drivers/nvdimm/namespace_devs.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
   4  */
   5 #include <linux/module.h>
   6 #include <linux/device.h>
   7 #include <linux/sort.h>
   8 #include <linux/slab.h>
   9 #include <linux/list.h>
  10 #include <linux/nd.h>
  11 #include "nd-core.h"
  12 #include "pmem.h"
  13 #include "pfn.h"
  14 #include "nd.h"
  15
  16 static void namespace_io_release(struct device *dev)
  17 {
  18         struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
  19
  20         kfree(nsio);
  21 }
  22
  23 static void namespace_pmem_release(struct device *dev)
  24 {
  25         struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
  26         struct nd_region *nd_region = to_nd_region(dev->parent);
  27
  28         if (nspm->id >= 0)
  29                 ida_simple_remove(&nd_region->ns_ida, nspm->id);
  30         kfree(nspm->alt_name);
  31         kfree(nspm->uuid);
  32         kfree(nspm);
  33 }
  34
  35 static bool is_namespace_pmem(const struct device *dev);
  36 static bool is_namespace_io(const struct device *dev);
  37
  38 static int is_uuid_busy(struct device *dev, void *data)
  39 {
  40         uuid_t *uuid1 = data, *uuid2 = NULL;
  41
  42         if (is_namespace_pmem(dev)) {
  43                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
  44
  45                 uuid2 = nspm->uuid;
  46         } else if (is_nd_btt(dev)) {
  47                 struct nd_btt *nd_btt = to_nd_btt(dev);
  48
  49                 uuid2 = nd_btt->uuid;
  50         } else if (is_nd_pfn(dev)) {
  51                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
  52
  53                 uuid2 = nd_pfn->uuid;
  54         }
  55
  56         if (uuid2 && uuid_equal(uuid1, uuid2))
  57                 return -EBUSY;
  58
  59         return 0;
  60 }
  61
  62 static int is_namespace_uuid_busy(struct device *dev, void *data)
  63 {
  64         if (is_nd_region(dev))
  65                 return device_for_each_child(dev, data, is_uuid_busy);
  66         return 0;
  67 }
  68
  69 /**
  70  * nd_is_uuid_unique - verify that no other namespace has @uuid
  71  * @dev: any device on a nvdimm_bus
  72  * @uuid: uuid to check
  73  */
  74 bool nd_is_uuid_unique(struct device *dev, uuid_t *uuid)
  75 {
  76         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
  77
  78         if (!nvdimm_bus)
  79                 return false;
  80         WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm_bus->dev));
  81         if (device_for_each_child(&nvdimm_bus->dev, uuid,
  82                                 is_namespace_uuid_busy) != 0)
  83                 return false;
  84         return true;
  85 }
  86
  87 bool pmem_should_map_pages(struct device *dev)
  88 {
  89         struct nd_region *nd_region = to_nd_region(dev->parent);
  90         struct nd_namespace_common *ndns = to_ndns(dev);
  91         struct nd_namespace_io *nsio;
  92
  93         if (!IS_ENABLED(CONFIG_ZONE_DEVICE))
  94                 return false;
  95
  96         if (!test_bit(ND_REGION_PAGEMAP, &nd_region->flags))
  97                 return false;
  98
  99         if (is_nd_pfn(dev) || is_nd_btt(dev))
 100                 return false;
 101
 102         if (ndns->force_raw)
 103                 return false;
 104
 105         nsio = to_nd_namespace_io(dev);
 106         if (region_intersects(nsio->res.start, resource_size(&nsio->res),
 107                                 IORESOURCE_SYSTEM_RAM,
 108                                 IORES_DESC_NONE) == REGION_MIXED)
 109                 return false;
 110
 111         return ARCH_MEMREMAP_PMEM == MEMREMAP_WB;
 112 }
 113 EXPORT_SYMBOL(pmem_should_map_pages);
 114
 115 unsigned int pmem_sector_size(struct nd_namespace_common *ndns)
 116 {
 117         if (is_namespace_pmem(&ndns->dev)) {
 118                 struct nd_namespace_pmem *nspm;
 119
 120                 nspm = to_nd_namespace_pmem(&ndns->dev);
 121                 if (nspm->lbasize == 0 || nspm->lbasize == 512)
 122                         /* default */;
 123                 else if (nspm->lbasize == 4096)
 124                         return 4096;
 125                 else
 126                         dev_WARN(&ndns->dev, "unsupported sector size: %ld\n",
 127                                         nspm->lbasize);
 128         }
 129
 130         /*
 131          * There is no namespace label (is_namespace_io()), or the label
 132          * indicates the default sector size.
 133          */
 134         return 512;
 135 }
 136 EXPORT_SYMBOL(pmem_sector_size);
 137
 138 const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,
 139                 char *name)
 140 {
 141         struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
 142         const char *suffix = NULL;
 143
 144         if (ndns->claim && is_nd_btt(ndns->claim))
 145                 suffix = "s";
 146
 147         if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev)) {
 148                 int nsidx = 0;
 149
 150                 if (is_namespace_pmem(&ndns->dev)) {
 151                         struct nd_namespace_pmem *nspm;
 152
 153                         nspm = to_nd_namespace_pmem(&ndns->dev);
 154                         nsidx = nspm->id;
 155                 }
 156
 157                 if (nsidx)
 158                         sprintf(name, "pmem%d.%d%s", nd_region->id, nsidx,
 159                                         suffix ? suffix : "");
 160                 else
 161                         sprintf(name, "pmem%d%s", nd_region->id,
 162                                         suffix ? suffix : "");
 163         } else {
 164                 return NULL;
 165         }
 166
 167         return name;
 168 }
 169 EXPORT_SYMBOL(nvdimm_namespace_disk_name);
 170
 171 const uuid_t *nd_dev_to_uuid(struct device *dev)
 172 {
 173         if (dev && is_namespace_pmem(dev)) {
 174                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 175
 176                 return nspm->uuid;
 177         }
 178         return &uuid_null;
 179 }
 180 EXPORT_SYMBOL(nd_dev_to_uuid);
 181
 182 static ssize_t nstype_show(struct device *dev,
 183                 struct device_attribute *attr, char *buf)
 184 {
 185         struct nd_region *nd_region = to_nd_region(dev->parent);
 186
 187         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
 188 }
 189 static DEVICE_ATTR_RO(nstype);
 190
 191 static ssize_t __alt_name_store(struct device *dev, const char *buf,
 192                 const size_t len)
 193 {
 194         char *input, *pos, *alt_name, **ns_altname;
 195         ssize_t rc;
 196
 197         if (is_namespace_pmem(dev)) {
 198                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 199
 200                 ns_altname = &nspm->alt_name;
 201         } else
 202                 return -ENXIO;
 203
 204         if (dev->driver || to_ndns(dev)->claim)
 205                 return -EBUSY;
 206
 207         input = kstrndup(buf, len, GFP_KERNEL);
 208         if (!input)
 209                 return -ENOMEM;
 210
 211         pos = strim(input);
 212         if (strlen(pos) + 1 > NSLABEL_NAME_LEN) {
 213                 rc = -EINVAL;
 214                 goto out;
 215         }
 216
 217         alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL);
 218         if (!alt_name) {
 219                 rc = -ENOMEM;
 220                 goto out;
 221         }
 222         kfree(*ns_altname);
 223         *ns_altname = alt_name;
 224         sprintf(*ns_altname, "%s", pos);
 225         rc = len;
 226
 227 out:
 228         kfree(input);
 229         return rc;
 230 }
 231
 232 static int nd_namespace_label_update(struct nd_region *nd_region,
 233                 struct device *dev)
 234 {
 235         dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim,
 236                         "namespace must be idle during label update\n");
 237         if (dev->driver || to_ndns(dev)->claim)
 238                 return 0;
 239
 240         /*
 241          * Only allow label writes that will result in a valid namespace
 242          * or deletion of an existing namespace.
 243          */
 244         if (is_namespace_pmem(dev)) {
 245                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 246                 resource_size_t size = resource_size(&nspm->nsio.res);
 247
 248                 if (size == 0 && nspm->uuid)
 249                         /* delete allocation */;
 250                 else if (!nspm->uuid)
 251                         return 0;
 252
 253                 return nd_pmem_namespace_label_update(nd_region, nspm, size);
 254         } else
 255                 return -ENXIO;
 256 }
 257
 258 static ssize_t alt_name_store(struct device *dev,
 259                 struct device_attribute *attr, const char *buf, size_t len)
 260 {
 261         struct nd_region *nd_region = to_nd_region(dev->parent);
 262         ssize_t rc;
 263
 264         device_lock(dev);
 265         nvdimm_bus_lock(dev);
 266         wait_nvdimm_bus_probe_idle(dev);
 267         rc = __alt_name_store(dev, buf, len);
 268         if (rc >= 0)
 269                 rc = nd_namespace_label_update(nd_region, dev);
 270         dev_dbg(dev, "%s(%zd)\n", rc < 0 ? "fail " : "", rc);
 271         nvdimm_bus_unlock(dev);
 272         device_unlock(dev);
 273
 274         return rc < 0 ? rc : len;
 275 }
 276
 277 static ssize_t alt_name_show(struct device *dev,
 278                 struct device_attribute *attr, char *buf)
 279 {
 280         char *ns_altname;
 281
 282         if (is_namespace_pmem(dev)) {
 283                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 284
 285                 ns_altname = nspm->alt_name;
 286         } else
 287                 return -ENXIO;
 288
 289         return sprintf(buf, "%s\n", ns_altname ? ns_altname : "");
 290 }
 291 static DEVICE_ATTR_RW(alt_name);
 292
 293 static int scan_free(struct nd_region *nd_region,
 294                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
 295                 resource_size_t n)
 296 {
 297         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 298         int rc = 0;
 299
 300         while (n) {
 301                 struct resource *res, *last;
 302
 303                 last = NULL;
 304                 for_each_dpa_resource(ndd, res)
 305                         if (strcmp(res->name, label_id->id) == 0)
 306                                 last = res;
 307                 res = last;
 308                 if (!res)
 309                         return 0;
 310
 311                 if (n >= resource_size(res)) {
 312                         n -= resource_size(res);
 313                         nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc);
 314                         nvdimm_free_dpa(ndd, res);
 315                         /* retry with last resource deleted */
 316                         continue;
 317                 }
 318
 319                 rc = adjust_resource(res, res->start, resource_size(res) - n);
 320                 if (rc == 0)
 321                         res->flags |= DPA_RESOURCE_ADJUSTED;
 322                 nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc);
 323                 break;
 324         }
 325
 326         return rc;
 327 }
 328
 329 /**
 330  * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
 331  * @nd_region: the set of dimms to reclaim @n bytes from
 332  * @label_id: unique identifier for the namespace consuming this dpa range
 333  * @n: number of bytes per-dimm to release
 334  *
 335  * Assumes resources are ordered.  Starting from the end try to
 336  * adjust_resource() the allocation to @n, but if @n is larger than the
 337  * allocation delete it and find the 'new' last allocation in the label
 338  * set.
