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                             unsigned int);
  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
 182         /*
 183          * Because freeing back to the cache may be deferred, it is not
 184          * safe to unload the module (and hence destroy the cache) until
 185          * this has completed. To prevent premature unloading, take an
 186          * extra module reference here and release only when the object
 187          * has been freed back to the cache.
 188          */
 189         __module_get(THIS_MODULE);
 190         INIT_WORK(&blkif->free_work, xen_blkif_deferred_free);
 191
 192         return blkif;
 193 }
 194
 195 static int xen_blkif_map(struct xen_blkif_ring *ring, grant_ref_t *gref,
 196                          unsigned int nr_grefs, unsigned int evtchn)
 197 {
 198         int err;
 199         struct xen_blkif *blkif = ring->blkif;
 200
 201         /* Already connected through? */
 202         if (ring->irq)
 203                 return 0;
 204
 205         err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
 206                                      &ring->blk_ring);
 207         if (err < 0)
 208                 return err;
 209
 210         switch (blkif->blk_protocol) {
 211         case BLKIF_PROTOCOL_NATIVE:
 212         {
 213                 struct blkif_sring *sring;
 214                 sring = (struct blkif_sring *)ring->blk_ring;
 215                 BACK_RING_INIT(&ring->blk_rings.native, sring,
 216                                XEN_PAGE_SIZE * nr_grefs);
 217                 break;
 218         }
 219         case BLKIF_PROTOCOL_X86_32:
 220         {
 221                 struct blkif_x86_32_sring *sring_x86_32;
 222                 sring_x86_32 = (struct blkif_x86_32_sring *)ring->blk_ring;
 223                 BACK_RING_INIT(&ring->blk_rings.x86_32, sring_x86_32,
 224                                XEN_PAGE_SIZE * nr_grefs);
 225                 break;
 226         }
 227         case BLKIF_PROTOCOL_X86_64:
 228         {
 229                 struct blkif_x86_64_sring *sring_x86_64;
 230                 sring_x86_64 = (struct blkif_x86_64_sring *)ring->blk_ring;
 231                 BACK_RING_INIT(&ring->blk_rings.x86_64, sring_x86_64,
 232                                XEN_PAGE_SIZE * nr_grefs);
 233                 break;
 234         }
 235         default:
 236                 BUG();
 237         }
 238
 239         err = bind_interdomain_evtchn_to_irqhandler_lateeoi(blkif->domid,
 240                         evtchn, xen_blkif_be_int, 0, "blkif-backend", ring);
 241         if (err < 0) {
 242                 xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
 243                 ring->blk_rings.common.sring = NULL;
 244                 return err;
 245         }
 246         ring->irq = err;
 247
 248         return 0;
 249 }
 250
 251 static int xen_blkif_disconnect(struct xen_blkif *blkif)
 252 {
 253         struct pending_req *req, *n;
 254         unsigned int j, r;
 255
 256         for (r = 0; r < blkif->nr_rings; r++) {
 257                 struct xen_blkif_ring *ring = &blkif->rings[r];
 258                 unsigned int i = 0;
 259
 260                 if (!ring->active)
 261                         continue;
 262
 263                 if (ring->xenblkd) {
 264                         kthread_stop(ring->xenblkd);
 265                         ring->xenblkd = NULL;
 266                         wake_up(&ring->shutdown_wq);
 267                 }
 268
 269                 /* The above kthread_stop() guarantees that at this point we
 270                  * don't have any discard_io or other_io requests. So, checking
 271                  * for inflight IO is enough.
