drivers/block/xen-blkback/xenbus.c

   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 <linux/module.h>
  12 #include <linux/kthread.h>
  13 #include <xen/events.h>
  14 #include <xen/grant_table.h>
  15 #include "common.h"
  16
  17 /* On the XenBus the max length of 'ring-ref%u'. */
  18 #define RINGREF_NAME_LEN (20)
  19
  20 struct backend_info {
  21         struct xenbus_device    *dev;
  22         struct xen_blkif        *blkif;
  23         struct xenbus_watch     backend_watch;
  24         unsigned                major;
  25         unsigned                minor;
  26         char                    *mode;
  27 };
  28
  29 static struct kmem_cache *xen_blkif_cachep;
  30 static void connect(struct backend_info *);
  31 static int connect_ring(struct backend_info *);
  32 static void backend_changed(struct xenbus_watch *, const char *,
  33                             const char *);
  34 static void xen_blkif_free(struct xen_blkif *blkif);
  35 static void xen_vbd_free(struct xen_vbd *vbd);
  36
  37 struct xenbus_device *xen_blkbk_xenbus(struct backend_info *be)
  38 {
  39         return be->dev;
  40 }
  41
  42 /*
  43  * The last request could free the device from softirq context and
  44  * xen_blkif_free() can sleep.
  45  */
  46 static void xen_blkif_deferred_free(struct work_struct *work)
  47 {
  48         struct xen_blkif *blkif;
  49
  50         blkif = container_of(work, struct xen_blkif, free_work);
  51         xen_blkif_free(blkif);
  52 }
  53
  54 static int blkback_name(struct xen_blkif *blkif, char *buf)
  55 {
  56         char *devpath, *devname;
  57         struct xenbus_device *dev = blkif->be->dev;
  58
  59         devpath = xenbus_read(XBT_NIL, dev->nodename, "dev", NULL);
  60         if (IS_ERR(devpath))
  61                 return PTR_ERR(devpath);
  62
  63         devname = strstr(devpath, "/dev/");
  64         if (devname != NULL)
  65                 devname += strlen("/dev/");
  66         else
  67                 devname  = devpath;
  68
  69         snprintf(buf, TASK_COMM_LEN, "%d.%s", blkif->domid, devname);
  70         kfree(devpath);
  71
  72         return 0;
  73 }
  74
  75 static void xen_update_blkif_status(struct xen_blkif *blkif)
  76 {
  77         int err;
  78         char name[TASK_COMM_LEN];
  79         struct xen_blkif_ring *ring;
  80         int i;
  81
  82         /* Not ready to connect? */
  83         if (!blkif->rings || !blkif->rings[0].irq || !blkif->vbd.bdev)
  84                 return;
  85
  86         /* Already connected? */
  87         if (blkif->be->dev->state == XenbusStateConnected)
  88                 return;
  89
  90         /* Attempt to connect: exit if we fail to. */
  91         connect(blkif->be);
  92         if (blkif->be->dev->state != XenbusStateConnected)
  93                 return;
  94
  95         err = blkback_name(blkif, name);
  96         if (err) {
  97                 xenbus_dev_error(blkif->be->dev, err, "get blkback dev name");
  98                 return;
  99         }
 100
 101         err = filemap_write_and_wait(blkif->vbd.bdev->bd_inode->i_mapping);
 102         if (err) {
 103                 xenbus_dev_error(blkif->be->dev, err, "block flush");
 104                 return;
 105         }
 106         invalidate_inode_pages2(blkif->vbd.bdev->bd_inode->i_mapping);
 107
 108         for (i = 0; i < blkif->nr_rings; i++) {
 109                 ring = &blkif->rings[i];
 110                 ring->xenblkd = kthread_run(xen_blkif_schedule, ring, "%s-%d", name, i);
 111                 if (IS_ERR(ring->xenblkd)) {
 112                         err = PTR_ERR(ring->xenblkd);
 113                         ring->xenblkd = NULL;
 114                         xenbus_dev_fatal(blkif->be->dev, err,
 115                                         "start %s-%d xenblkd", name, i);
 116                         goto out;
 117                 }
 118         }
 119         return;
 120
 121 out:
 122         while (--i >= 0) {
 123                 ring = &blkif->rings[i];
 124                 kthread_stop(ring->xenblkd);
 125         }
 126         return;
 127 }
 128
 129 static int xen_blkif_alloc_rings(struct xen_blkif *blkif)
 130 {
 131         unsigned int r;
 132
 133         blkif->rings = kcalloc(blkif->nr_rings, sizeof(struct xen_blkif_ring),
 134                                GFP_KERNEL);
 135         if (!blkif->rings)
 136                 return -ENOMEM;
 137
 138         for (r = 0; r < blkif->nr_rings; r++) {
 139                 struct xen_blkif_ring *ring = &blkif->rings[r];
 140
 141                 spin_lock_init(&ring->blk_ring_lock);
 142                 init_waitqueue_head(&ring->wq);
 143                 INIT_LIST_HEAD(&ring->pending_free);
 144                 INIT_LIST_HEAD(&ring->persistent_purge_list);
 145                 INIT_WORK(&ring->persistent_purge_work, xen_blkbk_unmap_purged_grants);
 146                 gnttab_page_cache_init(&ring->free_pages);
 147
 148                 spin_lock_init(&ring->pending_free_lock);
 149                 init_waitqueue_head(&ring->pending_free_wq);
 150                 init_waitqueue_head(&ring->shutdown_wq);
 151                 ring->blkif = blkif;
 152                 ring->st_print = jiffies;
 153                 ring->active = true;
 154         }
 155
 156         return 0;
 157 }
 158
 159 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
 160 {
 161         struct xen_blkif *blkif;
 162
 163         BUILD_BUG_ON(MAX_INDIRECT_PAGES > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
 164
 165         blkif = kmem_cache_zalloc(xen_blkif_cachep, GFP_KERNEL);
 166         if (!blkif)
 167                 return ERR_PTR(-ENOMEM);
 168
 169         blkif->domid = domid;
 170         atomic_set(&blkif->refcnt, 1);
 171         init_completion(&blkif->drain_complete);
 172
 173         /*
 174          * Because freeing back to the cache may be deferred, it is not
 175          * safe to unload the module (and hence destroy the cache) until
 176          * this has completed. To prevent premature unloading, take an
 177          * extra module reference here and release only when the object
 178          * has been freed back to the cache.
