drivers/nvdimm/namespace_devs.c

   1 /*
   2  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of version 2 of the GNU General Public License as
   6  * published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful, but
   9  * WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  */
  13 #include <linux/module.h>
  14 #include <linux/device.h>
  15 #include <linux/slab.h>
  16 #include <linux/nd.h>
  17 #include "nd-core.h"
  18 #include "nd.h"
  19
  20 static void namespace_io_release(struct device *dev)
  21 {
  22         struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
  23
  24         kfree(nsio);
  25 }
  26
  27 static void namespace_pmem_release(struct device *dev)
  28 {
  29         struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
  30
  31         kfree(nspm->alt_name);
  32         kfree(nspm->uuid);
  33         kfree(nspm);
  34 }
  35
  36 static void namespace_blk_release(struct device *dev)
  37 {
  38         struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
  39         struct nd_region *nd_region = to_nd_region(dev->parent);
  40
  41         if (nsblk->id >= 0)
  42                 ida_simple_remove(&nd_region->ns_ida, nsblk->id);
  43         kfree(nsblk->alt_name);
  44         kfree(nsblk->uuid);
  45         kfree(nsblk->res);
  46         kfree(nsblk);
  47 }
  48
  49 static struct device_type namespace_io_device_type = {
  50         .name = "nd_namespace_io",
  51         .release = namespace_io_release,
  52 };
  53
  54 static struct device_type namespace_pmem_device_type = {
  55         .name = "nd_namespace_pmem",
  56         .release = namespace_pmem_release,
  57 };
  58
  59 static struct device_type namespace_blk_device_type = {
  60         .name = "nd_namespace_blk",
  61         .release = namespace_blk_release,
  62 };
  63
  64 static bool is_namespace_pmem(struct device *dev)
  65 {
  66         return dev ? dev->type == &namespace_pmem_device_type : false;
  67 }
  68
  69 static bool is_namespace_blk(struct device *dev)
  70 {
  71         return dev ? dev->type == &namespace_blk_device_type : false;
  72 }
  73
  74 static bool is_namespace_io(struct device *dev)
  75 {
  76         return dev ? dev->type == &namespace_io_device_type : false;
  77 }
  78
  79 const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,
  80                 char *name)
  81 {
  82         struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
  83         const char *suffix = "";
  84
  85         if (ndns->claim && is_nd_btt(ndns->claim))
  86                 suffix = "s";
  87
  88         if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev))
  89                 sprintf(name, "pmem%d%s", nd_region->id, suffix);
  90         else if (is_namespace_blk(&ndns->dev)) {
  91                 struct nd_namespace_blk *nsblk;
  92
  93                 nsblk = to_nd_namespace_blk(&ndns->dev);
  94                 sprintf(name, "ndblk%d.%d%s", nd_region->id, nsblk->id, suffix);
  95         } else {
  96                 return NULL;
  97         }
  98
  99         return name;
 100 }
 101 EXPORT_SYMBOL(nvdimm_namespace_disk_name);
 102
 103 const u8 *nd_dev_to_uuid(struct device *dev)
 104 {
 105         static const u8 null_uuid[16];
 106
 107         if (!dev)
 108                 return null_uuid;
 109
 110         if (is_namespace_pmem(dev)) {
 111                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 112
 113                 return nspm->uuid;
 114         } else if (is_namespace_blk(dev)) {
 115                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 116
 117                 return nsblk->uuid;
 118         } else
 119                 return null_uuid;
 120 }
 121 EXPORT_SYMBOL(nd_dev_to_uuid);
 122
 123 static ssize_t nstype_show(struct device *dev,
 124                 struct device_attribute *attr, char *buf)
 125 {
 126         struct nd_region *nd_region = to_nd_region(dev->parent);
 127
 128         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
 129 }
 130 static DEVICE_ATTR_RO(nstype);
 131
 132 static ssize_t __alt_name_store(struct device *dev, const char *buf,
 133                 const size_t len)
 134 {
 135         char *input, *pos, *alt_name, **ns_altname;
 136         ssize_t rc;
 137
 138         if (is_namespace_pmem(dev)) {
 139                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 140
 141                 ns_altname = &nspm->alt_name;
 142         } else if (is_namespace_blk(dev)) {
 143                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 144
 145                 ns_altname = &nsblk->alt_name;
 146         } else
 147                 return -ENXIO;
 148
 149         if (dev->driver || to_ndns(dev)->claim)
 150                 return -EBUSY;
 151
 152         input = kmemdup(buf, len + 1, GFP_KERNEL);
 153         if (!input)
 154                 return -ENOMEM;
 155
 156         input[len] = '\0';
 157         pos = strim(input);
 158         if (strlen(pos) + 1 > NSLABEL_NAME_LEN) {
 159                 rc = -EINVAL;
 160                 goto out;
 161         }
 162
 163         alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL);
 164         if (!alt_name) {
 165                 rc = -ENOMEM;
 166                 goto out;
 167         }
 168         kfree(*ns_altname);
 169         *ns_altname = alt_name;
 170         sprintf(*ns_altname, "%s", pos);
 171         rc = len;
 172
 173 out:
 174         kfree(input);
 175         return rc;
 176 }
 177
 178 static resource_size_t nd_namespace_blk_size(struct nd_namespace_blk *nsblk)
 179 {
 180         struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
 181         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 182         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 183         struct nd_label_id label_id;
 184         resource_size_t size = 0;
 185         struct resource *res;
 186
 187         if (!nsblk->uuid)
 188                 return 0;
 189         nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
 190         for_each_dpa_resource(ndd, res)
 191                 if (strcmp(res->name, label_id.id) == 0)
 192                         size += resource_size(res);
 193         return size;
 194 }
 195
 196 static bool __nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
 197 {
 198         struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
 199         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 200         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 201         struct nd_label_id label_id;
 202         struct resource *res;
 203         int count, i;
 204
 205         if (!nsblk->uuid || !nsblk->lbasize || !ndd)
 206                 return false;
 207
 208         count = 0;
 209         nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
 210         for_each_dpa_resource(ndd, res) {
 211                 if (strcmp(res->name, label_id.id) != 0)
 212                         continue;
 213                 /*
 214                  * Resources with unacknoweldged adjustments indicate a
 215                  * failure to update labels
 216                  */
 217                 if (res->flags & DPA_RESOURCE_ADJUSTED)
 218                         return false;
 219                 count++;
 220         }
 221
 222         /* These values match after a successful label update */
 223         if (count != nsblk->num_resources)
 224                 return false;
 225
 226         for (i = 0; i < nsblk->num_resources; i++) {
 227                 struct resource *found = NULL;
 228
 229                 for_each_dpa_resource(ndd, res)
 230                         if (res == nsblk->res[i]) {
 231                                 found = res;
 232                                 break;
 233                         }
 234                 /* stale resource */
 235                 if (!found)
 236                         return false;
 237         }
 238
 239         return true;
 240 }
 241
 242 resource_size_t nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
 243 {
 244         resource_size_t size;
 245
 246         nvdimm_bus_lock(&nsblk->common.dev);
 247         size = __nd_namespace_blk_validate(nsblk);
 248         nvdimm_bus_unlock(&nsblk->common.dev);
 249
 250         return size;
 251 }
 252 EXPORT_SYMBOL(nd_namespace_blk_validate);
 253
 254
 255 static int nd_namespace_label_update(struct nd_region *nd_region,
 256                 struct device *dev)
 257 {
 258         dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim,
 259                         "namespace must be idle during label update\n");
 260         if (dev->driver || to_ndns(dev)->claim)
 261                 return 0;
 262
 263         /*
 264          * Only allow label writes that will result in a valid namespace
 265          * or deletion of an existing namespace.
