drivers/block/xen-blkback/xenbus.c

/* [<][>][^][v][top][bottom][index][help] */
This source file includes following definitions.
xen_blkbk_xenbus
xen_blkif_deferred_free
blkback_name
xen_update_blkif_status
xen_blkif_alloc_rings
xen_blkif_alloc
xen_blkif_map
xen_blkif_disconnect
xen_blkif_free
xen_blkif_interface_init
xenvbd_sysfs_addif
xenvbd_sysfs_delif
xen_vbd_free
xen_vbd_create
xen_blkbk_remove
xen_blkbk_flush_diskcache
xen_blkbk_discard
xen_blkbk_barrier
xen_blkbk_probe
backend_changed
frontend_changed
connect
read_per_ring_refs
connect_ring
xen_blkif_xenbus_init
   1 // SPDX-License-Identifier: GPL-2.0-or-later
   2 /*  Xenbus code for blkif backend
   3     Copyright (C) 2005 Rusty Russell <rusty@rustcorp.com.au>
   4     Copyright (C) 2005 XenSource Ltd
   5 
   6 
   7 */
   8 
   9 #define pr_fmt(fmt) "xen-blkback: " fmt
  10 
  11 #include <stdarg.h>
  12 #include <linux/module.h>
  13 #include <linux/kthread.h>
  14 #include <xen/events.h>
  15 #include <xen/grant_table.h>
  16 #include "common.h"
  17 
  18 /* On the XenBus the max length of 'ring-ref%u'. */
  19 #define RINGREF_NAME_LEN (20)
  20 
  21 struct backend_info {
  22         struct xenbus_device    *dev;
  23         struct xen_blkif        *blkif;
  24         struct xenbus_watch     backend_watch;
  25         unsigned                major;
  26         unsigned                minor;
  27         char                    *mode;
  28 };
  29 
  30 static struct kmem_cache *xen_blkif_cachep;
  31 static void connect(struct backend_info *);
  32 static int connect_ring(struct backend_info *);
  33 static void backend_changed(struct xenbus_watch *, const char *,
  34                             const char *);
  35 static void xen_blkif_free(struct xen_blkif *blkif);
  36 static void xen_vbd_free(struct xen_vbd *vbd);
  37 
  38 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be)
  39 {
  40         return be->dev;
  41 }
  42 
  43 /*
  44  * The last request could free the device from softirq context and
  45  * xen_blkif_free() can sleep.
  46  */
  47 static void xen_blkif_deferred_free(struct work_struct *work)
  48 {
  49         struct xen_blkif *blkif;
  50 
  51         blkif = container_of(work, struct xen_blkif, free_work);
  52         xen_blkif_free(blkif);
  53 }
  54 
  55 static int blkback_name(struct xen_blkif *blkif, char *buf)
  56 {
  57         char *devpath, *devname;
  58         struct xenbus_device *dev = blkif->be->dev;
  59 
  60         devpath = xenbus_read(XBT_NIL, dev->nodename, "dev", NULL);
  61         if (IS_ERR(devpath))
  62                 return PTR_ERR(devpath);
  63 
  64         devname = strstr(devpath, "/dev/");
  65         if (devname != NULL)
  66                 devname += strlen("/dev/");
  67         else
  68                 devname  = devpath;
  69 
  70         snprintf(buf, TASK_COMM_LEN, "%d.%s", blkif->domid, devname);
  71         kfree(devpath);
  72 
  73         return 0;
  74 }
  75 
  76 static void xen_update_blkif_status(struct xen_blkif *blkif)
  77 {
  78         int err;
  79         char name[TASK_COMM_LEN];
  80         struct xen_blkif_ring *ring;
  81         int i;
  82 
  83         /* Not ready to connect? */
  84         if (!blkif->rings || !blkif->rings[0].irq || !blkif->vbd.bdev)
  85                 return;
  86 
  87         /* Already connected? */
  88         if (blkif->be->dev->state == XenbusStateConnected)
  89                 return;
  90 
  91         /* Attempt to connect: exit if we fail to. */
  92         connect(blkif->be);
  93         if (blkif->be->dev->state != XenbusStateConnected)
  94                 return;
  95 
  96         err = blkback_name(blkif, name);
  97         if (err) {
  98                 xenbus_dev_error(blkif->be->dev, err, "get blkback dev name");
  99                 return;
 100         }
 101 
 102         err = filemap_write_and_wait(blkif->vbd.bdev->bd_inode->i_mapping);
 103         if (err) {
 104                 xenbus_dev_error(blkif->be->dev, err, "block flush");
 105                 return;
 106         }
 107         invalidate_inode_pages2(blkif->vbd.bdev->bd_inode->i_mapping);
 108 
 109         for (i = 0; i < blkif->nr_rings; i++) {
 110                 ring = &blkif->rings[i];
 111                 ring->xenblkd = kthread_run(xen_blkif_schedule, ring, "%s-%d", name, i);
 112                 if (IS_ERR(ring->xenblkd)) {
 113                         err = PTR_ERR(ring->xenblkd);
 114                         ring->xenblkd = NULL;
 115                         xenbus_dev_fatal(blkif->be->dev, err,
 116                                         "start %s-%d xenblkd", name, i);
 117                         goto out;
 118                 }
 119         }
 120         return;
 121 
 122 out:
 123         while (--i >= 0) {
 124                 ring = &blkif->rings[i];
 125                 kthread_stop(ring->xenblkd);
 126         }
 127         return;
 128 }
 129 
 130 static int xen_blkif_alloc_rings(struct xen_blkif *blkif)
 131 {
 132         unsigned int r;
 133 
 134         blkif->rings = kcalloc(blkif->nr_rings, sizeof(struct xen_blkif_ring),
 135                                GFP_KERNEL);
