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/pmem.h>
  17 #include <linux/nd.h>
  18 #include "nd-core.h"
  19 #include "nd.h"
  20
  21 static void namespace_io_release(struct device *dev)
  22 {
  23         struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
  24
  25         kfree(nsio);
  26 }
  27
  28 static void namespace_pmem_release(struct device *dev)
  29 {
  30         struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
  31
  32         kfree(nspm->alt_name);
  33         kfree(nspm->uuid);
  34         kfree(nspm);
  35 }
  36
  37 static void namespace_blk_release(struct device *dev)
  38 {
  39         struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
  40         struct nd_region *nd_region = to_nd_region(dev->parent);
  41
  42         if (nsblk->id >= 0)
  43                 ida_simple_remove(&nd_region->ns_ida, nsblk->id);
  44         kfree(nsblk->alt_name);
  45         kfree(nsblk->uuid);
  46         kfree(nsblk->res);
  47         kfree(nsblk);
  48 }
  49
  50 static struct device_type namespace_io_device_type = {
  51         .name = "nd_namespace_io",
  52         .release = namespace_io_release,
  53 };
  54
  55 static struct device_type namespace_pmem_device_type = {
  56         .name = "nd_namespace_pmem",
  57         .release = namespace_pmem_release,
  58 };
  59
  60 static struct device_type namespace_blk_device_type = {
  61         .name = "nd_namespace_blk",
  62         .release = namespace_blk_release,
  63 };
  64
  65 static bool is_namespace_pmem(struct device *dev)
  66 {
  67         return dev ? dev->type == &namespace_pmem_device_type : false;
  68 }
  69
  70 static bool is_namespace_blk(struct device *dev)
  71 {
  72         return dev ? dev->type == &namespace_blk_device_type : false;
  73 }
  74
  75 static bool is_namespace_io(struct device *dev)
  76 {
  77         return dev ? dev->type == &namespace_io_device_type : false;
  78 }
  79
  80 bool pmem_should_map_pages(struct device *dev)
  81 {
  82         struct nd_region *nd_region = to_nd_region(dev->parent);
  83
  84         if (!IS_ENABLED(CONFIG_ZONE_DEVICE))
  85                 return false;
  86
  87         if (!test_bit(ND_REGION_PAGEMAP, &nd_region->flags))
  88                 return false;
  89
  90         if (is_nd_pfn(dev) || is_nd_btt(dev))
  91                 return false;
  92
  93 #ifdef ARCH_MEMREMAP_PMEM
  94         return ARCH_MEMREMAP_PMEM == MEMREMAP_WB;
  95 #else
  96         return false;
  97 #endif
  98 }
  99 EXPORT_SYMBOL(pmem_should_map_pages);
 100
 101 const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,
 102                 char *name)
 103 {
 104         struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
 105         const char *suffix = NULL;
 106
 107         if (ndns->claim) {
 108                 if (is_nd_btt(ndns->claim))
 109                         suffix = "s";
 110                 else if (is_nd_pfn(ndns->claim))
 111                         suffix = "m";
 112                 else
 113                         dev_WARN_ONCE(&ndns->dev, 1,
 114                                         "unknown claim type by %s\n",
 115                                         dev_name(ndns->claim));
 116         }
 117
 118         if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev)) {
 119                 if (!suffix && pmem_should_map_pages(&ndns->dev))
 120                         suffix = "m";
 121                 sprintf(name, "pmem%d%s", nd_region->id, suffix ? suffix : "");
 122         } else if (is_namespace_blk(&ndns->dev)) {
 123                 struct nd_namespace_blk *nsblk;
 124
 125                 nsblk = to_nd_namespace_blk(&ndns->dev);
 126                 sprintf(name, "ndblk%d.%d%s", nd_region->id, nsblk->id,
 127                                 suffix ? suffix : "");
 128         } else {
 129                 return NULL;
 130         }
 131
 132         return name;
 133 }
 134 EXPORT_SYMBOL(nvdimm_namespace_disk_name);
 135
 136 const u8 *nd_dev_to_uuid(struct device *dev)
 137 {
 138         static const u8 null_uuid[16];
 139
 140         if (!dev)
 141                 return null_uuid;
 142
 143         if (is_namespace_pmem(dev)) {
 144                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 145
 146                 return nspm->uuid;
 147         } else if (is_namespace_blk(dev)) {
 148                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 149
 150                 return nsblk->uuid;
 151         } else
 152                 return null_uuid;
 153 }
 154 EXPORT_SYMBOL(nd_dev_to_uuid);
 155
 156 static ssize_t nstype_show(struct device *dev,
 157                 struct device_attribute *attr, char *buf)
 158 {
 159         struct nd_region *nd_region = to_nd_region(dev->parent);
 160
 161         return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
