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/kstrtox.h>
   6 #include <linux/module.h>
   7 #include <linux/device.h>
   8 #include <linux/sort.h>
   9 #include <linux/slab.h>
  10 #include <linux/list.h>
  11 #include <linux/nd.h>
  12 #include "nd-core.h"
  13 #include "pmem.h"
  14 #include "pfn.h"
  15 #include "nd.h"
  16
  17 static void namespace_io_release(struct device *dev)
  18 {
  19         struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
  20
  21         kfree(nsio);
  22 }
  23
  24 static void namespace_pmem_release(struct device *dev)
  25 {
  26         struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
  27         struct nd_region *nd_region = to_nd_region(dev->parent);
  28
  29         if (nspm->id >= 0)
  30                 ida_free(&nd_region->ns_ida, nspm->id);
  31         kfree(nspm->alt_name);
  32         kfree(nspm->uuid);
  33         kfree(nspm);
  34 }
  35
  36 static bool is_namespace_pmem(const struct device *dev);
  37 static bool is_namespace_io(const struct device *dev);
  38
  39 static int is_uuid_busy(struct device *dev, void *data)
  40 {
  41         uuid_t *uuid1 = data, *uuid2 = NULL;
  42
  43         if (is_namespace_pmem(dev)) {
  44                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
  45
  46                 uuid2 = nspm->uuid;
  47         } else if (is_nd_btt(dev)) {
  48                 struct nd_btt *nd_btt = to_nd_btt(dev);
  49
  50                 uuid2 = nd_btt->uuid;
  51         } else if (is_nd_pfn(dev)) {
  52                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
  53
  54                 uuid2 = nd_pfn->uuid;
  55         }
  56
  57         if (uuid2 && uuid_equal(uuid1, uuid2))
  58                 return -EBUSY;
  59
  60         return 0;
  61 }
  62
  63 static int is_namespace_uuid_busy(struct device *dev, void *data)
  64 {
  65         if (is_nd_region(dev))
  66                 return device_for_each_child(dev, data, is_uuid_busy);
  67         return 0;
  68 }
  69
  70 /**
  71  * nd_is_uuid_unique - verify that no other namespace has @uuid
  72  * @dev: any device on a nvdimm_bus
  73  * @uuid: uuid to check
  74  *
  75  * Returns: %true if the uuid is unique, %false if not
  76  */
  77 bool nd_is_uuid_unique(struct device *dev, uuid_t *uuid)
  78 {
  79         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
  80
  81         if (!nvdimm_bus)
  82                 return false;
  83         WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm_bus->dev));
  84         if (device_for_each_child(&nvdimm_bus->dev, uuid,
  85                                 is_namespace_uuid_busy) != 0)
  86                 return false;
  87         return true;
  88 }
  89
  90 bool pmem_should_map_pages(struct device *dev)
  91 {
  92         struct nd_region *nd_region = to_nd_region(dev->parent);
  93         struct nd_namespace_common *ndns = to_ndns(dev);
  94         struct nd_namespace_io *nsio;
  95
  96         if (!IS_ENABLED(CONFIG_ZONE_DEVICE))
  97                 return false;
  98
  99         if (!test_bit(ND_REGION_PAGEMAP, &nd_region->flags))
 100                 return false;
 101
 102         if (is_nd_pfn(dev) || is_nd_btt(dev))
 103                 return false;
 104
 105         if (ndns->force_raw)
 106                 return false;
 107
 108         nsio = to_nd_namespace_io(dev);
 109         if (region_intersects(nsio->res.start, resource_size(&nsio->res),
 110                                 IORESOURCE_SYSTEM_RAM,
 111                                 IORES_DESC_NONE) == REGION_MIXED)
 112                 return false;
 113
 114         return ARCH_MEMREMAP_PMEM == MEMREMAP_WB;
 115 }
 116 EXPORT_SYMBOL(pmem_should_map_pages);
 117
 118 unsigned int pmem_sector_size(struct nd_namespace_common *ndns)
 119 {
 120         if (is_namespace_pmem(&ndns->dev)) {
 121                 struct nd_namespace_pmem *nspm;
 122
 123                 nspm = to_nd_namespace_pmem(&ndns->dev);
 124                 if (nspm->lbasize == 0 || nspm->lbasize == 512)
 125                         /* default */;
 126                 else if (nspm->lbasize == 4096)
 127                         return 4096;
 128                 else
 129                         dev_WARN(&ndns->dev, "unsupported sector size: %ld\n",
 130                                         nspm->lbasize);
 131         }
 132
 133         /*
 134          * There is no namespace label (is_namespace_io()), or the label
 135          * indicates the default sector size.
 136          */
 137         return 512;
 138 }
 139 EXPORT_SYMBOL(pmem_sector_size);
 140
 141 const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,
 142                 char *name)
 143 {
 144         struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
 145         const char *suffix = NULL;
 146
 147         if (ndns->claim && is_nd_btt(ndns->claim))
 148                 suffix = "s";
 149
 150         if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev)) {
 151                 int nsidx = 0;
 152
 153                 if (is_namespace_pmem(&ndns->dev)) {
 154                         struct nd_namespace_pmem *nspm;
 155
 156                         nspm = to_nd_namespace_pmem(&ndns->dev);
 157                         nsidx = nspm->id;
 158                 }
 159
 160                 if (nsidx)
 161                         sprintf(name, "pmem%d.%d%s", nd_region->id, nsidx,
 162                                         suffix ? suffix : "");
 163                 else
 164                         sprintf(name, "pmem%d%s", nd_region->id,
 165                                         suffix ? suffix : "");
 166         } else {
 167                 return NULL;
 168         }
 169
 170         return name;
 171 }
 172 EXPORT_SYMBOL(nvdimm_namespace_disk_name);
 173
 174 const uuid_t *nd_dev_to_uuid(struct device *dev)
 175 {
 176         if (dev && is_namespace_pmem(dev)) {
 177                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 178
 179                 return nspm->uuid;
 180         }
 181         return &uuid_null;
 182 }
 183 EXPORT_SYMBOL(nd_dev_to_uuid);
 184
 185 static ssize_t nstype_show(struct device *dev,
 186                 struct device_attribute *attr, char *buf)
 187 {
 188         struct nd_region *nd_region = to_nd_region(dev->parent);
 189
 190         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
 191 }
 192 static DEVICE_ATTR_RO(nstype);
 193
 194 static ssize_t __alt_name_store(struct device *dev, const char *buf,
 195                 const size_t len)
 196 {
 197         char *input, *pos, *alt_name, **ns_altname;
 198         ssize_t rc;
 199
 200         if (is_namespace_pmem(dev)) {
 201                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 202
 203                 ns_altname = &nspm->alt_name;
 204         } else
 205                 return -ENXIO;
 206
 207         if (dev->driver || to_ndns(dev)->claim)
 208                 return -EBUSY;
 209
 210         input = kstrndup(buf, len, GFP_KERNEL);
 211         if (!input)
 212                 return -ENOMEM;
 213
 214         pos = strim(input);
 215         if (strlen(pos) + 1 > NSLABEL_NAME_LEN) {
 216                 rc = -EINVAL;
 217                 goto out;
 218         }
 219
 220         alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL);
 221         if (!alt_name) {
 222                 rc = -ENOMEM;
 223                 goto out;
 224         }
 225         kfree(*ns_altname);
 226         *ns_altname = alt_name;
 227         sprintf(*ns_altname, "%s", pos);
 228         rc = len;
 229
 230 out:
 231         kfree(input);
 232         return rc;
 233 }
 234
 235 static int nd_namespace_label_update(struct nd_region *nd_region,
 236                 struct device *dev)
 237 {
 238         dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim,
 239                         "namespace must be idle during label update\n");
 240         if (dev->driver || to_ndns(dev)->claim)
 241                 return 0;
 242
 243         /*
 244          * Only allow label writes that will result in a valid namespace
 245          * or deletion of an existing namespace.
 246          */
 247         if (is_namespace_pmem(dev)) {
 248                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 249                 resource_size_t size = resource_size(&nspm->nsio.res);
 250
 251                 if (size == 0 && nspm->uuid)
 252                         /* delete allocation */;
 253                 else if (!nspm->uuid)
 254                         return 0;
 255
 256                 return nd_pmem_namespace_label_update(nd_region, nspm, size);
 257         } else
 258                 return -ENXIO;
 259 }
 260
 261 static ssize_t alt_name_store(struct device *dev,
 262                 struct device_attribute *attr, const char *buf, size_t len)
 263 {
 264         struct nd_region *nd_region = to_nd_region(dev->parent);
 265         ssize_t rc;
 266
 267         device_lock(dev);
 268         nvdimm_bus_lock(dev);
 269         wait_nvdimm_bus_probe_idle(dev);
 270         rc = __alt_name_store(dev, buf, len);
 271         if (rc >= 0)
 272                 rc = nd_namespace_label_update(nd_region, dev);
 273         dev_dbg(dev, "%s(%zd)\n", rc < 0 ? "fail " : "", rc);
 274         nvdimm_bus_unlock(dev);
 275         device_unlock(dev);
 276
 277         return rc < 0 ? rc : len;
 278 }
 279
 280 static ssize_t alt_name_show(struct device *dev,
 281                 struct device_attribute *attr, char *buf)
 282 {
 283         char *ns_altname;
 284
 285         if (is_namespace_pmem(dev)) {
 286                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 287
 288                 ns_altname = nspm->alt_name;
 289         } else
 290                 return -ENXIO;
 291
 292         return sprintf(buf, "%s\n", ns_altname ? ns_altname : "");
 293 }
 294 static DEVICE_ATTR_RW(alt_name);
 295
 296 static int scan_free(struct nd_region *nd_region,
 297                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
 298                 resource_size_t n)
 299 {
 300         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 301         int rc = 0;
 302
 303         while (n) {
 304                 struct resource *res, *last;
 305
 306                 last = NULL;
 307                 for_each_dpa_resource(ndd, res)
 308                         if (strcmp(res->name, label_id->id) == 0)
 309                                 last = res;
 310                 res = last;
 311                 if (!res)
 312                         return 0;
 313
 314                 if (n >= resource_size(res)) {
 315                         n -= resource_size(res);
 316                         nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc);
 317                         nvdimm_free_dpa(ndd, res);
 318                         /* retry with last resource deleted */
 319                         continue;
 320                 }
 321
 322                 rc = adjust_resource(res, res->start, resource_size(res) - n);
 323                 if (rc == 0)
 324                         res->flags |= DPA_RESOURCE_ADJUSTED;
 325                 nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc);
 326                 break;
 327         }
 328
 329         return rc;
 330 }
 331
 332 /**
 333  * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
 334  * @nd_region: the set of dimms to reclaim @n bytes from
 335  * @label_id: unique identifier for the namespace consuming this dpa range
 336  * @n: number of bytes per-dimm to release
 337  *
 338  * Assumes resources are ordered.  Starting from the end try to
 339  * adjust_resource() the allocation to @n, but if @n is larger than the
 340  * allocation delete it and find the 'new' last allocation in the label
 341  * set.