 136         if (!blkif->rings)
 137                 return -ENOMEM;
 138 
 139         for (r = 0; r < blkif->nr_rings; r++) {
 140                 struct xen_blkif_ring *ring = &blkif->rings[r];
 141 
 142                 spin_lock_init(&ring->blk_ring_lock);
 143                 init_waitqueue_head(&ring->wq);
 144                 INIT_LIST_HEAD(&ring->pending_free);
 145                 INIT_LIST_HEAD(&ring->persistent_purge_list);
 146                 INIT_WORK(&ring->persistent_purge_work, xen_blkbk_unmap_purged_grants);
 147                 spin_lock_init(&ring->free_pages_lock);
 148                 INIT_LIST_HEAD(&ring->free_pages);
 149 
 150                 spin_lock_init(&ring->pending_free_lock);
 151                 init_waitqueue_head(&ring->pending_free_wq);
 152                 init_waitqueue_head(&ring->shutdown_wq);
 153                 ring->blkif = blkif;
 154                 ring->st_print = jiffies;
 155                 ring->active = true;
 156         }
 157 
 158         return 0;
 159 }
 160 
 161 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
 162 {
 163         struct xen_blkif *blkif;
 164 
 165         BUILD_BUG_ON(MAX_INDIRECT_PAGES > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
 166 
 167         blkif = kmem_cache_zalloc(xen_blkif_cachep, GFP_KERNEL);
 168         if (!blkif)
 169                 return ERR_PTR(-ENOMEM);
 170 
 171         blkif->domid = domid;
 172         atomic_set(&blkif->refcnt, 1);
 173         init_completion(&blkif->drain_complete);
 174 
 175         /*
 176          * Because freeing back to the cache may be deferred, it is not
 177          * safe to unload the module (and hence destroy the cache) until
 178          * this has completed. To prevent premature unloading, take an
 179          * extra module reference here and release only when the object
 180          * has been freed back to the cache.
 181          */
 182         __module_get(THIS_MODULE);
 183         INIT_WORK(&blkif->free_work, xen_blkif_deferred_free);
 184 
 185         return blkif;
 186 }
 187 
 188 static int xen_blkif_map(struct xen_blkif_ring *ring, grant_ref_t *gref,
 189                          unsigned int nr_grefs, unsigned int evtchn)
 190 {
 191         int err;
 192         struct xen_blkif *blkif = ring->blkif;
 193 
 194         /* Already connected through? */
 195         if (ring->irq)
 196                 return 0;
 197 
 198         err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
 199                                      &ring->blk_ring);
 200         if (err < 0)
 201                 return err;
 202 
 203         switch (blkif->blk_protocol) {
 204         case BLKIF_PROTOCOL_NATIVE:
 205         {
 206                 struct blkif_sring *sring;
 207                 sring = (struct blkif_sring *)ring->blk_ring;
 208                 BACK_RING_INIT(&ring->blk_rings.native, sring,
 209                                XEN_PAGE_SIZE * nr_grefs);
 210                 break;
 211         }
 212         case BLKIF_PROTOCOL_X86_32:
 213         {
 214                 struct blkif_x86_32_sring *sring_x86_32;
 215                 sring_x86_32 = (struct blkif_x86_32_sring *)ring->blk_ring;
 216                 BACK_RING_INIT(&ring->blk_rings.x86_32, sring_x86_32,
 217                                XEN_PAGE_SIZE * nr_grefs);
 218                 break;
 219         }
 220         case BLKIF_PROTOCOL_X86_64:
 221         {
 222                 struct blkif_x86_64_sring *sring_x86_64;
 223                 sring_x86_64 = (struct blkif_x86_64_sring *)ring->blk_ring;
 224                 BACK_RING_INIT(&ring->blk_rings.x86_64, sring_x86_64,
 225                                XEN_PAGE_SIZE * nr_grefs);
 226                 break;
 227         }
 228         default:
 229                 BUG();
 230         }
 231 
 232         err = bind_interdomain_evtchn_to_irqhandler(blkif->domid, evtchn,
 233                                                     xen_blkif_be_int, 0,
 234                                                     "blkif-backend", ring);
 235         if (err < 0) {
 236                 xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
 237                 ring->blk_rings.common.sring = NULL;
 238                 return err;
 239         }
 240         ring->irq = err;
 241 
 242         return 0;
 243 }
 244 
 245 static int xen_blkif_disconnect(struct xen_blkif *blkif)
 246 {
 247         struct pending_req *req, *n;
 248         unsigned int j, r;
 249         bool busy = false;
 250 
 251         for (r = 0; r < blkif->nr_rings; r++) {
 252                 struct xen_blkif_ring *ring = &blkif->rings[r];
 253                 unsigned int i = 0;
 254 
 255                 if (!ring->active)
 256                         continue;
 257 
 258                 if (ring->xenblkd) {
 259                         kthread_stop(ring->xenblkd);
 260                         wake_up(&ring->shutdown_wq);
 261                 }
 262 
 263                 /* The above kthread_stop() guarantees that at this point we
 264                  * don't have any discard_io or other_io requests. So, checking
 265                  * for inflight IO is enough.