 179          */
 180         __module_get(THIS_MODULE);
 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         const struct blkif_common_sring *sring_common;
 192         RING_IDX rsp_prod, req_prod;
 193         unsigned int size;
 194
 195         /* Already connected through? */
 196         if (ring->irq)
 197                 return 0;
 198
 199         err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
 200                                      &ring->blk_ring);
 201         if (err < 0)
 202                 return err;
 203
 204         sring_common = (struct blkif_common_sring *)ring->blk_ring;
 205         rsp_prod = READ_ONCE(sring_common->rsp_prod);
 206         req_prod = READ_ONCE(sring_common->req_prod);
 207
 208         switch (blkif->blk_protocol) {
 209         case BLKIF_PROTOCOL_NATIVE:
 210         {
 211                 struct blkif_sring *sring_native =
 212                         (struct blkif_sring *)ring->blk_ring;
 213
 214                 BACK_RING_ATTACH(&ring->blk_rings.native, sring_native,
 215                                  rsp_prod, XEN_PAGE_SIZE * nr_grefs);
 216                 size = __RING_SIZE(sring_native, 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                         (struct blkif_x86_32_sring *)ring->blk_ring;
 223
 224                 BACK_RING_ATTACH(&ring->blk_rings.x86_32, sring_x86_32,
 225                                  rsp_prod, XEN_PAGE_SIZE * nr_grefs);
 226                 size = __RING_SIZE(sring_x86_32, XEN_PAGE_SIZE * nr_grefs);
 227                 break;
 228         }
 229         case BLKIF_PROTOCOL_X86_64:
 230         {
 231                 struct blkif_x86_64_sring *sring_x86_64 =
 232                         (struct blkif_x86_64_sring *)ring->blk_ring;
 233
 234                 BACK_RING_ATTACH(&ring->blk_rings.x86_64, sring_x86_64,
 235                                  rsp_prod, XEN_PAGE_SIZE * nr_grefs);
 236                 size = __RING_SIZE(sring_x86_64, XEN_PAGE_SIZE * nr_grefs);
 237                 break;
 238         }
 239         default:
 240                 BUG();
 241         }
 242
 243         err = -EIO;
 244         if (req_prod - rsp_prod > size)
 245                 goto fail;
 246
 247         err = bind_interdomain_evtchn_to_irqhandler_lateeoi(blkif->be->dev,
 248                         evtchn, xen_blkif_be_int, 0, "blkif-backend", ring);
 249         if (err < 0)
 250                 goto fail;
 251         ring->irq = err;
 252
 253         return 0;
 254
 255 fail:
 256         xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
 257         ring->blk_rings.common.sring = NULL;
 258         return err;
 259 }
 260
 261 static int xen_blkif_disconnect(struct xen_blkif *blkif)
 262 {
 263         struct pending_req *req, *n;
 264         unsigned int j, r;
 265         bool busy = false;
 266
 267         for (r = 0; r < blkif->nr_rings; r++) {
 268                 struct xen_blkif_ring *ring = &blkif->rings[r];
 269                 unsigned int i = 0;
 270
 271                 if (!ring->active)
 272                         continue;
 273
 274                 if (ring->xenblkd) {
 275                         kthread_stop(ring->xenblkd);
 276                         ring->xenblkd = NULL;
 277                         wake_up(&ring->shutdown_wq);
 278                 }
 279
 280                 /* The above kthread_stop() guarantees that at this point we
 281                  * don't have any discard_io or other_io requests. So, checking
 282                  * for inflight IO is enough.
 283                  */
 284                 if (atomic_read(&ring->inflight) > 0) {
 285                         busy = true;
 286                         continue;
 287                 }
 288
 289                 if (ring->irq) {
 290                         unbind_from_irqhandler(ring->irq, ring);
 291                         ring->irq = 0;
 292                 }
 293
 294                 if (ring->blk_rings.common.sring) {
 295                         xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
 296                         ring->blk_rings.common.sring = NULL;
 297                 }
 298
 299                 /* Remove all persistent grants and the cache of ballooned pages. */
 300                 xen_blkbk_free_caches(ring);
 301
 302                 /* Check that there is no request in use */
 303                 list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
 304                         list_del(&req->free_list);
 305
 306                         for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
 307                                 kfree(req->segments[j]);
 308
 309                         for (j = 0; j < MAX_INDIRECT_PAGES; j++)
 310                                 kfree(req->indirect_pages[j]);
 311
 312                         kfree(req);
 313                         i++;
 314                 }
 315
 316                 BUG_ON(atomic_read(&ring->persistent_gnt_in_use) != 0);
 317                 BUG_ON(!list_empty(&ring->persistent_purge_list));
 318                 BUG_ON(!RB_EMPTY_ROOT(&ring->persistent_gnts));
 319                 BUG_ON(ring->free_pages.num_pages != 0);
 320                 BUG_ON(ring->persistent_gnt_c != 0);
 321                 WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
 322                 ring->active = false;
 323         }
 324         if (busy)
 325                 return -EBUSY;
 326
 327         blkif->nr_ring_pages = 0;
 328         /*
 329          * blkif->rings was allocated in connect_ring, so we should free it in
 330          * here.