 339  */
 340 static int shrink_dpa_allocation(struct nd_region *nd_region,
 341                 struct nd_label_id *label_id, resource_size_t n)
 342 {
 343         int i;
 344
 345         for (i = 0; i < nd_region->ndr_mappings; i++) {
 346                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 347                 int rc;
 348
 349                 rc = scan_free(nd_region, nd_mapping, label_id, n);
 350                 if (rc)
 351                         return rc;
 352         }
 353
 354         return 0;
 355 }
 356
 357 static resource_size_t init_dpa_allocation(struct nd_label_id *label_id,
 358                 struct nd_region *nd_region, struct nd_mapping *nd_mapping,
 359                 resource_size_t n)
 360 {
 361         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 362         struct resource *res;
 363         int rc = 0;
 364
 365         /* first resource allocation for this label-id or dimm */
 366         res = nvdimm_allocate_dpa(ndd, label_id, nd_mapping->start, n);
 367         if (!res)
 368                 rc = -EBUSY;
 369
 370         nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc);
 371         return rc ? n : 0;
 372 }
 373
 374
 375 /**
 376  * space_valid() - validate free dpa space against constraints
 377  * @nd_region: hosting region of the free space
 378  * @ndd: dimm device data for debug
 379  * @label_id: namespace id to allocate space
 380  * @prev: potential allocation that precedes free space
 381  * @next: allocation that follows the given free space range
 382  * @exist: first allocation with same id in the mapping
 383  * @n: range that must satisfied for pmem allocations
 384  * @valid: free space range to validate
 385  *
 386  * BLK-space is valid as long as it does not precede a PMEM
 387  * allocation in a given region. PMEM-space must be contiguous
 388  * and adjacent to an existing allocation (if one
 389  * exists).  If reserving PMEM any space is valid.
 390  */
 391 static void space_valid(struct nd_region *nd_region, struct nvdimm_drvdata *ndd,
 392                 struct nd_label_id *label_id, struct resource *prev,
 393                 struct resource *next, struct resource *exist,
 394                 resource_size_t n, struct resource *valid)
 395 {
 396         bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0;
 397         unsigned long align;
 398
 399         align = nd_region->align / nd_region->ndr_mappings;
 400         valid->start = ALIGN(valid->start, align);
 401         valid->end = ALIGN_DOWN(valid->end + 1, align) - 1;
 402
 403         if (valid->start >= valid->end)
 404                 goto invalid;
 405
 406         if (is_reserve)
 407                 return;
 408
 409         /* allocation needs to be contiguous, so this is all or nothing */
 410         if (resource_size(valid) < n)
 411                 goto invalid;
 412
 413         /* we've got all the space we need and no existing allocation */
 414         if (!exist)
 415                 return;
 416
 417         /* allocation needs to be contiguous with the existing namespace */
 418         if (valid->start == exist->end + 1
 419                         || valid->end == exist->start - 1)
 420                 return;
 421
 422  invalid:
 423         /* truncate @valid size to 0 */
 424         valid->end = valid->start - 1;
 425 }
 426
 427 enum alloc_loc {
 428         ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
 429 };
 430
 431 static resource_size_t scan_allocate(struct nd_region *nd_region,
 432                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
 433                 resource_size_t n)
 434 {
 435         resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1;
 436         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 437         struct resource *res, *exist = NULL, valid;
 438         const resource_size_t to_allocate = n;
 439         int first;
 440
 441         for_each_dpa_resource(ndd, res)
 442                 if (strcmp(label_id->id, res->name) == 0)
 443                         exist = res;
 444
 445         valid.start = nd_mapping->start;
 446         valid.end = mapping_end;
 447         valid.name = "free space";
 448  retry:
 449         first = 0;
 450         for_each_dpa_resource(ndd, res) {
 451                 struct resource *next = res->sibling, *new_res = NULL;
 452                 resource_size_t allocate, available = 0;
 453                 enum alloc_loc loc = ALLOC_ERR;
 454                 const char *action;
 455                 int rc = 0;
 456
 457                 /* ignore resources outside this nd_mapping */
 458                 if (res->start > mapping_end)
 459                         continue;
 460                 if (res->end < nd_mapping->start)
 461                         continue;
 462
 463                 /* space at the beginning of the mapping */
 464                 if (!first++ && res->start > nd_mapping->start) {
 465                         valid.start = nd_mapping->start;
 466                         valid.end = res->start - 1;
 467                         space_valid(nd_region, ndd, label_id, NULL, next, exist,
 468                                         to_allocate, &valid);
 469                         available = resource_size(&valid);
 470                         if (available)
 471                                 loc = ALLOC_BEFORE;
 472                 }
 473
 474                 /* space between allocations */
 475                 if (!loc && next) {
 476                         valid.start = res->start + resource_size(res);
 477                         valid.end = min(mapping_end, next->start - 1);
 478                         space_valid(nd_region, ndd, label_id, res, next, exist,
 479                                         to_allocate, &valid);
 480                         available = resource_size(&valid);
 481                         if (available)
 482                                 loc = ALLOC_MID;
 483                 }
 484
 485                 /* space at the end of the mapping */
 486                 if (!loc && !next) {
 487                         valid.start = res->start + resource_size(res);
 488                         valid.end = mapping_end;
 489                         space_valid(nd_region, ndd, label_id, res, next, exist,
 490                                         to_allocate, &valid);
 491                         available = resource_size(&valid);
 492                         if (available)
 493                                 loc = ALLOC_AFTER;
 494                 }
 495
 496                 if (!loc || !available)
 497                         continue;
 498                 allocate = min(available, n);
 499                 switch (loc) {
 500                 case ALLOC_BEFORE:
 501                         if (strcmp(res->name, label_id->id) == 0) {
 502                                 /* adjust current resource up */
 503                                 rc = adjust_resource(res, res->start - allocate,
 504                                                 resource_size(res) + allocate);
 505                                 action = "cur grow up";
 506                         } else
 507                                 action = "allocate";
 508                         break;
 509                 case ALLOC_MID:
 510                         if (strcmp(next->name, label_id->id) == 0) {
 511                                 /* adjust next resource up */
 512                                 rc = adjust_resource(next, next->start
 513                                                 - allocate, resource_size(next)
 514                                                 + allocate);
 515                                 new_res = next;
 516                                 action = "next grow up";
 517                         } else if (strcmp(res->name, label_id->id) == 0) {
 518                                 action = "grow down";
 519                         } else
 520                                 action = "allocate";
 521                         break;
 522                 case ALLOC_AFTER:
 523                         if (strcmp(res->name, label_id->id) == 0)
 524                                 action = "grow down";
 525                         else
 526                                 action = "allocate";
 527                         break;
 528                 default:
 529                         return n;
 530                 }
 531
 532                 if (strcmp(action, "allocate") == 0) {
 533                         new_res = nvdimm_allocate_dpa(ndd, label_id,
 534                                         valid.start, allocate);
 535                         if (!new_res)
 536                                 rc = -EBUSY;
 537                 } else if (strcmp(action, "grow down") == 0) {
 538                         /* adjust current resource down */
 539                         rc = adjust_resource(res, res->start, resource_size(res)
 540                                         + allocate);
 541                         if (rc == 0)
 542                                 res->flags |= DPA_RESOURCE_ADJUSTED;
 543                 }
 544
 545                 if (!new_res)
 546                         new_res = res;
 547
 548                 nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n",
 549                                 action, loc, rc);
 550
 551                 if (rc)
 552                         return n;
 553
 554                 n -= allocate;
 555                 if (n) {
 556                         /*
 557                          * Retry scan with newly inserted resources.