 162 }
 163 static DEVICE_ATTR_RO(nstype);
 164
 165 static ssize_t __alt_name_store(struct device *dev, const char *buf,
 166                 const size_t len)
 167 {
 168         char *input, *pos, *alt_name, **ns_altname;
 169         ssize_t rc;
 170
 171         if (is_namespace_pmem(dev)) {
 172                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 173
 174                 ns_altname = &nspm->alt_name;
 175         } else if (is_namespace_blk(dev)) {
 176                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 177
 178                 ns_altname = &nsblk->alt_name;
 179         } else
 180                 return -ENXIO;
 181
 182         if (dev->driver || to_ndns(dev)->claim)
 183                 return -EBUSY;
 184
 185         input = kmemdup(buf, len + 1, GFP_KERNEL);
 186         if (!input)
 187                 return -ENOMEM;
 188
 189         input[len] = '\0';
 190         pos = strim(input);
 191         if (strlen(pos) + 1 > NSLABEL_NAME_LEN) {
 192                 rc = -EINVAL;
 193                 goto out;
 194         }
 195
 196         alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL);
 197         if (!alt_name) {
 198                 rc = -ENOMEM;
 199                 goto out;
 200         }
 201         kfree(*ns_altname);
 202         *ns_altname = alt_name;
 203         sprintf(*ns_altname, "%s", pos);
 204         rc = len;
 205
 206 out:
 207         kfree(input);
 208         return rc;
 209 }
 210
 211 static resource_size_t nd_namespace_blk_size(struct nd_namespace_blk *nsblk)
 212 {
 213         struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
 214         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 215         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 216         struct nd_label_id label_id;
 217         resource_size_t size = 0;
 218         struct resource *res;
 219
 220         if (!nsblk->uuid)
 221                 return 0;
 222         nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
 223         for_each_dpa_resource(ndd, res)
 224                 if (strcmp(res->name, label_id.id) == 0)
 225                         size += resource_size(res);
 226         return size;
 227 }
 228
 229 static bool __nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
 230 {
 231         struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
 232         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
 233         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 234         struct nd_label_id label_id;
 235         struct resource *res;
 236         int count, i;
 237
 238         if (!nsblk->uuid || !nsblk->lbasize || !ndd)
 239                 return false;
 240
 241         count = 0;
 242         nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
 243         for_each_dpa_resource(ndd, res) {
 244                 if (strcmp(res->name, label_id.id) != 0)
 245                         continue;
 246                 /*
 247                  * Resources with unacknoweldged adjustments indicate a
 248                  * failure to update labels
 249                  */
 250                 if (res->flags & DPA_RESOURCE_ADJUSTED)
 251                         return false;
 252                 count++;
 253         }
 254
 255         /* These values match after a successful label update */
 256         if (count != nsblk->num_resources)
 257                 return false;
 258
 259         for (i = 0; i < nsblk->num_resources; i++) {
 260                 struct resource *found = NULL;
 261
 262                 for_each_dpa_resource(ndd, res)
 263                         if (res == nsblk->res[i]) {
 264                                 found = res;
 265                                 break;
 266                         }
 267                 /* stale resource */
 268                 if (!found)
 269                         return false;
 270         }
 271
 272         return true;
 273 }
 274
 275 resource_size_t nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
 276 {
 277         resource_size_t size;
 278
 279         nvdimm_bus_lock(&nsblk->common.dev);
 280         size = __nd_namespace_blk_validate(nsblk);
 281         nvdimm_bus_unlock(&nsblk->common.dev);
 282
 283         return size;
 284 }
 285 EXPORT_SYMBOL(nd_namespace_blk_validate);
 286
 287
 288 static int nd_namespace_label_update(struct nd_region *nd_region,
 289                 struct device *dev)
 290 {
 291         dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim,
 292                         "namespace must be idle during label update\n");
 293         if (dev->driver || to_ndns(dev)->claim)
 294                 return 0;
 295
 296         /*
 297          * Only allow label writes that will result in a valid namespace
 298          * or deletion of an existing namespace.