 331          */
 332         kfree(blkif->rings);
 333         blkif->rings = NULL;
 334         blkif->nr_rings = 0;
 335
 336         return 0;
 337 }
 338
 339 static void xen_blkif_free(struct xen_blkif *blkif)
 340 {
 341         WARN_ON(xen_blkif_disconnect(blkif));
 342         xen_vbd_free(&blkif->vbd);
 343         kfree(blkif->be->mode);
 344         kfree(blkif->be);
 345
 346         /* Make sure everything is drained before shutting down */
 347         kmem_cache_free(xen_blkif_cachep, blkif);
 348         module_put(THIS_MODULE);
 349 }
 350
 351 int __init xen_blkif_interface_init(void)
 352 {
 353         xen_blkif_cachep = kmem_cache_create("blkif_cache",
 354                                              sizeof(struct xen_blkif),
 355                                              0, 0, NULL);
 356         if (!xen_blkif_cachep)
 357                 return -ENOMEM;
 358
 359         return 0;
 360 }
 361
 362 void xen_blkif_interface_fini(void)
 363 {
 364         kmem_cache_destroy(xen_blkif_cachep);
 365         xen_blkif_cachep = NULL;
 366 }
 367
 368 /*
 369  *  sysfs interface for VBD I/O requests
 370  */
 371
 372 #define VBD_SHOW_ALLRING(name, format)                                  \
 373         static ssize_t show_##name(struct device *_dev,                 \
 374                                    struct device_attribute *attr,       \
 375                                    char *buf)                           \
 376         {                                                               \
 377                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 378                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 379                 struct xen_blkif *blkif = be->blkif;                    \
 380                 unsigned int i;                                         \
 381                 unsigned long long result = 0;                          \
 382                                                                         \
 383                 if (!blkif->rings)                              \
 384                         goto out;                                       \
 385                                                                         \
 386                 for (i = 0; i < blkif->nr_rings; i++) {         \
 387                         struct xen_blkif_ring *ring = &blkif->rings[i]; \
 388                                                                         \
 389                         result += ring->st_##name;                      \
 390                 }                                                       \
 391                                                                         \
 392 out:                                                                    \
 393                 return sprintf(buf, format, result);                    \
 394         }                                                               \
 395         static DEVICE_ATTR(name, 0444, show_##name, NULL)
 396
 397 VBD_SHOW_ALLRING(oo_req,  "%llu\n");
 398 VBD_SHOW_ALLRING(rd_req,  "%llu\n");
 399 VBD_SHOW_ALLRING(wr_req,  "%llu\n");
 400 VBD_SHOW_ALLRING(f_req,  "%llu\n");
 401 VBD_SHOW_ALLRING(ds_req,  "%llu\n");
 402 VBD_SHOW_ALLRING(rd_sect, "%llu\n");
 403 VBD_SHOW_ALLRING(wr_sect, "%llu\n");
 404
 405 static struct attribute *xen_vbdstat_attrs[] = {
 406         &dev_attr_oo_req.attr,
 407         &dev_attr_rd_req.attr,
 408         &dev_attr_wr_req.attr,
 409         &dev_attr_f_req.attr,
 410         &dev_attr_ds_req.attr,
 411         &dev_attr_rd_sect.attr,
 412         &dev_attr_wr_sect.attr,
 413         NULL
 414 };
 415
 416 static const struct attribute_group xen_vbdstat_group = {
 417         .name = "statistics",
 418         .attrs = xen_vbdstat_attrs,
 419 };
 420
 421 #define VBD_SHOW(name, format, args...)                                 \
 422         static ssize_t show_##name(struct device *_dev,                 \
 423                                    struct device_attribute *attr,       \
 424                                    char *buf)                           \
 425         {                                                               \
 426                 struct xenbus_device *dev = to_xenbus_device(_dev);     \
 427                 struct backend_info *be = dev_get_drvdata(&dev->dev);   \
 428                                                                         \
 429                 return sprintf(buf, format, ##args);                    \
 430         }                                                               \
 431         static DEVICE_ATTR(name, 0444, show_##name, NULL)
 432
 433 VBD_SHOW(physical_device, "%x:%x\n", be->major, be->minor);
 434 VBD_SHOW(mode, "%s\n", be->mode);
 435
 436 static int xenvbd_sysfs_addif(struct xenbus_device *dev)
 437 {
 438         int error;
 439
 440         error = device_create_file(&dev->dev, &dev_attr_physical_device);
 441         if (error)
 442                 goto fail1;
 443
 444         error = device_create_file(&dev->dev, &dev_attr_mode);
 445         if (error)
 446                 goto fail2;
 447
 448         error = sysfs_create_group(&dev->dev.kobj, &xen_vbdstat_group);
 449         if (error)
 450                 goto fail3;
 451
 452         return 0;
 453
 454 fail3:  sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 455 fail2:  device_remove_file(&dev->dev, &dev_attr_mode);
 456 fail1:  device_remove_file(&dev->dev, &dev_attr_physical_device);
 457         return error;
 458 }
 459
 460 static void xenvbd_sysfs_delif(struct xenbus_device *dev)
 461 {
 462         sysfs_remove_group(&dev->dev.kobj, &xen_vbdstat_group);
 463         device_remove_file(&dev->dev, &dev_attr_mode);
 464         device_remove_file(&dev->dev, &dev_attr_physical_device);
 465 }
 466
 467 static void xen_vbd_free(struct xen_vbd *vbd)
 468 {
 469         if (vbd->bdev)
 470                 blkdev_put(vbd->bdev, vbd->readonly ? FMODE_READ : FMODE_WRITE);
 471         vbd->bdev = NULL;
 472 }
 473
 474 /* Enable the persistent grants feature. */