 266                  */
 267                 if (atomic_read(&ring->inflight) > 0) {
 268                         busy = true;
 269                         continue;
 270                 }
 271 
 272                 if (ring->irq) {
 273                         unbind_from_irqhandler(ring->irq, ring);
 274                         ring->irq = 0;
 275                 }
 276 
 277                 if (ring->blk_rings.common.sring) {
 278                         xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
 279                         ring->blk_rings.common.sring = NULL;
 280                 }
 281 
 282                 /* Remove all persistent grants and the cache of ballooned pages. */
 283                 xen_blkbk_free_caches(ring);
 284 
 285                 /* Check that there is no request in use */
 286                 list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
 287                         list_del(&req->free_list);
 288 
 289                         for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
 290                                 kfree(req->segments[j]);
 291 
 292                         for (j = 0; j < MAX_INDIRECT_PAGES; j++)
 293                                 kfree(req->indirect_pages[j]);
 294 
 295                         kfree(req);
 296                         i++;
 297                 }
 298 
 299                 BUG_ON(atomic_read(&ring->persistent_gnt_in_use) != 0);
 300                 BUG_ON(!list_empty(&ring->persistent_purge_list));
 301                 BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts));
 302                 BUG_ON(!list_empty(&ring->free_pages));
 303                 BUG_ON(ring->free_pages_num != 0);
 304                 BUG_ON(ring->persistent_gnt_c != 0);
 305                 WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
 306                 ring->active = false;
 307         }
 308         if (busy)
 309                 return -EBUSY;
 310 
 311         blkif->nr_ring_pages = 0;
 312         /*
 313          * blkif->rings was allocated in connect_ring, so we should free it in
 314          * here.
 315          */
 316         kfree(blkif->rings);
 317         blkif->rings = NULL;
 318         blkif->nr_rings = 0;
 319 
 320         return 0;
 321 }
 322 
 323 static void xen_blkif_free(struct xen_blkif *blkif)
 324 {
 325         WARN_ON(xen_blkif_disconnect(blkif));
 326         xen_vbd_free(&blkif->vbd);
 327         kfree(blkif->be->mode);
 328         kfree(blkif->be);
 329 
 330         /* Make sure everything is drained before shutting down */
 331         kmem_cache_free(xen_blkif_cachep, blkif);
 332         module_put(THIS_MODULE);
 333 }
 334 
 335 int __init xen_blkif_interface_init(void)
 336 {
 337         xen_blkif_cachep = kmem_cache_create("blkif_cache",
 338                                              sizeof(struct xen_blkif),
 339                                              0, 0, NULL);
 340         if (!xen_blkif_cachep)
 341                 return -ENOMEM;
 342 
 343         return 0;
 344 }
 345 
 346 /*
 347  *  sysfs interface for VBD I/O requests
 348  */
 349 
 350 #define VBD_SHOW_ALLRING(name, format)                                  \
 351         static ssize_t show_##name(struct device *_dev,                 \
 352                                    struct device_attribute *attr,       \
 353                                    char *buf)                           \
 354         {                                                               \
 355                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 356                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 357                 struct xen_blkif *blkif = be->blkif;                    \
 358                 unsigned int i;                                         \
 359                 unsigned long long result = 0;                          \
 360                                                                         \
 361                 if (!blkif->rings)                              \
 362                         goto out;                                       \
 363                                                                         \
 364                 for (i = 0; i < blkif->nr_rings; i++) {         \
 365                         struct xen_blkif_ring *ring = &blkif->rings[i]; \
 366                                                                         \
 367                         result += ring->st_##name;                      \
 368                 }                                                       \
 369                                                                         \
 370 out:                                                                    \
 371                 return sprintf(buf, format, result);                    \
 372         }                                                               \
 373         static DEVICE_ATTR(name, 0444, show_##name, NULL)
 374 
 375 VBD_SHOW_ALLRING(oo_req,  "%llu\n");
 376 VBD_SHOW_ALLRING(rd_req,  "%llu\n");
 377 VBD_SHOW_ALLRING(wr_req,  "%llu\n");
 378 VBD_SHOW_ALLRING(f_req,  "%llu\n");
 379 VBD_SHOW_ALLRING(ds_req,  "%llu\n");
 380 VBD_SHOW_ALLRING(rd_sect, "%llu\n");
 381 VBD_SHOW_ALLRING(wr_sect, "%llu\n");
 382 
 383 static struct attribute *xen_vbdstat_attrs[] = {
 384         &dev_attr_oo_req.attr,
 385         &dev_attr_rd_req.attr,
 386         &dev_attr_wr_req.attr,
 387         &dev_attr_f_req.attr,
 388         &dev_attr_ds_req.attr,
 389         &dev_attr_rd_sect.attr,
 390         &dev_attr_wr_sect.attr,
 391         NULL