 266          */
 267         if (is_namespace_pmem(dev)) {
 268                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 269                 resource_size_t size = resource_size(&nspm->nsio.res);
 270
 271                 if (size == 0 && nspm->uuid)
 272                         /* delete allocation */;
 273                 else if (!nspm->uuid)
 274                         return 0;
 275
 276                 return nd_pmem_namespace_label_update(nd_region, nspm, size);
 277         } else if (is_namespace_blk(dev)) {
 278                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 279                 resource_size_t size = nd_namespace_blk_size(nsblk);
 280
 281                 if (size == 0 && nsblk->uuid)
 282                         /* delete allocation */;
 283                 else if (!nsblk->uuid || !nsblk->lbasize)
 284                         return 0;
 285
 286                 return nd_blk_namespace_label_update(nd_region, nsblk, size);
 287         } else
 288                 return -ENXIO;
 289 }
 290
 291 static ssize_t alt_name_store(struct device *dev,
 292                 struct device_attribute *attr, const char *buf, size_t len)
 293 {
 294         struct nd_region *nd_region = to_nd_region(dev->parent);
 295         ssize_t rc;
 296
 297         device_lock(dev);
 298         nvdimm_bus_lock(dev);
 299         wait_nvdimm_bus_probe_idle(dev);
 300         rc = __alt_name_store(dev, buf, len);
 301         if (rc >= 0)
 302                 rc = nd_namespace_label_update(nd_region, dev);
 303         dev_dbg(dev, "%s: %s(%zd)\n", __func__, rc < 0 ? "fail " : "", rc);
 304         nvdimm_bus_unlock(dev);
 305         device_unlock(dev);
 306
 307         return rc < 0 ? rc : len;
 308 }
 309
 310 static ssize_t alt_name_show(struct device *dev,
 311                 struct device_attribute *attr, char *buf)
 312 {
 313         char *ns_altname;
 314
 315         if (is_namespace_pmem(dev)) {
 316                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 317
 318                 ns_altname = nspm->alt_name;
 319         } else if (is_namespace_blk(dev)) {
 320                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 321
 322                 ns_altname = nsblk->alt_name;
 323         } else
 324                 return -ENXIO;
 325
 326         return sprintf(buf, "%s\n", ns_altname ? ns_altname : "");
 327 }
 328 static DEVICE_ATTR_RW(alt_name);
 329
 330 static int scan_free(struct nd_region *nd_region,
 331                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
 332                 resource_size_t n)
 333 {
 334         bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
 335         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 336         int rc = 0;
 337
 338         while (n) {
 339                 struct resource *res, *last;
 340                 resource_size_t new_start;
 341
 342                 last = NULL;
 343                 for_each_dpa_resource(ndd, res)
 344                         if (strcmp(res->name, label_id->id) == 0)
 345                                 last = res;
 346                 res = last;
 347                 if (!res)
 348                         return 0;
 349
 350                 if (n >= resource_size(res)) {
 351                         n -= resource_size(res);
 352                         nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc);
 353                         nvdimm_free_dpa(ndd, res);
 354                         /* retry with last resource deleted */
 355                         continue;
 356                 }
 357
 358                 /*
 359                  * Keep BLK allocations relegated to high DPA as much as
 360                  * possible
 361                  */
 362                 if (is_blk)
 363                         new_start = res->start + n;
 364                 else
 365                         new_start = res->start;
 366
 367                 rc = adjust_resource(res, new_start, resource_size(res) - n);
 368                 if (rc == 0)
 369                         res->flags |= DPA_RESOURCE_ADJUSTED;
 370                 nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc);
 371                 break;
 372         }
 373
 374         return rc;
 375 }
 376
 377 /**
 378  * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
 379  * @nd_region: the set of dimms to reclaim @n bytes from
 380  * @label_id: unique identifier for the namespace consuming this dpa range
 381  * @n: number of bytes per-dimm to release
 382  *
 383  * Assumes resources are ordered.  Starting from the end try to
 384  * adjust_resource() the allocation to @n, but if @n is larger than the
 385  * allocation delete it and find the 'new' last allocation in the label
 386  * set.
 387  */
 388 static int shrink_dpa_allocation(struct nd_region *nd_region,
 389                 struct nd_label_id *label_id, resource_size_t n)
 390 {
 391         int i;
 392
 393         for (i = 0; i < nd_region->ndr_mappings; i++) {
 394                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 395                 int rc;
 396
 397                 rc = scan_free(nd_region, nd_mapping, label_id, n);
 398                 if (rc)
 399                         return rc;
 400         }
 401
 402         return 0;
 403 }
 404
 405 static resource_size_t init_dpa_allocation(struct nd_label_id *label_id,
 406                 struct nd_region *nd_region, struct nd_mapping *nd_mapping,
 407                 resource_size_t n)
 408 {
 409         bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
 410         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 411         resource_size_t first_dpa;
 412         struct resource *res;
 413         int rc = 0;
 414
 415         /* allocate blk from highest dpa first */
 416         if (is_blk)
 417                 first_dpa = nd_mapping->start + nd_mapping->size - n;
 418         else
 419                 first_dpa = nd_mapping->start;
 420
 421         /* first resource allocation for this label-id or dimm */
 422         res = nvdimm_allocate_dpa(ndd, label_id, first_dpa, n);
 423         if (!res)
 424                 rc = -EBUSY;
 425
 426         nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc);
 427         return rc ? n : 0;
 428 }
 429
 430 static bool space_valid(bool is_pmem, bool is_reserve,
 431                 struct nd_label_id *label_id, struct resource *res)
 432 {
 433         /*
 434          * For BLK-space any space is valid, for PMEM-space, it must be
 435          * contiguous with an existing allocation unless we are
 436          * reserving pmem.
 437          */
 438         if (is_reserve || !is_pmem)
 439                 return true;
 440         if (!res || strcmp(res->name, label_id->id) == 0)
 441                 return true;
 442         return false;
 443 }
 444
 445 enum alloc_loc {
 446         ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
 447 };
 448
 449 static resource_size_t scan_allocate(struct nd_region *nd_region,
 450                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
 451                 resource_size_t n)
 452 {
 453         resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1;
 454         bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0;
 455         bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
 456         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 457         const resource_size_t to_allocate = n;
 458         struct resource *res;
 459         int first;
 460
 461  retry:
 462         first = 0;
 463         for_each_dpa_resource(ndd, res) {
 464                 resource_size_t allocate, available = 0, free_start, free_end;
 465                 struct resource *next = res->sibling, *new_res = NULL;
 466                 enum alloc_loc loc = ALLOC_ERR;
 467                 const char *action;
 468                 int rc = 0;
 469
 470                 /* ignore resources outside this nd_mapping */
 471                 if (res->start > mapping_end)
 472                         continue;
 473                 if (res->end < nd_mapping->start)
 474                         continue;
 475
 476                 /* space at the beginning of the mapping */
 477                 if (!first++ && res->start > nd_mapping->start) {
 478                         free_start = nd_mapping->start;
 479                         available = res->start - free_start;
 480                         if (space_valid(is_pmem, is_reserve, label_id, NULL))
 481                                 loc = ALLOC_BEFORE;
 482                 }
 483
 484                 /* space between allocations */
 485                 if (!loc && next) {
 486                         free_start = res->start + resource_size(res);
 487                         free_end = min(mapping_end, next->start - 1);
 488                         if (space_valid(is_pmem, is_reserve, label_id, res)
 489                                         && free_start < free_end) {
 490                                 available = free_end + 1 - free_start;
 491                                 loc = ALLOC_MID;
 492                         }
 493                 }
 494
 495                 /* space at the end of the mapping */
 496                 if (!loc && !next) {
 497                         free_start = res->start + resource_size(res);
 498                         free_end = mapping_end;
 499                         if (space_valid(is_pmem, is_reserve, label_id, res)
 500                                         && free_start < free_end) {
 501                                 available = free_end + 1 - free_start;
 502                                 loc = ALLOC_AFTER;
 503                         }
 504                 }
 505
 506                 if (!loc || !available)
 507                         continue;
 508                 allocate = min(available, n);
 509                 switch (loc) {
 510                 case ALLOC_BEFORE:
 511                         if (strcmp(res->name, label_id->id) == 0) {
 512                                 /* adjust current resource up */
 513                                 if (is_pmem && !is_reserve)
 514                                         return n;
 515                                 rc = adjust_resource(res, res->start - allocate,
 516                                                 resource_size(res) + allocate);
 517                                 action = "cur grow up";
 518                         } else
 519                                 action = "allocate";
 520                         break;
 521                 case ALLOC_MID:
 522                         if (strcmp(next->name, label_id->id) == 0) {
 523                                 /* adjust next resource up */
 524                                 if (is_pmem && !is_reserve)
 525                                         return n;
 526                                 rc = adjust_resource(next, next->start
 527                                                 - allocate, resource_size(next)
 528                                                 + allocate);
 529                                 new_res = next;
 530                                 action = "next grow up";
 531                         } else if (strcmp(res->name, label_id->id) == 0) {
 532                                 action = "grow down";
 533                         } else
 534                                 action = "allocate";
 535                         break;
 536                 case ALLOC_AFTER:
 537                         if (strcmp(res->name, label_id->id) == 0)
 538                                 action = "grow down";
 539                         else
 540                                 action = "allocate";
 541                         break;
 542                 default:
 543                         return n;
 544                 }
 545
 546                 if (strcmp(action, "allocate") == 0) {
 547                         /* BLK allocate bottom up */
 548                         if (!is_pmem)
 549                                 free_start += available - allocate;
 550                         else if (!is_reserve && free_start != nd_mapping->start)
 551                                 return n;
 552
 553                         new_res = nvdimm_allocate_dpa(ndd, label_id,
 554                                         free_start, allocate);
 555                         if (!new_res)
 556                                 rc = -EBUSY;
 557                 } else if (strcmp(action, "grow down") == 0) {
 558                         /* adjust current resource down */
 559                         rc = adjust_resource(res, res->start, resource_size(res)
 560                                         + allocate);
 561                         if (rc == 0)
 562                                 res->flags |= DPA_RESOURCE_ADJUSTED;
 563                 }
 564
 565                 if (!new_res)
 566                         new_res = res;
 567
 568                 nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n",
 569                                 action, loc, rc);
 570
 571                 if (rc)
 572                         return n;
 573
 574                 n -= allocate;
 575                 if (n) {
 576                         /*
 577                          * Retry scan with newly inserted resources.