 475 static bool feature_persistent = true;
 476 module_param(feature_persistent, bool, 0644);
 477 MODULE_PARM_DESC(feature_persistent,
 478                 "Enables the persistent grants feature");
 479
 480 static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
 481                           unsigned major, unsigned minor, int readonly,
 482                           int cdrom)
 483 {
 484         struct xen_vbd *vbd;
 485         struct block_device *bdev;
 486         struct request_queue *q;
 487
 488         vbd = &blkif->vbd;
 489         vbd->handle   = handle;
 490         vbd->readonly = readonly;
 491         vbd->type     = 0;
 492
 493         vbd->pdevice  = MKDEV(major, minor);
 494
 495         bdev = blkdev_get_by_dev(vbd->pdevice, vbd->readonly ?
 496                                  FMODE_READ : FMODE_WRITE, NULL);
 497
 498         if (IS_ERR(bdev)) {
 499                 pr_warn("xen_vbd_create: device %08x could not be opened\n",
 500                         vbd->pdevice);
 501                 return -ENOENT;
 502         }
 503
 504         vbd->bdev = bdev;
 505         if (vbd->bdev->bd_disk == NULL) {
 506                 pr_warn("xen_vbd_create: device %08x doesn't exist\n",
 507                         vbd->pdevice);
 508                 xen_vbd_free(vbd);
 509                 return -ENOENT;
 510         }
 511         vbd->size = vbd_sz(vbd);
 512
 513         if (vbd->bdev->bd_disk->flags & GENHD_FL_CD || cdrom)
 514                 vbd->type |= VDISK_CDROM;
 515         if (vbd->bdev->bd_disk->flags & GENHD_FL_REMOVABLE)
 516                 vbd->type |= VDISK_REMOVABLE;
 517
 518         q = bdev_get_queue(bdev);
 519         if (q && test_bit(QUEUE_FLAG_WC, &q->queue_flags))
 520                 vbd->flush_support = true;
 521
 522         if (q && blk_queue_secure_erase(q))
 523                 vbd->discard_secure = true;
 524
 525         vbd->feature_gnt_persistent = feature_persistent;
 526
 527         pr_debug("Successful creation of handle=%04x (dom=%u)\n",
 528                 handle, blkif->domid);
 529         return 0;
 530 }
 531
 532 static int xen_blkbk_remove(struct xenbus_device *dev)
 533 {
 534         struct backend_info *be = dev_get_drvdata(&dev->dev);
 535
 536         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 537
 538         if (be->major || be->minor)
 539                 xenvbd_sysfs_delif(dev);
 540
 541         if (be->backend_watch.node) {
 542                 unregister_xenbus_watch(&be->backend_watch);
 543                 kfree(be->backend_watch.node);
 544                 be->backend_watch.node = NULL;
 545         }
 546
 547         dev_set_drvdata(&dev->dev, NULL);
 548
 549         if (be->blkif) {
 550                 xen_blkif_disconnect(be->blkif);
 551
 552                 /* Put the reference we set in xen_blkif_alloc(). */
 553                 xen_blkif_put(be->blkif);
 554         }
 555
 556         return 0;
 557 }
 558
 559 int xen_blkbk_flush_diskcache(struct xenbus_transaction xbt,
 560                               struct backend_info *be, int state)
 561 {
 562         struct xenbus_device *dev = be->dev;
 563         int err;
 564
 565         err = xenbus_printf(xbt, dev->nodename, "feature-flush-cache",
 566                             "%d", state);
 567         if (err)
 568                 dev_warn(&dev->dev, "writing feature-flush-cache (%d)", err);
 569
 570         return err;
 571 }
 572
 573 static void xen_blkbk_discard(struct xenbus_transaction xbt, struct backend_info *be)
 574 {
 575         struct xenbus_device *dev = be->dev;
 576         struct xen_blkif *blkif = be->blkif;
 577         int err;
 578         int state = 0;
 579         struct block_device *bdev = be->blkif->vbd.bdev;
 580         struct request_queue *q = bdev_get_queue(bdev);
 581
 582         if (!xenbus_read_unsigned(dev->nodename, "discard-enable", 1))
 583                 return;
 584
 585         if (blk_queue_discard(q)) {
 586                 err = xenbus_printf(xbt, dev->nodename,
 587                         "discard-granularity", "%u",
 588                         q->limits.discard_granularity);
 589                 if (err) {
 590                         dev_warn(&dev->dev, "writing discard-granularity (%d)", err);
 591                         return;
 592                 }
 593                 err = xenbus_printf(xbt, dev->nodename,
 594                         "discard-alignment", "%u",
 595                         q->limits.discard_alignment);
 596                 if (err) {
 597                         dev_warn(&dev->dev, "writing discard-alignment (%d)", err);
 598                         return;
 599                 }
 600                 state = 1;
 601                 /* Optional. */
 602                 err = xenbus_printf(xbt, dev->nodename,
 603                                     "discard-secure", "%d",
 604                                     blkif->vbd.discard_secure);
 605                 if (err) {
 606                         dev_warn(&dev->dev, "writing discard-secure (%d)", err);
 607                         return;
 608                 }
 609         }
 610         err = xenbus_printf(xbt, dev->nodename, "feature-discard",
 611                             "%d", state);
 612         if (err)
 613                 dev_warn(&dev->dev, "writing feature-discard (%d)", err);
 614 }
 615
 616 int xen_blkbk_barrier(struct xenbus_transaction xbt,
 617                       struct backend_info *be, int state)
 618 {
 619         struct xenbus_device *dev = be->dev;
 620         int err;
 621
 622         err = xenbus_printf(xbt, dev->nodename, "feature-barrier",
 623                             "%d", state);
 624         if (err)
 625                 dev_warn(&dev->dev, "writing feature-barrier (%d)", err);
 626
 627         return err;
 628 }
 629
 630 /*
 631  * Entry point to this code when a new device is created.  Allocate the basic
 632  * structures, and watch the store waiting for the hotplug scripts to tell us
 633  * the device's physical major and minor numbers.  Switch to InitWait.