 342  *
 343  * Returns: %0 on success on -errno on error
 344  */
 345 static int shrink_dpa_allocation(struct nd_region *nd_region,
 346                 struct nd_label_id *label_id, resource_size_t n)
 347 {
 348         int i;
 349
 350         for (i = 0; i < nd_region->ndr_mappings; i++) {
 351                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 352                 int rc;
 353
 354                 rc = scan_free(nd_region, nd_mapping, label_id, n);
 355                 if (rc)
 356                         return rc;
 357         }
 358
 359         return 0;
 360 }
 361
 362 static resource_size_t init_dpa_allocation(struct nd_label_id *label_id,
 363                 struct nd_region *nd_region, struct nd_mapping *nd_mapping,
 364                 resource_size_t n)
 365 {
 366         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 367         struct resource *res;
 368         int rc = 0;
 369
 370         /* first resource allocation for this label-id or dimm */
 371         res = nvdimm_allocate_dpa(ndd, label_id, nd_mapping->start, n);
 372         if (!res)
 373                 rc = -EBUSY;
 374
 375         nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc);
 376         return rc ? n : 0;
 377 }
 378
 379
 380 /**
 381  * space_valid() - validate free dpa space against constraints
 382  * @nd_region: hosting region of the free space
 383  * @ndd: dimm device data for debug
 384  * @label_id: namespace id to allocate space
 385  * @prev: potential allocation that precedes free space
 386  * @next: allocation that follows the given free space range
 387  * @exist: first allocation with same id in the mapping
 388  * @n: range that must satisfied for pmem allocations
 389  * @valid: free space range to validate
 390  *
 391  * BLK-space is valid as long as it does not precede a PMEM
 392  * allocation in a given region. PMEM-space must be contiguous
 393  * and adjacent to an existing allocation (if one
 394  * exists).  If reserving PMEM any space is valid.
 395  */
 396 static void space_valid(struct nd_region *nd_region, struct nvdimm_drvdata *ndd,
 397                 struct nd_label_id *label_id, struct resource *prev,
 398                 struct resource *next, struct resource *exist,
 399                 resource_size_t n, struct resource *valid)
 400 {
 401         bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0;
 402         unsigned long align;
 403
 404         align = nd_region->align / nd_region->ndr_mappings;
 405         valid->start = ALIGN(valid->start, align);
 406         valid->end = ALIGN_DOWN(valid->end + 1, align) - 1;
 407
 408         if (valid->start >= valid->end)
 409                 goto invalid;
 410
 411         if (is_reserve)
 412                 return;
 413
 414         /* allocation needs to be contiguous, so this is all or nothing */
 415         if (resource_size(valid) < n)
 416                 goto invalid;
 417
 418         /* we've got all the space we need and no existing allocation */
 419         if (!exist)
 420                 return;
 421
 422         /* allocation needs to be contiguous with the existing namespace */
 423         if (valid->start == exist->end + 1
 424                         || valid->end == exist->start - 1)
 425                 return;
 426
 427  invalid:
 428         /* truncate @valid size to 0 */
 429         valid->end = valid->start - 1;
 430 }
 431
 432 enum alloc_loc {
 433         ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
 434 };
 435
 436 static resource_size_t scan_allocate(struct nd_region *nd_region,
 437                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
 438                 resource_size_t n)
 439 {
 440         resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1;
 441         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 442         struct resource *res, *exist = NULL, valid;
 443         const resource_size_t to_allocate = n;
 444         int first;
 445
 446         for_each_dpa_resource(ndd, res)
 447                 if (strcmp(label_id->id, res->name) == 0)
 448                         exist = res;
 449
 450         valid.start = nd_mapping->start;
 451         valid.end = mapping_end;
 452         valid.name = "free space";
 453  retry:
 454         first = 0;
 455         for_each_dpa_resource(ndd, res) {
 456                 struct resource *next = res->sibling, *new_res = NULL;
 457                 resource_size_t allocate, available = 0;
 458                 enum alloc_loc loc = ALLOC_ERR;
 459                 const char *action;
 460                 int rc = 0;
 461
 462                 /* ignore resources outside this nd_mapping */
 463                 if (res->start > mapping_end)
 464                         continue;
 465                 if (res->end < nd_mapping->start)
 466                         continue;
 467
 468                 /* space at the beginning of the mapping */
 469                 if (!first++ && res->start > nd_mapping->start) {
 470                         valid.start = nd_mapping->start;
 471                         valid.end = res->start - 1;
 472                         space_valid(nd_region, ndd, label_id, NULL, next, exist,
 473                                         to_allocate, &valid);
 474                         available = resource_size(&valid);
 475                         if (available)
 476                                 loc = ALLOC_BEFORE;
 477                 }
 478
 479                 /* space between allocations */
 480                 if (!loc && next) {
 481                         valid.start = res->start + resource_size(res);
 482                         valid.end = min(mapping_end, next->start - 1);
 483                         space_valid(nd_region, ndd, label_id, res, next, exist,
 484                                         to_allocate, &valid);
 485                         available = resource_size(&valid);
 486                         if (available)
 487                                 loc = ALLOC_MID;
 488                 }
 489
 490                 /* space at the end of the mapping */
 491                 if (!loc && !next) {
 492                         valid.start = res->start + resource_size(res);
 493                         valid.end = mapping_end;
 494                         space_valid(nd_region, ndd, label_id, res, next, exist,
 495                                         to_allocate, &valid);
 496                         available = resource_size(&valid);
 497                         if (available)
 498                                 loc = ALLOC_AFTER;
 499                 }
 500
 501                 if (!loc || !available)
 502                         continue;
 503                 allocate = min(available, n);
 504                 switch (loc) {
 505                 case ALLOC_BEFORE:
 506                         if (strcmp(res->name, label_id->id) == 0) {
 507                                 /* adjust current resource up */
 508                                 rc = adjust_resource(res, res->start - allocate,
 509                                                 resource_size(res) + allocate);
 510                                 action = "cur grow up";
 511                         } else
 512                                 action = "allocate";
 513                         break;
 514                 case ALLOC_MID:
 515                         if (strcmp(next->name, label_id->id) == 0) {
 516                                 /* adjust next resource up */
 517                                 rc = adjust_resource(next, next->start
 518                                                 - allocate, resource_size(next)
 519                                                 + allocate);
 520                                 new_res = next;
 521                                 action = "next grow up";
 522                         } else if (strcmp(res->name, label_id->id) == 0) {
 523                                 action = "grow down";
 524                         } else
 525                                 action = "allocate";
 526                         break;
 527                 case ALLOC_AFTER:
 528                         if (strcmp(res->name, label_id->id) == 0)
 529                                 action = "grow down";
 530                         else
 531                                 action = "allocate";
 532                         break;
 533                 default:
 534                         return n;
 535                 }
 536
 537                 if (strcmp(action, "allocate") == 0) {
 538                         new_res = nvdimm_allocate_dpa(ndd, label_id,
 539                                         valid.start, allocate);
 540                         if (!new_res)
 541                                 rc = -EBUSY;
 542                 } else if (strcmp(action, "grow down") == 0) {
 543                         /* adjust current resource down */
 544                         rc = adjust_resource(res, res->start, resource_size(res)
 545                                         + allocate);
 546                         if (rc == 0)
 547                                 res->flags |= DPA_RESOURCE_ADJUSTED;
 548                 }
 549
 550                 if (!new_res)
 551                         new_res = res;
 552
 553                 nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n",
 554                                 action, loc, rc);
 555
 556                 if (rc)
 557                         return n;
 558
 559                 n -= allocate;
 560                 if (n) {
 561                         /*
 562                          * Retry scan with newly inserted resources.