 272                  */
 273                 if (atomic_read(&ring->inflight) > 0)
 274                         return -EBUSY;
 275
 276                 if (ring->irq) {
 277                         unbind_from_irqhandler(ring->irq, ring);
 278                         ring->irq = 0;
 279                 }
 280
 281                 if (ring->blk_rings.common.sring) {
 282                         xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
 283                         ring->blk_rings.common.sring = NULL;
 284                 }
 285
 286                 /* Remove all persistent grants and the cache of ballooned pages. */
 287                 xen_blkbk_free_caches(ring);
 288
 289                 /* Check that there is no request in use */
 290                 list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
 291                         list_del(&req->free_list);
 292
 293                         for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
 294                                 kfree(req->segments[j]);
 295
 296                         for (j = 0; j < MAX_INDIRECT_PAGES; j++)
 297                                 kfree(req->indirect_pages[j]);
 298
 299                         kfree(req);
 300                         i++;
 301                 }
 302
 303                 BUG_ON(atomic_read(&ring->persistent_gnt_in_use) != 0);
 304                 BUG_ON(!list_empty(&ring->persistent_purge_list));
 305                 BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts));
 306                 BUG_ON(!list_empty(&ring->free_pages));
 307                 BUG_ON(ring->free_pages_num != 0);
 308                 BUG_ON(ring->persistent_gnt_c != 0);
 309                 WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
 310                 ring->active = false;
 311         }
 312         blkif->nr_ring_pages = 0;
 313         /*
 314          * blkif->rings was allocated in connect_ring, so we should free it in
 315          * here.
 316          */
 317         kfree(blkif->rings);
 318         blkif->rings = NULL;
 319         blkif->nr_rings = 0;
 320
 321         return 0;
 322 }
 323
 324 static void xen_blkif_free(struct xen_blkif *blkif)
 325 {
 326
 327         WARN_ON(xen_blkif_disconnect(blkif));
 328         xen_vbd_free(&blkif->vbd);
 329         kfree(blkif->be->mode);
 330         kfree(blkif->be);
 331
 332         /* Make sure everything is drained before shutting down */
 333         kmem_cache_free(xen_blkif_cachep, blkif);
 334         module_put(THIS_MODULE);
 335 }
 336
 337 int __init xen_blkif_interface_init(void)
 338 {
 339         xen_blkif_cachep = kmem_cache_create("blkif_cache",
 340                                              sizeof(struct xen_blkif),
 341                                              0, 0, NULL);
 342         if (!xen_blkif_cachep)
 343                 return -ENOMEM;
 344
 345         return 0;
 346 }
 347
 348 /*
 349  *  sysfs interface for VBD I/O requests
 350  */
 351
 352 #define VBD_SHOW_ALLRING(name, format)                                  \
 353         static ssize_t show_##name(struct device *_dev,                 \
 354                                    struct device_attribute *attr,       \
 355                                    char *buf)                           \
 356         {                                                               \
 357                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 358                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 359                 struct xen_blkif *blkif = be->blkif;                    \
 360                 unsigned int i;                                         \
 361                 unsigned long long result = 0;                          \
 362                                                                         \
 363                 if (!blkif->rings)                              \
 364                         goto out;                                       \
 365                                                                         \
 366                 for (i = 0; i < blkif->nr_rings; i++) {         \
 367                         struct xen_blkif_ring *ring = &blkif->rings[i]; \
 368                                                                         \
 369                         result += ring->st_##name;                      \
 370                 }                                                       \
 371                                                                         \
 372 out:                                                                    \
 373                 return sprintf(buf, format, result);                    \
 374         }                                                               \
 375         static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 376
 377 VBD_SHOW_ALLRING(oo_req,  "%llu\n");
 378 VBD_SHOW_ALLRING(rd_req,  "%llu\n");
 379 VBD_SHOW_ALLRING(wr_req,  "%llu\n");
 380 VBD_SHOW_ALLRING(f_req,  "%llu\n");
 381 VBD_SHOW_ALLRING(ds_req,  "%llu\n");
 382 VBD_SHOW_ALLRING(rd_sect, "%llu\n");
 383 VBD_SHOW_ALLRING(wr_sect, "%llu\n");
 384
 385 static struct attribute *xen_vbdstat_attrs[] = {
 386         &dev_attr_oo_req.attr,
 387         &dev_attr_rd_req.attr,
 388         &dev_attr_wr_req.attr,
 389         &dev_attr_f_req.attr,
 390         &dev_attr_ds_req.attr,
 391         &dev_attr_rd_sect.attr,
 392         &dev_attr_wr_sect.attr,
 393         NULL
 394 };
 395