 299          */
 300         if (is_namespace_pmem(dev)) {
 301                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 302                 resource_size_t size = resource_size(&nspm->nsio.res);
 303
 304                 if (size == 0 && nspm->uuid)
 305                         /* delete allocation */;
 306                 else if (!nspm->uuid)
 307                         return 0;
 308
 309                 return nd_pmem_namespace_label_update(nd_region, nspm, size);
 310         } else if (is_namespace_blk(dev)) {
 311                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 312                 resource_size_t size = nd_namespace_blk_size(nsblk);
 313
 314                 if (size == 0 && nsblk->uuid)
 315                         /* delete allocation */;
 316                 else if (!nsblk->uuid || !nsblk->lbasize)
 317                         return 0;
 318
 319                 return nd_blk_namespace_label_update(nd_region, nsblk, size);
 320         } else
 321                 return -ENXIO;
 322 }
 323
 324 static ssize_t alt_name_store(struct device *dev,
 325                 struct device_attribute *attr, const char *buf, size_t len)
 326 {
 327         struct nd_region *nd_region = to_nd_region(dev->parent);
 328         ssize_t rc;
 329
 330         device_lock(dev);
 331         nvdimm_bus_lock(dev);
 332         wait_nvdimm_bus_probe_idle(dev);
 333         rc = __alt_name_store(dev, buf, len);
 334         if (rc >= 0)
 335                 rc = nd_namespace_label_update(nd_region, dev);
 336         dev_dbg(dev, "%s: %s(%zd)\n", __func__, rc < 0 ? "fail " : "", rc);
 337         nvdimm_bus_unlock(dev);
 338         device_unlock(dev);
 339
 340         return rc < 0 ? rc : len;
 341 }
 342
 343 static ssize_t alt_name_show(struct device *dev,
 344                 struct device_attribute *attr, char *buf)
 345 {
 346         char *ns_altname;
 347
 348         if (is_namespace_pmem(dev)) {
 349                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 350
 351                 ns_altname = nspm->alt_name;
 352         } else if (is_namespace_blk(dev)) {
 353                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 354
 355                 ns_altname = nsblk->alt_name;
 356         } else
 357                 return -ENXIO;
 358
 359         return sprintf(buf, "%s\n", ns_altname ? ns_altname : "");
 360 }
 361 static DEVICE_ATTR_RW(alt_name);
 362
 363 static int scan_free(struct nd_region *nd_region,
 364                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
 365                 resource_size_t n)
 366 {
 367         bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
 368         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 369         int rc = 0;
 370
 371         while (n) {
 372                 struct resource *res, *last;
 373                 resource_size_t new_start;
 374
 375                 last = NULL;
 376                 for_each_dpa_resource(ndd, res)
 377                         if (strcmp(res->name, label_id->id) == 0)
 378                                 last = res;
 379                 res = last;
 380                 if (!res)
 381                         return 0;
 382
 383                 if (n >= resource_size(res)) {
 384                         n -= resource_size(res);
 385                         nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc);
 386                         nvdimm_free_dpa(ndd, res);
 387                         /* retry with last resource deleted */
 388                         continue;
 389                 }
 390
 391                 /*
 392                  * Keep BLK allocations relegated to high DPA as much as
 393                  * possible
 394                  */
 395                 if (is_blk)
 396                         new_start = res->start + n;
 397                 else
 398                         new_start = res->start;
 399
 400                 rc = adjust_resource(res, new_start, resource_size(res) - n);
 401                 if (rc == 0)
 402                         res->flags |= DPA_RESOURCE_ADJUSTED;
 403                 nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc);
 404                 break;
 405         }
 406
 407         return rc;
 408 }
 409
 410 /**
 411  * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
 412  * @nd_region: the set of dimms to reclaim @n bytes from
 413  * @label_id: unique identifier for the namespace consuming this dpa range
 414  * @n: number of bytes per-dimm to release
 415  *
 416  * Assumes resources are ordered.  Starting from the end try to
 417  * adjust_resource() the allocation to @n, but if @n is larger than the
 418  * allocation delete it and find the 'new' last allocation in the label
 419  * set.
 420  */
 421 static int shrink_dpa_allocation(struct nd_region *nd_region,
 422                 struct nd_label_id *label_id, resource_size_t n)
 423 {
 424         int i;
 425
 426         for (i = 0; i < nd_region->ndr_mappings; i++) {
 427                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 428                 int rc;
 429
 430                 rc = scan_free(nd_region, nd_mapping, label_id, n);
 431                 if (rc)
 432                         return rc;
 433         }
 434
 435         return 0;
 436 }
 437
 438 static resource_size_t init_dpa_allocation(struct nd_label_id *label_id,
 439                 struct nd_region *nd_region, struct nd_mapping *nd_mapping,
 440                 resource_size_t n)
 441 {
 442         bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
 443         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 444         resource_size_t first_dpa;
 445         struct resource *res;
 446         int rc = 0;
 447
 448         /* allocate blk from highest dpa first */
 449         if (is_blk)
 450                 first_dpa = nd_mapping->start + nd_mapping->size - n;
 451         else
 452                 first_dpa = nd_mapping->start;
 453
 454         /* first resource allocation for this label-id or dimm */
 455         res = nvdimm_allocate_dpa(ndd, label_id, first_dpa, n);
 456         if (!res)
 457                 rc = -EBUSY;
 458
 459         nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc);
 460         return rc ? n : 0;
 461 }
 462
 463 static bool space_valid(bool is_pmem, bool is_reserve,
 464                 struct nd_label_id *label_id, struct resource *res)
 465 {
 466         /*
 467          * For BLK-space any space is valid, for PMEM-space, it must be
 468          * contiguous with an existing allocation unless we are
 469          * reserving pmem.