 578                          * For example, if we did an ALLOC_BEFORE
 579                          * insertion there may also have been space
 580                          * available for an ALLOC_AFTER insertion, so we
 581                          * need to check this same resource again
 582                          */
 583                         goto retry;
 584                 } else
 585                         return 0;
 586         }
 587
 588         /*
 589          * If we allocated nothing in the BLK case it may be because we are in
 590          * an initial "pmem-reserve pass".  Only do an initial BLK allocation
 591          * when none of the DPA space is reserved.
 592          */
 593         if ((is_pmem || !ndd->dpa.child) && n == to_allocate)
 594                 return init_dpa_allocation(label_id, nd_region, nd_mapping, n);
 595         return n;
 596 }
 597
 598 static int merge_dpa(struct nd_region *nd_region,
 599                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id)
 600 {
 601         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 602         struct resource *res;
 603
 604         if (strncmp("pmem", label_id->id, 4) == 0)
 605                 return 0;
 606  retry:
 607         for_each_dpa_resource(ndd, res) {
 608                 int rc;
 609                 struct resource *next = res->sibling;
 610                 resource_size_t end = res->start + resource_size(res);
 611
 612                 if (!next || strcmp(res->name, label_id->id) != 0
 613                                 || strcmp(next->name, label_id->id) != 0
 614                                 || end != next->start)
 615                         continue;
 616                 end += resource_size(next);
 617                 nvdimm_free_dpa(ndd, next);
 618                 rc = adjust_resource(res, res->start, end - res->start);
 619                 nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc);
 620                 if (rc)
 621                         return rc;
 622                 res->flags |= DPA_RESOURCE_ADJUSTED;
 623                 goto retry;
 624         }
 625
 626         return 0;
 627 }
 628
 629 static int __reserve_free_pmem(struct device *dev, void *data)
 630 {
 631         struct nvdimm *nvdimm = data;
 632         struct nd_region *nd_region;
 633         struct nd_label_id label_id;
 634         int i;
 635
 636         if (!is_nd_pmem(dev))
 637                 return 0;
 638
 639         nd_region = to_nd_region(dev);
 640         if (nd_region->ndr_mappings == 0)
 641                 return 0;
 642
 643         memset(&label_id, 0, sizeof(label_id));
 644         strcat(label_id.id, "pmem-reserve");
 645         for (i = 0; i < nd_region->ndr_mappings; i++) {
 646                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 647                 resource_size_t n, rem = 0;
 648
 649                 if (nd_mapping->nvdimm != nvdimm)
 650                         continue;
 651
 652                 n = nd_pmem_available_dpa(nd_region, nd_mapping, &rem);
 653                 if (n == 0)
 654                         return 0;
 655                 rem = scan_allocate(nd_region, nd_mapping, &label_id, n);
 656                 dev_WARN_ONCE(&nd_region->dev, rem,
 657                                 "pmem reserve underrun: %#llx of %#llx bytes\n",
 658                                 (unsigned long long) n - rem,
 659                                 (unsigned long long) n);
 660                 return rem ? -ENXIO : 0;
 661         }
 662
 663         return 0;
 664 }
 665
 666 static void release_free_pmem(struct nvdimm_bus *nvdimm_bus,
 667                 struct nd_mapping *nd_mapping)
 668 {
 669         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 670         struct resource *res, *_res;
 671
 672         for_each_dpa_resource_safe(ndd, res, _res)
 673                 if (strcmp(res->name, "pmem-reserve") == 0)
 674                         nvdimm_free_dpa(ndd, res);
 675 }
 676
 677 static int reserve_free_pmem(struct nvdimm_bus *nvdimm_bus,
 678                 struct nd_mapping *nd_mapping)
 679 {
 680         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 681         int rc;
 682
 683         rc = device_for_each_child(&nvdimm_bus->dev, nvdimm,
 684                         __reserve_free_pmem);
 685         if (rc)
 686                 release_free_pmem(nvdimm_bus, nd_mapping);
 687         return rc;
 688 }
 689
 690 /**
 691  * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
 692  * @nd_region: the set of dimms to allocate @n more bytes from
 693  * @label_id: unique identifier for the namespace consuming this dpa range
 694  * @n: number of bytes per-dimm to add to the existing allocation
 695  *
 696  * Assumes resources are ordered.  For BLK regions, first consume
 697  * BLK-only available DPA free space, then consume PMEM-aliased DPA
 698  * space starting at the highest DPA.  For PMEM regions start
 699  * allocations from the start of an interleave set and end at the first
 700  * BLK allocation or the end of the interleave set, whichever comes
 701  * first.