 563                          * For example, if we did an ALLOC_BEFORE
 564                          * insertion there may also have been space
 565                          * available for an ALLOC_AFTER insertion, so we
 566                          * need to check this same resource again
 567                          */
 568                         goto retry;
 569                 } else
 570                         return 0;
 571         }
 572
 573         if (n == to_allocate)
 574                 return init_dpa_allocation(label_id, nd_region, nd_mapping, n);
 575         return n;
 576 }
 577
 578 static int merge_dpa(struct nd_region *nd_region,
 579                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id)
 580 {
 581         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 582         struct resource *res;
 583
 584         if (strncmp("pmem", label_id->id, 4) == 0)
 585                 return 0;
 586  retry:
 587         for_each_dpa_resource(ndd, res) {
 588                 int rc;
 589                 struct resource *next = res->sibling;
 590                 resource_size_t end = res->start + resource_size(res);
 591
 592                 if (!next || strcmp(res->name, label_id->id) != 0
 593                                 || strcmp(next->name, label_id->id) != 0
 594                                 || end != next->start)
 595                         continue;
 596                 end += resource_size(next);
 597                 nvdimm_free_dpa(ndd, next);
 598                 rc = adjust_resource(res, res->start, end - res->start);
 599                 nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc);
 600                 if (rc)
 601                         return rc;
 602                 res->flags |= DPA_RESOURCE_ADJUSTED;
 603                 goto retry;
 604         }
 605
 606         return 0;
 607 }
 608
 609 int __reserve_free_pmem(struct device *dev, void *data)
 610 {
 611         struct nvdimm *nvdimm = data;
 612         struct nd_region *nd_region;
 613         struct nd_label_id label_id;
 614         int i;
 615
 616         if (!is_memory(dev))
 617                 return 0;
 618
 619         nd_region = to_nd_region(dev);
 620         if (nd_region->ndr_mappings == 0)
 621                 return 0;
 622
 623         memset(&label_id, 0, sizeof(label_id));
 624         strcat(label_id.id, "pmem-reserve");
 625         for (i = 0; i < nd_region->ndr_mappings; i++) {
 626                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 627                 resource_size_t n, rem = 0;
 628
 629                 if (nd_mapping->nvdimm != nvdimm)
 630                         continue;
 631
 632                 n = nd_pmem_available_dpa(nd_region, nd_mapping);
 633                 if (n == 0)
 634                         return 0;
 635                 rem = scan_allocate(nd_region, nd_mapping, &label_id, n);
 636                 dev_WARN_ONCE(&nd_region->dev, rem,
 637                                 "pmem reserve underrun: %#llx of %#llx bytes\n",
 638                                 (unsigned long long) n - rem,
 639                                 (unsigned long long) n);
 640                 return rem ? -ENXIO : 0;
 641         }
 642
 643         return 0;
 644 }
 645
 646 void release_free_pmem(struct nvdimm_bus *nvdimm_bus,
 647                 struct nd_mapping *nd_mapping)
 648 {
 649         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 650         struct resource *res, *_res;
 651
 652         for_each_dpa_resource_safe(ndd, res, _res)
 653                 if (strcmp(res->name, "pmem-reserve") == 0)
 654                         nvdimm_free_dpa(ndd, res);
 655 }
 656
 657 /**
 658  * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
 659  * @nd_region: the set of dimms to allocate @n more bytes from
 660  * @label_id: unique identifier for the namespace consuming this dpa range
 661  * @n: number of bytes per-dimm to add to the existing allocation
 662  *
 663  * Assumes resources are ordered.  For BLK regions, first consume
 664  * BLK-only available DPA free space, then consume PMEM-aliased DPA
 665  * space starting at the highest DPA.  For PMEM regions start
 666  * allocations from the start of an interleave set and end at the first
 667  * BLK allocation or the end of the interleave set, whichever comes
 668  * first.
 669  *
 670  * Returns: %0 on success on -errno on error
 671  */
 672 static int grow_dpa_allocation(struct nd_region *nd_region,
 673                 struct nd_label_id *label_id, resource_size_t n)
 674 {
 675         int i;
 676
 677         for (i = 0; i < nd_region->ndr_mappings; i++) {
 678                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 679                 resource_size_t rem = n;
 680                 int rc;
 681
 682                 rem = scan_allocate(nd_region, nd_mapping, label_id, rem);
 683                 dev_WARN_ONCE(&nd_region->dev, rem,
 684                                 "allocation underrun: %#llx of %#llx bytes\n",
 685                                 (unsigned long long) n - rem,
 686                                 (unsigned long long) n);
 687                 if (rem)
 688                         return -ENXIO;
 689
 690                 rc = merge_dpa(nd_region, nd_mapping, label_id);
 691                 if (rc)
 692                         return rc;
 693         }
 694
 695         return 0;
 696 }
 697
 698 static void nd_namespace_pmem_set_resource(struct nd_region *nd_region,
 699                 struct nd_namespace_pmem *nspm, resource_size_t size)
 700 {
 701         struct resource *res = &nspm->nsio.res;
 702         resource_size_t offset = 0;
 703
 704         if (size && !nspm->uuid) {
 705                 WARN_ON_ONCE(1);
 706                 size = 0;
 707         }
 708
 709         if (size && nspm->uuid) {
 710                 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 711                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 712                 struct nd_label_id label_id;
 713                 struct resource *res;
 714
 715                 if (!ndd) {
 716                         size = 0;
 717                         goto out;
 718                 }
 719
 720                 nd_label_gen_id(&label_id, nspm->uuid, 0);
 721
 722                 /* calculate a spa offset from the dpa allocation offset */
 723                 for_each_dpa_resource(ndd, res)
 724                         if (strcmp(res->name, label_id.id) == 0) {
 725                                 offset = (res->start - nd_mapping->start)
 726                                         * nd_region->ndr_mappings;
 727                                 goto out;
 728                         }
 729
 730                 WARN_ON_ONCE(1);
 731                 size = 0;
 732         }
 733
 734  out:
 735         res->start = nd_region->ndr_start + offset;
 736         res->end = res->start + size - 1;
 737 }
 738
 739 static bool uuid_not_set(const uuid_t *uuid, struct device *dev,
 740                          const char *where)
 741 {
 742         if (!uuid) {
 743                 dev_dbg(dev, "%s: uuid not set\n", where);
 744                 return true;
 745         }
 746         return false;
 747 }
 748
 749 static ssize_t __size_store(struct device *dev, unsigned long long val)
 750 {
 751         resource_size_t allocated = 0, available = 0;
 752         struct nd_region *nd_region = to_nd_region(dev->parent);
 753         struct nd_namespace_common *ndns = to_ndns(dev);
 754         struct nd_mapping *nd_mapping;
 755         struct nvdimm_drvdata *ndd;
 756         struct nd_label_id label_id;
 757         u32 flags = 0, remainder;
 758         int rc, i, id = -1;
 759         uuid_t *uuid = NULL;
 760
 761         if (dev->driver || ndns->claim)
 762                 return -EBUSY;
 763
 764         if (is_namespace_pmem(dev)) {
 765                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 766
 767                 uuid = nspm->uuid;
 768                 id = nspm->id;
 769         }
 770
 771         /*
 772          * We need a uuid for the allocation-label and dimm(s) on which
 773          * to store the label.
 774          */
 775         if (uuid_not_set(uuid, dev, __func__))
 776                 return -ENXIO;
 777         if (nd_region->ndr_mappings == 0) {
 778                 dev_dbg(dev, "not associated with dimm(s)\n");
 779                 return -ENXIO;
 780         }
 781
 782         div_u64_rem(val, nd_region->align, &remainder);
 783         if (remainder) {
 784                 dev_dbg(dev, "%llu is not %ldK aligned\n", val,
 785                                 nd_region->align / SZ_1K);
 786                 return -EINVAL;
 787         }
 788
 789         nd_label_gen_id(&label_id, uuid, flags);
 790         for (i = 0; i < nd_region->ndr_mappings; i++) {
 791                 nd_mapping = &nd_region->mapping[i];
 792                 ndd = to_ndd(nd_mapping);
 793
 794                 /*
 795                  * All dimms in an interleave set, need to be enabled
 796                  * for the size to be changed.
 797                  */
 798                 if (!ndd)
 799                         return -ENXIO;
 800
 801                 allocated += nvdimm_allocated_dpa(ndd, &label_id);
 802         }
 803         available = nd_region_allocatable_dpa(nd_region);
 804
 805         if (val > available + allocated)
 806                 return -ENOSPC;
 807
 808         if (val == allocated)
 809                 return 0;
 810
 811         val = div_u64(val, nd_region->ndr_mappings);
 812         allocated = div_u64(allocated, nd_region->ndr_mappings);
 813         if (val < allocated)
 814                 rc = shrink_dpa_allocation(nd_region, &label_id,
 815                                 allocated - val);
 816         else
 817                 rc = grow_dpa_allocation(nd_region, &label_id, val - allocated);
 818
 819         if (rc)
 820                 return rc;
 821
 822         if (is_namespace_pmem(dev)) {
 823                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 824
 825                 nd_namespace_pmem_set_resource(nd_region, nspm,
 826                                 val * nd_region->ndr_mappings);
 827         }
 828
 829         /*
 830          * Try to delete the namespace if we deleted all of its
 831          * allocation, this is not the seed or 0th device for the
 832          * region, and it is not actively claimed by a btt, pfn, or dax
 833          * instance.