 396 static const struct attribute_group xen_vbdstat_group = {
 397         .name = "statistics",
 398         .attrs = xen_vbdstat_attrs,
 399 };
 400
 401 #define VBD_SHOW(name, format, args...)                                 \
 402         static ssize_t show_##name(struct device *_dev,                 \
 403                                    struct device_attribute *attr,       \
 404                                    char *buf)                           \
 405         {                                                               \
 406                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 407                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 408                                                                         \
 409                 return sprintf(buf, format, ##args);                    \
 410         }                                                               \
 411         static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
 412
 413 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
 414 VBD_SHOW(mode, "%s\n", be->mode);
 415
 416 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
 417 {
 418         int error;
 419
 420         error = device_create_file(&dev->dev, &dev_attr_physical_device);
 421         if (error)
 422                 goto fail1;
 423
 424         error = device_create_file(&dev->dev, &dev_attr_mode);
 425         if (error)
 426                 goto fail2;
 427
 428         error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
 429         if (error)
 430                 goto fail3;
 431
 432         return 0;
 433
 434 fail3:  sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 435 fail2:  device_remove_file(&dev->dev, &dev_attr_mode);
 436 fail1:  device_remove_file(&dev->dev, &dev_attr_physical_device);
 437         return error;
 438 }
 439
 440 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
 441 {
 442         sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 443         device_remove_file(&dev->dev, &dev_attr_mode);
 444         device_remove_file(&dev->dev, &dev_attr_physical_device);
 445 }
 446
 447
 448 static void xen_vbd_free(struct xen_vbd *vbd)
 449 {
 450         if (vbd->bdev)
 451                 blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
 452         vbd->bdev = NULL;
 453 }
 454
 455 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
 456                           unsigned major, unsigned minor, int readonly,
 457                           int cdrom)
 458 {
 459         struct xen_vbd *vbd;
 460         struct block_device *bdev;
 461         struct request_queue *q;
 462
 463         vbd = &blkif->vbd;
 464         vbd->handle   = handle;
 465         vbd->readonly = readonly;
 466         vbd->type     = 0;
 467
 468         vbd->pdevice  = MKDEV(major, minor);
 469
 470         bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
 471                                  FMODE_READ : FMODE_WRITE, NULL);
 472
 473         if (IS_ERR(bdev)) {
 474                 pr_warn("xen_vbd_create: device %08x could not be opened\n",
 475                         vbd->pdevice);
 476                 return -ENOENT;
 477         }
 478
 479         vbd->bdev = bdev;
 480         if (vbd->bdev->bd_disk == NULL) {
 481                 pr_warn("xen_vbd_create: device %08x doesn't exist\n",
 482                         vbd->pdevice);
 483                 xen_vbd_free(vbd);
 484                 return -ENOENT;
 485         }
 486         vbd->size = vbd_sz(vbd);
 487
 488         if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
 489                 vbd->type |= VDISK_CDROM;
 490         if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
 491                 vbd->type |= VDISK_REMOVABLE;
 492
 493         q = bdev_get_queue(bdev);
 494         if (q && test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 495                 vbd->flush_support = true;
 496
 497         if (q && blk_queue_secure_erase(q))
 498                 vbd->discard_secure = true;
 499
 500         pr_debug("Successful creation of handle=%04x (dom=%u)\n",
 501                 handle, blkif->domid);
 502         return 0;
 503 }
 504 static int xen_blkbk_remove(struct xenbus_device *dev)
 505 {
 506         struct backend_info *be = dev_get_drvdata(&dev->dev);
 507
 508         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 509
 510         if (be->major || be->minor)
 511                 xenvbd_sysfs_delif(dev);
 512
 513         if (be->backend_watch.node) {
 514                 unregister_xenbus_watch(&be->backend_watch);
 515                 kfree(be->backend_watch.node);
 516                 be->backend_watch.node = NULL;
 517         }
 518
 519         dev_set_drvdata(&dev->dev, NULL);
 520
 521         if (be->blkif)
 522                 xen_blkif_disconnect(be->blkif);
 523
 524         /* Put the reference we set in xen_blkif_alloc(). */
 525         xen_blkif_put(be->blkif);
 526         return 0;
 527 }
 528
 529 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
 530                               struct backend_info *be, int state)
 531 {
 532         struct xenbus_device *dev = be->dev;
 533         int err;
 534
 535         err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
 536                             "%d", state);
 537         if (err)
 538                 dev_warn(&dev->dev, "writing feature-flush-cache (%d)", err);
 539
 540         return err;
 541 }
 542