 634  */
 635 static int xen_blkbk_probe(struct xenbus_device *dev,
 636                            const struct xenbus_device_id *id)
 637 {
 638         int err;
 639         struct backend_info *be = kzalloc(sizeof(struct backend_info),
 640                                           GFP_KERNEL);
 641
 642         /* match the pr_debug in xen_blkbk_remove */
 643         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 644
 645         if (!be) {
 646                 xenbus_dev_fatal(dev, -ENOMEM,
 647                                  "allocating backend structure");
 648                 return -ENOMEM;
 649         }
 650         be->dev = dev;
 651         dev_set_drvdata(&dev->dev, be);
 652
 653         be->blkif = xen_blkif_alloc(dev->otherend_id);
 654         if (IS_ERR(be->blkif)) {
 655                 err = PTR_ERR(be->blkif);
 656                 be->blkif = NULL;
 657                 xenbus_dev_fatal(dev, err, "creating block interface");
 658                 goto fail;
 659         }
 660
 661         err = xenbus_printf(XBT_NIL, dev->nodename,
 662                             "feature-max-indirect-segments", "%u",
 663                             MAX_INDIRECT_SEGMENTS);
 664         if (err)
 665                 dev_warn(&dev->dev,
 666                          "writing %s/feature-max-indirect-segments (%d)",
 667                          dev->nodename, err);
 668
 669         /* Multi-queue: advertise how many queues are supported by us.*/
 670         err = xenbus_printf(XBT_NIL, dev->nodename,
 671                             "multi-queue-max-queues", "%u", xenblk_max_queues);
 672         if (err)
 673                 pr_warn("Error writing multi-queue-max-queues\n");
 674
 675         /* setup back pointer */
 676         be->blkif->be = be;
 677
 678         err = xenbus_watch_pathfmt(dev, &be->backend_watch, NULL,
 679                                    backend_changed,
 680                                    "%s/%s", dev->nodename, "physical-device");
 681         if (err)
 682                 goto fail;
 683
 684         err = xenbus_printf(XBT_NIL, dev->nodename, "max-ring-page-order", "%u",
 685                             xen_blkif_max_ring_order);
 686         if (err)
 687                 pr_warn("%s write out 'max-ring-page-order' failed\n", __func__);
 688
 689         err = xenbus_switch_state(dev, XenbusStateInitWait);
 690         if (err)
 691                 goto fail;
 692
 693         return 0;
 694
 695 fail:
 696         pr_warn("%s failed\n", __func__);
 697         xen_blkbk_remove(dev);
 698         return err;
 699 }
 700
 701 /*
 702  * Callback received when the hotplug scripts have placed the physical-device
 703  * node.  Read it and the mode node, and create a vbd.  If the frontend is
 704  * ready, connect.
 705  */
 706 static void backend_changed(struct xenbus_watch *watch,
 707                             const char *path, const char *token)
 708 {
 709         int err;
 710         unsigned major;
 711         unsigned minor;
 712         struct backend_info *be
 713                 = container_of(watch, struct backend_info, backend_watch);
 714         struct xenbus_device *dev = be->dev;
 715         int cdrom = 0;
 716         unsigned long handle;
 717         char *device_type;
 718
 719         pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 720
 721         err = xenbus_scanf(XBT_NIL, dev->nodename, "physical-device", "%x:%x",
 722                            &major, &minor);
 723         if (XENBUS_EXIST_ERR(err)) {
 724                 /*
 725                  * Since this watch will fire once immediately after it is
 726                  * registered, we expect this.  Ignore it, and wait for the
 727                  * hotplug scripts.