 834          */
 835         if (val == 0 && id != 0 && nd_region->ns_seed != dev && !ndns->claim)
 836                 nd_device_unregister(dev, ND_ASYNC);
 837
 838         return rc;
 839 }
 840
 841 static ssize_t size_store(struct device *dev,
 842                 struct device_attribute *attr, const char *buf, size_t len)
 843 {
 844         struct nd_region *nd_region = to_nd_region(dev->parent);
 845         unsigned long long val;
 846         int rc;
 847
 848         rc = kstrtoull(buf, 0, &val);
 849         if (rc)
 850                 return rc;
 851
 852         device_lock(dev);
 853         nvdimm_bus_lock(dev);
 854         wait_nvdimm_bus_probe_idle(dev);
 855         rc = __size_store(dev, val);
 856         if (rc >= 0)
 857                 rc = nd_namespace_label_update(nd_region, dev);
 858
 859         /* setting size zero == 'delete namespace' */
 860         if (rc == 0 && val == 0 && is_namespace_pmem(dev)) {
 861                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 862
 863                 kfree(nspm->uuid);
 864                 nspm->uuid = NULL;
 865         }
 866
 867         dev_dbg(dev, "%llx %s (%d)\n", val, rc < 0 ? "fail" : "success", rc);
 868
 869         nvdimm_bus_unlock(dev);
 870         device_unlock(dev);
 871
 872         return rc < 0 ? rc : len;
 873 }
 874
 875 resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
 876 {
 877         struct device *dev = &ndns->dev;
 878
 879         if (is_namespace_pmem(dev)) {
 880                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 881
 882                 return resource_size(&nspm->nsio.res);
 883         } else if (is_namespace_io(dev)) {
 884                 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
 885
 886                 return resource_size(&nsio->res);
 887         } else
 888                 WARN_ONCE(1, "unknown namespace type\n");
 889         return 0;
 890 }
 891
 892 resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
 893 {
 894         resource_size_t size;
 895
 896         nvdimm_bus_lock(&ndns->dev);
 897         size = __nvdimm_namespace_capacity(ndns);
 898         nvdimm_bus_unlock(&ndns->dev);
 899
 900         return size;
 901 }
 902 EXPORT_SYMBOL(nvdimm_namespace_capacity);
 903
 904 bool nvdimm_namespace_locked(struct nd_namespace_common *ndns)
 905 {
 906         int i;
 907         bool locked = false;
 908         struct device *dev = &ndns->dev;
 909         struct nd_region *nd_region = to_nd_region(dev->parent);
 910
 911         for (i = 0; i < nd_region->ndr_mappings; i++) {
 912                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 913                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
 914
 915                 if (test_bit(NDD_LOCKED, &nvdimm->flags)) {
 916                         dev_dbg(dev, "%s locked\n", nvdimm_name(nvdimm));
 917                         locked = true;
 918                 }
 919         }
 920         return locked;
 921 }
 922 EXPORT_SYMBOL(nvdimm_namespace_locked);
 923
 924 static ssize_t size_show(struct device *dev,
 925                 struct device_attribute *attr, char *buf)
 926 {
 927         return sprintf(buf, "%llu\n", (unsigned long long)
 928                         nvdimm_namespace_capacity(to_ndns(dev)));
 929 }
 930 static DEVICE_ATTR(size, 0444, size_show, size_store);
 931
 932 static uuid_t *namespace_to_uuid(struct device *dev)
 933 {
 934         if (is_namespace_pmem(dev)) {
 935                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 936
 937                 return nspm->uuid;
 938         }
 939         return ERR_PTR(-ENXIO);
 940 }
 941
 942 static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
 943                          char *buf)
 944 {
 945         uuid_t *uuid = namespace_to_uuid(dev);
 946
 947         if (IS_ERR(uuid))
 948                 return PTR_ERR(uuid);
 949         if (uuid)
 950                 return sprintf(buf, "%pUb\n", uuid);
 951         return sprintf(buf, "\n");
 952 }
 953
 954 /**
 955  * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
 956  * @nd_region: parent region so we can updates all dimms in the set
 957  * @dev: namespace type for generating label_id
 958  * @new_uuid: incoming uuid
 959  * @old_uuid: reference to the uuid storage location in the namespace object
 960  *
 961  * Returns: %0 on success on -errno on error
 962  */
 963 static int namespace_update_uuid(struct nd_region *nd_region,
 964                                  struct device *dev, uuid_t *new_uuid,
 965                                  uuid_t **old_uuid)
 966 {
 967         struct nd_label_id old_label_id;
 968         struct nd_label_id new_label_id;
 969         int i;
 970
 971         if (!nd_is_uuid_unique(dev, new_uuid))
 972                 return -EINVAL;
 973
 974         if (*old_uuid == NULL)
 975                 goto out;
 976
 977         /*
 978          * If we've already written a label with this uuid, then it's
 979          * too late to rename because we can't reliably update the uuid
 980          * without losing the old namespace.  Userspace must delete this
 981          * namespace to abandon the old uuid.
 982          */
 983         for (i = 0; i < nd_region->ndr_mappings; i++) {
 984                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 985
 986                 /*
 987                  * This check by itself is sufficient because old_uuid
 988                  * would be NULL above if this uuid did not exist in the
 989                  * currently written set.
 990                  *
 991                  * FIXME: can we delete uuid with zero dpa allocated?
 992                  */
 993                 if (list_empty(&nd_mapping->labels))
 994                         return -EBUSY;
 995         }
 996
 997         nd_label_gen_id(&old_label_id, *old_uuid, 0);
 998         nd_label_gen_id(&new_label_id, new_uuid, 0);
 999         for (i = 0; i < nd_region->ndr_mappings; i++) {
1000                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1001                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1002                 struct nd_label_ent *label_ent;
1003                 struct resource *res;
1004
1005                 for_each_dpa_resource(ndd, res)
1006                         if (strcmp(res->name, old_label_id.id) == 0)
1007                                 sprintf((void *) res->name, "%s",
1008                                                 new_label_id.id);
1009
1010                 mutex_lock(&nd_mapping->lock);
1011                 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1012                         struct nd_namespace_label *nd_label = label_ent->label;
1013                         struct nd_label_id label_id;
1014                         uuid_t uuid;
1015
1016                         if (!nd_label)
1017                                 continue;
1018                         nsl_get_uuid(ndd, nd_label, &uuid);
1019                         nd_label_gen_id(&label_id, &uuid,
1020                                         nsl_get_flags(ndd, nd_label));
1021                         if (strcmp(old_label_id.id, label_id.id) == 0)
1022                                 set_bit(ND_LABEL_REAP, &label_ent->flags);
1023                 }
1024                 mutex_unlock(&nd_mapping->lock);
1025         }
1026         kfree(*old_uuid);
1027  out:
1028         *old_uuid = new_uuid;
1029         return 0;
1030 }
1031
1032 static ssize_t uuid_store(struct device *dev,
1033                 struct device_attribute *attr, const char *buf, size_t len)
1034 {
1035         struct nd_region *nd_region = to_nd_region(dev->parent);
1036         uuid_t *uuid = NULL;
1037         uuid_t **ns_uuid;
1038         ssize_t rc = 0;
1039
1040         if (is_namespace_pmem(dev)) {
1041                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1042
1043                 ns_uuid = &nspm->uuid;
1044         } else
1045                 return -ENXIO;
1046
1047         device_lock(dev);
1048         nvdimm_bus_lock(dev);
1049         wait_nvdimm_bus_probe_idle(dev);
1050         if (to_ndns(dev)->claim)
1051                 rc = -EBUSY;
1052         if (rc >= 0)
1053                 rc = nd_uuid_store(dev, &uuid, buf, len);
1054         if (rc >= 0)
1055                 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
1056         if (rc >= 0)
1057                 rc = nd_namespace_label_update(nd_region, dev);
1058         else
1059                 kfree(uuid);
1060         dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
1061                         buf[len - 1] == '\n' ? "" : "\n");
1062         nvdimm_bus_unlock(dev);
1063         device_unlock(dev);
1064
1065         return rc < 0 ? rc : len;
1066 }
1067 static DEVICE_ATTR_RW(uuid);
1068
1069 static ssize_t resource_show(struct device *dev,
1070                 struct device_attribute *attr, char *buf)
1071 {
1072         struct resource *res;
1073
1074         if (is_namespace_pmem(dev)) {
1075                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1076
1077                 res = &nspm->nsio.res;
1078         } else if (is_namespace_io(dev)) {
1079                 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1080
1081                 res = &nsio->res;
1082         } else
1083                 return -ENXIO;
1084
1085         /* no address to convey if the namespace has no allocation */
1086         if (resource_size(res) == 0)
1087                 return -ENXIO;
1088         return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
1089 }
1090 static DEVICE_ATTR_ADMIN_RO(resource);
1091
1092 static const unsigned long pmem_lbasize_supported[] = { 512, 4096, 0 };
1093
1094 static ssize_t sector_size_show(struct device *dev,
1095                 struct device_attribute *attr, char *buf)
1096 {
1097         if (is_namespace_pmem(dev)) {
1098                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1099
1100                 return nd_size_select_show(nspm->lbasize,
1101                                 pmem_lbasize_supported, buf);
1102         }
1103         return -ENXIO;
1104 }
1105
1106 static ssize_t sector_size_store(struct device *dev,
1107                 struct device_attribute *attr, const char *buf, size_t len)
1108 {
1109         struct nd_region *nd_region = to_nd_region(dev->parent);
1110         const unsigned long *supported;
1111         unsigned long *lbasize;
1112         ssize_t rc = 0;
1113
1114         if (is_namespace_pmem(dev)) {
1115                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1116
1117                 lbasize = &nspm->lbasize;
1118                 supported = pmem_lbasize_supported;
1119         } else
1120                 return -ENXIO;
1121
1122         device_lock(dev);
1123         nvdimm_bus_lock(dev);
1124         if (to_ndns(dev)->claim)
1125                 rc = -EBUSY;
1126         if (rc >= 0)
1127                 rc = nd_size_select_store(dev, buf, lbasize, supported);
1128         if (rc >= 0)
1129                 rc = nd_namespace_label_update(nd_region, dev);
1130         dev_dbg(dev, "result: %zd %s: %s%s", rc, rc < 0 ? "tried" : "wrote",
1131                         buf, buf[len - 1] == '\n' ? "" : "\n");
1132         nvdimm_bus_unlock(dev);
1133         device_unlock(dev);
1134
1135         return rc ? rc : len;
1136 }
1137 static DEVICE_ATTR_RW(sector_size);
1138
1139 static ssize_t dpa_extents_show(struct device *dev,
1140                 struct device_attribute *attr, char *buf)
1141 {
1142         struct nd_region *nd_region = to_nd_region(dev->parent);
1143         struct nd_label_id label_id;
1144         uuid_t *uuid = NULL;
1145         int count = 0, i;
1146         u32 flags = 0;
1147
1148         nvdimm_bus_lock(dev);
1149         if (is_namespace_pmem(dev)) {
1150                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1151
1152                 uuid = nspm->uuid;
1153                 flags = 0;
1154         }
1155
1156         if (!uuid)
1157                 goto out;
1158
1159         nd_label_gen_id(&label_id, uuid, flags);
1160         for (i = 0; i < nd_region->ndr_mappings; i++) {
1161                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1162                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1163                 struct resource *res;
1164
1165                 for_each_dpa_resource(ndd, res)
1166                         if (strcmp(res->name, label_id.id) == 0)
1167                                 count++;
1168         }
1169  out:
1170         nvdimm_bus_unlock(dev);
1171
1172         return sprintf(buf, "%d\n", count);
1173 }
1174 static DEVICE_ATTR_RO(dpa_extents);
1175
1176 static int btt_claim_class(struct device *dev)
1177 {
1178         struct nd_region *nd_region = to_nd_region(dev->parent);
1179         int i, loop_bitmask = 0;
1180
1181         for (i = 0; i < nd_region->ndr_mappings; i++) {
1182                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1183                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1184                 struct nd_namespace_index *nsindex;
1185
1186                 /*
1187                  * If any of the DIMMs do not support labels the only
1188                  * possible BTT format is v1.