 543 static void xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
 544 {
 545         struct xenbus_device *dev = be->dev;
 546         struct xen_blkif *blkif = be->blkif;
 547         int err;
 548         int state = 0, discard_enable;
 549         struct block_device *bdev = be->blkif->vbd.bdev;
 550         struct request_queue *q = bdev_get_queue(bdev);
 551
 552         err = xenbus_scanf(XBT_NIL, dev->nodename, "discard-enable", "%d",
 553                            &discard_enable);
 554         if (err == 1 && !discard_enable)
 555                 return;
 556
 557         if (blk_queue_discard(q)) {
 558                 err = xenbus_printf(xbt, dev->nodename,
 559                         "discard-granularity", "%u",
 560                         q->limits.discard_granularity);
 561                 if (err) {
 562                         dev_warn(&dev->dev, "writing discard-granularity (%d)", err);
 563                         return;
 564                 }
 565                 err = xenbus_printf(xbt, dev->nodename,
 566                         "discard-alignment", "%u",
 567                         q->limits.discard_alignment);
 568                 if (err) {
 569                         dev_warn(&dev->dev, "writing discard-alignment (%d)", err);
 570                         return;
 571                 }
 572                 state = 1;
 573                 /* Optional. */
 574                 err = xenbus_printf(xbt, dev->nodename,
 575                                     "discard-secure", "%d",
 576                                     blkif->vbd.discard_secure);
 577                 if (err) {
 578                         dev_warn(&dev->dev, "writing discard-secure (%d)", err);
 579                         return;
 580                 }
 581         }
 582         err = xenbus_printf(xbt, dev->nodename, "feature-discard",
 583                             "%d", state);
 584         if (err)
 585                 dev_warn(&dev->dev, "writing feature-discard (%d)", err);
 586 }
 587 int xen_blkbk_barrier(struct xenbus_transaction xbt,
 588                       struct backend_info *be, int state)
 589 {
 590         struct xenbus_device *dev = be->dev;
 591         int err;
 592
 593         err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
 594                             "%d", state);
 595         if (err)
 596                 dev_warn(&dev->dev, "writing feature-barrier (%d)", err);
 597
 598         return err;
 599 }
 600
 601 /*
 602  * Entry point to this code when a new device is created.  Allocate the basic
 603  * structures, and watch the store waiting for the hotplug scripts to tell us
 604  * the device's physical major and minor numbers.  Switch to InitWait.
 605  */
 606 static int xen_blkbk_probe(struct xenbus_device *dev,
 607                            const struct xenbus_device_id *id)
 608 {
 609         int err;
 610         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 611                                           GFP_KERNEL);
 612
 613         /* match the pr_debug in xen_blkbk_remove */
 614         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 615
 616         if (!be) {
 617                 xenbus_dev_fatal(dev, -ENOMEM,
 618                                  "allocating backend structure");
 619                 return -ENOMEM;
 620         }
 621         be->dev = dev;
 622         dev_set_drvdata(&dev->dev, be);
 623
 624         be->blkif = xen_blkif_alloc(dev->otherend_id);
 625         if (IS_ERR(be->blkif)) {
 626                 err = PTR_ERR(be->blkif);
 627                 be->blkif = NULL;
 628                 xenbus_dev_fatal(dev, err, "creating block interface");
 629                 goto fail;
 630         }
 631
 632         err = xenbus_printf(XBT_NIL, dev->nodename,
 633                             "feature-max-indirect-segments", "%u",
 634                             MAX_INDIRECT_SEGMENTS);
 635         if (err)
 636                 dev_warn(&dev->dev,
 637                          "writing %s/feature-max-indirect-segments (%d)",
 638                          dev->nodename, err);
 639
 640         /* Multi-queue: advertise how many queues are supported by us.*/
 641         err = xenbus_printf(XBT_NIL, dev->nodename,
 642                             "multi-queue-max-queues", "%u", xenblk_max_queues);
 643         if (err)
 644                 pr_warn("Error writing multi-queue-max-queues\n");
 645
 646         /* setup back pointer */
 647         be->blkif->be = be;
 648
 649         err = xenbus_watch_pathfmt(dev, &be->backend_watch, NULL,
 650                                    backend_changed,
 651                                    "%s/%s", dev->nodename, "physical-device");
 652         if (err)
 653                 goto fail;
 654
 655         err = xenbus_printf(XBT_NIL, dev->nodename, "max-ring-page-order", "%u",
 656                             xen_blkif_max_ring_order);
 657         if (err)
 658                 pr_warn("%s write out 'max-ring-page-order' failed\n", __func__);
 659
 660         err = xenbus_switch_state(dev, XenbusStateInitWait);
 661         if (err)
 662                 goto fail;
 663
 664         return 0;
 665
 666 fail:
 667         pr_warn("%s failed\n", __func__);
 668         xen_blkbk_remove(dev);
 669         return err;
 670 }
 671
 672
 673 /*
 674  * Callback received when the hotplug scripts have placed the physical-device
 675  * node.  Read it and the mode node, and create a vbd.  If the frontend is
 676  * ready, connect.