 702  */
 703 static int grow_dpa_allocation(struct nd_region *nd_region,
 704                 struct nd_label_id *label_id, resource_size_t n)
 705 {
 706         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
 707         bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
 708         int i;
 709
 710         for (i = 0; i < nd_region->ndr_mappings; i++) {
 711                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 712                 resource_size_t rem = n;
 713                 int rc, j;
 714
 715                 /*
 716                  * In the BLK case try once with all unallocated PMEM
 717                  * reserved, and once without
 718                  */
 719                 for (j = is_pmem; j < 2; j++) {
 720                         bool blk_only = j == 0;
 721
 722                         if (blk_only) {
 723                                 rc = reserve_free_pmem(nvdimm_bus, nd_mapping);
 724                                 if (rc)
 725                                         return rc;
 726                         }
 727                         rem = scan_allocate(nd_region, nd_mapping,
 728                                         label_id, rem);
 729                         if (blk_only)
 730                                 release_free_pmem(nvdimm_bus, nd_mapping);
 731
 732                         /* try again and allow encroachments into PMEM */
 733                         if (rem == 0)
 734                                 break;
 735                 }
 736
 737                 dev_WARN_ONCE(&nd_region->dev, rem,
 738                                 "allocation underrun: %#llx of %#llx bytes\n",
 739                                 (unsigned long long) n - rem,
 740                                 (unsigned long long) n);
 741                 if (rem)
 742                         return -ENXIO;
 743
 744                 rc = merge_dpa(nd_region, nd_mapping, label_id);
 745                 if (rc)
 746                         return rc;
 747         }
 748
 749         return 0;
 750 }
 751
 752 static void nd_namespace_pmem_set_size(struct nd_region *nd_region,
 753                 struct nd_namespace_pmem *nspm, resource_size_t size)
 754 {
 755         struct resource *res = &nspm->nsio.res;
 756
 757         res->start = nd_region->ndr_start;
 758         res->end = nd_region->ndr_start + size - 1;
 759 }
 760
 761 static ssize_t __size_store(struct device *dev, unsigned long long val)
 762 {
 763         resource_size_t allocated = 0, available = 0;
 764         struct nd_region *nd_region = to_nd_region(dev->parent);
 765         struct nd_mapping *nd_mapping;
 766         struct nvdimm_drvdata *ndd;
 767         struct nd_label_id label_id;
 768         u32 flags = 0, remainder;
 769         u8 *uuid = NULL;
 770         int rc, i;
 771
 772         if (dev->driver || to_ndns(dev)->claim)
 773                 return -EBUSY;
 774
 775         if (is_namespace_pmem(dev)) {
 776                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 777
 778                 uuid = nspm->uuid;
 779         } else if (is_namespace_blk(dev)) {
 780                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 781
 782                 uuid = nsblk->uuid;
 783                 flags = NSLABEL_FLAG_LOCAL;
 784         }
 785
 786         /*
 787          * We need a uuid for the allocation-label and dimm(s) on which
 788          * to store the label.
 789          */
 790         if (!uuid || nd_region->ndr_mappings == 0)
 791                 return -ENXIO;
 792
 793         div_u64_rem(val, SZ_4K * nd_region->ndr_mappings, &remainder);
 794         if (remainder) {
 795                 dev_dbg(dev, "%llu is not %dK aligned\n", val,
 796                                 (SZ_4K * nd_region->ndr_mappings) / SZ_1K);
 797                 return -EINVAL;
 798         }
 799
 800         nd_label_gen_id(&label_id, uuid, flags);
 801         for (i = 0; i < nd_region->ndr_mappings; i++) {
 802                 nd_mapping = &nd_region->mapping[i];
 803                 ndd = to_ndd(nd_mapping);
 804
 805                 /*
 806                  * All dimms in an interleave set, or the base dimm for a blk
 807                  * region, need to be enabled for the size to be changed.
 808                  */
 809                 if (!ndd)
 810                         return -ENXIO;
 811
 812                 allocated += nvdimm_allocated_dpa(ndd, &label_id);
 813         }
 814         available = nd_region_available_dpa(nd_region);
 815
 816         if (val > available + allocated)
 817                 return -ENOSPC;
 818
 819         if (val == allocated)
 820                 return 0;
 821
 822         val = div_u64(val, nd_region->ndr_mappings);
 823         allocated = div_u64(allocated, nd_region->ndr_mappings);
 824         if (val < allocated)
 825                 rc = shrink_dpa_allocation(nd_region, &label_id,
 826                                 allocated - val);
 827         else
 828                 rc = grow_dpa_allocation(nd_region, &label_id, val - allocated);
 829
 830         if (rc)
 831                 return rc;
 832
 833         if (is_namespace_pmem(dev)) {
 834                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 835
 836                 nd_namespace_pmem_set_size(nd_region, nspm,
 837                                 val * nd_region->ndr_mappings);
 838         } else if (is_namespace_blk(dev)) {
 839                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 840
 841                 /*
 842                  * Try to delete the namespace if we deleted all of its
 843                  * allocation, this is not the seed device for the
 844                  * region, and it is not actively claimed by a btt
 845                  * instance.
 846                  */
 847                 if (val == 0 && nd_region->ns_seed != dev
 848                                 && !nsblk->common.claim)
 849                         nd_device_unregister(dev, ND_ASYNC);
 850         }
 851
 852         return rc;
 853 }
 854
 855 static ssize_t size_store(struct device *dev,
 856                 struct device_attribute *attr, const char *buf, size_t len)
 857 {
 858         struct nd_region *nd_region = to_nd_region(dev->parent);
 859         unsigned long long val;
 860         u8 **uuid = NULL;
 861         int rc;
 862
 863         rc = kstrtoull(buf, 0, &val);
 864         if (rc)
 865                 return rc;
 866
 867         device_lock(dev);
 868         nvdimm_bus_lock(dev);
 869         wait_nvdimm_bus_probe_idle(dev);
 870         rc = __size_store(dev, val);
 871         if (rc >= 0)
 872                 rc = nd_namespace_label_update(nd_region, dev);
 873
 874         if (is_namespace_pmem(dev)) {
 875                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 876
 877                 uuid = &nspm->uuid;
 878         } else if (is_namespace_blk(dev)) {
 879                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 880
 881                 uuid = &nsblk->uuid;
 882         }
 883
 884         if (rc == 0 && val == 0 && uuid) {
 885                 /* setting size zero == 'delete namespace' */
 886                 kfree(*uuid);
 887                 *uuid = NULL;
 888         }
 889
 890         dev_dbg(dev, "%s: %llx %s (%d)\n", __func__, val, rc < 0
 891                         ? "fail" : "success", rc);
 892
 893         nvdimm_bus_unlock(dev);
 894         device_unlock(dev);
 895
 896         return rc < 0 ? rc : len;
 897 }
 898
 899 resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
 900 {
 901         struct device *dev = &ndns->dev;
 902
 903         if (is_namespace_pmem(dev)) {
 904                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 905
 906                 return resource_size(&nspm->nsio.res);
 907         } else if (is_namespace_blk(dev)) {
 908                 return nd_namespace_blk_size(to_nd_namespace_blk(dev));
 909         } else if (is_namespace_io(dev)) {
 910                 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
 911
 912                 return resource_size(&nsio->res);
 913         } else
 914                 WARN_ONCE(1, "unknown namespace type\n");
 915         return 0;
 916 }
 917
 918 resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
 919 {
 920         resource_size_t size;
 921
 922         nvdimm_bus_lock(&ndns->dev);
 923         size = __nvdimm_namespace_capacity(ndns);
 924         nvdimm_bus_unlock(&ndns->dev);
 925
 926         return size;
 927 }
 928 EXPORT_SYMBOL(nvdimm_namespace_capacity);
 929
 930 static ssize_t size_show(struct device *dev,
 931                 struct device_attribute *attr, char *buf)
 932 {
 933         return sprintf(buf, "%llu\n", (unsigned long long)
 934                         nvdimm_namespace_capacity(to_ndns(dev)));
 935 }
 936 static DEVICE_ATTR(size, S_IRUGO, size_show, size_store);
 937
 938 static ssize_t uuid_show(struct device *dev,
 939                 struct device_attribute *attr, char *buf)
 940 {
 941         u8 *uuid;
 942
 943         if (is_namespace_pmem(dev)) {
 944                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 945
 946                 uuid = nspm->uuid;
 947         } else if (is_namespace_blk(dev)) {
 948                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 949
 950                 uuid = nsblk->uuid;
 951         } else
 952                 return -ENXIO;
 953
 954         if (uuid)
 955                 return sprintf(buf, "%pUb\n", uuid);
 956         return sprintf(buf, "\n");
 957 }
 958
 959 /**
 960  * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
 961  * @nd_region: parent region so we can updates all dimms in the set
 962  * @dev: namespace type for generating label_id
 963  * @new_uuid: incoming uuid
 964  * @old_uuid: reference to the uuid storage location in the namespace object
 965  */
 966 static int namespace_update_uuid(struct nd_region *nd_region,
 967                 struct device *dev, u8 *new_uuid, u8 **old_uuid)
 968 {
 969         u32 flags = is_namespace_blk(dev) ? NSLABEL_FLAG_LOCAL : 0;
 970         struct nd_label_id old_label_id;
 971         struct nd_label_id new_label_id;
 972         int i;
 973
 974         if (!nd_is_uuid_unique(dev, new_uuid))
 975                 return -EINVAL;
 976
 977         if (*old_uuid == NULL)
 978                 goto out;
 979
 980         /*
 981          * If we've already written a label with this uuid, then it's
 982          * too late to rename because we can't reliably update the uuid
 983          * without losing the old namespace.  Userspace must delete this
 984          * namespace to abandon the old uuid.