1189                  */
1190                 if (!ndd) {
1191                         loop_bitmask = 0;
1192                         break;
1193                 }
1194
1195                 nsindex = to_namespace_index(ndd, ndd->ns_current);
1196                 if (nsindex == NULL)
1197                         loop_bitmask |= 1;
1198                 else {
1199                         /* check whether existing labels are v1.1 or v1.2 */
1200                         if (__le16_to_cpu(nsindex->major) == 1
1201                                         && __le16_to_cpu(nsindex->minor) == 1)
1202                                 loop_bitmask |= 2;
1203                         else
1204                                 loop_bitmask |= 4;
1205                 }
1206         }
1207         /*
1208          * If nsindex is null loop_bitmask's bit 0 will be set, and if an index
1209          * block is found, a v1.1 label for any mapping will set bit 1, and a
1210          * v1.2 label will set bit 2.
1211          *
1212          * At the end of the loop, at most one of the three bits must be set.
1213          * If multiple bits were set, it means the different mappings disagree
1214          * about their labels, and this must be cleaned up first.
1215          *
1216          * If all the label index blocks are found to agree, nsindex of NULL
1217          * implies labels haven't been initialized yet, and when they will,
1218          * they will be of the 1.2 format, so we can assume BTT2.0
1219          *
1220          * If 1.1 labels are found, we enforce BTT1.1, and if 1.2 labels are
1221          * found, we enforce BTT2.0
1222          *
1223          * If the loop was never entered, default to BTT1.1 (legacy namespaces)
1224          */
1225         switch (loop_bitmask) {
1226         case 0:
1227         case 2:
1228                 return NVDIMM_CCLASS_BTT;
1229         case 1:
1230         case 4:
1231                 return NVDIMM_CCLASS_BTT2;
1232         default:
1233                 return -ENXIO;
1234         }
1235 }
1236
1237 static ssize_t holder_show(struct device *dev,
1238                 struct device_attribute *attr, char *buf)
1239 {
1240         struct nd_namespace_common *ndns = to_ndns(dev);
1241         ssize_t rc;
1242
1243         device_lock(dev);
1244         rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
1245         device_unlock(dev);
1246
1247         return rc;
1248 }
1249 static DEVICE_ATTR_RO(holder);
1250
1251 static int __holder_class_store(struct device *dev, const char *buf)
1252 {
1253         struct nd_namespace_common *ndns = to_ndns(dev);
1254
1255         if (dev->driver || ndns->claim)
1256                 return -EBUSY;
1257
1258         if (sysfs_streq(buf, "btt")) {
1259                 int rc = btt_claim_class(dev);
1260
1261                 if (rc < NVDIMM_CCLASS_NONE)
1262                         return rc;
1263                 ndns->claim_class = rc;
1264         } else if (sysfs_streq(buf, "pfn"))
1265                 ndns->claim_class = NVDIMM_CCLASS_PFN;
1266         else if (sysfs_streq(buf, "dax"))
1267                 ndns->claim_class = NVDIMM_CCLASS_DAX;
1268         else if (sysfs_streq(buf, ""))
1269                 ndns->claim_class = NVDIMM_CCLASS_NONE;
1270         else
1271                 return -EINVAL;
1272
1273         return 0;
1274 }
1275
1276 static ssize_t holder_class_store(struct device *dev,
1277                 struct device_attribute *attr, const char *buf, size_t len)
1278 {
1279         struct nd_region *nd_region = to_nd_region(dev->parent);
1280         int rc;
1281
1282         device_lock(dev);
1283         nvdimm_bus_lock(dev);
1284         wait_nvdimm_bus_probe_idle(dev);
1285         rc = __holder_class_store(dev, buf);
1286         if (rc >= 0)
1287                 rc = nd_namespace_label_update(nd_region, dev);
1288         dev_dbg(dev, "%s(%d)\n", rc < 0 ? "fail " : "", rc);
1289         nvdimm_bus_unlock(dev);
1290         device_unlock(dev);
1291
1292         return rc < 0 ? rc : len;
1293 }
1294
1295 static ssize_t holder_class_show(struct device *dev,
1296                 struct device_attribute *attr, char *buf)
1297 {
1298         struct nd_namespace_common *ndns = to_ndns(dev);
1299         ssize_t rc;
1300
1301         device_lock(dev);
1302         if (ndns->claim_class == NVDIMM_CCLASS_NONE)
1303                 rc = sprintf(buf, "\n");
1304         else if ((ndns->claim_class == NVDIMM_CCLASS_BTT) ||
1305                         (ndns->claim_class == NVDIMM_CCLASS_BTT2))
1306                 rc = sprintf(buf, "btt\n");
1307         else if (ndns->claim_class == NVDIMM_CCLASS_PFN)
1308                 rc = sprintf(buf, "pfn\n");
1309         else if (ndns->claim_class == NVDIMM_CCLASS_DAX)
1310                 rc = sprintf(buf, "dax\n");
1311         else
1312                 rc = sprintf(buf, "<unknown>\n");
1313         device_unlock(dev);
1314
1315         return rc;
1316 }
1317 static DEVICE_ATTR_RW(holder_class);
1318
1319 static ssize_t mode_show(struct device *dev,
1320                 struct device_attribute *attr, char *buf)
1321 {
1322         struct nd_namespace_common *ndns = to_ndns(dev);
1323         struct device *claim;
1324         char *mode;
1325         ssize_t rc;
1326
1327         device_lock(dev);
1328         claim = ndns->claim;
1329         if (claim && is_nd_btt(claim))
1330                 mode = "safe";
1331         else if (claim && is_nd_pfn(claim))
1332                 mode = "memory";
1333         else if (claim && is_nd_dax(claim))
1334                 mode = "dax";
1335         else if (!claim && pmem_should_map_pages(dev))
1336                 mode = "memory";
1337         else
1338                 mode = "raw";
1339         rc = sprintf(buf, "%s\n", mode);
1340         device_unlock(dev);
1341
1342         return rc;
1343 }
1344 static DEVICE_ATTR_RO(mode);
1345
1346 static ssize_t force_raw_store(struct device *dev,
1347                 struct device_attribute *attr, const char *buf, size_t len)
1348 {
1349         bool force_raw;
1350         int rc = kstrtobool(buf, &force_raw);
1351
1352         if (rc)
1353                 return rc;
1354
1355         to_ndns(dev)->force_raw = force_raw;
1356         return len;
1357 }
1358
1359 static ssize_t force_raw_show(struct device *dev,
1360                 struct device_attribute *attr, char *buf)
1361 {
1362         return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
1363 }
1364 static DEVICE_ATTR_RW(force_raw);
1365
1366 static struct attribute *nd_namespace_attributes[] = {
1367         &dev_attr_nstype.attr,
1368         &dev_attr_size.attr,
1369         &dev_attr_mode.attr,
1370         &dev_attr_uuid.attr,
1371         &dev_attr_holder.attr,
1372         &dev_attr_resource.attr,
1373         &dev_attr_alt_name.attr,
1374         &dev_attr_force_raw.attr,
1375         &dev_attr_sector_size.attr,
1376         &dev_attr_dpa_extents.attr,
1377         &dev_attr_holder_class.attr,
1378         NULL,
1379 };
1380
1381 static umode_t namespace_visible(struct kobject *kobj,
1382                 struct attribute *a, int n)
1383 {
1384         struct device *dev = container_of(kobj, struct device, kobj);
1385
1386         if (is_namespace_pmem(dev)) {
1387                 if (a == &dev_attr_size.attr)
1388                         return 0644;
1389
1390                 return a->mode;
1391         }
1392
1393         /* base is_namespace_io() attributes */
1394         if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr ||
1395             a == &dev_attr_holder.attr || a == &dev_attr_holder_class.attr ||
1396             a == &dev_attr_force_raw.attr || a == &dev_attr_mode.attr ||
1397             a == &dev_attr_resource.attr)
1398                 return a->mode;
1399
1400         return 0;
1401 }
1402
1403 static struct attribute_group nd_namespace_attribute_group = {
1404         .attrs = nd_namespace_attributes,
1405         .is_visible = namespace_visible,
1406 };
1407
1408 static const struct attribute_group *nd_namespace_attribute_groups[] = {
1409         &nd_device_attribute_group,
1410         &nd_namespace_attribute_group,
1411         &nd_numa_attribute_group,
1412         NULL,
1413 };
1414
1415 static const struct device_type namespace_io_device_type = {
1416         .name = "nd_namespace_io",
1417         .release = namespace_io_release,
1418         .groups = nd_namespace_attribute_groups,
1419 };
1420
1421 static const struct device_type namespace_pmem_device_type = {
1422         .name = "nd_namespace_pmem",
1423         .release = namespace_pmem_release,
1424         .groups = nd_namespace_attribute_groups,
1425 };
1426
1427 static bool is_namespace_pmem(const struct device *dev)
1428 {
1429         return dev ? dev->type == &namespace_pmem_device_type : false;
1430 }
1431
1432 static bool is_namespace_io(const struct device *dev)
1433 {
1434         return dev ? dev->type == &namespace_io_device_type : false;
1435 }
1436
1437 struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
1438 {
1439         struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
1440         struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
1441         struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
1442         struct nd_namespace_common *ndns = NULL;
1443         resource_size_t size;
1444
1445         if (nd_btt || nd_pfn || nd_dax) {
1446                 if (nd_btt)
1447                         ndns = nd_btt->ndns;
1448                 else if (nd_pfn)
1449                         ndns = nd_pfn->ndns;
1450                 else if (nd_dax)
1451                         ndns = nd_dax->nd_pfn.ndns;
1452
1453                 if (!ndns)
1454                         return ERR_PTR(-ENODEV);
1455
1456                 /*
1457                  * Flush any in-progess probes / removals in the driver
1458                  * for the raw personality of this namespace.