 677  */
 678 static void backend_changed(struct xenbus_watch *watch,
 679                             const char **vec, unsigned int len)
 680 {
 681         int err;
 682         unsigned major;
 683         unsigned minor;
 684         struct backend_info *be
 685                 = container_of(watch, struct backend_info, backend_watch);
 686         struct xenbus_device *dev = be->dev;
 687         int cdrom = 0;
 688         unsigned long handle;
 689         char *device_type;
 690
 691         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 692
 693         err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
 694                            &major, &minor);
 695         if (XENBUS_EXIST_ERR(err)) {
 696                 /*
 697                  * Since this watch will fire once immediately after it is
 698                  * registered, we expect this.  Ignore it, and wait for the
 699                  * hotplug scripts.
 700                  */
 701                 return;
 702         }
 703         if (err != 2) {
 704                 xenbus_dev_fatal(dev, err, "reading physical-device");
 705                 return;
 706         }
 707
 708         if (be->major | be->minor) {
 709                 if (be->major != major || be->minor != minor)
 710                         pr_warn("changing physical device (from %x:%x to %x:%x) not supported.\n",
 711                                 be->major, be->minor, major, minor);
 712                 return;
 713         }
 714
 715         be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
 716         if (IS_ERR(be->mode)) {
 717                 err = PTR_ERR(be->mode);
 718                 be->mode = NULL;
 719                 xenbus_dev_fatal(dev, err, "reading mode");
 720                 return;
 721         }
 722
 723         device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
 724         if (!IS_ERR(device_type)) {
 725                 cdrom = strcmp(device_type, "cdrom") == 0;
 726                 kfree(device_type);
 727         }
 728
 729         /* Front end dir is a number, which is used as the handle. */
 730         err = kstrtoul(strrchr(dev->otherend, '/') + 1, 0, &handle);
 731         if (err) {
 732                 kfree(be->mode);
 733                 be->mode = NULL;
 734                 return;
 735         }
 736
 737         be->major = major;
 738         be->minor = minor;
 739
 740         err = xen_vbd_create(be->blkif, handle, major, minor,
 741                              !strchr(be->mode, 'w'), cdrom);
 742
 743         if (err)
 744                 xenbus_dev_fatal(dev, err, "creating vbd structure");
 745         else {
 746                 err = xenvbd_sysfs_addif(dev);
 747                 if (err) {
 748                         xen_vbd_free(&be->blkif->vbd);
 749                         xenbus_dev_fatal(dev, err, "creating sysfs entries");
 750                 }
 751         }
 752
 753         if (err) {
 754                 kfree(be->mode);
 755                 be->mode = NULL;
 756                 be->major = 0;
 757                 be->minor = 0;
 758         } else {
 759                 /* We're potentially connected now */
 760                 xen_update_blkif_status(be->blkif);
 761         }
 762 }
 763
 764
 765 /*
 766  * Callback received when the frontend's state changes.