 392 };
 393 
 394 static const struct attribute_group xen_vbdstat_group = {
 395         .name = "statistics",
 396         .attrs = xen_vbdstat_attrs,
 397 };
 398 
 399 #define VBD_SHOW(name, format, args...)                                 \
 400         static ssize_t show_##name(struct device *_dev,                 \
 401                                    struct device_attribute *attr,       \
 402                                    char *buf)                           \
 403         {                                                               \
 404                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 405                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 406                                                                         \
 407                 return sprintf(buf, format, ##args);                    \
 408         }                                                               \
 409         static DEVICE_ATTR(name, 0444, show_##name, NULL)
 410 
 411 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
 412 VBD_SHOW(mode, "%s\n", be->mode);
 413 
 414 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
 415 {
 416         int error;
 417 
 418         error = device_create_file(&dev->dev, &dev_attr_physical_device);
 419         if (error)
 420                 goto fail1;
 421 
 422         error = device_create_file(&dev->dev, &dev_attr_mode);
 423         if (error)
 424                 goto fail2;
 425 
 426         error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
 427         if (error)
 428                 goto fail3;
 429 
 430         return 0;
 431 
 432 fail3:  sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 433 fail2:  device_remove_file(&dev->dev, &dev_attr_mode);
 434 fail1:  device_remove_file(&dev->dev, &dev_attr_physical_device);
 435         return error;
 436 }
 437 
 438 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
 439 {
 440         sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 441         device_remove_file(&dev->dev, &dev_attr_mode);
 442         device_remove_file(&dev->dev, &dev_attr_physical_device);
 443 }
 444 
 445 
 446 static void xen_vbd_free(struct xen_vbd *vbd)
 447 {
 448         if (vbd->bdev)
 449                 blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
 450         vbd->bdev = NULL;
 451 }
 452 
 453 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
 454                           unsigned major, unsigned minor, int readonly,
 455                           int cdrom)
 456 {
 457         struct xen_vbd *vbd;
 458         struct block_device *bdev;
 459         struct request_queue *q;
 460 
 461         vbd = &blkif->vbd;
 462         vbd->handle   = handle;
 463         vbd->readonly = readonly;
 464         vbd->type     = 0;
 465 
 466         vbd->pdevice  = MKDEV(major, minor);
 467 
 468         bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
 469                                  FMODE_READ : FMODE_WRITE, NULL);
 470 
 471         if (IS_ERR(bdev)) {
 472                 pr_warn("xen_vbd_create: device %08x could not be opened\n",
 473                         vbd->pdevice);
 474                 return -ENOENT;
 475         }
 476 
 477         vbd->bdev = bdev;
 478         if (vbd->bdev->bd_disk == NULL) {
 479                 pr_warn("xen_vbd_create: device %08x doesn't exist\n",
 480                         vbd->pdevice);
 481                 xen_vbd_free(vbd);
 482                 return -ENOENT;
 483         }
 484         vbd->size = vbd_sz(vbd);
 485 
 486         if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
 487                 vbd->type |= VDISK_CDROM;
 488         if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
 489                 vbd->type |= VDISK_REMOVABLE;
 490 
 491         q = bdev_get_queue(bdev);
 492         if (q && test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 493                 vbd->flush_support = true;
 494 
 495         if (q && blk_queue_secure_erase(q))
 496                 vbd->discard_secure = true;
 497 
 498         pr_debug("Successful creation of handle=%04x (dom=%u)\n",
 499                 handle, blkif->domid);
 500         return 0;
 501 }
 502 static int xen_blkbk_remove(struct xenbus_device *dev)
 503 {
 504         struct backend_info *be = dev_get_drvdata(&dev->dev);
 505 
 506         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 507 
 508         if (be->major || be->minor)
 509                 xenvbd_sysfs_delif(dev);
 510 
 511         if (be->backend_watch.node) {
 512                 unregister_xenbus_watch(&be->backend_watch);
 513                 kfree(be->backend_watch.node);
 514                 be->backend_watch.node = NULL;
 515         }
 516 
 517         dev_set_drvdata(&dev->dev, NULL);
 518 
 519         if (be->blkif) {
 520                 xen_blkif_disconnect(be->blkif);
 521 
 522                 /* Put the reference we set in xen_blkif_alloc(). */
 523                 xen_blkif_put(be->blkif);
 524         }
 525 
 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;
 549         struct block_device *bdev = be->blkif->vbd.bdev;
 550         struct request_queue *q = bdev_get_queue(bdev);
 551 
 552         if (!xenbus_read_unsigned(dev->nodename, "discard-enable", 1))
 553                 return;
 554 
 555         if (blk_queue_discard(q)) {
 556                 err = xenbus_printf(xbt, dev->nodename,
 557                         "discard-granularity", "%u",
 558                         q->limits.discard_granularity);
 559                 if (err) {