1459                  */
1460                 device_lock(&ndns->dev);
1461                 device_unlock(&ndns->dev);
1462                 if (ndns->dev.driver) {
1463                         dev_dbg(&ndns->dev, "is active, can't bind %s\n",
1464                                         dev_name(dev));
1465                         return ERR_PTR(-EBUSY);
1466                 }
1467                 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != dev,
1468                                         "host (%s) vs claim (%s) mismatch\n",
1469                                         dev_name(dev),
1470                                         dev_name(ndns->claim)))
1471                         return ERR_PTR(-ENXIO);
1472         } else {
1473                 ndns = to_ndns(dev);
1474                 if (ndns->claim) {
1475                         dev_dbg(dev, "claimed by %s, failing probe\n",
1476                                 dev_name(ndns->claim));
1477
1478                         return ERR_PTR(-ENXIO);
1479                 }
1480         }
1481
1482         if (nvdimm_namespace_locked(ndns))
1483                 return ERR_PTR(-EACCES);
1484
1485         size = nvdimm_namespace_capacity(ndns);
1486         if (size < ND_MIN_NAMESPACE_SIZE) {
1487                 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
1488                                 &size, ND_MIN_NAMESPACE_SIZE);
1489                 return ERR_PTR(-ENODEV);
1490         }
1491
1492         /*
1493          * Note, alignment validation for fsdax and devdax mode
1494          * namespaces happens in nd_pfn_validate() where infoblock
1495          * padding parameters can be applied.
1496          */
1497         if (pmem_should_map_pages(dev)) {
1498                 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
1499                 struct resource *res = &nsio->res;
1500
1501                 if (!IS_ALIGNED(res->start | (res->end + 1),
1502                                         memremap_compat_align())) {
1503                         dev_err(&ndns->dev, "%pr misaligned, unable to map\n", res);
1504                         return ERR_PTR(-EOPNOTSUPP);
1505                 }
1506         }
1507
1508         if (is_namespace_pmem(&ndns->dev)) {
1509                 struct nd_namespace_pmem *nspm;
1510
1511                 nspm = to_nd_namespace_pmem(&ndns->dev);
1512                 if (uuid_not_set(nspm->uuid, &ndns->dev, __func__))
1513                         return ERR_PTR(-ENODEV);
1514         }
1515
1516         return ndns;
1517 }
1518 EXPORT_SYMBOL(nvdimm_namespace_common_probe);
1519
1520 int devm_namespace_enable(struct device *dev, struct nd_namespace_common *ndns,
1521                 resource_size_t size)
1522 {
1523         return devm_nsio_enable(dev, to_nd_namespace_io(&ndns->dev), size);
1524 }
1525 EXPORT_SYMBOL_GPL(devm_namespace_enable);
1526
1527 void devm_namespace_disable(struct device *dev, struct nd_namespace_common *ndns)
1528 {
1529         devm_nsio_disable(dev, to_nd_namespace_io(&ndns->dev));
1530 }
1531 EXPORT_SYMBOL_GPL(devm_namespace_disable);
1532
1533 static struct device **create_namespace_io(struct nd_region *nd_region)
1534 {
1535         struct nd_namespace_io *nsio;
1536         struct device *dev, **devs;
1537         struct resource *res;
1538
1539         nsio = kzalloc(sizeof(*nsio), GFP_KERNEL);
1540         if (!nsio)
1541                 return NULL;
1542
1543         devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1544         if (!devs) {
1545                 kfree(nsio);
1546                 return NULL;
1547         }
1548
1549         dev = &nsio->common.dev;
1550         dev->type = &namespace_io_device_type;
1551         dev->parent = &nd_region->dev;
1552         res = &nsio->res;
1553         res->name = dev_name(&nd_region->dev);
1554         res->flags = IORESOURCE_MEM;
1555         res->start = nd_region->ndr_start;
1556         res->end = res->start + nd_region->ndr_size - 1;
1557
1558         devs[0] = dev;
1559         return devs;
1560 }
1561
1562 static bool has_uuid_at_pos(struct nd_region *nd_region, const uuid_t *uuid,
1563                             u64 cookie, u16 pos)
1564 {
1565         struct nd_namespace_label *found = NULL;
1566         int i;
1567
1568         for (i = 0; i < nd_region->ndr_mappings; i++) {
1569                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1570                 struct nd_interleave_set *nd_set = nd_region->nd_set;
1571                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1572                 struct nd_label_ent *label_ent;
1573                 bool found_uuid = false;
1574
1575                 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1576                         struct nd_namespace_label *nd_label = label_ent->label;
1577                         u16 position;
1578
1579                         if (!nd_label)
1580                                 continue;
1581                         position = nsl_get_position(ndd, nd_label);
1582
1583                         if (!nsl_validate_isetcookie(ndd, nd_label, cookie))
1584                                 continue;
1585
1586                         if (!nsl_uuid_equal(ndd, nd_label, uuid))
1587                                 continue;
1588
1589                         if (!nsl_validate_type_guid(ndd, nd_label,
1590                                                     &nd_set->type_guid))
1591                                 continue;
1592
1593                         if (found_uuid) {
1594                                 dev_dbg(ndd->dev, "duplicate entry for uuid\n");
1595                                 return false;
1596                         }
1597                         found_uuid = true;
1598                         if (!nsl_validate_nlabel(nd_region, ndd, nd_label))
1599                                 continue;
1600                         if (position != pos)
1601                                 continue;
1602                         found = nd_label;
1603                         break;
1604                 }
1605                 if (found)
1606                         break;
1607         }
1608         return found != NULL;
1609 }
1610
1611 static int select_pmem_id(struct nd_region *nd_region, const uuid_t *pmem_id)
1612 {
1613         int i;
1614
1615         for (i = 0; i < nd_region->ndr_mappings; i++) {
1616                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1617                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1618                 struct nd_namespace_label *nd_label = NULL;
1619                 u64 hw_start, hw_end, pmem_start, pmem_end;
1620                 struct nd_label_ent *label_ent;
1621
1622                 lockdep_assert_held(&nd_mapping->lock);
1623                 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1624                         nd_label = label_ent->label;
1625                         if (!nd_label)
1626                                 continue;
1627                         if (nsl_uuid_equal(ndd, nd_label, pmem_id))
1628                                 break;
1629                         nd_label = NULL;
1630                 }
1631
1632                 if (!nd_label) {
1633                         WARN_ON(1);
1634                         return -EINVAL;
1635                 }
1636
1637                 /*
1638                  * Check that this label is compliant with the dpa
1639                  * range published in NFIT
1640                  */
1641                 hw_start = nd_mapping->start;
1642                 hw_end = hw_start + nd_mapping->size;
1643                 pmem_start = nsl_get_dpa(ndd, nd_label);
1644                 pmem_end = pmem_start + nsl_get_rawsize(ndd, nd_label);
1645                 if (pmem_start >= hw_start && pmem_start < hw_end
1646                                 && pmem_end <= hw_end && pmem_end > hw_start)
1647                         /* pass */;
1648                 else {
1649                         dev_dbg(&nd_region->dev, "%s invalid label for %pUb\n",
1650                                 dev_name(ndd->dev),
1651                                 nsl_uuid_raw(ndd, nd_label));
1652                         return -EINVAL;
1653                 }
1654
1655                 /* move recently validated label to the front of the list */
1656                 list_move(&label_ent->list, &nd_mapping->labels);
1657         }
1658         return 0;
1659 }
1660
1661 /**
1662  * create_namespace_pmem - validate interleave set labelling, retrieve label0
1663  * @nd_region: region with mappings to validate
1664  * @nd_mapping: container of dpa-resource-root + labels
1665  * @nd_label: target pmem namespace label to evaluate
1666  *
1667  * Returns: the created &struct device on success or ERR_PTR(-errno) on error
1668  */
1669 static struct device *create_namespace_pmem(struct nd_region *nd_region,
1670                                             struct nd_mapping *nd_mapping,
1671                                             struct nd_namespace_label *nd_label)
1672 {
1673         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1674         struct nd_namespace_index *nsindex =
1675                 to_namespace_index(ndd, ndd->ns_current);
1676         u64 cookie = nd_region_interleave_set_cookie(nd_region, nsindex);
1677         u64 altcookie = nd_region_interleave_set_altcookie(nd_region);
1678         struct nd_label_ent *label_ent;
1679         struct nd_namespace_pmem *nspm;
1680         resource_size_t size = 0;
1681         struct resource *res;
1682         struct device *dev;
1683         uuid_t uuid;
1684         int rc = 0;
1685         u16 i;
1686
1687         if (cookie == 0) {
1688                 dev_dbg(&nd_region->dev, "invalid interleave-set-cookie\n");
1689                 return ERR_PTR(-ENXIO);
1690         }
1691
1692         if (!nsl_validate_isetcookie(ndd, nd_label, cookie)) {
1693                 dev_dbg(&nd_region->dev, "invalid cookie in label: %pUb\n",
1694                         nsl_uuid_raw(ndd, nd_label));
1695                 if (!nsl_validate_isetcookie(ndd, nd_label, altcookie))
1696                         return ERR_PTR(-EAGAIN);
1697
1698                 dev_dbg(&nd_region->dev, "valid altcookie in label: %pUb\n",
1699                         nsl_uuid_raw(ndd, nd_label));
1700         }
1701
1702         nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1703         if (!nspm)
1704                 return ERR_PTR(-ENOMEM);
1705
1706         nspm->id = -1;
1707         dev = &nspm->nsio.common.dev;
1708         dev->type = &namespace_pmem_device_type;
1709         dev->parent = &nd_region->dev;
1710         res = &nspm->nsio.res;
1711         res->name = dev_name(&nd_region->dev);
1712         res->flags = IORESOURCE_MEM;
1713
1714         for (i = 0; i < nd_region->ndr_mappings; i++) {
1715                 nsl_get_uuid(ndd, nd_label, &uuid);
1716                 if (has_uuid_at_pos(nd_region, &uuid, cookie, i))
1717                         continue;
1718                 if (has_uuid_at_pos(nd_region, &uuid, altcookie, i))
1719                         continue;
1720                 break;
1721         }
1722
1723         if (i < nd_region->ndr_mappings) {
1724                 struct nvdimm *nvdimm = nd_region->mapping[i].nvdimm;
1725
1726                 /*
1727                  * Give up if we don't find an instance of a uuid at each
1728                  * position (from 0 to nd_region->ndr_mappings - 1), or if we
1729                  * find a dimm with two instances of the same uuid.