 558                          * For example, if we did an ALLOC_BEFORE
 559                          * insertion there may also have been space
 560                          * available for an ALLOC_AFTER insertion, so we
 561                          * need to check this same resource again
 562                          */
 563                         goto retry;
 564                 } else
 565                         return 0;
 566         }
 567
 568         if (n == to_allocate)
 569                 return init_dpa_allocation(label_id, nd_region, nd_mapping, n);
 570         return n;
 571 }
 572
 573 static int merge_dpa(struct nd_region *nd_region,
 574                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id)
 575 {
 576         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 577         struct resource *res;
 578
 579         if (strncmp("pmem", label_id->id, 4) == 0)
 580                 return 0;
 581  retry:
 582         for_each_dpa_resource(ndd, res) {
 583                 int rc;
 584                 struct resource *next = res->sibling;
 585                 resource_size_t end = res->start + resource_size(res);
 586
 587                 if (!next || strcmp(res->name, label_id->id) != 0
 588                                 || strcmp(next->name, label_id->id) != 0
 589                                 || end != next->start)
 590                         continue;
 591                 end += resource_size(next);
 592                 nvdimm_free_dpa(ndd, next);
 593                 rc = adjust_resource(res, res->start, end - res->start);
 594                 nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc);
 595                 if (rc)
 596                         return rc;
 597                 res->flags |= DPA_RESOURCE_ADJUSTED;
 598                 goto retry;
 599         }
 600
 601         return 0;
 602 }
 603
 604 int __reserve_free_pmem(struct device *dev, void *data)
 605 {
 606         struct nvdimm *nvdimm = data;
 607         struct nd_region *nd_region;
 608         struct nd_label_id label_id;
 609         int i;
 610
 611         if (!is_memory(dev))
 612                 return 0;
 613
 614         nd_region = to_nd_region(dev);
 615         if (nd_region->ndr_mappings == 0)
 616                 return 0;
 617
 618         memset(&label_id, 0, sizeof(label_id));
 619         strcat(label_id.id, "pmem-reserve");
 620         for (i = 0; i < nd_region->ndr_mappings; i++) {
 621                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 622                 resource_size_t n, rem = 0;
 623
 624                 if (nd_mapping->nvdimm != nvdimm)
 625                         continue;
 626
 627                 n = nd_pmem_available_dpa(nd_region, nd_mapping);
 628                 if (n == 0)
 629                         return 0;
 630                 rem = scan_allocate(nd_region, nd_mapping, &label_id, n);
 631                 dev_WARN_ONCE(&nd_region->dev, rem,
 632                                 "pmem reserve underrun: %#llx of %#llx bytes\n",
 633                                 (unsigned long long) n - rem,
 634                                 (unsigned long long) n);
 635                 return rem ? -ENXIO : 0;
 636         }
 637
 638         return 0;
 639 }
 640
 641 void release_free_pmem(struct nvdimm_bus *nvdimm_bus,
 642                 struct nd_mapping *nd_mapping)
 643 {
 644         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 645         struct resource *res, *_res;
 646
 647         for_each_dpa_resource_safe(ndd, res, _res)
 648                 if (strcmp(res->name, "pmem-reserve") == 0)
 649                         nvdimm_free_dpa(ndd, res);
 650 }
 651
 652 /**
 653  * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
 654  * @nd_region: the set of dimms to allocate @n more bytes from
 655  * @label_id: unique identifier for the namespace consuming this dpa range
 656  * @n: number of bytes per-dimm to add to the existing allocation
 657  *
 658  * Assumes resources are ordered.  For BLK regions, first consume
 659  * BLK-only available DPA free space, then consume PMEM-aliased DPA
 660  * space starting at the highest DPA.  For PMEM regions start
 661  * allocations from the start of an interleave set and end at the first
 662  * BLK allocation or the end of the interleave set, whichever comes
 663  * first.
 664  */
 665 static int grow_dpa_allocation(struct nd_region *nd_region,
 666                 struct nd_label_id *label_id, resource_size_t n)
 667 {
 668         int i;
 669
 670         for (i = 0; i < nd_region->ndr_mappings; i++) {
 671                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 672                 resource_size_t rem = n;
 673                 int rc;
 674
 675                 rem = scan_allocate(nd_region, nd_mapping, label_id, rem);
 676                 dev_WARN_ONCE(&nd_region->dev, rem,
 677                                 "allocation underrun: %#llx of %#llx bytes\n",
 678                                 (unsigned long long) n - rem,
 679                                 (unsigned long long) n);
 680                 if (rem)
 681                         return -ENXIO;
 682
 683                 rc = merge_dpa(nd_region, nd_mapping, label_id);
 684                 if (rc)
 685                         return rc;
 686         }
 687
 688         return 0;
 689 }
 690
 691 static void nd_namespace_pmem_set_resource(struct nd_region *nd_region,
 692                 struct nd_namespace_pmem *nspm, resource_size_t size)
 693 {
 694         struct resource *res = &nspm->nsio.res;
 695         resource_size_t offset = 0;
 696
 697         if (size && !nspm->uuid) {
 698                 WARN_ON_ONCE(1);
 699                 size = 0;
 700         }
 701
 702         if (size && nspm->uuid) {
 703                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 704                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 705                 struct nd_label_id label_id;
 706                 struct resource *res;
 707
 708                 if (!ndd) {
 709                         size = 0;
 710                         goto out;
 711                 }
 712
 713                 nd_label_gen_id(&label_id, nspm->uuid, 0);
 714
 715                 /* calculate a spa offset from the dpa allocation offset */
 716                 for_each_dpa_resource(ndd, res)
 717                         if (strcmp(res->name, label_id.id) == 0) {
 718                                 offset = (res->start - nd_mapping->start)
 719                                         * nd_region->ndr_mappings;
 720                                 goto out;
 721                         }
 722
 723                 WARN_ON_ONCE(1);
 724                 size = 0;
 725         }
 726
 727  out:
 728         res->start = nd_region->ndr_start + offset;
 729         res->end = res->start + size - 1;
 730 }
 731
 732 static bool uuid_not_set(const uuid_t *uuid, struct device *dev,
 733                          const char *where)
 734 {
 735         if (!uuid) {
 736                 dev_dbg(dev, "%s: uuid not set\n", where);
 737                 return true;
 738         }
 739         return false;
 740 }
 741
 742 static ssize_t __size_store(struct device *dev, unsigned long long val)
 743 {
 744         resource_size_t allocated = 0, available = 0;
 745         struct nd_region *nd_region = to_nd_region(dev->parent);
 746         struct nd_namespace_common *ndns = to_ndns(dev);
 747         struct nd_mapping *nd_mapping;
 748         struct nvdimm_drvdata *ndd;
 749         struct nd_label_id label_id;
 750         u32 flags = 0, remainder;
 751         int rc, i, id = -1;
 752         uuid_t *uuid = NULL;
 753
 754         if (dev->driver || ndns->claim)
 755                 return -EBUSY;
 756
 757         if (is_namespace_pmem(dev)) {
 758                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 759
 760                 uuid = nspm->uuid;
 761                 id = nspm->id;
 762         }
 763
 764         /*
 765          * We need a uuid for the allocation-label and dimm(s) on which
 766          * to store the label.
 767          */
 768         if (uuid_not_set(uuid, dev, __func__))
 769                 return -ENXIO;
 770         if (nd_region->ndr_mappings == 0) {
 771                 dev_dbg(dev, "not associated with dimm(s)\n");
 772                 return -ENXIO;
 773         }
 774
 775         div_u64_rem(val, nd_region->align, &remainder);
 776         if (remainder) {
 777                 dev_dbg(dev, "%llu is not %ldK aligned\n", val,
 778                                 nd_region->align / SZ_1K);
 779                 return -EINVAL;
 780         }
 781
 782         nd_label_gen_id(&label_id, uuid, flags);
 783         for (i = 0; i < nd_region->ndr_mappings; i++) {
 784                 nd_mapping = &nd_region->mapping[i];
 785                 ndd = to_ndd(nd_mapping);
 786
 787                 /*
 788                  * All dimms in an interleave set, need to be enabled
 789                  * for the size to be changed.
 790                  */
 791                 if (!ndd)
 792                         return -ENXIO;
 793
 794                 allocated += nvdimm_allocated_dpa(ndd, &label_id);
 795         }
 796         available = nd_region_allocatable_dpa(nd_region);
 797
 798         if (val > available + allocated)
 799                 return -ENOSPC;
 800
 801         if (val == allocated)
 802                 return 0;
 803
 804         val = div_u64(val, nd_region->ndr_mappings);
 805         allocated = div_u64(allocated, nd_region->ndr_mappings);
 806         if (val < allocated)
 807                 rc = shrink_dpa_allocation(nd_region, &label_id,
 808                                 allocated - val);
 809         else
 810                 rc = grow_dpa_allocation(nd_region, &label_id, val - allocated);
 811
 812         if (rc)
 813                 return rc;
 814
 815         if (is_namespace_pmem(dev)) {
 816                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 817
 818                 nd_namespace_pmem_set_resource(nd_region, nspm,
 819                                 val * nd_region->ndr_mappings);
 820         }
 821
 822         /*
 823          * Try to delete the namespace if we deleted all of its
 824          * allocation, this is not the seed or 0th device for the
 825          * region, and it is not actively claimed by a btt, pfn, or dax
 826          * instance.