 560                         dev_warn(&dev->dev, "writing discard-granularity (%d)", err);
 561                         return;
 562                 }
 563                 err = xenbus_printf(xbt, dev->nodename,
 564                         "discard-alignment", "%u",
 565                         q->limits.discard_alignment);
 566                 if (err) {
 567                         dev_warn(&dev->dev, "writing discard-alignment (%d)", err);
 568                         return;
 569                 }
 570                 state = 1;
 571                 /* Optional. */
 572                 err = xenbus_printf(xbt, dev->nodename,
 573                                     "discard-secure", "%d",
 574                                     blkif->vbd.discard_secure);
 575                 if (err) {
 576                         dev_warn(&dev->dev, "writing discard-secure (%d)", err);
 577                         return;
 578                 }
 579         }
 580         err = xenbus_printf(xbt, dev->nodename, "feature-discard",
 581                             "%d", state);
 582         if (err)
 583                 dev_warn(&dev->dev, "writing feature-discard (%d)", err);
 584 }
 585 int xen_blkbk_barrier(struct xenbus_transaction xbt,
 586                       struct backend_info *be, int state)
 587 {
 588         struct xenbus_device *dev = be->dev;
 589         int err;
 590 
 591         err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
 592                             "%d", state);
 593         if (err)
 594                 dev_warn(&dev->dev, "writing feature-barrier (%d)", err);
 595 
 596         return err;
 597 }
 598 
 599 /*
 600  * Entry point to this code when a new device is created.  Allocate the basic
 601  * structures, and watch the store waiting for the hotplug scripts to tell us
 602  * the device's physical major and minor numbers.  Switch to InitWait.
 603  */
 604 static int xen_blkbk_probe(struct xenbus_device *dev,
 605                            const struct xenbus_device_id *id)
 606 {
 607         int err;
 608         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 609                                           GFP_KERNEL);
 610 
 611         /* match the pr_debug in xen_blkbk_remove */
 612         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 613 
 614         if (!be) {
 615                 xenbus_dev_fatal(dev, -ENOMEM,
 616                                  "allocating backend structure");
 617                 return -ENOMEM;
 618         }
 619         be->dev = dev;
 620         dev_set_drvdata(&dev->dev, be);
 621 
 622         be->blkif = xen_blkif_alloc(dev->otherend_id);
 623         if (IS_ERR(be->blkif)) {
 624                 err = PTR_ERR(be->blkif);
 625                 be->blkif = NULL;
 626                 xenbus_dev_fatal(dev, err, "creating block interface");
 627                 goto fail;
 628         }
 629 
 630         err = xenbus_printf(XBT_NIL, dev->nodename,
 631                             "feature-max-indirect-segments", "%u",
 632                             MAX_INDIRECT_SEGMENTS);
 633         if (err)
 634                 dev_warn(&dev->dev,
 635                          "writing %s/feature-max-indirect-segments (%d)",
 636                          dev->nodename, err);
 637 
 638         /* Multi-queue: advertise how many queues are supported by us.*/
 639         err = xenbus_printf(XBT_NIL, dev->nodename,
 640                             "multi-queue-max-queues", "%u", xenblk_max_queues);
 641         if (err)
 642                 pr_warn("Error writing multi-queue-max-queues\n");
 643 
 644         /* setup back pointer */
 645         be->blkif->be = be;
 646 
 647         err = xenbus_watch_pathfmt(dev, &be->backend_watch, backend_changed,
 648                                    "%s/%s", dev->nodename, "physical-device");
 649         if (err)
 650                 goto fail;
 651 
 652         err = xenbus_printf(XBT_NIL, dev->nodename, "max-ring-page-order", "%u",
 653                             xen_blkif_max_ring_order);
 654         if (err)
 655                 pr_warn("%s write out 'max-ring-page-order' failed\n", __func__);
 656 
 657         err = xenbus_switch_state(dev, XenbusStateInitWait);
 658         if (err)
 659                 goto fail;
 660 
 661         return 0;
 662 
 663 fail:
 664         pr_warn("%s failed\n", __func__);
 665         xen_blkbk_remove(dev);
 666         return err;
 667 }
 668 
 669 
 670 /*
 671  * Callback received when the hotplug scripts have placed the physical-device
 672  * node.  Read it and the mode node, and create a vbd.  If the frontend is
 673  * ready, connect.
 674  */
 675 static void backend_changed(struct xenbus_watch *watch,
 676                             const char *path, const char *token)
 677 {
 678         int err;
 679         unsigned major;
 680         unsigned minor;
 681         struct backend_info *be
 682                 = container_of(watch, struct backend_info, backend_watch);
 683         struct xenbus_device *dev = be->dev;
 684         int cdrom = 0;
 685         unsigned long handle;
 686         char *device_type;
 687 
 688         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 689 
 690         err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
 691                            &major, &minor);
 692         if (XENBUS_EXIST_ERR(err)) {
 693                 /*
 694                  * Since this watch will fire once immediately after it is
 695                  * registered, we expect this.  Ignore it, and wait for the
 696                  * hotplug scripts.