1730                  */
1731                 dev_err(&nd_region->dev, "%s missing label for %pUb\n",
1732                         nvdimm_name(nvdimm), nsl_uuid_raw(ndd, nd_label));
1733                 rc = -EINVAL;
1734                 goto err;
1735         }
1736
1737         /*
1738          * Fix up each mapping's 'labels' to have the validated pmem label for
1739          * that position at labels[0], and NULL at labels[1].  In the process,
1740          * check that the namespace aligns with interleave-set.
1741          */
1742         nsl_get_uuid(ndd, nd_label, &uuid);
1743         rc = select_pmem_id(nd_region, &uuid);
1744         if (rc)
1745                 goto err;
1746
1747         /* Calculate total size and populate namespace properties from label0 */
1748         for (i = 0; i < nd_region->ndr_mappings; i++) {
1749                 struct nd_namespace_label *label0;
1750                 struct nvdimm_drvdata *ndd;
1751
1752                 nd_mapping = &nd_region->mapping[i];
1753                 label_ent = list_first_entry_or_null(&nd_mapping->labels,
1754                                 typeof(*label_ent), list);
1755                 label0 = label_ent ? label_ent->label : NULL;
1756
1757                 if (!label0) {
1758                         WARN_ON(1);
1759                         continue;
1760                 }
1761
1762                 ndd = to_ndd(nd_mapping);
1763                 size += nsl_get_rawsize(ndd, label0);
1764                 if (nsl_get_position(ndd, label0) != 0)
1765                         continue;
1766                 WARN_ON(nspm->alt_name || nspm->uuid);
1767                 nspm->alt_name = kmemdup(nsl_ref_name(ndd, label0),
1768                                          NSLABEL_NAME_LEN, GFP_KERNEL);
1769                 nsl_get_uuid(ndd, label0, &uuid);
1770                 nspm->uuid = kmemdup(&uuid, sizeof(uuid_t), GFP_KERNEL);
1771                 nspm->lbasize = nsl_get_lbasize(ndd, label0);
1772                 nspm->nsio.common.claim_class =
1773                         nsl_get_claim_class(ndd, label0);
1774         }
1775
1776         if (!nspm->alt_name || !nspm->uuid) {
1777                 rc = -ENOMEM;
1778                 goto err;
1779         }
1780
1781         nd_namespace_pmem_set_resource(nd_region, nspm, size);
1782
1783         return dev;
1784  err:
1785         namespace_pmem_release(dev);
1786         switch (rc) {
1787         case -EINVAL:
1788                 dev_dbg(&nd_region->dev, "invalid label(s)\n");
1789                 break;
1790         default:
1791                 dev_dbg(&nd_region->dev, "unexpected err: %d\n", rc);
1792                 break;
1793         }
1794         return ERR_PTR(rc);
1795 }
1796
1797 static struct device *nd_namespace_pmem_create(struct nd_region *nd_region)
1798 {
1799         struct nd_namespace_pmem *nspm;
1800         struct resource *res;
1801         struct device *dev;
1802
1803         if (!is_memory(&nd_region->dev))
1804                 return NULL;
1805
1806         nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1807         if (!nspm)
1808                 return NULL;
1809
1810         dev = &nspm->nsio.common.dev;
1811         dev->type = &namespace_pmem_device_type;
1812         dev->parent = &nd_region->dev;
1813         res = &nspm->nsio.res;
1814         res->name = dev_name(&nd_region->dev);
1815         res->flags = IORESOURCE_MEM;
1816
1817         nspm->id = ida_alloc(&nd_region->ns_ida, GFP_KERNEL);
1818         if (nspm->id < 0) {
1819                 kfree(nspm);
1820                 return NULL;
1821         }
1822         dev_set_name(dev, "namespace%d.%d", nd_region->id, nspm->id);
1823         nd_namespace_pmem_set_resource(nd_region, nspm, 0);
1824
1825         return dev;
1826 }
1827
1828 static struct lock_class_key nvdimm_namespace_key;
1829
1830 void nd_region_create_ns_seed(struct nd_region *nd_region)
1831 {
1832         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1833
1834         if (nd_region_to_nstype(nd_region) == ND_DEVICE_NAMESPACE_IO)
1835                 return;
1836
1837         nd_region->ns_seed = nd_namespace_pmem_create(nd_region);
1838
1839         /*
1840          * Seed creation failures are not fatal, provisioning is simply
1841          * disabled until memory becomes available
1842          */
1843         if (!nd_region->ns_seed)
1844                 dev_err(&nd_region->dev, "failed to create namespace\n");
1845         else {
1846                 device_initialize(nd_region->ns_seed);
1847                 lockdep_set_class(&nd_region->ns_seed->mutex,
1848                                   &nvdimm_namespace_key);
1849                 nd_device_register(nd_region->ns_seed);
1850         }
1851 }
1852
1853 void nd_region_create_dax_seed(struct nd_region *nd_region)
1854 {
1855         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1856         nd_region->dax_seed = nd_dax_create(nd_region);
1857         /*
1858          * Seed creation failures are not fatal, provisioning is simply
1859          * disabled until memory becomes available
1860          */
1861         if (!nd_region->dax_seed)
1862                 dev_err(&nd_region->dev, "failed to create dax namespace\n");
1863 }
1864
1865 void nd_region_create_pfn_seed(struct nd_region *nd_region)
1866 {
1867         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1868         nd_region->pfn_seed = nd_pfn_create(nd_region);
1869         /*
1870          * Seed creation failures are not fatal, provisioning is simply
1871          * disabled until memory becomes available
1872          */
1873         if (!nd_region->pfn_seed)
1874                 dev_err(&nd_region->dev, "failed to create pfn namespace\n");
1875 }
1876
1877 void nd_region_create_btt_seed(struct nd_region *nd_region)
1878 {
1879         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1880         nd_region->btt_seed = nd_btt_create(nd_region);
1881         /*
1882          * Seed creation failures are not fatal, provisioning is simply
1883          * disabled until memory becomes available
1884          */
1885         if (!nd_region->btt_seed)
1886                 dev_err(&nd_region->dev, "failed to create btt namespace\n");
1887 }
1888
1889 static int add_namespace_resource(struct nd_region *nd_region,
1890                 struct nd_namespace_label *nd_label, struct device **devs,
1891                 int count)
1892 {
1893         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
1894         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1895         int i;
1896
1897         for (i = 0; i < count; i++) {
1898                 uuid_t *uuid = namespace_to_uuid(devs[i]);
1899
1900                 if (IS_ERR(uuid)) {
1901                         WARN_ON(1);
1902                         continue;
1903                 }
1904
1905                 if (!nsl_uuid_equal(ndd, nd_label, uuid))
1906                         continue;
1907                 dev_err(&nd_region->dev,
1908                         "error: conflicting extents for uuid: %pUb\n", uuid);
1909                 return -ENXIO;
1910         }
1911
1912         return i;
1913 }
1914
1915 static int cmp_dpa(const void *a, const void *b)
1916 {
1917         const struct device *dev_a = *(const struct device **) a;
1918         const struct device *dev_b = *(const struct device **) b;
1919         struct nd_namespace_pmem *nspm_a, *nspm_b;
1920
1921         if (is_namespace_io(dev_a))
1922                 return 0;
1923
1924         nspm_a = to_nd_namespace_pmem(dev_a);
1925         nspm_b = to_nd_namespace_pmem(dev_b);
1926
1927         return memcmp(&nspm_a->nsio.res.start, &nspm_b->nsio.res.start,
1928                         sizeof(resource_size_t));
1929 }
1930
1931 static struct device **scan_labels(struct nd_region *nd_region)
1932 {
1933         int i, count = 0;
1934         struct device *dev, **devs;
1935         struct nd_label_ent *label_ent, *e;
1936         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
1937         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1938         resource_size_t map_end = nd_mapping->start + nd_mapping->size - 1;
1939
1940         devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
1941         if (!devs)
1942                 return NULL;
1943
1944         /* "safe" because create_namespace_pmem() might list_move() label_ent */
1945         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
1946                 struct nd_namespace_label *nd_label = label_ent->label;
1947                 struct device **__devs;
1948
1949                 if (!nd_label)
1950                         continue;
1951
1952                 /* skip labels that describe extents outside of the region */
1953                 if (nsl_get_dpa(ndd, nd_label) < nd_mapping->start ||
1954                     nsl_get_dpa(ndd, nd_label) > map_end)
1955                         continue;
1956
1957                 i = add_namespace_resource(nd_region, nd_label, devs, count);
1958                 if (i < 0)
1959                         goto err;
1960                 if (i < count)
1961                         continue;
1962                 if (count) {
1963                         __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
1964                         if (!__devs)
1965                                 goto err;
1966                         memcpy(__devs, devs, sizeof(dev) * count);
1967                         kfree(devs);
1968                         devs = __devs;
1969                 }
1970
1971                 dev = create_namespace_pmem(nd_region, nd_mapping, nd_label);
1972                 if (IS_ERR(dev)) {
1973                         switch (PTR_ERR(dev)) {