 827          */
 828         if (val == 0 && id != 0 && nd_region->ns_seed != dev && !ndns->claim)
 829                 nd_device_unregister(dev, ND_ASYNC);
 830
 831         return rc;
 832 }
 833
 834 static ssize_t size_store(struct device *dev,
 835                 struct device_attribute *attr, const char *buf, size_t len)
 836 {
 837         struct nd_region *nd_region = to_nd_region(dev->parent);
 838         unsigned long long val;
 839         int rc;
 840
 841         rc = kstrtoull(buf, 0, &val);
 842         if (rc)
 843                 return rc;
 844
 845         device_lock(dev);
 846         nvdimm_bus_lock(dev);
 847         wait_nvdimm_bus_probe_idle(dev);
 848         rc = __size_store(dev, val);
 849         if (rc >= 0)
 850                 rc = nd_namespace_label_update(nd_region, dev);
 851
 852         /* setting size zero == 'delete namespace' */
 853         if (rc == 0 && val == 0 && is_namespace_pmem(dev)) {
 854                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 855
 856                 kfree(nspm->uuid);
 857                 nspm->uuid = NULL;
 858         }
 859
 860         dev_dbg(dev, "%llx %s (%d)\n", val, rc < 0 ? "fail" : "success", rc);
 861
 862         nvdimm_bus_unlock(dev);
 863         device_unlock(dev);
 864
 865         return rc < 0 ? rc : len;
 866 }
 867
 868 resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
 869 {
 870         struct device *dev = &ndns->dev;
 871
 872         if (is_namespace_pmem(dev)) {
 873                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 874
 875                 return resource_size(&nspm->nsio.res);
 876         } else if (is_namespace_io(dev)) {
 877                 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
 878
 879                 return resource_size(&nsio->res);
 880         } else
 881                 WARN_ONCE(1, "unknown namespace type\n");
 882         return 0;
 883 }
 884
 885 resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
 886 {
 887         resource_size_t size;
 888
 889         nvdimm_bus_lock(&ndns->dev);
 890         size = __nvdimm_namespace_capacity(ndns);
 891         nvdimm_bus_unlock(&ndns->dev);
 892
 893         return size;
 894 }
 895 EXPORT_SYMBOL(nvdimm_namespace_capacity);
 896
 897 bool nvdimm_namespace_locked(struct nd_namespace_common *ndns)
 898 {
 899         int i;
 900         bool locked = false;
 901         struct device *dev = &ndns->dev;
 902         struct nd_region *nd_region = to_nd_region(dev->parent);
 903
 904         for (i = 0; i < nd_region->ndr_mappings; i++) {
 905                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 906                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
 907
 908                 if (test_bit(NDD_LOCKED, &nvdimm->flags)) {
 909                         dev_dbg(dev, "%s locked\n", nvdimm_name(nvdimm));
 910                         locked = true;
 911                 }
 912         }
 913         return locked;
 914 }
 915 EXPORT_SYMBOL(nvdimm_namespace_locked);
 916
 917 static ssize_t size_show(struct device *dev,
 918                 struct device_attribute *attr, char *buf)
 919 {
 920         return sprintf(buf, "%llu\n", (unsigned long long)
 921                         nvdimm_namespace_capacity(to_ndns(dev)));
 922 }
 923 static DEVICE_ATTR(size, 0444, size_show, size_store);
 924
 925 static uuid_t *namespace_to_uuid(struct device *dev)
 926 {
 927         if (is_namespace_pmem(dev)) {
 928                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 929
 930                 return nspm->uuid;
 931         }
 932         return ERR_PTR(-ENXIO);
 933 }
 934
 935 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
 936                          char *buf)
 937 {
 938         uuid_t *uuid = namespace_to_uuid(dev);
 939
 940         if (IS_ERR(uuid))
 941                 return PTR_ERR(uuid);
 942         if (uuid)
 943                 return sprintf(buf, "%pUb\n", uuid);
 944         return sprintf(buf, "\n");
 945 }
 946
 947 /**
 948  * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
 949  * @nd_region: parent region so we can updates all dimms in the set
 950  * @dev: namespace type for generating label_id
 951  * @new_uuid: incoming uuid
 952  * @old_uuid: reference to the uuid storage location in the namespace object
 953  */
 954 static int namespace_update_uuid(struct nd_region *nd_region,
 955                                  struct device *dev, uuid_t *new_uuid,
 956                                  uuid_t **old_uuid)
 957 {
 958         struct nd_label_id old_label_id;
 959         struct nd_label_id new_label_id;
 960         int i;
 961
 962         if (!nd_is_uuid_unique(dev, new_uuid))
 963                 return -EINVAL;
 964
 965         if (*old_uuid == NULL)
 966                 goto out;
 967
 968         /*
 969          * If we've already written a label with this uuid, then it's
 970          * too late to rename because we can't reliably update the uuid
 971          * without losing the old namespace.  Userspace must delete this
 972          * namespace to abandon the old uuid.
 973          */
 974         for (i = 0; i < nd_region->ndr_mappings; i++) {
 975                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 976
 977                 /*
 978                  * This check by itself is sufficient because old_uuid
 979                  * would be NULL above if this uuid did not exist in the
 980                  * currently written set.
 981                  *
 982                  * FIXME: can we delete uuid with zero dpa allocated?
 983                  */
 984                 if (list_empty(&nd_mapping->labels))
 985                         return -EBUSY;
 986         }
 987
 988         nd_label_gen_id(&old_label_id, *old_uuid, 0);
 989         nd_label_gen_id(&new_label_id, new_uuid, 0);
 990         for (i = 0; i < nd_region->ndr_mappings; i++) {
 991                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 992                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 993                 struct nd_label_ent *label_ent;
 994                 struct resource *res;
 995
 996                 for_each_dpa_resource(ndd, res)
 997                         if (strcmp(res->name, old_label_id.id) == 0)
 998                                 sprintf((void *) res->name, "%s",
 999                                                 new_label_id.id);
1000
1001                 mutex_lock(&nd_mapping->lock);
1002                 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1003                         struct nd_namespace_label *nd_label = label_ent->label;
1004                         struct nd_label_id label_id;
1005                         uuid_t uuid;
1006
1007                         if (!nd_label)
1008                                 continue;
1009                         nsl_get_uuid(ndd, nd_label, &uuid);
1010                         nd_label_gen_id(&label_id, &uuid,
1011                                         nsl_get_flags(ndd, nd_label));
1012                         if (strcmp(old_label_id.id, label_id.id) == 0)
1013                                 set_bit(ND_LABEL_REAP, &label_ent->flags);
1014                 }
1015                 mutex_unlock(&nd_mapping->lock);
1016         }
1017         kfree(*old_uuid);
1018  out:
1019         *old_uuid = new_uuid;
1020         return 0;
1021 }
1022
1023 static ssize_t uuid_store(struct device *dev,
1024                 struct device_attribute *attr, const char *buf, size_t len)
1025 {
1026         struct nd_region *nd_region = to_nd_region(dev->parent);
1027         uuid_t *uuid = NULL;
1028         uuid_t **ns_uuid;
1029         ssize_t rc = 0;
1030
1031         if (is_namespace_pmem(dev)) {
1032                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1033
1034                 ns_uuid = &nspm->uuid;
1035         } else
1036                 return -ENXIO;
1037
1038         device_lock(dev);
1039         nvdimm_bus_lock(dev);
1040         wait_nvdimm_bus_probe_idle(dev);
1041         if (to_ndns(dev)->claim)
1042                 rc = -EBUSY;
1043         if (rc >= 0)
1044                 rc = nd_uuid_store(dev, &uuid, buf, len);
1045         if (rc >= 0)
1046                 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
1047         if (rc >= 0)
1048                 rc = nd_namespace_label_update(nd_region, dev);
1049         else
1050                 kfree(uuid);
1051         dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
1052                         buf[len - 1] == '\n' ? "" : "\n");
1053         nvdimm_bus_unlock(dev);
1054         device_unlock(dev);
1055
1056         return rc < 0 ? rc : len;
1057 }
1058 static DEVICE_ATTR_RW(uuid);
1059
1060 static ssize_t resource_show(struct device *dev,
1061                 struct device_attribute *attr, char *buf)
1062 {
1063         struct resource *res;
1064
1065         if (is_namespace_pmem(dev)) {
1066                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1067
1068                 res = &nspm->nsio.res;
1069         } else if (is_namespace_io(dev)) {
1070                 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1071
1072                 res = &nsio->res;
1073         } else
1074                 return -ENXIO;
1075
1076         /* no address to convey if the namespace has no allocation */
1077         if (resource_size(res) == 0)
1078                 return -ENXIO;
1079         return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
1080 }
1081 static DEVICE_ATTR_ADMIN_RO(resource);
1082
1083 static const unsigned long pmem_lbasize_supported[] = { 512, 4096, 0 };
1084
1085 static ssize_t sector_size_show(struct device *dev,
1086                 struct device_attribute *attr, char *buf)
1087 {
1088         if (is_namespace_pmem(dev)) {
1089                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1090
1091                 return nd_size_select_show(nspm->lbasize,
1092                                 pmem_lbasize_supported, buf);
1093         }
1094         return -ENXIO;
1095 }
1096
1097 static ssize_t sector_size_store(struct device *dev,
1098                 struct device_attribute *attr, const char *buf, size_t len)
1099 {
1100         struct nd_region *nd_region = to_nd_region(dev->parent);
1101         const unsigned long *supported;
1102         unsigned long *lbasize;
1103         ssize_t rc = 0;
1104
1105         if (is_namespace_pmem(dev)) {
1106                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1107
1108                 lbasize = &nspm->lbasize;
1109                 supported = pmem_lbasize_supported;
1110         } else
1111                 return -ENXIO;
1112
1113         device_lock(dev);
1114         nvdimm_bus_lock(dev);
1115         if (to_ndns(dev)->claim)
1116                 rc = -EBUSY;
1117         if (rc >= 0)
1118                 rc = nd_size_select_store(dev, buf, lbasize, supported);
1119         if (rc >= 0)
1120                 rc = nd_namespace_label_update(nd_region, dev);
1121         dev_dbg(dev, "result: %zd %s: %s%s", rc, rc < 0 ? "tried" : "wrote",
1122                         buf, buf[len - 1] == '\n' ? "" : "\n");
1123         nvdimm_bus_unlock(dev);
1124         device_unlock(dev);
1125
1126         return rc ? rc : len;
1127 }
1128 static DEVICE_ATTR_RW(sector_size);
1129
1130 static ssize_t dpa_extents_show(struct device *dev,
1131                 struct device_attribute *attr, char *buf)
1132 {
1133         struct nd_region *nd_region = to_nd_region(dev->parent);
1134         struct nd_label_id label_id;
1135         uuid_t *uuid = NULL;
1136         int count = 0, i;
1137         u32 flags = 0;
1138
1139         nvdimm_bus_lock(dev);
1140         if (is_namespace_pmem(dev)) {
1141                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1142
1143                 uuid = nspm->uuid;
1144                 flags = 0;
1145         }
1146
1147         if (!uuid)
1148                 goto out;
1149
1150         nd_label_gen_id(&label_id, uuid, flags);
1151         for (i = 0; i < nd_region->ndr_mappings; i++) {
1152                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1153                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1154                 struct resource *res;
1155
1156                 for_each_dpa_resource(ndd, res)
1157                         if (strcmp(res->name, label_id.id) == 0)
1158                                 count++;
1159         }
1160  out:
1161         nvdimm_bus_unlock(dev);
1162
1163         return sprintf(buf, "%d\n", count);
1164 }
1165 static DEVICE_ATTR_RO(dpa_extents);
1166
1167 static int btt_claim_class(struct device *dev)
1168 {
1169         struct nd_region *nd_region = to_nd_region(dev->parent);
1170         int i, loop_bitmask = 0;
1171
1172         for (i = 0; i < nd_region->ndr_mappings; i++) {
1173                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1174                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1175                 struct nd_namespace_index *nsindex;
1176
1177                 /*
1178                  * If any of the DIMMs do not support labels the only
1179                  * possible BTT format is v1.