 470          */
 471         if (is_reserve || !is_pmem)
 472                 return true;
 473         if (!res || strcmp(res->name, label_id->id) == 0)
 474                 return true;
 475         return false;
 476 }
 477
 478 enum alloc_loc {
 479         ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
 480 };
 481
 482 static resource_size_t scan_allocate(struct nd_region *nd_region,
 483                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
 484                 resource_size_t n)
 485 {
 486         resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1;
 487         bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0;
 488         bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
 489         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 490         const resource_size_t to_allocate = n;
 491         struct resource *res;
 492         int first;
 493
 494  retry:
 495         first = 0;
 496         for_each_dpa_resource(ndd, res) {
 497                 resource_size_t allocate, available = 0, free_start, free_end;
 498                 struct resource *next = res->sibling, *new_res = NULL;
 499                 enum alloc_loc loc = ALLOC_ERR;
 500                 const char *action;
 501                 int rc = 0;
 502
 503                 /* ignore resources outside this nd_mapping */
 504                 if (res->start > mapping_end)
 505                         continue;
 506                 if (res->end < nd_mapping->start)
 507                         continue;
 508
 509                 /* space at the beginning of the mapping */
 510                 if (!first++ && res->start > nd_mapping->start) {
 511                         free_start = nd_mapping->start;
 512                         available = res->start - free_start;
 513                         if (space_valid(is_pmem, is_reserve, label_id, NULL))
 514                                 loc = ALLOC_BEFORE;
 515                 }
 516
 517                 /* space between allocations */
 518                 if (!loc && next) {
 519                         free_start = res->start + resource_size(res);
 520                         free_end = min(mapping_end, next->start - 1);
 521                         if (space_valid(is_pmem, is_reserve, label_id, res)
 522                                         && free_start < free_end) {
 523                                 available = free_end + 1 - free_start;
 524                                 loc = ALLOC_MID;
 525                         }
 526                 }
 527
 528                 /* space at the end of the mapping */
 529                 if (!loc && !next) {
 530                         free_start = res->start + resource_size(res);
 531                         free_end = mapping_end;
 532                         if (space_valid(is_pmem, is_reserve, label_id, res)
 533                                         && free_start < free_end) {
 534                                 available = free_end + 1 - free_start;
 535                                 loc = ALLOC_AFTER;
 536                         }
 537                 }
 538
 539                 if (!loc || !available)
 540                         continue;
 541                 allocate = min(available, n);
 542                 switch (loc) {
 543                 case ALLOC_BEFORE:
 544                         if (strcmp(res->name, label_id->id) == 0) {
 545                                 /* adjust current resource up */
 546                                 if (is_pmem && !is_reserve)
 547                                         return n;
 548                                 rc = adjust_resource(res, res->start - allocate,
 549                                                 resource_size(res) + allocate);
 550                                 action = "cur grow up";
 551                         } else
 552                                 action = "allocate";
 553                         break;
 554                 case ALLOC_MID:
 555                         if (strcmp(next->name, label_id->id) == 0) {
 556                                 /* adjust next resource up */
 557                                 if (is_pmem && !is_reserve)
 558                                         return n;
 559                                 rc = adjust_resource(next, next->start
 560                                                 - allocate, resource_size(next)
 561                                                 + allocate);
 562                                 new_res = next;
 563                                 action = "next grow up";
 564                         } else if (strcmp(res->name, label_id->id) == 0) {
 565                                 action = "grow down";
 566                         } else
 567                                 action = "allocate";
 568                         break;
 569                 case ALLOC_AFTER:
 570                         if (strcmp(res->name, label_id->id) == 0)
 571                                 action = "grow down";
 572                         else
 573                                 action = "allocate";
 574                         break;
 575                 default:
 576                         return n;
 577                 }
 578
 579                 if (strcmp(action, "allocate") == 0) {
 580                         /* BLK allocate bottom up */
 581                         if (!is_pmem)
 582                                 free_start += available - allocate;
 583                         else if (!is_reserve && free_start != nd_mapping->start)
 584                                 return n;
 585
 586                         new_res = nvdimm_allocate_dpa(ndd, label_id,
 587                                         free_start, allocate);
 588                         if (!new_res)
 589                                 rc = -EBUSY;
 590                 } else if (strcmp(action, "grow down") == 0) {
 591                         /* adjust current resource down */
 592                         rc = adjust_resource(res, res->start, resource_size(res)
 593                                         + allocate);
 594                         if (rc == 0)
 595                                 res->flags |= DPA_RESOURCE_ADJUSTED;
 596                 }
 597
 598                 if (!new_res)
 599                         new_res = res;
 600
 601                 nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n",
 602                                 action, loc, rc);
 603
 604                 if (rc)
 605                         return n;
 606
 607                 n -= allocate;
 608                 if (n) {
 609                         /*
 610                          * Retry scan with newly inserted resources.
 611                          * For example, if we did an ALLOC_BEFORE
 612                          * insertion there may also have been space
 613                          * available for an ALLOC_AFTER insertion, so we
 614                          * need to check this same resource again
 615                          */
 616                         goto retry;
 617                 } else
 618                         return 0;
 619         }
 620
 621         /*
 622          * If we allocated nothing in the BLK case it may be because we are in
 623          * an initial "pmem-reserve pass".  Only do an initial BLK allocation
 624          * when none of the DPA space is reserved.