 728                  */
 729                 return;
 730         }
 731         if (err != 2) {
 732                 xenbus_dev_fatal(dev, err, "reading physical-device");
 733                 return;
 734         }
 735
 736         if (be->major | be->minor) {
 737                 if (be->major != major || be->minor != minor)
 738                         pr_warn("changing physical device (from %x:%x to %x:%x) not supported.\n",
 739                                 be->major, be->minor, major, minor);
 740                 return;
 741         }
 742
 743         be->mode = xenbus_read(XBT_NIL, dev->nodename, "mode", NULL);
 744         if (IS_ERR(be->mode)) {
 745                 err = PTR_ERR(be->mode);
 746                 be->mode = NULL;
 747                 xenbus_dev_fatal(dev, err, "reading mode");
 748                 return;
 749         }
 750
 751         device_type = xenbus_read(XBT_NIL, dev->otherend, "device-type", NULL);
 752         if (!IS_ERR(device_type)) {
 753                 cdrom = strcmp(device_type, "cdrom") == 0;
 754                 kfree(device_type);
 755         }
 756
 757         /* Front end dir is a number, which is used as the handle. */
 758         err = kstrtoul(strrchr(dev->otherend, '/') + 1, 0, &handle);
 759         if (err) {
 760                 kfree(be->mode);
 761                 be->mode = NULL;
 762                 return;
 763         }
 764
 765         be->major = major;
 766         be->minor = minor;
 767
 768         err = xen_vbd_create(be->blkif, handle, major, minor,
 769                              !strchr(be->mode, 'w'), cdrom);
 770
 771         if (err)
 772                 xenbus_dev_fatal(dev, err, "creating vbd structure");
 773         else {
 774                 err = xenvbd_sysfs_addif(dev);
 775                 if (err) {
 776                         xen_vbd_free(&be->blkif->vbd);
 777                         xenbus_dev_fatal(dev, err, "creating sysfs entries");
 778                 }
 779         }
 780
 781         if (err) {
 782                 kfree(be->mode);
 783                 be->mode = NULL;
 784                 be->major = 0;
 785                 be->minor = 0;
 786         } else {
 787                 /* We're potentially connected now */
 788                 xen_update_blkif_status(be->blkif);
 789         }
 790 }
 791
 792 /*
 793  * Callback received when the frontend's state changes.
 794  */
 795 static void frontend_changed(struct xenbus_device *dev,
 796                              enum xenbus_state frontend_state)
 797 {
 798         struct backend_info *be = dev_get_drvdata(&dev->dev);
 799         int err;
 800
 801         pr_debug("%s %p %s\n", __func__, dev, xenbus_strstate(frontend_state));
 802
 803         switch (frontend_state) {
 804         case XenbusStateInitialising:
 805                 if (dev->state == XenbusStateClosed) {
 806                         pr_info("%s: prepare for reconnect\n", dev->nodename);
 807                         xenbus_switch_state(dev, XenbusStateInitWait);
 808                 }
 809                 break;
 810
 811         case XenbusStateInitialised:
 812         case XenbusStateConnected:
 813                 /*
 814                  * Ensure we connect even when two watches fire in
 815                  * close succession and we miss the intermediate value
 816                  * of frontend_state.
 817                  */
 818                 if (dev->state == XenbusStateConnected)
 819                         break;
 820
 821                 /*
 822                  * Enforce precondition before potential leak point.
 823                  * xen_blkif_disconnect() is idempotent.
 824                  */
 825                 err = xen_blkif_disconnect(be->blkif);
 826                 if (err) {
 827                         xenbus_dev_fatal(dev, err, "pending I/O");
 828                         break;
 829                 }
 830
 831                 err = connect_ring(be);
 832                 if (err) {
 833                         /*
 834                          * Clean up so that memory resources can be used by
 835                          * other devices. connect_ring reported already error.
 836                          */
 837                         xen_blkif_disconnect(be->blkif);
 838                         break;
 839                 }
 840                 xen_update_blkif_status(be->blkif);
 841                 break;
 842
 843         case XenbusStateClosing:
 844                 xenbus_switch_state(dev, XenbusStateClosing);
 845                 break;
 846
 847         case XenbusStateClosed:
 848                 xen_blkif_disconnect(be->blkif);
 849                 xenbus_switch_state(dev, XenbusStateClosed);
 850                 if (xenbus_dev_is_online(dev))
 851                         break;
 852                 fallthrough;
 853                 /* if not online */
 854         case XenbusStateUnknown:
 855                 /* implies xen_blkif_disconnect() via xen_blkbk_remove() */
 856                 device_unregister(&dev->dev);
 857                 break;
 858
 859         default:
 860                 xenbus_dev_fatal(dev, -EINVAL, "saw state %d at frontend",
 861                                  frontend_state);
 862                 break;
 863         }
 864 }
 865
 866 /* Once a memory pressure is detected, squeeze free page pools for a while. */
 867 static unsigned int buffer_squeeze_duration_ms = 10;
 868 module_param_named(buffer_squeeze_duration_ms,
 869                 buffer_squeeze_duration_ms, int, 0644);
 870 MODULE_PARM_DESC(buffer_squeeze_duration_ms,
 871 "Duration in ms to squeeze pages buffer when a memory pressure is detected");
 872
 873 /*
 874  * Callback received when the memory pressure is detected.
 875  */
 876 static void reclaim_memory(struct xenbus_device *dev)
 877 {
 878         struct backend_info *be = dev_get_drvdata(&dev->dev);
 879
 880         if (!be)
 881                 return;
 882         be->blkif->buffer_squeeze_end = jiffies +
 883                 msecs_to_jiffies(buffer_squeeze_duration_ms);
 884 }
 885
 886 /* ** Connection ** */
 887
 888 /*
 889  * Write the physical details regarding the block device to the store, and
 890  * switch to Connected state.