 767  */
 768 static void frontend_changed(struct xenbus_device *dev,
 769                              enum xenbus_state frontend_state)
 770 {
 771         struct backend_info *be = dev_get_drvdata(&dev->dev);
 772         int err;
 773
 774         pr_debug("%s %p %s\n", __func__, dev, xenbus_strstate(frontend_state));
 775
 776         switch (frontend_state) {
 777         case XenbusStateInitialising:
 778                 if (dev->state == XenbusStateClosed) {
 779                         pr_info("%s: prepare for reconnect\n", dev->nodename);
 780                         xenbus_switch_state(dev, XenbusStateInitWait);
 781                 }
 782                 break;
 783
 784         case XenbusStateInitialised:
 785         case XenbusStateConnected:
 786                 /*
 787                  * Ensure we connect even when two watches fire in
 788                  * close succession and we miss the intermediate value
 789                  * of frontend_state.
 790                  */
 791                 if (dev->state == XenbusStateConnected)
 792                         break;
 793
 794                 /*
 795                  * Enforce precondition before potential leak point.
 796                  * xen_blkif_disconnect() is idempotent.
 797                  */
 798                 err = xen_blkif_disconnect(be->blkif);
 799                 if (err) {
 800                         xenbus_dev_fatal(dev, err, "pending I/O");
 801                         break;
 802                 }
 803
 804                 err = connect_ring(be);
 805                 if (err) {
 806                         /*
 807                          * Clean up so that memory resources can be used by
 808                          * other devices. connect_ring reported already error.
 809                          */
 810                         xen_blkif_disconnect(be->blkif);
 811                         break;
 812                 }
 813                 xen_update_blkif_status(be->blkif);
 814                 break;
 815
 816         case XenbusStateClosing:
 817                 xenbus_switch_state(dev, XenbusStateClosing);
 818                 break;
 819
 820         case XenbusStateClosed:
 821                 xen_blkif_disconnect(be->blkif);
 822                 xenbus_switch_state(dev, XenbusStateClosed);
 823                 if (xenbus_dev_is_online(dev))
 824                         break;
 825                 /* fall through if not online */
 826         case XenbusStateUnknown:
 827                 /* implies xen_blkif_disconnect() via xen_blkbk_remove() */
 828                 device_unregister(&dev->dev);
 829                 break;
 830
 831         default:
 832                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 833                                  frontend_state);
 834                 break;
 835         }
 836 }
 837
 838
 839 /* ** Connection ** */
 840
 841
 842 /*
 843  * Write the physical details regarding the block device to the store, and
 844  * switch to Connected state.
 845  */
 846 static void connect(struct backend_info *be)
 847 {
 848         struct xenbus_transaction xbt;
 849         int err;
 850         struct xenbus_device *dev = be->dev;
 851
 852         pr_debug("%s %s\n", __func__, dev->otherend);
 853
 854         /* Supply the information about the device the frontend needs */
 855 again:
 856         err = xenbus_transaction_start(&xbt);
 857         if (err) {
 858                 xenbus_dev_fatal(dev, err, "starting transaction");
 859                 return;
 860         }
 861
 862         /* If we can't advertise it is OK. */
 863         xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
 864
 865         xen_blkbk_discard(xbt, be);
 866
 867         xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
 868
 869         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u", 1);
 870         if (err) {
 871                 xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
 872                                  dev->nodename);
 873                 goto abort;
 874         }
 875
 876         err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
 877                             (unsigned long long)vbd_sz(&be->blkif->vbd));
 878         if (err) {
 879                 xenbus_dev_fatal(dev, err, "writing %s/sectors",
 880                                  dev->nodename);
 881                 goto abort;
 882         }
 883
 884         /* FIXME: use a typename instead */
 885         err = xenbus_printf(xbt, dev->nodename, "info", "%u",
 886                             be->blkif->vbd.type |
 887                             (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
 888         if (err) {
 889                 xenbus_dev_fatal(dev, err, "writing %s/info",
 890                                  dev->nodename);
 891                 goto abort;
 892         }
 893         err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
 894                             (unsigned long)
 895                             bdev_logical_block_size(be->blkif->vbd.bdev));
 896         if (err) {
 897                 xenbus_dev_fatal(dev, err, "writing %s/sector-size",
 898                                  dev->nodename);
 899                 goto abort;
 900         }
 901         err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
 902                             bdev_physical_block_size(be->blkif->vbd.bdev));
 903         if (err)
 904                 xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
 905                                  dev->nodename);
 906
 907         err = xenbus_transaction_end(xbt, 0);
 908         if (err == -EAGAIN)
 909                 goto again;
 910         if (err)
 911                 xenbus_dev_fatal(dev, err, "ending transaction");
 912
 913         err = xenbus_switch_state(dev, XenbusStateConnected);
 914         if (err)
 915                 xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
 916                                  dev->nodename);
 917
 918         return;
 919  abort:
 920         xenbus_transaction_end(xbt, 1);
 921 }
 922
 923 /*
 924  * Each ring may have multi pages, depends on "ring-page-order".