 625          */
 626         if ((is_pmem || !ndd->dpa.child) && n == to_allocate)
 627                 return init_dpa_allocation(label_id, nd_region, nd_mapping, n);
 628         return n;
 629 }
 630
 631 static int merge_dpa(struct nd_region *nd_region,
 632                 struct nd_mapping *nd_mapping, struct nd_label_id *label_id)
 633 {
 634         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 635         struct resource *res;
 636
 637         if (strncmp("pmem", label_id->id, 4) == 0)
 638                 return 0;
 639  retry:
 640         for_each_dpa_resource(ndd, res) {
 641                 int rc;
 642                 struct resource *next = res->sibling;
 643                 resource_size_t end = res->start + resource_size(res);
 644
 645                 if (!next || strcmp(res->name, label_id->id) != 0
 646                                 || strcmp(next->name, label_id->id) != 0
 647                                 || end != next->start)
 648                         continue;
 649                 end += resource_size(next);
 650                 nvdimm_free_dpa(ndd, next);
 651                 rc = adjust_resource(res, res->start, end - res->start);
 652                 nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc);
 653                 if (rc)
 654                         return rc;
 655                 res->flags |= DPA_RESOURCE_ADJUSTED;
 656                 goto retry;
 657         }
 658
 659         return 0;
 660 }
 661
 662 static int __reserve_free_pmem(struct device *dev, void *data)
 663 {
 664         struct nvdimm *nvdimm = data;
 665         struct nd_region *nd_region;
 666         struct nd_label_id label_id;
 667         int i;
 668
 669         if (!is_nd_pmem(dev))
 670                 return 0;
 671
 672         nd_region = to_nd_region(dev);
 673         if (nd_region->ndr_mappings == 0)
 674                 return 0;
 675
 676         memset(&label_id, 0, sizeof(label_id));
 677         strcat(label_id.id, "pmem-reserve");
 678         for (i = 0; i < nd_region->ndr_mappings; i++) {
 679                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 680                 resource_size_t n, rem = 0;
 681
 682                 if (nd_mapping->nvdimm != nvdimm)
 683                         continue;
 684
 685                 n = nd_pmem_available_dpa(nd_region, nd_mapping, &rem);
 686                 if (n == 0)
 687                         return 0;
 688                 rem = scan_allocate(nd_region, nd_mapping, &label_id, n);
 689                 dev_WARN_ONCE(&nd_region->dev, rem,
 690                                 "pmem reserve underrun: %#llx of %#llx bytes\n",
 691                                 (unsigned long long) n - rem,
 692                                 (unsigned long long) n);
 693                 return rem ? -ENXIO : 0;
 694         }
 695
 696         return 0;
 697 }
 698
 699 static void release_free_pmem(struct nvdimm_bus *nvdimm_bus,
 700                 struct nd_mapping *nd_mapping)
 701 {
 702         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
 703         struct resource *res, *_res;
 704
 705         for_each_dpa_resource_safe(ndd, res, _res)
 706                 if (strcmp(res->name, "pmem-reserve") == 0)
 707                         nvdimm_free_dpa(ndd, res);
 708 }
 709
 710 static int reserve_free_pmem(struct nvdimm_bus *nvdimm_bus,
 711                 struct nd_mapping *nd_mapping)
 712 {
 713         struct nvdimm *nvdimm = nd_mapping->nvdimm;
 714         int rc;
 715
 716         rc = device_for_each_child(&nvdimm_bus->dev, nvdimm,
 717                         __reserve_free_pmem);
 718         if (rc)
 719                 release_free_pmem(nvdimm_bus, nd_mapping);
 720         return rc;
 721 }
 722
 723 /**
 724  * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
 725  * @nd_region: the set of dimms to allocate @n more bytes from
 726  * @label_id: unique identifier for the namespace consuming this dpa range
 727  * @n: number of bytes per-dimm to add to the existing allocation
 728  *
 729  * Assumes resources are ordered.  For BLK regions, first consume
 730  * BLK-only available DPA free space, then consume PMEM-aliased DPA
 731  * space starting at the highest DPA.  For PMEM regions start
 732  * allocations from the start of an interleave set and end at the first
 733  * BLK allocation or the end of the interleave set, whichever comes
 734  * first.