 697                  */
 698                 return;
 699         }
 700         if (err != 2) {
 701                 xenbus_dev_fatal(dev, err, "reading physical-device");
 702                 return;
 703         }
 704 
 705         if (be->major | be->minor) {
 706                 if (be->major != major || be->minor != minor)
 707                         pr_warn("changing physical device (from %x:%x to %x:%x) not supported.\n",
 708                                 be->major, be->minor, major, minor);
 709                 return;
 710         }
 711 
 712         be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
 713         if (IS_ERR(be->mode)) {
 714                 err = PTR_ERR(be->mode);
 715                 be->mode = NULL;
 716                 xenbus_dev_fatal(dev, err, "reading mode");
 717                 return;
 718         }
 719 
 720         device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
 721         if (!IS_ERR(device_type)) {
 722                 cdrom = strcmp(device_type, "cdrom") == 0;
 723                 kfree(device_type);
 724         }
 725 
 726         /* Front end dir is a number, which is used as the handle. */
 727         err = kstrtoul(strrchr(dev->otherend, '/') + 1, 0, &handle);
 728         if (err) {
 729                 kfree(be->mode);
 730                 be->mode = NULL;
 731                 return;
 732         }
 733 
 734         be->major = major;
 735         be->minor = minor;
 736 
 737         err = xen_vbd_create(be->blkif, handle, major, minor,
 738                              !strchr(be->mode, 'w'), cdrom);
 739 
 740         if (err)
 741                 xenbus_dev_fatal(dev, err, "creating vbd structure");
 742         else {
 743                 err = xenvbd_sysfs_addif(dev);
 744                 if (err) {
 745                         xen_vbd_free(&be->blkif->vbd);
 746                         xenbus_dev_fatal(dev, err, "creating sysfs entries");
 747                 }
 748         }
 749 
 750         if (err) {
 751                 kfree(be->mode);
 752                 be->mode = NULL;
 753                 be->major = 0;
 754                 be->minor = 0;
 755         } else {
 756                 /* We're potentially connected now */
 757                 xen_update_blkif_status(be->blkif);
 758         }
 759 }
 760 
 761 
 762 /*
 763  * Callback received when the frontend's state changes.
 764  */
 765 static void frontend_changed(struct xenbus_device *dev,
 766                              enum xenbus_state frontend_state)
 767 {
 768         struct backend_info *be = dev_get_drvdata(&dev->dev);
 769         int err;
 770 
 771         pr_debug("%s %p %s\n", __func__, dev, xenbus_strstate(frontend_state));
 772 
 773         switch (frontend_state) {
 774         case XenbusStateInitialising:
 775                 if (dev->state == XenbusStateClosed) {
 776                         pr_info("%s: prepare for reconnect\n", dev->nodename);
 777                         xenbus_switch_state(dev, XenbusStateInitWait);
 778                 }
 779                 break;
 780 
 781         case XenbusStateInitialised:
 782         case XenbusStateConnected:
 783                 /*
 784                  * Ensure we connect even when two watches fire in
 785                  * close succession and we miss the intermediate value
 786                  * of frontend_state.
 787                  */
 788                 if (dev->state == XenbusStateConnected)
 789                         break;
 790 
 791                 /*
 792                  * Enforce precondition before potential leak point.
 793                  * xen_blkif_disconnect() is idempotent.
 794                  */
 795                 err = xen_blkif_disconnect(be->blkif);
 796                 if (err) {
 797                         xenbus_dev_fatal(dev, err, "pending I/O");
 798                         break;
 799                 }
 800 
 801                 err = connect_ring(be);
 802                 if (err) {
 803                         /*
 804                          * Clean up so that memory resources can be used by
 805                          * other devices. connect_ring reported already error.
 806                          */
 807                         xen_blkif_disconnect(be->blkif);
 808                         break;
 809                 }
 810                 xen_update_blkif_status(be->blkif);
 811                 break;
 812 
 813         case XenbusStateClosing:
 814                 xenbus_switch_state(dev, XenbusStateClosing);
 815                 break;
 816 
 817         case XenbusStateClosed:
 818                 xen_blkif_disconnect(be->blkif);
 819                 xenbus_switch_state(dev, XenbusStateClosed);
 820                 if (xenbus_dev_is_online(dev))
 821                         break;
 822                 /* fall through */
 823                 /* if not online */
 824         case XenbusStateUnknown:
 825                 /* implies xen_blkif_disconnect() via xen_blkbk_remove() */
 826                 device_unregister(&dev->dev);
 827                 break;
 828 
 829         default:
 830                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 831                                  frontend_state);
 832                 break;
 833         }
 834 }
 835 
 836 
 837 /* ** Connection ** */
 838 
 839 
 840 /*
 841  * Write the physical details regarding the block device to the store, and
 842  * switch to Connected state.