1180                  */
1181                 if (!ndd) {
1182                         loop_bitmask = 0;
1183                         break;
1184                 }
1185
1186                 nsindex = to_namespace_index(ndd, ndd->ns_current);
1187                 if (nsindex == NULL)
1188                         loop_bitmask |= 1;
1189                 else {
1190                         /* check whether existing labels are v1.1 or v1.2 */
1191                         if (__le16_to_cpu(nsindex->major) == 1
1192                                         && __le16_to_cpu(nsindex->minor) == 1)
1193                                 loop_bitmask |= 2;
1194                         else
1195                                 loop_bitmask |= 4;
1196                 }
1197         }
1198         /*
1199          * If nsindex is null loop_bitmask's bit 0 will be set, and if an index
1200          * block is found, a v1.1 label for any mapping will set bit 1, and a
1201          * v1.2 label will set bit 2.
1202          *
1203          * At the end of the loop, at most one of the three bits must be set.
1204          * If multiple bits were set, it means the different mappings disagree
1205          * about their labels, and this must be cleaned up first.
1206          *
1207          * If all the label index blocks are found to agree, nsindex of NULL
1208          * implies labels haven't been initialized yet, and when they will,
1209          * they will be of the 1.2 format, so we can assume BTT2.0
1210          *
1211          * If 1.1 labels are found, we enforce BTT1.1, and if 1.2 labels are
1212          * found, we enforce BTT2.0
1213          *
1214          * If the loop was never entered, default to BTT1.1 (legacy namespaces)
1215          */
1216         switch (loop_bitmask) {
1217         case 0:
1218         case 2:
1219                 return NVDIMM_CCLASS_BTT;
1220         case 1:
1221         case 4:
1222                 return NVDIMM_CCLASS_BTT2;
1223         default:
1224                 return -ENXIO;
1225         }
1226 }
1227
1228 static ssize_t holder_show(struct device *dev,
1229                 struct device_attribute *attr, char *buf)
1230 {
1231         struct nd_namespace_common *ndns = to_ndns(dev);
1232         ssize_t rc;
1233
1234         device_lock(dev);
1235         rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
1236         device_unlock(dev);
1237
1238         return rc;
1239 }
1240 static DEVICE_ATTR_RO(holder);
1241
1242 static int __holder_class_store(struct device *dev, const char *buf)
1243 {
1244         struct nd_namespace_common *ndns = to_ndns(dev);
1245
1246         if (dev->driver || ndns->claim)
1247                 return -EBUSY;
1248
1249         if (sysfs_streq(buf, "btt")) {
1250                 int rc = btt_claim_class(dev);
1251
1252                 if (rc < NVDIMM_CCLASS_NONE)
1253                         return rc;
1254                 ndns->claim_class = rc;
1255         } else if (sysfs_streq(buf, "pfn"))
1256                 ndns->claim_class = NVDIMM_CCLASS_PFN;
1257         else if (sysfs_streq(buf, "dax"))
1258                 ndns->claim_class = NVDIMM_CCLASS_DAX;
1259         else if (sysfs_streq(buf, ""))
1260                 ndns->claim_class = NVDIMM_CCLASS_NONE;
1261         else
1262                 return -EINVAL;
1263
1264         return 0;
1265 }
1266
1267 static ssize_t holder_class_store(struct device *dev,
1268                 struct device_attribute *attr, const char *buf, size_t len)
1269 {
1270         struct nd_region *nd_region = to_nd_region(dev->parent);
1271         int rc;
1272
1273         device_lock(dev);
1274         nvdimm_bus_lock(dev);
1275         wait_nvdimm_bus_probe_idle(dev);
1276         rc = __holder_class_store(dev, buf);
1277         if (rc >= 0)
1278                 rc = nd_namespace_label_update(nd_region, dev);
1279         dev_dbg(dev, "%s(%d)\n", rc < 0 ? "fail " : "", rc);
1280         nvdimm_bus_unlock(dev);
1281         device_unlock(dev);
1282
1283         return rc < 0 ? rc : len;
1284 }
1285
1286 static ssize_t holder_class_show(struct device *dev,
1287                 struct device_attribute *attr, char *buf)
1288 {
1289         struct nd_namespace_common *ndns = to_ndns(dev);
1290         ssize_t rc;
1291
1292         device_lock(dev);
1293         if (ndns->claim_class == NVDIMM_CCLASS_NONE)
1294                 rc = sprintf(buf, "\n");
1295         else if ((ndns->claim_class == NVDIMM_CCLASS_BTT) ||
1296                         (ndns->claim_class == NVDIMM_CCLASS_BTT2))
1297                 rc = sprintf(buf, "btt\n");
1298         else if (ndns->claim_class == NVDIMM_CCLASS_PFN)
1299                 rc = sprintf(buf, "pfn\n");
1300         else if (ndns->claim_class == NVDIMM_CCLASS_DAX)
1301                 rc = sprintf(buf, "dax\n");
1302         else
1303                 rc = sprintf(buf, "<unknown>\n");
1304         device_unlock(dev);
1305
1306         return rc;
1307 }
1308 static DEVICE_ATTR_RW(holder_class);
1309
1310 static ssize_t mode_show(struct device *dev,
1311                 struct device_attribute *attr, char *buf)
1312 {
1313         struct nd_namespace_common *ndns = to_ndns(dev);
1314         struct device *claim;
1315         char *mode;
1316         ssize_t rc;
1317
1318         device_lock(dev);
1319         claim = ndns->claim;
1320         if (claim && is_nd_btt(claim))
1321                 mode = "safe";
1322         else if (claim && is_nd_pfn(claim))
1323                 mode = "memory";
1324         else if (claim && is_nd_dax(claim))
1325                 mode = "dax";
1326         else if (!claim && pmem_should_map_pages(dev))
1327                 mode = "memory";
1328         else
1329                 mode = "raw";
1330         rc = sprintf(buf, "%s\n", mode);
1331         device_unlock(dev);
1332
1333         return rc;
1334 }
1335 static DEVICE_ATTR_RO(mode);
1336
1337 static ssize_t force_raw_store(struct device *dev,
1338                 struct device_attribute *attr, const char *buf, size_t len)
1339 {
1340         bool force_raw;
1341         int rc = strtobool(buf, &force_raw);
1342
1343         if (rc)
1344                 return rc;
1345
1346         to_ndns(dev)->force_raw = force_raw;
1347         return len;
1348 }
1349
1350 static ssize_t force_raw_show(struct device *dev,
1351                 struct device_attribute *attr, char *buf)
1352 {
1353         return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
1354 }
1355 static DEVICE_ATTR_RW(force_raw);
1356
1357 static struct attribute *nd_namespace_attributes[] = {
1358         &dev_attr_nstype.attr,
1359         &dev_attr_size.attr,
1360         &dev_attr_mode.attr,
1361         &dev_attr_uuid.attr,
1362         &dev_attr_holder.attr,
1363         &dev_attr_resource.attr,
1364         &dev_attr_alt_name.attr,
1365         &dev_attr_force_raw.attr,
1366         &dev_attr_sector_size.attr,
1367         &dev_attr_dpa_extents.attr,
1368         &dev_attr_holder_class.attr,
1369         NULL,
1370 };
1371
1372 static umode_t namespace_visible(struct kobject *kobj,
1373                 struct attribute *a, int n)
1374 {
1375         struct device *dev = container_of(kobj, struct device, kobj);
1376
1377         if (is_namespace_pmem(dev)) {
1378                 if (a == &dev_attr_size.attr)
1379                         return 0644;
1380
1381                 return a->mode;
1382         }
1383
1384         /* base is_namespace_io() attributes */
1385         if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr ||
1386             a == &dev_attr_holder.attr || a == &dev_attr_holder_class.attr ||
1387             a == &dev_attr_force_raw.attr || a == &dev_attr_mode.attr ||
1388             a == &dev_attr_resource.attr)
1389                 return a->mode;
1390
1391         return 0;
1392 }
1393
1394 static struct attribute_group nd_namespace_attribute_group = {
1395         .attrs = nd_namespace_attributes,
1396         .is_visible = namespace_visible,
1397 };
1398
1399 static const struct attribute_group *nd_namespace_attribute_groups[] = {
1400         &nd_device_attribute_group,
1401         &nd_namespace_attribute_group,
1402         &nd_numa_attribute_group,
1403         NULL,
1404 };
1405
1406 static const struct device_type namespace_io_device_type = {
1407         .name = "nd_namespace_io",
1408         .release = namespace_io_release,
1409         .groups = nd_namespace_attribute_groups,
1410 };
1411
1412 static const struct device_type namespace_pmem_device_type = {
1413         .name = "nd_namespace_pmem",
1414         .release = namespace_pmem_release,
1415         .groups = nd_namespace_attribute_groups,
1416 };
1417
1418 static bool is_namespace_pmem(const struct device *dev)
1419 {
1420         return dev ? dev->type == &namespace_pmem_device_type : false;
1421 }
1422
1423 static bool is_namespace_io(const struct device *dev)
1424 {
1425         return dev ? dev->type == &namespace_io_device_type : false;
1426 }
1427
1428 struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
1429 {
1430         struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
1431         struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
1432         struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
1433         struct nd_namespace_common *ndns = NULL;
1434         resource_size_t size;
1435
1436         if (nd_btt || nd_pfn || nd_dax) {
1437                 if (nd_btt)
1438                         ndns = nd_btt->ndns;
1439                 else if (nd_pfn)
1440                         ndns = nd_pfn->ndns;
1441                 else if (nd_dax)
1442                         ndns = nd_dax->nd_pfn.ndns;
1443
1444                 if (!ndns)
1445                         return ERR_PTR(-ENODEV);
1446
1447                 /*
1448                  * Flush any in-progess probes / removals in the driver
1449                  * for the raw personality of this namespace.