 985          */
 986         for (i = 0; i < nd_region->ndr_mappings; i++) {
 987                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 988
 989                 /*
 990                  * This check by itself is sufficient because old_uuid
 991                  * would be NULL above if this uuid did not exist in the
 992                  * currently written set.
 993                  *
 994                  * FIXME: can we delete uuid with zero dpa allocated?
 995                  */
 996                 if (nd_mapping->labels)
 997                         return -EBUSY;
 998         }
 999
1000         nd_label_gen_id(&old_label_id, *old_uuid, flags);
1001         nd_label_gen_id(&new_label_id, new_uuid, flags);
1002         for (i = 0; i < nd_region->ndr_mappings; i++) {
1003                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1004                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1005                 struct resource *res;
1006
1007                 for_each_dpa_resource(ndd, res)
1008                         if (strcmp(res->name, old_label_id.id) == 0)
1009                                 sprintf((void *) res->name, "%s",
1010                                                 new_label_id.id);
1011         }
1012         kfree(*old_uuid);
1013  out:
1014         *old_uuid = new_uuid;
1015         return 0;
1016 }
1017
1018 static ssize_t uuid_store(struct device *dev,
1019                 struct device_attribute *attr, const char *buf, size_t len)
1020 {
1021         struct nd_region *nd_region = to_nd_region(dev->parent);
1022         u8 *uuid = NULL;
1023         ssize_t rc = 0;
1024         u8 **ns_uuid;
1025
1026         if (is_namespace_pmem(dev)) {
1027                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1028
1029                 ns_uuid = &nspm->uuid;
1030         } else if (is_namespace_blk(dev)) {
1031                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1032
1033                 ns_uuid = &nsblk->uuid;
1034         } else
1035                 return -ENXIO;
1036
1037         device_lock(dev);
1038         nvdimm_bus_lock(dev);
1039         wait_nvdimm_bus_probe_idle(dev);
1040         if (to_ndns(dev)->claim)
1041                 rc = -EBUSY;
1042         if (rc >= 0)
1043                 rc = nd_uuid_store(dev, &uuid, buf, len);
1044         if (rc >= 0)
1045                 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
1046         if (rc >= 0)
1047                 rc = nd_namespace_label_update(nd_region, dev);
1048         else
1049                 kfree(uuid);
1050         dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
1051                         rc, buf, buf[len - 1] == '\n' ? "" : "\n");
1052         nvdimm_bus_unlock(dev);
1053         device_unlock(dev);
1054
1055         return rc < 0 ? rc : len;
1056 }
1057 static DEVICE_ATTR_RW(uuid);
1058
1059 static ssize_t resource_show(struct device *dev,
1060                 struct device_attribute *attr, char *buf)
1061 {
1062         struct resource *res;
1063
1064         if (is_namespace_pmem(dev)) {
1065                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1066
1067                 res = &nspm->nsio.res;
1068         } else if (is_namespace_io(dev)) {
1069                 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1070
1071                 res = &nsio->res;
1072         } else
1073                 return -ENXIO;
1074
1075         /* no address to convey if the namespace has no allocation */
1076         if (resource_size(res) == 0)
1077                 return -ENXIO;
1078         return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
1079 }
1080 static DEVICE_ATTR_RO(resource);
1081
1082 static const unsigned long ns_lbasize_supported[] = { 512, 520, 528,
1083         4096, 4104, 4160, 4224, 0 };
1084
1085 static ssize_t sector_size_show(struct device *dev,
1086                 struct device_attribute *attr, char *buf)
1087 {
1088         struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1089
1090         if (!is_namespace_blk(dev))
1091                 return -ENXIO;
1092
1093         return nd_sector_size_show(nsblk->lbasize, ns_lbasize_supported, buf);
1094 }
1095
1096 static ssize_t sector_size_store(struct device *dev,
1097                 struct device_attribute *attr, const char *buf, size_t len)
1098 {
1099         struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1100         struct nd_region *nd_region = to_nd_region(dev->parent);
1101         ssize_t rc = 0;
1102
1103         if (!is_namespace_blk(dev))
1104                 return -ENXIO;
1105
1106         device_lock(dev);
1107         nvdimm_bus_lock(dev);
1108         if (to_ndns(dev)->claim)
1109                 rc = -EBUSY;
1110         if (rc >= 0)
1111                 rc = nd_sector_size_store(dev, buf, &nsblk->lbasize,
1112                                 ns_lbasize_supported);
1113         if (rc >= 0)
1114                 rc = nd_namespace_label_update(nd_region, dev);
1115         dev_dbg(dev, "%s: result: %zd %s: %s%s", __func__,
1116                         rc, rc < 0 ? "tried" : "wrote", buf,
1117                         buf[len - 1] == '\n' ? "" : "\n");
1118         nvdimm_bus_unlock(dev);
1119         device_unlock(dev);
1120
1121         return rc ? rc : len;
1122 }
1123 static DEVICE_ATTR_RW(sector_size);
1124
1125 static ssize_t dpa_extents_show(struct device *dev,
1126                 struct device_attribute *attr, char *buf)
1127 {
1128         struct nd_region *nd_region = to_nd_region(dev->parent);
1129         struct nd_label_id label_id;
1130         int count = 0, i;
1131         u8 *uuid = NULL;
1132         u32 flags = 0;
1133
1134         nvdimm_bus_lock(dev);
1135         if (is_namespace_pmem(dev)) {
1136                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1137
1138                 uuid = nspm->uuid;
1139                 flags = 0;
1140         } else if (is_namespace_blk(dev)) {
1141                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1142
1143                 uuid = nsblk->uuid;
1144                 flags = NSLABEL_FLAG_LOCAL;
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 ssize_t holder_show(struct device *dev,
1168                 struct device_attribute *attr, char *buf)
1169 {
1170         struct nd_namespace_common *ndns = to_ndns(dev);
1171         ssize_t rc;
1172
1173         device_lock(dev);
1174         rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
1175         device_unlock(dev);
1176
1177         return rc;
1178 }
1179 static DEVICE_ATTR_RO(holder);
1180
1181 static ssize_t force_raw_store(struct device *dev,
1182                 struct device_attribute *attr, const char *buf, size_t len)
1183 {
1184         bool force_raw;
1185         int rc = strtobool(buf, &force_raw);
1186
1187         if (rc)
1188                 return rc;
1189
1190         to_ndns(dev)->force_raw = force_raw;
1191         return len;
1192 }
1193
1194 static ssize_t force_raw_show(struct device *dev,
1195                 struct device_attribute *attr, char *buf)
1196 {
1197         return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
1198 }
1199 static DEVICE_ATTR_RW(force_raw);
1200
1201 static struct attribute *nd_namespace_attributes[] = {
1202         &dev_attr_nstype.attr,
1203         &dev_attr_size.attr,
1204         &dev_attr_uuid.attr,
1205         &dev_attr_holder.attr,
1206         &dev_attr_resource.attr,
1207         &dev_attr_alt_name.attr,
1208         &dev_attr_force_raw.attr,
1209         &dev_attr_sector_size.attr,
1210         &dev_attr_dpa_extents.attr,
1211         NULL,
1212 };
1213
1214 static umode_t namespace_visible(struct kobject *kobj,
1215                 struct attribute *a, int n)
1216 {
1217         struct device *dev = container_of(kobj, struct device, kobj);
1218
1219         if (a == &dev_attr_resource.attr) {
1220                 if (is_namespace_blk(dev))
1221                         return 0;
1222                 return a->mode;
1223         }
1224
1225         if (is_namespace_pmem(dev) || is_namespace_blk(dev)) {
1226                 if (a == &dev_attr_size.attr)
1227                         return S_IWUSR | S_IRUGO;
1228
1229                 if (is_namespace_pmem(dev) && a == &dev_attr_sector_size.attr)
1230                         return 0;
1231
1232                 return a->mode;
1233         }
1234
1235         if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr
1236                         || a == &dev_attr_holder.attr
1237                         || a == &dev_attr_force_raw.attr)
1238                 return a->mode;
1239
1240         return 0;
1241 }
1242
1243 static struct attribute_group nd_namespace_attribute_group = {
1244         .attrs = nd_namespace_attributes,
1245         .is_visible = namespace_visible,
1246 };
1247
1248 static const struct attribute_group *nd_namespace_attribute_groups[] = {
1249         &nd_device_attribute_group,
1250         &nd_namespace_attribute_group,
1251         &nd_numa_attribute_group,
1252         NULL,
1253 };
1254
1255 struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
1256 {
1257         struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
1258         struct nd_namespace_common *ndns;
1259         resource_size_t size;
1260
1261         if (nd_btt) {
1262                 ndns = nd_btt->ndns;
1263                 if (!ndns)
1264                         return ERR_PTR(-ENODEV);
1265
1266                 /*
1267                  * Flush any in-progess probes / removals in the driver
1268                  * for the raw personality of this namespace.