 843  */
 844 static void connect(struct backend_info *be)
 845 {
 846         struct xenbus_transaction xbt;
 847         int err;
 848         struct xenbus_device *dev = be->dev;
 849 
 850         pr_debug("%s %s\n", __func__, dev->otherend);
 851 
 852         /* Supply the information about the device the frontend needs */
 853 again:
 854         err = xenbus_transaction_start(&xbt);
 855         if (err) {
 856                 xenbus_dev_fatal(dev, err, "starting transaction");
 857                 return;
 858         }
 859 
 860         /* If we can't advertise it is OK. */
 861         xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
 862 
 863         xen_blkbk_discard(xbt, be);
 864 
 865         xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
 866 
 867         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u", 1);
 868         if (err) {
 869                 xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
 870                                  dev->nodename);
 871                 goto abort;
 872         }
 873 
 874         err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
 875                             (unsigned long long)vbd_sz(&be->blkif->vbd));
 876         if (err) {
 877                 xenbus_dev_fatal(dev, err, "writing %s/sectors",
 878                                  dev->nodename);
 879                 goto abort;
 880         }
 881 
 882         /* FIXME: use a typename instead */
 883         err = xenbus_printf(xbt, dev->nodename, "info", "%u",
 884                             be->blkif->vbd.type |
 885                             (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
 886         if (err) {
 887                 xenbus_dev_fatal(dev, err, "writing %s/info",
 888                                  dev->nodename);
 889                 goto abort;
 890         }
 891         err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
 892                             (unsigned long)
 893                             bdev_logical_block_size(be->blkif->vbd.bdev));
 894         if (err) {
 895                 xenbus_dev_fatal(dev, err, "writing %s/sector-size",
 896                                  dev->nodename);
 897                 goto abort;
 898         }
 899         err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
 900                             bdev_physical_block_size(be->blkif->vbd.bdev));
 901         if (err)
 902                 xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
 903                                  dev->nodename);
 904 
 905         err = xenbus_transaction_end(xbt, 0);
 906         if (err == -EAGAIN)
 907                 goto again;
 908         if (err)
 909                 xenbus_dev_fatal(dev, err, "ending transaction");
 910 
 911         err = xenbus_switch_state(dev, XenbusStateConnected);
 912         if (err)
 913                 xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
 914                                  dev->nodename);
 915 
 916         return;
 917  abort:
 918         xenbus_transaction_end(xbt, 1);
 919 }
 920 
 921 /*
 922  * Each ring may have multi pages, depends on "ring-page-order".
 923  */
 924 static int read_per_ring_refs(struct xen_blkif_ring *ring, const char *dir)
 925 {
 926         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 927         struct pending_req *req, *n;
 928         int err, i, j;
 929         struct xen_blkif *blkif = ring->blkif;
 930         struct xenbus_device *dev = blkif->be->dev;
 931         unsigned int nr_grefs, evtchn;
 932 
 933         err = xenbus_scanf(XBT_NIL, dir, "event-channel", "%u",
 934                           &evtchn);
 935         if (err != 1) {
 936                 err = -EINVAL;
 937                 xenbus_dev_fatal(dev, err, "reading %s/event-channel", dir);
 938                 return err;
 939         }
 940 
 941         nr_grefs = blkif->nr_ring_pages;
 942 
 943         if (unlikely(!nr_grefs)) {
 944                 WARN_ON(true);
 945                 return -EINVAL;
 946         }
 947 
 948         for (i = 0; i < nr_grefs; i++) {
 949                 char ring_ref_name[RINGREF_NAME_LEN];
 950 
 951                 snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
 952                 err = xenbus_scanf(XBT_NIL, dir, ring_ref_name,
 953                                    "%u", &ring_ref[i]);
 954 
 955                 if (err != 1) {
 956                         if (nr_grefs == 1)
 957                                 break;
 958 
 959                         err = -EINVAL;
 960                         xenbus_dev_fatal(dev, err, "reading %s/%s",
 961                                          dir, ring_ref_name);
 962                         return err;
 963                 }
 964         }
 965 
 966         if (err != 1) {
 967                 WARN_ON(nr_grefs != 1);
 968 
 969                 err = xenbus_scanf(XBT_NIL, dir, "ring-ref", "%u",
 970                                    &ring_ref[0]);
 971                 if (err != 1) {
 972                         err = -EINVAL;
 973                         xenbus_dev_fatal(dev, err, "reading %s/ring-ref", dir);
 974                         return err;
 975                 }
 976         }
 977 
 978         err = -ENOMEM;
 979         for (i = 0; i < nr_grefs * XEN_BLKIF_REQS_PER_PAGE; i++) {
 980                 req = kzalloc(sizeof(*req), GFP_KERNEL);
 981                 if (!req)
 982                         goto fail;
 983                 list_add_tail(&req->free_list, &ring->pending_free);
 984                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
 985                         req->segments[j] = kzalloc(sizeof(*req->segments[0]), GFP_KERNEL);
 986                         if (!req->segments[j])
 987                                 goto fail;
 988                 }
 989                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
 990                         req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
 991                                                          GFP_KERNEL);
 992                         if (!req->indirect_pages[j])