1450                  */
1451                 device_lock(&ndns->dev);
1452                 device_unlock(&ndns->dev);
1453                 if (ndns->dev.driver) {
1454                         dev_dbg(&ndns->dev, "is active, can't bind %s\n",
1455                                         dev_name(dev));
1456                         return ERR_PTR(-EBUSY);
1457                 }
1458                 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != dev,
1459                                         "host (%s) vs claim (%s) mismatch\n",
1460                                         dev_name(dev),
1461                                         dev_name(ndns->claim)))
1462                         return ERR_PTR(-ENXIO);
1463         } else {
1464                 ndns = to_ndns(dev);
1465                 if (ndns->claim) {
1466                         dev_dbg(dev, "claimed by %s, failing probe\n",
1467                                 dev_name(ndns->claim));
1468
1469                         return ERR_PTR(-ENXIO);
1470                 }
1471         }
1472
1473         if (nvdimm_namespace_locked(ndns))
1474                 return ERR_PTR(-EACCES);
1475
1476         size = nvdimm_namespace_capacity(ndns);
1477         if (size < ND_MIN_NAMESPACE_SIZE) {
1478                 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
1479                                 &size, ND_MIN_NAMESPACE_SIZE);
1480                 return ERR_PTR(-ENODEV);
1481         }
1482
1483         /*
1484          * Note, alignment validation for fsdax and devdax mode
1485          * namespaces happens in nd_pfn_validate() where infoblock
1486          * padding parameters can be applied.
1487          */
1488         if (pmem_should_map_pages(dev)) {
1489                 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
1490                 struct resource *res = &nsio->res;
1491
1492                 if (!IS_ALIGNED(res->start | (res->end + 1),
1493                                         memremap_compat_align())) {
1494                         dev_err(&ndns->dev, "%pr misaligned, unable to map\n", res);
1495                         return ERR_PTR(-EOPNOTSUPP);
1496                 }
1497         }
1498
1499         if (is_namespace_pmem(&ndns->dev)) {
1500                 struct nd_namespace_pmem *nspm;
1501
1502                 nspm = to_nd_namespace_pmem(&ndns->dev);
1503                 if (uuid_not_set(nspm->uuid, &ndns->dev, __func__))
1504                         return ERR_PTR(-ENODEV);
1505         }
1506
1507         return ndns;
1508 }
1509 EXPORT_SYMBOL(nvdimm_namespace_common_probe);
1510
1511 int devm_namespace_enable(struct device *dev, struct nd_namespace_common *ndns,
1512                 resource_size_t size)
1513 {
1514         return devm_nsio_enable(dev, to_nd_namespace_io(&ndns->dev), size);
1515 }
1516 EXPORT_SYMBOL_GPL(devm_namespace_enable);
1517
1518 void devm_namespace_disable(struct device *dev, struct nd_namespace_common *ndns)
1519 {
1520         devm_nsio_disable(dev, to_nd_namespace_io(&ndns->dev));
1521 }
1522 EXPORT_SYMBOL_GPL(devm_namespace_disable);
1523
1524 static struct device **create_namespace_io(struct nd_region *nd_region)
1525 {
1526         struct nd_namespace_io *nsio;
1527         struct device *dev, **devs;
1528         struct resource *res;
1529
1530         nsio = kzalloc(sizeof(*nsio), GFP_KERNEL);
1531         if (!nsio)
1532                 return NULL;
1533
1534         devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1535         if (!devs) {
1536                 kfree(nsio);
1537                 return NULL;
1538         }
1539
1540         dev = &nsio->common.dev;
1541         dev->type = &namespace_io_device_type;
1542         dev->parent = &nd_region->dev;
1543         res = &nsio->res;
1544         res->name = dev_name(&nd_region->dev);
1545         res->flags = IORESOURCE_MEM;
1546         res->start = nd_region->ndr_start;
1547         res->end = res->start + nd_region->ndr_size - 1;
1548
1549         devs[0] = dev;
1550         return devs;
1551 }
1552
1553 static bool has_uuid_at_pos(struct nd_region *nd_region, const uuid_t *uuid,
1554                             u64 cookie, u16 pos)
1555 {
1556         struct nd_namespace_label *found = NULL;
1557         int i;
1558
1559         for (i = 0; i < nd_region->ndr_mappings; i++) {
1560                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1561                 struct nd_interleave_set *nd_set = nd_region->nd_set;
1562                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1563                 struct nd_label_ent *label_ent;
1564                 bool found_uuid = false;
1565
1566                 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1567                         struct nd_namespace_label *nd_label = label_ent->label;
1568                         u16 position;
1569
1570                         if (!nd_label)
1571                                 continue;
1572                         position = nsl_get_position(ndd, nd_label);
1573
1574                         if (!nsl_validate_isetcookie(ndd, nd_label, cookie))
1575                                 continue;
1576
1577                         if (!nsl_uuid_equal(ndd, nd_label, uuid))
1578                                 continue;
1579
1580                         if (!nsl_validate_type_guid(ndd, nd_label,
1581                                                     &nd_set->type_guid))
1582                                 continue;
1583
1584                         if (found_uuid) {
1585                                 dev_dbg(ndd->dev, "duplicate entry for uuid\n");
1586                                 return false;
1587                         }
1588                         found_uuid = true;
1589                         if (!nsl_validate_nlabel(nd_region, ndd, nd_label))
1590                                 continue;
1591                         if (position != pos)
1592                                 continue;
1593                         found = nd_label;
1594                         break;
1595                 }
1596                 if (found)
1597                         break;
1598         }
1599         return found != NULL;
1600 }
1601
1602 static int select_pmem_id(struct nd_region *nd_region, const uuid_t *pmem_id)
1603 {
1604         int i;
1605
1606         if (!pmem_id)
1607                 return -ENODEV;
1608
1609         for (i = 0; i < nd_region->ndr_mappings; i++) {
1610                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1611                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1612                 struct nd_namespace_label *nd_label = NULL;
1613                 u64 hw_start, hw_end, pmem_start, pmem_end;
1614                 struct nd_label_ent *label_ent;
1615
1616                 lockdep_assert_held(&nd_mapping->lock);
1617                 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1618                         nd_label = label_ent->label;
1619                         if (!nd_label)
1620                                 continue;
1621                         if (nsl_uuid_equal(ndd, nd_label, pmem_id))
1622                                 break;
1623                         nd_label = NULL;
1624                 }
1625
1626                 if (!nd_label) {
1627                         WARN_ON(1);
1628                         return -EINVAL;
1629                 }
1630
1631                 /*
1632                  * Check that this label is compliant with the dpa
1633                  * range published in NFIT
1634                  */
1635                 hw_start = nd_mapping->start;
1636                 hw_end = hw_start + nd_mapping->size;
1637                 pmem_start = nsl_get_dpa(ndd, nd_label);
1638                 pmem_end = pmem_start + nsl_get_rawsize(ndd, nd_label);
1639                 if (pmem_start >= hw_start && pmem_start < hw_end
1640                                 && pmem_end <= hw_end && pmem_end > hw_start)
1641                         /* pass */;
1642                 else {
1643                         dev_dbg(&nd_region->dev, "%s invalid label for %pUb\n",
1644                                 dev_name(ndd->dev),
1645                                 nsl_uuid_raw(ndd, nd_label));
1646                         return -EINVAL;
1647                 }
1648
1649                 /* move recently validated label to the front of the list */
1650                 list_move(&label_ent->list, &nd_mapping->labels);
1651         }
1652         return 0;
1653 }
1654
1655 /**
1656  * create_namespace_pmem - validate interleave set labelling, retrieve label0
1657  * @nd_region: region with mappings to validate
1658  * @nspm: target namespace to create
1659  * @nd_label: target pmem namespace label to evaluate
1660  */
1661 static struct device *create_namespace_pmem(struct nd_region *nd_region,
1662                                             struct nd_mapping *nd_mapping,
1663                                             struct nd_namespace_label *nd_label)
1664 {
1665         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1666         struct nd_namespace_index *nsindex =
1667                 to_namespace_index(ndd, ndd->ns_current);
1668         u64 cookie = nd_region_interleave_set_cookie(nd_region, nsindex);
1669         u64 altcookie = nd_region_interleave_set_altcookie(nd_region);
1670         struct nd_label_ent *label_ent;
1671         struct nd_namespace_pmem *nspm;
1672         resource_size_t size = 0;
1673         struct resource *res;
1674         struct device *dev;
1675         uuid_t uuid;
1676         int rc = 0;
1677         u16 i;
1678
1679         if (cookie == 0) {
1680                 dev_dbg(&nd_region->dev, "invalid interleave-set-cookie\n");
1681                 return ERR_PTR(-ENXIO);
1682         }
1683
1684         if (!nsl_validate_isetcookie(ndd, nd_label, cookie)) {
1685                 dev_dbg(&nd_region->dev, "invalid cookie in label: %pUb\n",
1686                         nsl_uuid_raw(ndd, nd_label));
1687                 if (!nsl_validate_isetcookie(ndd, nd_label, altcookie))
1688                         return ERR_PTR(-EAGAIN);
1689
1690                 dev_dbg(&nd_region->dev, "valid altcookie in label: %pUb\n",
1691                         nsl_uuid_raw(ndd, nd_label));
1692         }
1693
1694         nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1695         if (!nspm)
1696                 return ERR_PTR(-ENOMEM);
1697
1698         nspm->id = -1;
1699         dev = &nspm->nsio.common.dev;
1700         dev->type = &namespace_pmem_device_type;
1701         dev->parent = &nd_region->dev;
1702         res = &nspm->nsio.res;
1703         res->name = dev_name(&nd_region->dev);
1704         res->flags = IORESOURCE_MEM;
1705
1706         for (i = 0; i < nd_region->ndr_mappings; i++) {
1707                 nsl_get_uuid(ndd, nd_label, &uuid);
1708                 if (has_uuid_at_pos(nd_region, &uuid, cookie, i))
1709                         continue;
1710                 if (has_uuid_at_pos(nd_region, &uuid, altcookie, i))
1711                         continue;
1712                 break;
1713         }
1714
1715         if (i < nd_region->ndr_mappings) {
1716                 struct nvdimm *nvdimm = nd_region->mapping[i].nvdimm;
1717
1718                 /*
1719                  * Give up if we don't find an instance of a uuid at each
1720                  * position (from 0 to nd_region->ndr_mappings - 1), or if we
1721                  * find a dimm with two instances of the same uuid.