1269                  */
1270                 device_lock(&ndns->dev);
1271                 device_unlock(&ndns->dev);
1272                 if (ndns->dev.driver) {
1273                         dev_dbg(&ndns->dev, "is active, can't bind %s\n",
1274                                         dev_name(&nd_btt->dev));
1275                         return ERR_PTR(-EBUSY);
1276                 }
1277                 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != &nd_btt->dev,
1278                                         "host (%s) vs claim (%s) mismatch\n",
1279                                         dev_name(&nd_btt->dev),
1280                                         dev_name(ndns->claim)))
1281                         return ERR_PTR(-ENXIO);
1282         } else {
1283                 ndns = to_ndns(dev);
1284                 if (ndns->claim) {
1285                         dev_dbg(dev, "claimed by %s, failing probe\n",
1286                                 dev_name(ndns->claim));
1287
1288                         return ERR_PTR(-ENXIO);
1289                 }
1290         }
1291
1292         size = nvdimm_namespace_capacity(ndns);
1293         if (size < ND_MIN_NAMESPACE_SIZE) {
1294                 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
1295                                 &size, ND_MIN_NAMESPACE_SIZE);
1296                 return ERR_PTR(-ENODEV);
1297         }
1298
1299         if (is_namespace_pmem(&ndns->dev)) {
1300                 struct nd_namespace_pmem *nspm;
1301
1302                 nspm = to_nd_namespace_pmem(&ndns->dev);
1303                 if (!nspm->uuid) {
1304                         dev_dbg(&ndns->dev, "%s: uuid not set\n", __func__);
1305                         return ERR_PTR(-ENODEV);
1306                 }
1307         } else if (is_namespace_blk(&ndns->dev)) {
1308                 struct nd_namespace_blk *nsblk;
1309
1310                 nsblk = to_nd_namespace_blk(&ndns->dev);
1311                 if (!nd_namespace_blk_validate(nsblk))
1312                         return ERR_PTR(-ENODEV);
1313         }
1314
1315         return ndns;
1316 }
1317 EXPORT_SYMBOL(nvdimm_namespace_common_probe);
1318
1319 static struct device **create_namespace_io(struct nd_region *nd_region)
1320 {
1321         struct nd_namespace_io *nsio;
1322         struct device *dev, **devs;
1323         struct resource *res;
1324
1325         nsio = kzalloc(sizeof(*nsio), GFP_KERNEL);
1326         if (!nsio)
1327                 return NULL;
1328
1329         devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1330         if (!devs) {
1331                 kfree(nsio);
1332                 return NULL;
1333         }
1334
1335         dev = &nsio->common.dev;
1336         dev->type = &namespace_io_device_type;
1337         dev->parent = &nd_region->dev;
1338         res = &nsio->res;
1339         res->name = dev_name(&nd_region->dev);
1340         res->flags = IORESOURCE_MEM;
1341         res->start = nd_region->ndr_start;
1342         res->end = res->start + nd_region->ndr_size - 1;
1343
1344         devs[0] = dev;
1345         return devs;
1346 }
1347
1348 static bool has_uuid_at_pos(struct nd_region *nd_region, u8 *uuid,
1349                 u64 cookie, u16 pos)
1350 {
1351         struct nd_namespace_label *found = NULL;
1352         int i;
1353
1354         for (i = 0; i < nd_region->ndr_mappings; i++) {
1355                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1356                 struct nd_namespace_label *nd_label;
1357                 bool found_uuid = false;
1358                 int l;
1359
1360                 for_each_label(l, nd_label, nd_mapping->labels) {
1361                         u64 isetcookie = __le64_to_cpu(nd_label->isetcookie);
1362                         u16 position = __le16_to_cpu(nd_label->position);
1363                         u16 nlabel = __le16_to_cpu(nd_label->nlabel);
1364
1365                         if (isetcookie != cookie)
1366                                 continue;
1367
1368                         if (memcmp(nd_label->uuid, uuid, NSLABEL_UUID_LEN) != 0)
1369                                 continue;
1370
1371                         if (found_uuid) {
1372                                 dev_dbg(to_ndd(nd_mapping)->dev,
1373                                                 "%s duplicate entry for uuid\n",
1374                                                 __func__);
1375                                 return false;
1376                         }
1377                         found_uuid = true;
1378                         if (nlabel != nd_region->ndr_mappings)
1379                                 continue;
1380                         if (position != pos)
1381                                 continue;
1382                         found = nd_label;
1383                         break;
1384                 }
1385                 if (found)
1386                         break;
1387         }
1388         return found != NULL;
1389 }
1390
1391 static int select_pmem_id(struct nd_region *nd_region, u8 *pmem_id)
1392 {
1393         struct nd_namespace_label *select = NULL;
1394         int i;
1395
1396         if (!pmem_id)
1397                 return -ENODEV;
1398
1399         for (i = 0; i < nd_region->ndr_mappings; i++) {
1400                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1401                 struct nd_namespace_label *nd_label;
1402                 u64 hw_start, hw_end, pmem_start, pmem_end;
1403                 int l;
1404
1405                 for_each_label(l, nd_label, nd_mapping->labels)
1406                         if (memcmp(nd_label->uuid, pmem_id, NSLABEL_UUID_LEN) == 0)
1407                                 break;
1408
1409                 if (!nd_label) {
1410                         WARN_ON(1);
1411                         return -EINVAL;
1412                 }
1413
1414                 select = nd_label;
1415                 /*
1416                  * Check that this label is compliant with the dpa
1417                  * range published in NFIT
1418                  */
1419                 hw_start = nd_mapping->start;
1420                 hw_end = hw_start + nd_mapping->size;
1421                 pmem_start = __le64_to_cpu(select->dpa);
1422                 pmem_end = pmem_start + __le64_to_cpu(select->rawsize);
1423                 if (pmem_start == hw_start && pmem_end <= hw_end)
1424                         /* pass */;
1425                 else
1426                         return -EINVAL;
1427
1428                 nd_mapping->labels[0] = select;
1429                 nd_mapping->labels[1] = NULL;
1430         }
1431         return 0;
1432 }
1433
1434 /**
1435  * find_pmem_label_set - validate interleave set labelling, retrieve label0
1436  * @nd_region: region with mappings to validate
1437  */
1438 static int find_pmem_label_set(struct nd_region *nd_region,
1439                 struct nd_namespace_pmem *nspm)
1440 {
1441         u64 cookie = nd_region_interleave_set_cookie(nd_region);
1442         struct nd_namespace_label *nd_label;
1443         u8 select_id[NSLABEL_UUID_LEN];
1444         resource_size_t size = 0;
1445         u8 *pmem_id = NULL;
1446         int rc = -ENODEV, l;
1447         u16 i;
1448
1449         if (cookie == 0)
1450                 return -ENXIO;
1451
1452         /*
1453          * Find a complete set of labels by uuid.  By definition we can start
1454          * with any mapping as the reference label
1455          */
1456         for_each_label(l, nd_label, nd_region->mapping[0].labels) {
1457                 u64 isetcookie = __le64_to_cpu(nd_label->isetcookie);
1458
1459                 if (isetcookie != cookie)
1460                         continue;
1461
1462                 for (i = 0; nd_region->ndr_mappings; i++)
1463                         if (!has_uuid_at_pos(nd_region, nd_label->uuid,
1464                                                 cookie, i))
1465                                 break;
1466                 if (i < nd_region->ndr_mappings) {
1467                         /*
1468                          * Give up if we don't find an instance of a
1469                          * uuid at each position (from 0 to
1470                          * nd_region->ndr_mappings - 1), or if we find a
1471                          * dimm with two instances of the same uuid.