 925  */
 926 static int read_per_ring_refs(struct xen_blkif_ring *ring, const char *dir)
 927 {
 928         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 929         struct pending_req *req, *n;
 930         int err, i, j;
 931         struct xen_blkif *blkif = ring->blkif;
 932         struct xenbus_device *dev = blkif->be->dev;
 933         unsigned int ring_page_order, nr_grefs, evtchn;
 934
 935         err = xenbus_scanf(XBT_NIL, dir, "event-channel", "%u",
 936                           &evtchn);
 937         if (err != 1) {
 938                 err = -EINVAL;
 939                 xenbus_dev_fatal(dev, err, "reading %s/event-channel", dir);
 940                 return err;
 941         }
 942
 943         err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-page-order", "%u",
 944                           &ring_page_order);
 945         if (err != 1) {
 946                 err = xenbus_scanf(XBT_NIL, dir, "ring-ref", "%u", &ring_ref[0]);
 947                 if (err != 1) {
 948                         err = -EINVAL;
 949                         xenbus_dev_fatal(dev, err, "reading %s/ring-ref", dir);
 950                         return err;
 951                 }
 952                 nr_grefs = 1;
 953         } else {
 954                 unsigned int i;
 955
 956                 if (ring_page_order > xen_blkif_max_ring_order) {
 957                         err = -EINVAL;
 958                         xenbus_dev_fatal(dev, err, "%s/request %d ring page order exceed max:%d",
 959                                          dir, ring_page_order,
 960                                          xen_blkif_max_ring_order);
 961                         return err;
 962                 }
 963
 964                 nr_grefs = 1 << ring_page_order;
 965                 for (i = 0; i < nr_grefs; i++) {
 966                         char ring_ref_name[RINGREF_NAME_LEN];
 967
 968                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
 969                         err = xenbus_scanf(XBT_NIL, dir, ring_ref_name,
 970                                            "%u", &ring_ref[i]);
 971                         if (err != 1) {
 972                                 err = -EINVAL;
 973                                 xenbus_dev_fatal(dev, err, "reading %s/%s",
 974                                                  dir, ring_ref_name);
 975                                 return err;
 976                         }
 977                 }
 978         }
 979         blkif->nr_ring_pages = nr_grefs;
 980
 981         err = -ENOMEM;
 982         for (i = 0; i < nr_grefs * XEN_BLKIF_REQS_PER_PAGE; i++) {
 983                 req = kzalloc(sizeof(*req), GFP_KERNEL);
 984                 if (!req)
 985                         goto fail;
 986                 list_add_tail(&req->free_list, &ring->pending_free);
 987                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
 988                         req->segments[j] = kzalloc(sizeof(*req->segments[0]), GFP_KERNEL);
 989                         if (!req->segments[j])
 990                                 goto fail;
 991                 }
 992                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
 993                         req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
 994                                                          GFP_KERNEL);
 995                         if (!req->indirect_pages[j])
 996                                 goto fail;
 997                 }
 998         }
 999
1000         /* Map the shared frame, irq etc. */
1001         err = xen_blkif_map(ring, ring_ref, nr_grefs, evtchn);
1002         if (err) {
1003                 xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
1004                 goto fail;
1005         }
1006
1007         return 0;
1008
1009 fail:
1010         list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
1011                 list_del(&req->free_list);
1012                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
1013                         if (!req->segments[j])
1014                                 break;
1015                         kfree(req->segments[j]);
1016                 }
1017                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1018                         if (!req->indirect_pages[j])
1019                                 break;
1020                         kfree(req->indirect_pages[j]);
1021                 }