1722                  */
1723                 dev_err(&nd_region->dev, "%s missing label for %pUb\n",
1724                         nvdimm_name(nvdimm), nsl_uuid_raw(ndd, nd_label));
1725                 rc = -EINVAL;
1726                 goto err;
1727         }
1728
1729         /*
1730          * Fix up each mapping's 'labels' to have the validated pmem label for
1731          * that position at labels[0], and NULL at labels[1].  In the process,
1732          * check that the namespace aligns with interleave-set.
1733          */
1734         nsl_get_uuid(ndd, nd_label, &uuid);
1735         rc = select_pmem_id(nd_region, &uuid);
1736         if (rc)
1737                 goto err;
1738
1739         /* Calculate total size and populate namespace properties from label0 */
1740         for (i = 0; i < nd_region->ndr_mappings; i++) {
1741                 struct nd_namespace_label *label0;
1742                 struct nvdimm_drvdata *ndd;
1743
1744                 nd_mapping = &nd_region->mapping[i];
1745                 label_ent = list_first_entry_or_null(&nd_mapping->labels,
1746                                 typeof(*label_ent), list);
1747                 label0 = label_ent ? label_ent->label : NULL;
1748
1749                 if (!label0) {
1750                         WARN_ON(1);
1751                         continue;
1752                 }
1753
1754                 ndd = to_ndd(nd_mapping);
1755                 size += nsl_get_rawsize(ndd, label0);
1756                 if (nsl_get_position(ndd, label0) != 0)
1757                         continue;
1758                 WARN_ON(nspm->alt_name || nspm->uuid);
1759                 nspm->alt_name = kmemdup(nsl_ref_name(ndd, label0),
1760                                          NSLABEL_NAME_LEN, GFP_KERNEL);
1761                 nsl_get_uuid(ndd, label0, &uuid);
1762                 nspm->uuid = kmemdup(&uuid, sizeof(uuid_t), GFP_KERNEL);
1763                 nspm->lbasize = nsl_get_lbasize(ndd, label0);
1764                 nspm->nsio.common.claim_class =
1765                         nsl_get_claim_class(ndd, label0);
1766         }
1767
1768         if (!nspm->alt_name || !nspm->uuid) {
1769                 rc = -ENOMEM;
1770                 goto err;
1771         }
1772
1773         nd_namespace_pmem_set_resource(nd_region, nspm, size);
1774
1775         return dev;
1776  err:
1777         namespace_pmem_release(dev);
1778         switch (rc) {
1779         case -EINVAL:
1780                 dev_dbg(&nd_region->dev, "invalid label(s)\n");
1781                 break;
1782         case -ENODEV:
1783                 dev_dbg(&nd_region->dev, "label not found\n");
1784                 break;
1785         default:
1786                 dev_dbg(&nd_region->dev, "unexpected err: %d\n", rc);
1787                 break;
1788         }
1789         return ERR_PTR(rc);
1790 }
1791
1792 static struct device *nd_namespace_pmem_create(struct nd_region *nd_region)
1793 {
1794         struct nd_namespace_pmem *nspm;
1795         struct resource *res;
1796         struct device *dev;
1797
1798         if (!is_memory(&nd_region->dev))
1799                 return NULL;
1800
1801         nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1802         if (!nspm)
1803                 return NULL;
1804
1805         dev = &nspm->nsio.common.dev;
1806         dev->type = &namespace_pmem_device_type;
1807         dev->parent = &nd_region->dev;
1808         res = &nspm->nsio.res;
1809         res->name = dev_name(&nd_region->dev);
1810         res->flags = IORESOURCE_MEM;
1811
1812         nspm->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
1813         if (nspm->id < 0) {
1814                 kfree(nspm);
1815                 return NULL;
1816         }
1817         dev_set_name(dev, "namespace%d.%d", nd_region->id, nspm->id);
1818         nd_namespace_pmem_set_resource(nd_region, nspm, 0);
1819
1820         return dev;
1821 }
1822
1823 static struct lock_class_key nvdimm_namespace_key;
1824
1825 void nd_region_create_ns_seed(struct nd_region *nd_region)
1826 {
1827         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1828
1829         if (nd_region_to_nstype(nd_region) == ND_DEVICE_NAMESPACE_IO)
1830                 return;
1831
1832         nd_region->ns_seed = nd_namespace_pmem_create(nd_region);
1833
1834         /*
1835          * Seed creation failures are not fatal, provisioning is simply
1836          * disabled until memory becomes available
1837          */
1838         if (!nd_region->ns_seed)
1839                 dev_err(&nd_region->dev, "failed to create namespace\n");
1840         else {
1841                 device_initialize(nd_region->ns_seed);
1842                 lockdep_set_class(&nd_region->ns_seed->mutex,
1843                                   &nvdimm_namespace_key);
1844                 nd_device_register(nd_region->ns_seed);
1845         }
1846 }
1847
1848 void nd_region_create_dax_seed(struct nd_region *nd_region)
1849 {
1850         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1851         nd_region->dax_seed = nd_dax_create(nd_region);
1852         /*
1853          * Seed creation failures are not fatal, provisioning is simply
1854          * disabled until memory becomes available
1855          */
1856         if (!nd_region->dax_seed)
1857                 dev_err(&nd_region->dev, "failed to create dax namespace\n");
1858 }
1859
1860 void nd_region_create_pfn_seed(struct nd_region *nd_region)
1861 {
1862         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1863         nd_region->pfn_seed = nd_pfn_create(nd_region);
1864         /*
1865          * Seed creation failures are not fatal, provisioning is simply
1866          * disabled until memory becomes available
1867          */
1868         if (!nd_region->pfn_seed)
1869                 dev_err(&nd_region->dev, "failed to create pfn namespace\n");
1870 }
1871
1872 void nd_region_create_btt_seed(struct nd_region *nd_region)
1873 {
1874         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1875         nd_region->btt_seed = nd_btt_create(nd_region);
1876         /*
1877          * Seed creation failures are not fatal, provisioning is simply
1878          * disabled until memory becomes available
1879          */
1880         if (!nd_region->btt_seed)
1881                 dev_err(&nd_region->dev, "failed to create btt namespace\n");
1882 }
1883
1884 static int add_namespace_resource(struct nd_region *nd_region,
1885                 struct nd_namespace_label *nd_label, struct device **devs,
1886                 int count)
1887 {
1888         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
1889         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1890         int i;
1891
1892         for (i = 0; i < count; i++) {
1893                 uuid_t *uuid = namespace_to_uuid(devs[i]);
1894
1895                 if (IS_ERR(uuid)) {
1896                         WARN_ON(1);
1897                         continue;
1898                 }
1899
1900                 if (!nsl_uuid_equal(ndd, nd_label, uuid))
1901                         continue;
1902                 dev_err(&nd_region->dev,
1903                         "error: conflicting extents for uuid: %pUb\n", uuid);
1904                 return -ENXIO;
1905         }
1906
1907         return i;
1908 }
1909
1910 static int cmp_dpa(const void *a, const void *b)
1911 {
1912         const struct device *dev_a = *(const struct device **) a;
1913         const struct device *dev_b = *(const struct device **) b;
1914         struct nd_namespace_pmem *nspm_a, *nspm_b;
1915
1916         if (is_namespace_io(dev_a))
1917                 return 0;
1918
1919         nspm_a = to_nd_namespace_pmem(dev_a);
1920         nspm_b = to_nd_namespace_pmem(dev_b);
1921
1922         return memcmp(&nspm_a->nsio.res.start, &nspm_b->nsio.res.start,
1923                         sizeof(resource_size_t));
1924 }
1925
1926 static struct device **scan_labels(struct nd_region *nd_region)
1927 {
1928         int i, count = 0;
1929         struct device *dev, **devs = NULL;
1930         struct nd_label_ent *label_ent, *e;
1931         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
1932         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1933         resource_size_t map_end = nd_mapping->start + nd_mapping->size - 1;
1934
1935         /* "safe" because create_namespace_pmem() might list_move() label_ent */
1936         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
1937                 struct nd_namespace_label *nd_label = label_ent->label;
1938                 struct device **__devs;
1939
1940                 if (!nd_label)
1941                         continue;
1942
1943                 /* skip labels that describe extents outside of the region */
1944                 if (nsl_get_dpa(ndd, nd_label) < nd_mapping->start ||
1945                     nsl_get_dpa(ndd, nd_label) > map_end)
1946                         continue;
1947
1948                 i = add_namespace_resource(nd_region, nd_label, devs, count);
1949                 if (i < 0)
1950                         goto err;
1951                 if (i < count)
1952                         continue;
1953                 __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
1954                 if (!__devs)
1955                         goto err;
1956                 memcpy(__devs, devs, sizeof(dev) * count);
1957                 kfree(devs);
1958                 devs = __devs;
1959
1960                 dev = create_namespace_pmem(nd_region, nd_mapping, nd_label);
1961                 if (IS_ERR(dev)) {
1962                         switch (PTR_ERR(dev)) {
1963                         case -EAGAIN:
1964                                 /* skip invalid labels */
1965                                 continue;
1966                         case -ENODEV:
1967                                 /* fallthrough to seed creation */