 891  */
 892 static void connect(struct backend_info *be)
 893 {
 894         struct xenbus_transaction xbt;
 895         int err;
 896         struct xenbus_device *dev = be->dev;
 897
 898         pr_debug("%s %s\n", __func__, dev->otherend);
 899
 900         /* Supply the information about the device the frontend needs */
 901 again:
 902         err = xenbus_transaction_start(&xbt);
 903         if (err) {
 904                 xenbus_dev_fatal(dev, err, "starting transaction");
 905                 return;
 906         }
 907
 908         /* If we can't advertise it is OK. */
 909         xen_blkbk_flush_diskcache(xbt, be, be->blkif->vbd.flush_support);
 910
 911         xen_blkbk_discard(xbt, be);
 912
 913         xen_blkbk_barrier(xbt, be, be->blkif->vbd.flush_support);
 914
 915         err = xenbus_printf(xbt, dev->nodename, "feature-persistent", "%u",
 916                         be->blkif->vbd.feature_gnt_persistent);
 917         if (err) {
 918                 xenbus_dev_fatal(dev, err, "writing %s/feature-persistent",
 919                                  dev->nodename);
 920                 goto abort;
 921         }
 922
 923         err = xenbus_printf(xbt, dev->nodename, "sectors", "%llu",
 924                             (unsigned long long)vbd_sz(&be->blkif->vbd));
 925         if (err) {
 926                 xenbus_dev_fatal(dev, err, "writing %s/sectors",
 927                                  dev->nodename);
 928                 goto abort;
 929         }
 930
 931         /* FIXME: use a typename instead */
 932         err = xenbus_printf(xbt, dev->nodename, "info", "%u",
 933                             be->blkif->vbd.type |
 934                             (be->blkif->vbd.readonly ? VDISK_READONLY : 0));
 935         if (err) {
 936                 xenbus_dev_fatal(dev, err, "writing %s/info",
 937                                  dev->nodename);
 938                 goto abort;
 939         }
 940         err = xenbus_printf(xbt, dev->nodename, "sector-size", "%lu",
 941                             (unsigned long)
 942                             bdev_logical_block_size(be->blkif->vbd.bdev));
 943         if (err) {
 944                 xenbus_dev_fatal(dev, err, "writing %s/sector-size",
 945                                  dev->nodename);
 946                 goto abort;
 947         }
 948         err = xenbus_printf(xbt, dev->nodename, "physical-sector-size", "%u",
 949                             bdev_physical_block_size(be->blkif->vbd.bdev));
 950         if (err)
 951                 xenbus_dev_error(dev, err, "writing %s/physical-sector-size",
 952                                  dev->nodename);
 953
 954         err = xenbus_transaction_end(xbt, 0);
 955         if (err == -EAGAIN)
 956                 goto again;
 957         if (err)
 958                 xenbus_dev_fatal(dev, err, "ending transaction");
 959
 960         err = xenbus_switch_state(dev, XenbusStateConnected);
 961         if (err)
 962                 xenbus_dev_fatal(dev, err, "%s: switching to Connected state",
 963                                  dev->nodename);
 964
 965         return;
 966  abort:
 967         xenbus_transaction_end(xbt, 1);
 968 }
 969
 970 /*
 971  * Each ring may have multi pages, depends on "ring-page-order".
 972  */
 973 static int read_per_ring_refs(struct xen_blkif_ring *ring, const char *dir)
 974 {
 975         unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
 976         struct pending_req *req, *n;
 977         int err, i, j;
 978         struct xen_blkif *blkif = ring->blkif;
 979         struct xenbus_device *dev = blkif->be->dev;
 980         unsigned int nr_grefs, evtchn;
 981
 982         err = xenbus_scanf(XBT_NIL, dir, "event-channel", "%u",
 983                           &evtchn);
 984         if (err != 1) {
 985                 err = -EINVAL;
 986                 xenbus_dev_fatal(dev, err, "reading %s/event-channel", dir);
 987                 return err;
 988         }
 989
 990         nr_grefs = blkif->nr_ring_pages;
 991
 992         if (unlikely(!nr_grefs)) {
 993                 WARN_ON(true);
 994                 return -EINVAL;
 995         }
 996
 997         for (i = 0; i < nr_grefs; i++) {
 998                 char ring_ref_name[RINGREF_NAME_LEN];
 999
1000                 if (blkif->multi_ref)
1001                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
1002                 else {
1003                         WARN_ON(i != 0);
1004                         snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref");
1005                 }
1006
1007                 err = xenbus_scanf(XBT_NIL, dir, ring_ref_name,
1008                                    "%u", &ring_ref[i]);
1009
1010                 if (err != 1) {
1011                         err = -EINVAL;
1012                         xenbus_dev_fatal(dev, err, "reading %s/%s",
1013                                          dir, ring_ref_name);
1014                         return err;
1015                 }
1016         }
1017
1018         err = -ENOMEM;
1019         for (i = 0; i < nr_grefs * XEN_BLKIF_REQS_PER_PAGE; i++) {
1020                 req = kzalloc(sizeof(*req), GFP_KERNEL);
1021                 if (!req)
1022                         goto fail;
1023                 list_add_tail(&req->free_list, &ring->pending_free);
1024                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
1025                         req->segments[j] = kzalloc(sizeof(*req->segments[0]), GFP_KERNEL);
1026                         if (!req->segments[j])
1027                                 goto fail;
1028                 }
1029                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1030                         req->indirect_pages[j] = kzalloc(sizeof(*req->indirect_pages[0]),
1031                                                          GFP_KERNEL);
1032                         if (!req->indirect_pages[j])
1033                                 goto fail;
1034                 }
1035         }
1036
1037         /* Map the shared frame, irq etc. */
1038         err = xen_blkif_map(ring, ring_ref, nr_grefs, evtchn);
1039         if (err) {