1022                 kfree(req);
1023         }
1024         return err;
1025 }
1026
1027 static int connect_ring(struct backend_info *be)
1028 {
1029         struct xenbus_device *dev = be->dev;
1030         unsigned int pers_grants;
1031         char protocol[64] = "";
1032         int err, i;
1033         char *xspath;
1034         size_t xspathsize;
1035         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1036         unsigned int requested_num_queues = 0;
1037
1038         pr_debug("%s %s\n", __func__, dev->otherend);
1039
1040         be->blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
1041         err = xenbus_scanf(XBT_NIL, dev->otherend, "protocol",
1042                            "%63s", protocol);
1043         if (err <= 0)
1044                 strcpy(protocol, "unspecified, assuming default");
1045         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
1046                 be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
1047         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
1048                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
1049         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
1050                 be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
1051         else {
1052                 xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
1053                 return -ENOSYS;
1054         }
1055         err = xenbus_scanf(XBT_NIL, dev->otherend,
1056                            "feature-persistent", "%u", &pers_grants);
1057         if (err <= 0)
1058                 pers_grants = 0;
1059
1060         be->blkif->vbd.feature_gnt_persistent = pers_grants;
1061         be->blkif->vbd.overflow_max_grants = 0;
1062
1063         /*
1064          * Read the number of hardware queues from frontend.
1065          */
1066         err = xenbus_scanf(XBT_NIL, dev->otherend, "multi-queue-num-queues",
1067                            "%u", &requested_num_queues);
1068         if (err < 0) {
1069                 requested_num_queues = 1;
1070         } else {
1071                 if (requested_num_queues > xenblk_max_queues
1072                     || requested_num_queues == 0) {
1073                         /* Buggy or malicious guest. */
1074                         xenbus_dev_fatal(dev, err,
1075                                         "guest requested %u queues, exceeding the maximum of %u.",
1076                                         requested_num_queues, xenblk_max_queues);
1077                         return -ENOSYS;
1078                 }
1079         }
1080         be->blkif->nr_rings = requested_num_queues;
1081         if (xen_blkif_alloc_rings(be->blkif))
1082                 return -ENOMEM;
1083
1084         pr_info("%s: using %d queues, protocol %d (%s) %s\n", dev->nodename,
1085                  be->blkif->nr_rings, be->blkif->blk_protocol, protocol,
1086                  pers_grants ? "persistent grants" : "");
1087
1088         if (be->blkif->nr_rings == 1)
1089                 return read_per_ring_refs(&be->blkif->rings[0], dev->otherend);
1090         else {
1091                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1092                 xspath = kmalloc(xspathsize, GFP_KERNEL);
1093                 if (!xspath) {
1094                         xenbus_dev_fatal(dev, -ENOMEM, "reading ring references");
1095                         return -ENOMEM;
1096                 }
1097
1098                 for (i = 0; i < be->blkif->nr_rings; i++) {
1099                         memset(xspath, 0, xspathsize);
1100                         snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, i);
1101                         err = read_per_ring_refs(&be->blkif->rings[i], xspath);
1102                         if (err) {
1103                                 kfree(xspath);
1104                                 return err;
1105                         }
1106                 }
1107                 kfree(xspath);
1108         }
1109         return 0;
1110 }
1111
1112 static const struct xenbus_device_id xen_blkbk_ids[] = {
1113         { "vbd" },
1114         { "" }
1115 };
1116
1117 static struct xenbus_driver xen_blkbk_driver = {
1118         .ids  = xen_blkbk_ids,
1119         .probe = xen_blkbk_probe,
1120         .remove = xen_blkbk_remove,
1121         .otherend_changed = frontend_changed
1122 };
1123
1124 int xen_blkif_xenbus_init(void)
1125 {
1126         return xenbus_register_backend(&xen_blkbk_driver);
1127 }