 993                                 goto fail;
 994                 }
 995         }
 996 
 997         /* Map the shared frame, irq etc. */
 998         err = xen_blkif_map(ring, ring_ref, nr_grefs, evtchn);
 999         if (err) {
1000                 xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
1001                 goto fail;
1002         }
1003 
1004         return 0;
1005 
1006 fail:
1007         list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
1008                 list_del(&req->free_list);
1009                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
1010                         if (!req->segments[j])
1011                                 break;
1012                         kfree(req->segments[j]);
1013                 }
1014                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1015                         if (!req->indirect_pages[j])
1016                                 break;
1017                         kfree(req->indirect_pages[j]);
1018                 }
1019                 kfree(req);
1020         }
1021         return err;
1022 }
1023 
1024 static int connect_ring(struct backend_info *be)
1025 {
1026         struct xenbus_device *dev = be->dev;
1027         struct xen_blkif *blkif = be->blkif;
1028         unsigned int pers_grants;
1029         char protocol[64] = "";
1030         int err, i;
1031         char *xspath;
1032         size_t xspathsize;
1033         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1034         unsigned int requested_num_queues = 0;
1035         unsigned int ring_page_order;
1036 
1037         pr_debug("%s %s\n", __func__, dev->otherend);
1038 
1039         blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
1040         err = xenbus_scanf(XBT_NIL, dev->otherend, "protocol",
1041                            "%63s", protocol);
1042         if (err <= 0)
1043                 strcpy(protocol, "unspecified, assuming default");
1044         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
1045                 blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
1046         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
1047                 blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
1048         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
1049                 blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
1050         else {
1051                 xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
1052                 return -ENOSYS;
1053         }
1054         pers_grants = xenbus_read_unsigned(dev->otherend, "feature-persistent",
1055                                            0);
1056         blkif->vbd.feature_gnt_persistent = pers_grants;
1057         blkif->vbd.overflow_max_grants = 0;
1058 
1059         /*
1060          * Read the number of hardware queues from frontend.
1061          */
1062         requested_num_queues = xenbus_read_unsigned(dev->otherend,
1063                                                     "multi-queue-num-queues",
1064                                                     1);
1065         if (requested_num_queues > xenblk_max_queues
1066             || requested_num_queues == 0) {
1067                 /* Buggy or malicious guest. */
1068                 xenbus_dev_fatal(dev, err,
1069                                 "guest requested %u queues, exceeding the maximum of %u.",
1070                                 requested_num_queues, xenblk_max_queues);
1071                 return -ENOSYS;
1072         }
1073         blkif->nr_rings = requested_num_queues;
1074         if (xen_blkif_alloc_rings(blkif))
1075                 return -ENOMEM;
1076 
1077         pr_info("%s: using %d queues, protocol %d (%s) %s\n", dev->nodename,
1078                  blkif->nr_rings, blkif->blk_protocol, protocol,
1079                  pers_grants ? "persistent grants" : "");
1080 
1081         ring_page_order = xenbus_read_unsigned(dev->otherend,
1082                                                "ring-page-order", 0);
1083 
1084         if (ring_page_order > xen_blkif_max_ring_order) {
1085                 err = -EINVAL;
1086                 xenbus_dev_fatal(dev, err,
1087                                  "requested ring page order %d exceed max:%d",
1088                                  ring_page_order,
1089                                  xen_blkif_max_ring_order);
1090                 return err;
1091         }
1092 
1093         blkif->nr_ring_pages = 1 << ring_page_order;
1094 
1095         if (blkif->nr_rings == 1)
1096                 return read_per_ring_refs(&blkif->rings[0], dev->otherend);
1097         else {
1098                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1099                 xspath = kmalloc(xspathsize, GFP_KERNEL);
1100                 if (!xspath) {
1101                         xenbus_dev_fatal(dev, -ENOMEM, "reading ring references");
1102                         return -ENOMEM;
1103                 }
1104 
1105                 for (i = 0; i < blkif->nr_rings; i++) {
1106                         memset(xspath, 0, xspathsize);
1107                         snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, i);
1108                         err = read_per_ring_refs(&blkif->rings[i], xspath);
1109                         if (err) {
1110                                 kfree(xspath);
1111                                 return err;
1112                         }
1113                 }
1114                 kfree(xspath);
1115         }
1116         return 0;
1117 }
1118 
1119 static const struct xenbus_device_id xen_blkbk_ids[] = {
1120         { "vbd" },
1121         { "" }
1122 };
1123 
1124 static struct xenbus_driver xen_blkbk_driver = {
1125         .ids  = xen_blkbk_ids,
1126         .probe = xen_blkbk_probe,
1127         .remove = xen_blkbk_remove,
1128         .otherend_changed = frontend_changed
1129 };
1130 
1131 int xen_blkif_xenbus_init(void)
1132 {
1133         return xenbus_register_backend(&xen_blkbk_driver);
1134 }
/* [<][>][^][v][top][bottom][index][help] */
root/drivers/block/xen-blkback/xenbus.c

DEFINITIONS