1040                 xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
1041                 goto fail;
1042         }
1043
1044         return 0;
1045
1046 fail:
1047         list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
1048                 list_del(&req->free_list);
1049                 for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++) {
1050                         if (!req->segments[j])
1051                                 break;
1052                         kfree(req->segments[j]);
1053                 }
1054                 for (j = 0; j < MAX_INDIRECT_PAGES; j++) {
1055                         if (!req->indirect_pages[j])
1056                                 break;
1057                         kfree(req->indirect_pages[j]);
1058                 }
1059                 kfree(req);
1060         }
1061         return err;
1062 }
1063
1064 static int connect_ring(struct backend_info *be)
1065 {
1066         struct xenbus_device *dev = be->dev;
1067         struct xen_blkif *blkif = be->blkif;
1068         char protocol[64] = "";
1069         int err, i;
1070         char *xspath;
1071         size_t xspathsize;
1072         const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
1073         unsigned int requested_num_queues = 0;
1074         unsigned int ring_page_order;
1075
1076         pr_debug("%s %s\n", __func__, dev->otherend);
1077
1078         blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
1079         err = xenbus_scanf(XBT_NIL, dev->otherend, "protocol",
1080                            "%63s", protocol);
1081         if (err <= 0)
1082                 strcpy(protocol, "unspecified, assuming default");
1083         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
1084                 blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
1085         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
1086                 blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
1087         else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
1088                 blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
1089         else {
1090                 xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
1091                 return -ENOSYS;
1092         }
1093         if (blkif->vbd.feature_gnt_persistent)
1094                 blkif->vbd.feature_gnt_persistent =
1095                         xenbus_read_unsigned(dev->otherend,
1096                                         "feature-persistent", 0);
1097
1098         blkif->vbd.overflow_max_grants = 0;
1099
1100         /*
1101          * Read the number of hardware queues from frontend.
1102          */
1103         requested_num_queues = xenbus_read_unsigned(dev->otherend,
1104                                                     "multi-queue-num-queues",
1105                                                     1);
1106         if (requested_num_queues > xenblk_max_queues
1107             || requested_num_queues == 0) {
1108                 /* Buggy or malicious guest. */
1109                 xenbus_dev_fatal(dev, err,
1110                                 "guest requested %u queues, exceeding the maximum of %u.",
1111                                 requested_num_queues, xenblk_max_queues);
1112                 return -ENOSYS;
1113         }
1114         blkif->nr_rings = requested_num_queues;
1115         if (xen_blkif_alloc_rings(blkif))
1116                 return -ENOMEM;
1117
1118         pr_info("%s: using %d queues, protocol %d (%s) %s\n", dev->nodename,
1119                  blkif->nr_rings, blkif->blk_protocol, protocol,
1120                  blkif->vbd.feature_gnt_persistent ? "persistent grants" : "");
1121
1122         err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-page-order", "%u",
1123                            &ring_page_order);
1124         if (err != 1) {
1125                 blkif->nr_ring_pages = 1;
1126                 blkif->multi_ref = false;
1127         } else if (ring_page_order <= xen_blkif_max_ring_order) {
1128                 blkif->nr_ring_pages = 1 << ring_page_order;
1129                 blkif->multi_ref = true;
1130         } else {
1131                 err = -EINVAL;
1132                 xenbus_dev_fatal(dev, err,
1133                                  "requested ring page order %d exceed max:%d",
1134                                  ring_page_order,
1135                                  xen_blkif_max_ring_order);
1136                 return err;
1137         }
1138
1139         if (blkif->nr_rings == 1)
1140                 return read_per_ring_refs(&blkif->rings[0], dev->otherend);
1141         else {
1142                 xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
1143                 xspath = kmalloc(xspathsize, GFP_KERNEL);
1144                 if (!xspath) {
1145                         xenbus_dev_fatal(dev, -ENOMEM, "reading ring references");
1146                         return -ENOMEM;
1147                 }
1148
1149                 for (i = 0; i < blkif->nr_rings; i++) {
1150                         memset(xspath, 0, xspathsize);
1151                         snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, i);
1152                         err = read_per_ring_refs(&blkif->rings[i], xspath);
1153                         if (err) {
1154                                 kfree(xspath);
1155                                 return err;
1156                         }
1157                 }
1158                 kfree(xspath);
1159         }
1160         return 0;
1161 }
1162
1163 static const struct xenbus_device_id xen_blkbk_ids[] = {
1164         { "vbd" },
1165         { "" }
1166 };
1167
1168 static struct xenbus_driver xen_blkbk_driver = {
1169         .ids  = xen_blkbk_ids,
1170         .probe = xen_blkbk_probe,
1171         .remove = xen_blkbk_remove,
1172         .otherend_changed = frontend_changed,
1173         .allow_rebind = true,
1174         .reclaim_memory = reclaim_memory,
1175 };
1176
1177 int xen_blkif_xenbus_init(void)
1178 {
1179         return xenbus_register_backend(&xen_blkbk_driver);
1180 }
1181
1182 void xen_blkif_xenbus_fini(void)
1183 {
1184         xenbus_unregister_driver(&xen_blkbk_driver);
1185 }