1968                                 break;
1969                         default:
1970                                 goto err;
1971                         }
1972                 } else
1973                         devs[count++] = dev;
1974
1975         }
1976
1977         dev_dbg(&nd_region->dev, "discovered %d namespace%s\n", count,
1978                 count == 1 ? "" : "s");
1979
1980         if (count == 0) {
1981                 struct nd_namespace_pmem *nspm;
1982
1983                 /* Publish a zero-sized namespace for userspace to configure. */
1984                 nd_mapping_free_labels(nd_mapping);
1985
1986                 devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
1987                 if (!devs)
1988                         goto err;
1989
1990                 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1991                 if (!nspm)
1992                         goto err;
1993                 dev = &nspm->nsio.common.dev;
1994                 dev->type = &namespace_pmem_device_type;
1995                 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
1996                 dev->parent = &nd_region->dev;
1997                 devs[count++] = dev;
1998         } else if (is_memory(&nd_region->dev)) {
1999                 /* clean unselected labels */
2000                 for (i = 0; i < nd_region->ndr_mappings; i++) {
2001                         struct list_head *l, *e;
2002                         LIST_HEAD(list);
2003                         int j;
2004
2005                         nd_mapping = &nd_region->mapping[i];
2006                         if (list_empty(&nd_mapping->labels)) {
2007                                 WARN_ON(1);
2008                                 continue;
2009                         }
2010
2011                         j = count;
2012                         list_for_each_safe(l, e, &nd_mapping->labels) {
2013                                 if (!j--)
2014                                         break;
2015                                 list_move_tail(l, &list);
2016                         }
2017                         nd_mapping_free_labels(nd_mapping);
2018                         list_splice_init(&list, &nd_mapping->labels);
2019                 }
2020         }
2021
2022         if (count > 1)
2023                 sort(devs, count, sizeof(struct device *), cmp_dpa, NULL);
2024
2025         return devs;
2026
2027  err:
2028         if (devs) {
2029                 for (i = 0; devs[i]; i++)
2030                         namespace_pmem_release(devs[i]);
2031                 kfree(devs);
2032         }
2033         return NULL;
2034 }
2035
2036 static struct device **create_namespaces(struct nd_region *nd_region)
2037 {
2038         struct nd_mapping *nd_mapping;
2039         struct device **devs;
2040         int i;
2041
2042         if (nd_region->ndr_mappings == 0)
2043                 return NULL;
2044
2045         /* lock down all mappings while we scan labels */
2046         for (i = 0; i < nd_region->ndr_mappings; i++) {
2047                 nd_mapping = &nd_region->mapping[i];
2048                 mutex_lock_nested(&nd_mapping->lock, i);
2049         }
2050
2051         devs = scan_labels(nd_region);
2052
2053         for (i = 0; i < nd_region->ndr_mappings; i++) {
2054                 int reverse = nd_region->ndr_mappings - 1 - i;
2055
2056                 nd_mapping = &nd_region->mapping[reverse];
2057                 mutex_unlock(&nd_mapping->lock);
2058         }
2059
2060         return devs;
2061 }
2062
2063 static void deactivate_labels(void *region)
2064 {
2065         struct nd_region *nd_region = region;
2066         int i;
2067
2068         for (i = 0; i < nd_region->ndr_mappings; i++) {
2069                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2070                 struct nvdimm_drvdata *ndd = nd_mapping->ndd;
2071                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2072
2073                 mutex_lock(&nd_mapping->lock);
2074                 nd_mapping_free_labels(nd_mapping);
2075                 mutex_unlock(&nd_mapping->lock);
2076
2077                 put_ndd(ndd);
2078                 nd_mapping->ndd = NULL;
2079                 if (ndd)
2080                         atomic_dec(&nvdimm->busy);
2081         }
2082 }
2083
2084 static int init_active_labels(struct nd_region *nd_region)
2085 {
2086         int i, rc = 0;
2087
2088         for (i = 0; i < nd_region->ndr_mappings; i++) {
2089                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2090                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2091                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2092                 struct nd_label_ent *label_ent;
2093                 int count, j;
2094
2095                 /*
2096                  * If the dimm is disabled then we may need to prevent
2097                  * the region from being activated.
2098                  */
2099                 if (!ndd) {
2100                         if (test_bit(NDD_LOCKED, &nvdimm->flags))
2101                                 /* fail, label data may be unreadable */;
2102                         else if (test_bit(NDD_LABELING, &nvdimm->flags))
2103                                 /* fail, labels needed to disambiguate dpa */;
2104                         else
2105                                 continue;
2106
2107                         dev_err(&nd_region->dev, "%s: is %s, failing probe\n",
2108                                         dev_name(&nd_mapping->nvdimm->dev),
2109                                         test_bit(NDD_LOCKED, &nvdimm->flags)
2110                                         ? "locked" : "disabled");
2111                         rc = -ENXIO;
2112                         goto out;
2113                 }
2114                 nd_mapping->ndd = ndd;
2115                 atomic_inc(&nvdimm->busy);
2116                 get_ndd(ndd);
2117
2118                 count = nd_label_active_count(ndd);
2119                 dev_dbg(ndd->dev, "count: %d\n", count);
2120                 if (!count)
2121                         continue;
2122                 for (j = 0; j < count; j++) {
2123                         struct nd_namespace_label *label;
2124
2125                         label_ent = kzalloc(sizeof(*label_ent), GFP_KERNEL);
2126                         if (!label_ent)
2127                                 break;
2128                         label = nd_label_active(ndd, j);
2129                         label_ent->label = label;
2130
2131                         mutex_lock(&nd_mapping->lock);
2132                         list_add_tail(&label_ent->list, &nd_mapping->labels);
2133                         mutex_unlock(&nd_mapping->lock);
2134                 }
2135
2136                 if (j < count)
2137                         break;
2138         }
2139
2140         if (i < nd_region->ndr_mappings)
2141                 rc = -ENOMEM;
2142
2143 out:
2144         if (rc) {
2145                 deactivate_labels(nd_region);
2146                 return rc;
2147         }
2148
2149         return devm_add_action_or_reset(&nd_region->dev, deactivate_labels,
2150                                         nd_region);
2151 }
2152
2153 int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
2154 {
2155         struct device **devs = NULL;
2156         int i, rc = 0, type;
2157
2158         *err = 0;
2159         nvdimm_bus_lock(&nd_region->dev);
2160         rc = init_active_labels(nd_region);
2161         if (rc) {
2162                 nvdimm_bus_unlock(&nd_region->dev);
2163                 return rc;
2164         }
2165
2166         type = nd_region_to_nstype(nd_region);
2167         switch (type) {
2168         case ND_DEVICE_NAMESPACE_IO:
2169                 devs = create_namespace_io(nd_region);
2170                 break;
2171         case ND_DEVICE_NAMESPACE_PMEM:
2172                 devs = create_namespaces(nd_region);
2173                 break;
2174         default:
2175                 break;
2176         }
2177         nvdimm_bus_unlock(&nd_region->dev);
2178
2179         if (!devs)
2180                 return -ENODEV;
2181
2182         for (i = 0; devs[i]; i++) {
2183                 struct device *dev = devs[i];
2184                 int id;
2185
2186                 if (type == ND_DEVICE_NAMESPACE_PMEM) {
2187                         struct nd_namespace_pmem *nspm;
2188
2189                         nspm = to_nd_namespace_pmem(dev);
2190                         id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2191                                             GFP_KERNEL);
2192                         nspm->id = id;
2193                 } else
2194                         id = i;
2195
2196                 if (id < 0)
2197                         break;
2198                 dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
2199                 device_initialize(dev);
2200                 lockdep_set_class(&dev->mutex, &nvdimm_namespace_key);
2201                 nd_device_register(dev);
2202         }
2203         if (i)
2204                 nd_region->ns_seed = devs[0];
2205
2206         if (devs[i]) {
2207                 int j;
2208
2209                 for (j = i; devs[j]; j++) {
2210                         struct device *dev = devs[j];
2211
2212                         device_initialize(dev);
2213                         put_device(dev);
2214                 }
2215                 *err = j - i;
2216                 /*
2217                  * All of the namespaces we tried to register failed, so
2218                  * fail region activation.
2219                  */
2220                 if (*err == 0)
2221                         rc = -ENODEV;
2222         }
2223         kfree(devs);
2224
2225         if (rc == -ENODEV)
2226                 return rc;
2227
2228         return i;
2229 }