 735  */
 736 static int grow_dpa_allocation(struct nd_region *nd_region,
 737                 struct nd_label_id *label_id, resource_size_t n)
 738 {
 739         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
 740         bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
 741         int i;
 742
 743         for (i = 0; i < nd_region->ndr_mappings; i++) {
 744                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
 745                 resource_size_t rem = n;
 746                 int rc, j;
 747
 748                 /*
 749                  * In the BLK case try once with all unallocated PMEM
 750                  * reserved, and once without
 751                  */
 752                 for (j = is_pmem; j < 2; j++) {
 753                         bool blk_only = j == 0;
 754
 755                         if (blk_only) {
 756                                 rc = reserve_free_pmem(nvdimm_bus, nd_mapping);
 757                                 if (rc)
 758                                         return rc;
 759                         }
 760                         rem = scan_allocate(nd_region, nd_mapping,
 761                                         label_id, rem);
 762                         if (blk_only)
 763                                 release_free_pmem(nvdimm_bus, nd_mapping);
 764
 765                         /* try again and allow encroachments into PMEM */
 766                         if (rem == 0)
 767                                 break;
 768                 }
 769
 770                 dev_WARN_ONCE(&nd_region->dev, rem,
 771                                 "allocation underrun: %#llx of %#llx bytes\n",
 772                                 (unsigned long long) n - rem,
 773                                 (unsigned long long) n);
 774                 if (rem)
 775                         return -ENXIO;
 776
 777                 rc = merge_dpa(nd_region, nd_mapping, label_id);
 778                 if (rc)
 779                         return rc;
 780         }
 781
 782         return 0;
 783 }
 784
 785 static void nd_namespace_pmem_set_size(struct nd_region *nd_region,
 786                 struct nd_namespace_pmem *nspm, resource_size_t size)
 787 {
 788         struct resource *res = &nspm->nsio.res;
 789
 790         res->start = nd_region->ndr_start;
 791         res->end = nd_region->ndr_start + size - 1;
 792 }
 793
 794 static ssize_t __size_store(struct device *dev, unsigned long long val)
 795 {
 796         resource_size_t allocated = 0, available = 0;
 797         struct nd_region *nd_region = to_nd_region(dev->parent);
 798         struct nd_mapping *nd_mapping;
 799         struct nvdimm_drvdata *ndd;
 800         struct nd_label_id label_id;
 801         u32 flags = 0, remainder;
 802         u8 *uuid = NULL;
 803         int rc, i;
 804
 805         if (dev->driver || to_ndns(dev)->claim)
 806                 return -EBUSY;
 807
 808         if (is_namespace_pmem(dev)) {
 809                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 810
 811                 uuid = nspm->uuid;
 812         } else if (is_namespace_blk(dev)) {
 813                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 814
 815                 uuid = nsblk->uuid;
 816                 flags = NSLABEL_FLAG_LOCAL;
 817         }
 818
 819         /*
 820          * We need a uuid for the allocation-label and dimm(s) on which
 821          * to store the label.
 822          */
 823         if (!uuid || nd_region->ndr_mappings == 0)
 824                 return -ENXIO;
 825
 826         div_u64_rem(val, SZ_4K * nd_region->ndr_mappings, &remainder);
 827         if (remainder) {
 828                 dev_dbg(dev, "%llu is not %dK aligned\n", val,
 829                                 (SZ_4K * nd_region->ndr_mappings) / SZ_1K);
 830                 return -EINVAL;
 831         }
 832
 833         nd_label_gen_id(&label_id, uuid, flags);
 834         for (i = 0; i < nd_region->ndr_mappings; i++) {
 835                 nd_mapping = &nd_region->mapping[i];
 836                 ndd = to_ndd(nd_mapping);
 837
 838                 /*
 839                  * All dimms in an interleave set, or the base dimm for a blk
 840                  * region, need to be enabled for the size to be changed.
 841                  */
 842                 if (!ndd)
 843                         return -ENXIO;
 844
 845                 allocated += nvdimm_allocated_dpa(ndd, &label_id);
 846         }
 847         available = nd_region_available_dpa(nd_region);
 848
 849         if (val > available + allocated)
 850                 return -ENOSPC;
 851
 852         if (val == allocated)
 853                 return 0;
 854
 855         val = div_u64(val, nd_region->ndr_mappings);
 856         allocated = div_u64(allocated, nd_region->ndr_mappings);
 857         if (val < allocated)
 858                 rc = shrink_dpa_allocation(nd_region, &label_id,
 859                                 allocated - val);
 860         else
 861                 rc = grow_dpa_allocation(nd_region, &label_id, val - allocated);
 862
 863         if (rc)
 864                 return rc;
 865
 866         if (is_namespace_pmem(dev)) {
 867                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 868
 869                 nd_namespace_pmem_set_size(nd_region, nspm,
 870                                 val * nd_region->ndr_mappings);
 871         } else if (is_namespace_blk(dev)) {
 872                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 873
 874                 /*
 875                  * Try to delete the namespace if we deleted all of its
 876                  * allocation, this is not the seed device for the
 877                  * region, and it is not actively claimed by a btt
 878                  * instance.
