (!req->segments[j])
1008                                 break;
1009                         kfree(req->segments[j]);
1010                 }
1011                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1012                         if (!req->indirect_pages[j])
1013                                 break;
1014                         kfree(req->indirect_pages[j]);
1015                 }
1016                 kfree(req);
1017         }
1018         return -ENOMEM;
1019
1020 }
1021
1022 static int connect_ring(struct backend_info *be)
1023 {
1024         struct xenbus_device *dev = be->dev;
1025         unsigned int pers_grants;
1026         char protocol[64] = "";
1027         int err, i;
1028         char *xspath;
1029         size_t xspathsize;
1030         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1031         unsigned int requested_num_queues = 0;
1032
1033         pr_debug("%s %s\n", __func__, dev->otherend);
1034
1035         be->blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
1036         err = xenbus_scanf(XBT_NIL, dev->otherend, "protocol",
1037                            "%63s", protocol);
1038         if (err <= 0)
1039                 strcpy(protocol, "unspecified, assuming default");
1040         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
1041                 be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
1042         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
1043                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
1044         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
1045                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
1046         else {
1047                 xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
1048                 return -ENOSYS;
1049         }
1050         pers_grants = xenbus_read_unsigned(dev->otherend, "feature-persistent",
1051                                            0);
1052         be->blkif->vbd.feature_gnt_persistent = pers_grants;
1053         be->blkif->vbd.overflow_max_grants = 0;
1054
1055         /*
1056          * Read the number of hardware queues from frontend.
1057          */
1058         requested_num_queues = xenbus_read_unsigned(dev->otherend,
1059                                                     "multi-queue-num-queues",
1060                                                     1);
1061         if (requested_num_queues > xenblk_max_queues
1062             || requested_num_queues == 0) {
1063                 /* Buggy or malicious guest. */
1064                 xenbus_dev_fatal(dev, err,
1065                                 "guest requested %u queues, exceeding the maximum of %u.",
1066                                 requested_num_queues, xenblk_max_queues);
1067                 return -ENOSYS;
1068         }
1069         be->blkif->nr_rings = requested_num_queues;
1070         if (xen_blkif_alloc_rings(be->blkif))
1071                 return -ENOMEM;
1072
1073         pr_info("%s: using %d queues, protocol %d (%s) %s\n", dev->nodename,
1074                  be->blkif->nr_rings, be->blkif->blk_protocol, protocol,
1075                  pers_grants ? "persistent grants" : "");
1076
1077         if (be->blkif->nr_rings == 1)
1078                 return read_per_ring_refs(&be->blkif->rings[0], dev->otherend);
1079         else {
1080                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1081                 xspath = kmalloc(xspathsize, GFP_KERNEL);
1082                 if (!xspath) {
1083                         xenbus_dev_fatal(dev, -ENOMEM, "reading ring references");
1084                         return -ENOMEM;
1085                 }
1086
1087                 for (i = 0; i < be->blkif->nr_rings; i++) {
1088                         memset(xspath, 0, xspathsize);
1089                         snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, i);
1090                         err = read_per_ring_refs(&be->blkif->rings[i], xspath);
1091                         if (err) {
1092                                 kfree(xspath);
1093                                 return err;
1094                         }
1095                 }
1096                 kfree(xspath);
1097         }
1098         return 0;
1099 }
1100
1101 static const struct xenbus_device_id xen_blkbk_ids[] = {
1102         { "vbd" },
1103         { "" }
1104 };
1105
1106 static struct xenbus_driver xen_blkbk_driver = {
1107         .ids  = xen_blkbk_ids,
1108         .probe = xen_blkbk_probe,
1109         .remove = xen_blkbk_remove,
1110         .otherend_changed = frontend_changed
1111 };
1112
1113 int xen_blkif_xenbus_init(void)
1114 {
1115         return xenbus_register_backend(&xen_blkbk_driver);
1116 }