1472                          */
1473                         rc = -EINVAL;
1474                         goto err;
1475                 } else if (pmem_id) {
1476                         /*
1477                          * If there is more than one valid uuid set, we
1478                          * need userspace to clean this up.
1479                          */
1480                         rc = -EBUSY;
1481                         goto err;
1482                 }
1483                 memcpy(select_id, nd_label->uuid, NSLABEL_UUID_LEN);
1484                 pmem_id = select_id;
1485         }
1486
1487         /*
1488          * Fix up each mapping's 'labels' to have the validated pmem label for
1489          * that position at labels[0], and NULL at labels[1].  In the process,
1490          * check that the namespace aligns with interleave-set.  We know
1491          * that it does not overlap with any blk namespaces by virtue of
1492          * the dimm being enabled (i.e. nd_label_reserve_dpa()
1493          * succeeded).
1494          */
1495         rc = select_pmem_id(nd_region, pmem_id);
1496         if (rc)
1497                 goto err;
1498
1499         /* Calculate total size and populate namespace properties from label0 */
1500         for (i = 0; i < nd_region->ndr_mappings; i++) {
1501                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1502                 struct nd_namespace_label *label0 = nd_mapping->labels[0];
1503
1504                 size += __le64_to_cpu(label0->rawsize);
1505                 if (__le16_to_cpu(label0->position) != 0)
1506                         continue;
1507                 WARN_ON(nspm->alt_name || nspm->uuid);
1508                 nspm->alt_name = kmemdup((void __force *) label0->name,
1509                                 NSLABEL_NAME_LEN, GFP_KERNEL);
1510                 nspm->uuid = kmemdup((void __force *) label0->uuid,
1511                                 NSLABEL_UUID_LEN, GFP_KERNEL);
1512         }
1513
1514         if (!nspm->alt_name || !nspm->uuid) {
1515                 rc = -ENOMEM;
1516                 goto err;
1517         }
1518
1519         nd_namespace_pmem_set_size(nd_region, nspm, size);
1520
1521         return 0;
1522  err:
1523         switch (rc) {
1524         case -EINVAL:
1525                 dev_dbg(&nd_region->dev, "%s: invalid label(s)\n", __func__);
1526                 break;
1527         case -ENODEV:
1528                 dev_dbg(&nd_region->dev, "%s: label not found\n", __func__);
1529                 break;
1530         default:
1531                 dev_dbg(&nd_region->dev, "%s: unexpected err: %d\n",
1532                                 __func__, rc);
1533                 break;
1534         }
1535         return rc;
1536 }
1537
1538 static struct device **create_namespace_pmem(struct nd_region *nd_region)
1539 {
1540         struct nd_namespace_pmem *nspm;
1541         struct device *dev, **devs;
1542         struct resource *res;
1543         int rc;
1544
1545         nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1546         if (!nspm)
1547                 return NULL;
1548
1549         dev = &nspm->nsio.common.dev;
1550         dev->type = &namespace_pmem_device_type;
1551         dev->parent = &nd_region->dev;
1552         res = &nspm->nsio.res;
1553         res->name = dev_name(&nd_region->dev);
1554         res->flags = IORESOURCE_MEM;
1555         rc = find_pmem_label_set(nd_region, nspm);
1556         if (rc == -ENODEV) {
1557                 int i;
1558
1559                 /* Pass, try to permit namespace creation... */
1560                 for (i = 0; i < nd_region->ndr_mappings; i++) {
1561                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1562
1563                         kfree(nd_mapping->labels);
1564                         nd_mapping->labels = NULL;
1565                 }
1566
1567                 /* Publish a zero-sized namespace for userspace to configure. */
1568                 nd_namespace_pmem_set_size(nd_region, nspm, 0);
1569
1570                 rc = 0;
1571         } else if (rc)
1572                 goto err;
1573
1574         devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1575         if (!devs)
1576                 goto err;
1577
1578         devs[0] = dev;
1579         return devs;
1580
1581  err:
1582         namespace_pmem_release(&nspm->nsio.common.dev);
1583         return NULL;
1584 }
1585
1586 struct resource *nsblk_add_resource(struct nd_region *nd_region,
1587                 struct nvdimm_drvdata *ndd, struct nd_namespace_blk *nsblk,
1588                 resource_size_t start)
1589 {
1590         struct nd_label_id label_id;
1591         struct resource *res;
1592
1593         nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
1594         res = krealloc(nsblk->res,
1595                         sizeof(void *) * (nsblk->num_resources + 1),
1596                         GFP_KERNEL);
1597         if (!res)
1598                 return NULL;
1599         nsblk->res = (struct resource **) res;
1600         for_each_dpa_resource(ndd, res)
1601                 if (strcmp(res->name, label_id.id) == 0
1602                                 && res->start == start) {
1603                         nsblk->res[nsblk->num_resources++] = res;
1604                         return res;
1605                 }
1606         return NULL;
1607 }
1608
1609 static struct device *nd_namespace_blk_create(struct nd_region *nd_region)
1610 {
1611         struct nd_namespace_blk *nsblk;
1612         struct device *dev;
1613
1614         if (!is_nd_blk(&nd_region->dev))
1615                 return NULL;
1616
1617         nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
1618         if (!nsblk)
1619                 return NULL;
1620
1621         dev = &nsblk->common.dev;
1622         dev->type = &namespace_blk_device_type;
1623         nsblk->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
1624         if (nsblk->id < 0) {
1625                 kfree(nsblk);
1626                 return NULL;
1627         }
1628         dev_set_name(dev, "namespace%d.%d", nd_region->id, nsblk->id);
1629         dev->parent = &nd_region->dev;
1630         dev->groups = nd_namespace_attribute_groups;
1631
1632         return &nsblk->common.dev;
1633 }
1634
1635 void nd_region_create_blk_seed(struct nd_region *nd_region)
1636 {
1637         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1638         nd_region->ns_seed = nd_namespace_blk_create(nd_region);
1639         /*
1640          * Seed creation failures are not fatal, provisioning is simply
1641          * disabled until memory becomes available
1642          */
1643         if (!nd_region->ns_seed)
1644                 dev_err(&nd_region->dev, "failed to create blk namespace\n");
1645         else
1646                 nd_device_register(nd_region->ns_seed);
1647 }
1648
1649 void nd_region_create_btt_seed(struct nd_region *nd_region)
1650 {
1651         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1652         nd_region->btt_seed = nd_btt_create(nd_region);
1653         /*
1654          * Seed creation failures are not fatal, provisioning is simply
1655          * disabled until memory becomes available
1656          */
1657         if (!nd_region->btt_seed)
1658                 dev_err(&nd_region->dev, "failed to create btt namespace\n");
1659 }
1660
1661 static struct device **create_namespace_blk(struct nd_region *nd_region)
1662 {
1663         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
1664         struct nd_namespace_label *nd_label;
1665         struct device *dev, **devs = NULL;
1666         struct nd_namespace_blk *nsblk;
1667         struct nvdimm_drvdata *ndd;
1668         int i, l, count = 0;
1669         struct resource *res;
1670
1671         if (nd_region->ndr_mappings == 0)
1672                 return NULL;
1673
1674         ndd = to_ndd(nd_mapping);
1675         for_each_label(l, nd_label, nd_mapping->labels) {
1676                 u32 flags = __le32_to_cpu(nd_label->flags);
1677                 char *name[NSLABEL_NAME_LEN];
1678                 struct device **__devs;
1679
1680                 if (flags & NSLABEL_FLAG_LOCAL)
1681                         /* pass */;
1682                 else
1683                         continue;
1684
1685                 for (i = 0; i < count; i++) {