 879                  */
 880                 if (val == 0 && nd_region->ns_seed != dev
 881                                 && !nsblk->common.claim)
 882                         nd_device_unregister(dev, ND_ASYNC);
 883         }
 884
 885         return rc;
 886 }
 887
 888 static ssize_t size_store(struct device *dev,
 889                 struct device_attribute *attr, const char *buf, size_t len)
 890 {
 891         struct nd_region *nd_region = to_nd_region(dev->parent);
 892         unsigned long long val;
 893         u8 **uuid = NULL;
 894         int rc;
 895
 896         rc = kstrtoull(buf, 0, &val);
 897         if (rc)
 898                 return rc;
 899
 900         device_lock(dev);
 901         nvdimm_bus_lock(dev);
 902         wait_nvdimm_bus_probe_idle(dev);
 903         rc = __size_store(dev, val);
 904         if (rc >= 0)
 905                 rc = nd_namespace_label_update(nd_region, dev);
 906
 907         if (is_namespace_pmem(dev)) {
 908                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 909
 910                 uuid = &nspm->uuid;
 911         } else if (is_namespace_blk(dev)) {
 912                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 913
 914                 uuid = &nsblk->uuid;
 915         }
 916
 917         if (rc == 0 && val == 0 && uuid) {
 918                 /* setting size zero == 'delete namespace' */
 919                 kfree(*uuid);
 920                 *uuid = NULL;
 921         }
 922
 923         dev_dbg(dev, "%s: %llx %s (%d)\n", __func__, val, rc < 0
 924                         ? "fail" : "success", rc);
 925
 926         nvdimm_bus_unlock(dev);
 927         device_unlock(dev);
 928
 929         return rc < 0 ? rc : len;
 930 }
 931
 932 resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
 933 {
 934         struct device *dev = &ndns->dev;
 935
 936         if (is_namespace_pmem(dev)) {
 937                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 938
 939                 return resource_size(&nspm->nsio.res);
 940         } else if (is_namespace_blk(dev)) {
 941                 return nd_namespace_blk_size(to_nd_namespace_blk(dev));
 942         } else if (is_namespace_io(dev)) {
 943                 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
 944
 945                 return resource_size(&nsio->res);
 946         } else
 947                 WARN_ONCE(1, "unknown namespace type\n");
 948         return 0;
 949 }
 950
 951 resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
 952 {
 953         resource_size_t size;
 954
 955         nvdimm_bus_lock(&ndns->dev);
 956         size = __nvdimm_namespace_capacity(ndns);
 957         nvdimm_bus_unlock(&ndns->dev);
 958
 959         return size;
 960 }
 961 EXPORT_SYMBOL(nvdimm_namespace_capacity);
 962
 963 static ssize_t size_show(struct device *dev,
 964                 struct device_attribute *attr, char *buf)
 965 {
 966         return sprintf(buf, "%llu\n", (unsigned long long)
 967                         nvdimm_namespace_capacity(to_ndns(dev)));
 968 }
 969 static DEVICE_ATTR(size, S_IRUGO, size_show, size_store);
 970
 971 static ssize_t uuid_show(struct device *dev,
 972                 struct device_attribute *attr, char *buf)
 973 {
 974         u8 *uuid;
 975
 976         if (is_namespace_pmem(dev)) {
 977                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
 978
 979                 uuid = nspm->uuid;
 980         } else if (is_namespace_blk(dev)) {
 981                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
 982
 983                 uuid = nsblk->uuid;
 984         } else
 985                 return -ENXIO;
 986
 987         if (uuid)
 988                 return sprintf(buf, "%pUb\n", uuid);
 989         return sprintf(buf, "\n");
 990 }
 991
 992 /**
 993  * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
 994  * @nd_region: parent region so we can updates all dimms in the set
 995  * @dev: namespace type for generating label_id
 996  * @new_uuid: incoming uuid
 997  * @old_uuid: reference to the uuid storage location in the namespace object
 998  */
 999 static int namespace_update_uuid(struct nd_region *nd_region,
1000                 struct device *dev, u8 *new_uuid, u8 **old_uuid)
1001 {
1002         u32 flags = is_namespace_blk(dev) ? NSLABEL_FLAG_LOCAL : 0;
1003         struct nd_label_id old_label_id;
1004         struct nd_label_id new_label_id;
1005         int i;
1006
1007         if (!nd_is_uuid_unique(dev, new_uuid))
1008                 return -EINVAL;
1009
1010         if (*old_uuid == NULL)
1011                 goto out;
1012
1013         /*
1014          * If we've already written a label with this uuid, then it's
1015          * too late to rename because we can't reliably update the uuid
1016          * without losing the old namespace.  Userspace must delete this
1017          * namespace to abandon the old uuid.
1018          */
1019         for (i = 0; i < nd_region->ndr_mappings; i++) {
1020                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1021
1022                 /*
1023                  * This check by itself is sufficient because old_uuid
1024                  * would be NULL above if this uuid did not exist in the
1025                  * currently written set.
1026                  *
1027                  * FIXME: can we delete uuid with zero dpa allocated?