1974                         case -EAGAIN:
1975                                 /* skip invalid labels */
1976                                 continue;
1977                         default:
1978                                 goto err;
1979                         }
1980                 } else
1981                         devs[count++] = dev;
1982
1983         }
1984
1985         dev_dbg(&nd_region->dev, "discovered %d namespace%s\n", count,
1986                 count == 1 ? "" : "s");
1987
1988         if (count == 0) {
1989                 struct nd_namespace_pmem *nspm;
1990
1991                 /* Publish a zero-sized namespace for userspace to configure. */
1992                 nd_mapping_free_labels(nd_mapping);
1993                 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1994                 if (!nspm)
1995                         goto err;
1996                 dev = &nspm->nsio.common.dev;
1997                 dev->type = &namespace_pmem_device_type;
1998                 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
1999                 dev->parent = &nd_region->dev;
2000                 devs[count++] = dev;
2001         } else if (is_memory(&nd_region->dev)) {
2002                 /* clean unselected labels */
2003                 for (i = 0; i < nd_region->ndr_mappings; i++) {
2004                         struct list_head *l, *e;
2005                         LIST_HEAD(list);
2006                         int j;
2007
2008                         nd_mapping = &nd_region->mapping[i];
2009                         if (list_empty(&nd_mapping->labels)) {
2010                                 WARN_ON(1);
2011                                 continue;
2012                         }
2013
2014                         j = count;
2015                         list_for_each_safe(l, e, &nd_mapping->labels) {
2016                                 if (!j--)
2017                                         break;
2018                                 list_move_tail(l, &list);
2019                         }
2020                         nd_mapping_free_labels(nd_mapping);
2021                         list_splice_init(&list, &nd_mapping->labels);
2022                 }
2023         }
2024
2025         if (count > 1)
2026                 sort(devs, count, sizeof(struct device *), cmp_dpa, NULL);
2027
2028         return devs;
2029
2030  err:
2031         for (i = 0; devs[i]; i++)
2032                 namespace_pmem_release(devs[i]);
2033         kfree(devs);
2034
2035         return NULL;
2036 }
2037
2038 static struct device **create_namespaces(struct nd_region *nd_region)
2039 {
2040         struct nd_mapping *nd_mapping;
2041         struct device **devs;
2042         int i;
2043
2044         if (nd_region->ndr_mappings == 0)
2045                 return NULL;
2046
2047         /* lock down all mappings while we scan labels */
2048         for (i = 0; i < nd_region->ndr_mappings; i++) {
2049                 nd_mapping = &nd_region->mapping[i];
2050                 mutex_lock_nested(&nd_mapping->lock, i);
2051         }
2052
2053         devs = scan_labels(nd_region);
2054
2055         for (i = 0; i < nd_region->ndr_mappings; i++) {
2056                 int reverse = nd_region->ndr_mappings - 1 - i;
2057
2058                 nd_mapping = &nd_region->mapping[reverse];
2059                 mutex_unlock(&nd_mapping->lock);
2060         }
2061
2062         return devs;
2063 }
2064
2065 static void deactivate_labels(void *region)
2066 {
2067         struct nd_region *nd_region = region;
2068         int i;
2069
2070         for (i = 0; i < nd_region->ndr_mappings; i++) {
2071                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2072                 struct nvdimm_drvdata *ndd = nd_mapping->ndd;
2073                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2074
2075                 mutex_lock(&nd_mapping->lock);
2076                 nd_mapping_free_labels(nd_mapping);
2077                 mutex_unlock(&nd_mapping->lock);
2078
2079                 put_ndd(ndd);
2080                 nd_mapping->ndd = NULL;
2081                 if (ndd)
2082                         atomic_dec(&nvdimm->busy);
2083         }
2084 }
2085
2086 static int init_active_labels(struct nd_region *nd_region)
2087 {
2088         int i, rc = 0;
2089
2090         for (i = 0; i < nd_region->ndr_mappings; i++) {
2091                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2092                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2093                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2094                 struct nd_label_ent *label_ent;
2095                 int count, j;
2096
2097                 /*
2098                  * If the dimm is disabled then we may need to prevent
2099                  * the region from being activated.
2100                  */
2101                 if (!ndd) {
2102                         if (test_bit(NDD_LOCKED, &nvdimm->flags))
2103                                 /* fail, label data may be unreadable */;
2104                         else if (test_bit(NDD_LABELING, &nvdimm->flags))
2105                                 /* fail, labels needed to disambiguate dpa */;
2106                         else
2107                                 continue;
2108
2109                         dev_err(&nd_region->dev, "%s: is %s, failing probe\n",
2110                                         dev_name(&nd_mapping->nvdimm->dev),
2111                                         test_bit(NDD_LOCKED, &nvdimm->flags)
2112                                         ? "locked" : "disabled");
2113                         rc = -ENXIO;
2114                         goto out;
2115                 }
2116                 nd_mapping->ndd = ndd;
2117                 atomic_inc(&nvdimm->busy);
2118                 get_ndd(ndd);
2119
2120                 count = nd_label_active_count(ndd);
2121                 dev_dbg(ndd->dev, "count: %d\n", count);
2122                 if (!count)
2123                         continue;
2124                 for (j = 0; j < count; j++) {
2125                         struct nd_namespace_label *label;
2126
2127                         label_ent = kzalloc(sizeof(*label_ent), GFP_KERNEL);
2128                         if (!label_ent)
2129                                 break;
2130                         label = nd_label_active(ndd, j);
2131                         label_ent->label = label;
2132
2133                         mutex_lock(&nd_mapping->lock);
2134                         list_add_tail(&label_ent->list, &nd_mapping->labels);
2135                         mutex_unlock(&nd_mapping->lock);
2136                 }
2137
2138                 if (j < count)
2139                         break;
2140         }
2141
2142         if (i < nd_region->ndr_mappings)
2143                 rc = -ENOMEM;
2144
2145 out:
2146         if (rc) {
2147                 deactivate_labels(nd_region);
2148                 return rc;
2149         }
2150
2151         return devm_add_action_or_reset(&nd_region->dev, deactivate_labels,
2152                                         nd_region);
2153 }
2154
2155 int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
2156 {
2157         struct device **devs = NULL;
2158         int i, rc = 0, type;
2159
2160         *err = 0;
2161         nvdimm_bus_lock(&nd_region->dev);
2162         rc = init_active_labels(nd_region);
2163         if (rc) {
2164                 nvdimm_bus_unlock(&nd_region->dev);
2165                 return rc;
2166         }
2167
2168         type = nd_region_to_nstype(nd_region);
2169         switch (type) {
2170         case ND_DEVICE_NAMESPACE_IO:
2171                 devs = create_namespace_io(nd_region);
2172                 break;
2173         case ND_DEVICE_NAMESPACE_PMEM:
2174                 devs = create_namespaces(nd_region);
2175                 break;
2176         default:
2177                 break;
2178         }
2179         nvdimm_bus_unlock(&nd_region->dev);
2180
2181         if (!devs)
2182                 return -ENODEV;
2183
2184         for (i = 0; devs[i]; i++) {
2185                 struct device *dev = devs[i];
2186                 int id;
2187
2188                 if (type == ND_DEVICE_NAMESPACE_PMEM) {
2189                         struct nd_namespace_pmem *nspm;
2190
2191                         nspm = to_nd_namespace_pmem(dev);
2192                         id = ida_alloc(&nd_region->ns_ida, GFP_KERNEL);
2193                         nspm->id = id;
2194                 } else
2195                         id = i;
2196
2197                 if (id < 0)
2198                         break;
2199                 dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
2200                 device_initialize(dev);
2201                 lockdep_set_class(&dev->mutex, &nvdimm_namespace_key);
2202                 nd_device_register(dev);
2203         }
2204         if (i)
2205                 nd_region->ns_seed = devs[0];
2206
2207         if (devs[i]) {
2208                 int j;
2209
2210                 for (j = i; devs[j]; j++) {
2211                         struct device *dev = devs[j];
2212
2213                         device_initialize(dev);
2214                         put_device(dev);
2215                 }
2216                 *err = j - i;
2217                 /*
2218                  * All of the namespaces we tried to register failed, so
2219                  * fail region activation.
2220                  */
2221                 if (*err == 0)
2222                         rc = -ENODEV;
2223         }
2224         kfree(devs);
2225
2226         if (rc == -ENODEV)
2227                 return rc;
2228
2229         return i;
2230 }