1686                         nsblk = to_nd_namespace_blk(devs[i]);
1687                         if (memcmp(nsblk->uuid, nd_label->uuid,
1688                                                 NSLABEL_UUID_LEN) == 0) {
1689                                 res = nsblk_add_resource(nd_region, ndd, nsblk,
1690                                                 __le64_to_cpu(nd_label->dpa));
1691                                 if (!res)
1692                                         goto err;
1693                                 nd_dbg_dpa(nd_region, ndd, res, "%s assign\n",
1694                                         dev_name(&nsblk->common.dev));
1695                                 break;
1696                         }
1697                 }
1698                 if (i < count)
1699                         continue;
1700                 __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
1701                 if (!__devs)
1702                         goto err;
1703                 memcpy(__devs, devs, sizeof(dev) * count);
1704                 kfree(devs);
1705                 devs = __devs;
1706
1707                 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
1708                 if (!nsblk)
1709                         goto err;
1710                 dev = &nsblk->common.dev;
1711                 dev->type = &namespace_blk_device_type;
1712                 dev->parent = &nd_region->dev;
1713                 dev_set_name(dev, "namespace%d.%d", nd_region->id, count);
1714                 devs[count++] = dev;
1715                 nsblk->id = -1;
1716                 nsblk->lbasize = __le64_to_cpu(nd_label->lbasize);
1717                 nsblk->uuid = kmemdup(nd_label->uuid, NSLABEL_UUID_LEN,
1718                                 GFP_KERNEL);
1719                 if (!nsblk->uuid)
1720                         goto err;
1721                 memcpy(name, nd_label->name, NSLABEL_NAME_LEN);
1722                 if (name[0])
1723                         nsblk->alt_name = kmemdup(name, NSLABEL_NAME_LEN,
1724                                         GFP_KERNEL);
1725                 res = nsblk_add_resource(nd_region, ndd, nsblk,
1726                                 __le64_to_cpu(nd_label->dpa));
1727                 if (!res)
1728                         goto err;
1729                 nd_dbg_dpa(nd_region, ndd, res, "%s assign\n",
1730                                 dev_name(&nsblk->common.dev));
1731         }
1732
1733         dev_dbg(&nd_region->dev, "%s: discovered %d blk namespace%s\n",
1734                         __func__, count, count == 1 ? "" : "s");
1735
1736         if (count == 0) {
1737                 /* Publish a zero-sized namespace for userspace to configure. */
1738                 for (i = 0; i < nd_region->ndr_mappings; i++) {
1739                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1740
1741                         kfree(nd_mapping->labels);
1742                         nd_mapping->labels = NULL;
1743                 }
1744
1745                 devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
1746                 if (!devs)
1747                         goto err;
1748                 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
1749                 if (!nsblk)
1750                         goto err;
1751                 dev = &nsblk->common.dev;
1752                 dev->type = &namespace_blk_device_type;
1753                 dev->parent = &nd_region->dev;
1754                 devs[count++] = dev;
1755         }
1756
1757         return devs;
1758
1759 err:
1760         for (i = 0; i < count; i++) {
1761                 nsblk = to_nd_namespace_blk(devs[i]);
1762                 namespace_blk_release(&nsblk->common.dev);
1763         }
1764         kfree(devs);
1765         return NULL;
1766 }
1767
1768 static int init_active_labels(struct nd_region *nd_region)
1769 {
1770         int i;
1771
1772         for (i = 0; i < nd_region->ndr_mappings; i++) {
1773                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1774                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1775                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1776                 int count, j;
1777
1778                 /*
1779                  * If the dimm is disabled then prevent the region from
1780                  * being activated if it aliases DPA.
1781                  */
1782                 if (!ndd) {
1783                         if ((nvdimm->flags & NDD_ALIASING) == 0)
1784                                 return 0;
1785                         dev_dbg(&nd_region->dev, "%s: is disabled, failing probe\n",
1786                                         dev_name(&nd_mapping->nvdimm->dev));
1787                         return -ENXIO;
1788                 }
1789                 nd_mapping->ndd = ndd;
1790                 atomic_inc(&nvdimm->busy);
1791                 get_ndd(ndd);
1792
1793                 count = nd_label_active_count(ndd);
1794                 dev_dbg(ndd->dev, "%s: %d\n", __func__, count);
1795                 if (!count)
1796                         continue;
1797                 nd_mapping->labels = kcalloc(count + 1, sizeof(void *),
1798                                 GFP_KERNEL);
1799                 if (!nd_mapping->labels)
1800                         return -ENOMEM;
1801                 for (j = 0; j < count; j++) {
1802                         struct nd_namespace_label *label;
1803
1804                         label = nd_label_active(ndd, j);
1805                         nd_mapping->labels[j] = label;
1806                 }
1807         }
1808
1809         return 0;
1810 }
1811
1812 int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
1813 {
1814         struct device **devs = NULL;
1815         int i, rc = 0, type;
1816
1817         *err = 0;
1818         nvdimm_bus_lock(&nd_region->dev);
1819         rc = init_active_labels(nd_region);
1820         if (rc) {
1821                 nvdimm_bus_unlock(&nd_region->dev);
1822                 return rc;
1823         }
1824
1825         type = nd_region_to_nstype(nd_region);
1826         switch (type) {
1827         case ND_DEVICE_NAMESPACE_IO:
1828                 devs = create_namespace_io(nd_region);
1829                 break;
1830         case ND_DEVICE_NAMESPACE_PMEM:
1831                 devs = create_namespace_pmem(nd_region);
1832                 break;
1833         case ND_DEVICE_NAMESPACE_BLK:
1834                 devs = create_namespace_blk(nd_region);
1835                 break;
1836         default:
1837                 break;
1838         }
1839         nvdimm_bus_unlock(&nd_region->dev);
1840
1841         if (!devs)
1842                 return -ENODEV;
1843
1844         for (i = 0; devs[i]; i++) {
1845                 struct device *dev = devs[i];
1846                 int id;
1847
1848                 if (type == ND_DEVICE_NAMESPACE_BLK) {
1849                         struct nd_namespace_blk *nsblk;
1850
1851                         nsblk = to_nd_namespace_blk(dev);
1852                         id = ida_simple_get(&nd_region->ns_ida, 0, 0,
1853                                         GFP_KERNEL);
1854                         nsblk->id = id;
1855                 } else
1856                         id = i;
1857
1858                 if (id < 0)
1859                         break;
1860                 dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
1861                 dev->groups = nd_namespace_attribute_groups;
1862                 nd_device_register(dev);
1863         }
1864         if (i)
1865                 nd_region->ns_seed = devs[0];
1866
1867         if (devs[i]) {
1868                 int j;
1869
1870                 for (j = i; devs[j]; j++) {
1871                         struct device *dev = devs[j];
1872
1873                         device_initialize(dev);
1874                         put_device(dev);
1875                 }
1876                 *err = j - i;
1877                 /*
1878                  * All of the namespaces we tried to register failed, so
1879                  * fail region activation.
1880                  */
1881                 if (*err == 0)
1882                         rc = -ENODEV;
1883         }
1884         kfree(devs);
1885
1886         if (rc == -ENODEV)
1887                 return rc;
1888
1889         return i;
1890 }