1028                  */
1029                 if (nd_mapping->labels)
1030                         return -EBUSY;
1031         }
1032
1033         nd_label_gen_id(&old_label_id, *old_uuid, flags);
1034         nd_label_gen_id(&new_label_id, new_uuid, flags);
1035         for (i = 0; i < nd_region->ndr_mappings; i++) {
1036                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1037                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1038                 struct resource *res;
1039
1040                 for_each_dpa_resource(ndd, res)
1041                         if (strcmp(res->name, old_label_id.id) == 0)
1042                                 sprintf((void *) res->name, "%s",
1043                                                 new_label_id.id);
1044         }
1045         kfree(*old_uuid);
1046  out:
1047         *old_uuid = new_uuid;
1048         return 0;
1049 }
1050
1051 static ssize_t uuid_store(struct device *dev,
1052                 struct device_attribute *attr, const char *buf, size_t len)
1053 {
1054         struct nd_region *nd_region = to_nd_region(dev->parent);
1055         u8 *uuid = NULL;
1056         ssize_t rc = 0;
1057         u8 **ns_uuid;
1058
1059         if (is_namespace_pmem(dev)) {
1060                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1061
1062                 ns_uuid = &nspm->uuid;
1063         } else if (is_namespace_blk(dev)) {
1064                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1065
1066                 ns_uuid = &nsblk->uuid;
1067         } else
1068                 return -ENXIO;
1069
1070         device_lock(dev);
1071         nvdimm_bus_lock(dev);
1072         wait_nvdimm_bus_probe_idle(dev);
1073         if (to_ndns(dev)->claim)
1074                 rc = -EBUSY;
1075         if (rc >= 0)
1076                 rc = nd_uuid_store(dev, &uuid, buf, len);
1077         if (rc >= 0)
1078                 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
1079         if (rc >= 0)
1080                 rc = nd_namespace_label_update(nd_region, dev);
1081         else
1082                 kfree(uuid);
1083         dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
1084                         rc, buf, buf[len - 1] == '\n' ? "" : "\n");
1085         nvdimm_bus_unlock(dev);
1086         device_unlock(dev);
1087
1088         return rc < 0 ? rc : len;
1089 }
1090 static DEVICE_ATTR_RW(uuid);
1091
1092 static ssize_t resource_show(struct device *dev,
1093                 struct device_attribute *attr, char *buf)
1094 {
1095         struct resource *res;
1096
1097         if (is_namespace_pmem(dev)) {
1098                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1099
1100                 res = &nspm->nsio.res;
1101         } else if (is_namespace_io(dev)) {
1102                 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1103
1104                 res = &nsio->res;
1105         } else
1106                 return -ENXIO;
1107
1108         /* no address to convey if the namespace has no allocation */
1109         if (resource_size(res) == 0)
1110                 return -ENXIO;
1111         return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
1112 }
1113 static DEVICE_ATTR_RO(resource);
1114
1115 static const unsigned long ns_lbasize_supported[] = { 512, 520, 528,
1116         4096, 4104, 4160, 4224, 0 };
1117
1118 static ssize_t sector_size_show(struct device *dev,
1119                 struct device_attribute *attr, char *buf)
1120 {
1121         struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1122
1123         if (!is_namespace_blk(dev))
1124                 return -ENXIO;
1125
1126         return nd_sector_size_show(nsblk->lbasize, ns_lbasize_supported, buf);
1127 }
1128
1129 static ssize_t sector_size_store(struct device *dev,
1130                 struct device_attribute *attr, const char *buf, size_t len)
1131 {
1132         struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1133         struct nd_region *nd_region = to_nd_region(dev->parent);
1134         ssize_t rc = 0;
1135
1136         if (!is_namespace_blk(dev))
1137                 return -ENXIO;
1138
1139         device_lock(dev);
1140         nvdimm_bus_lock(dev);
1141         if (to_ndns(dev)->claim)
1142                 rc = -EBUSY;
1143         if (rc >= 0)
1144                 rc = nd_sector_size_store(dev, buf, &nsblk->lbasize,
1145                                 ns_lbasize_supported);
1146         if (rc >= 0)
1147                 rc = nd_namespace_label_update(nd_region, dev);
1148         dev_dbg(dev, "%s: result: %zd %s: %s%s", __func__,
1149                         rc, rc < 0 ? "tried" : "wrote", buf,
1150                         buf[len - 1] == '\n' ? "" : "\n");
1151         nvdimm_bus_unlock(dev);
1152         device_unlock(dev);
1153
1154         return rc ? rc : len;
1155 }
1156 static DEVICE_ATTR_RW(sector_size);
1157
1158 static ssize_t dpa_extents_show(struct device *dev,
1159                 struct device_attribute *attr, char *buf)
1160 {
1161         struct nd_region *nd_region = to_nd_region(dev->parent);
1162         struct nd_label_id label_id;
1163         int count = 0, i;
1164         u8 *uuid = NULL;
1165         u32 flags = 0;
1166
1167         nvdimm_bus_lock(dev);
1168         if (is_namespace_pmem(dev)) {
1169                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1170
1171                 uuid = nspm->uuid;
1172                 flags = 0;
1173         } else if (is_namespace_blk(dev)) {
1174                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1175
1176                 uuid = nsblk->uuid;
1177                 flags = NSLABEL_FLAG_LOCAL;
1178         }
1179
1180         if (!uuid)
1181                 goto out;
1182
1183         nd_label_gen_id(&label_id, uuid, flags);
1184         for (i = 0; i < nd_region->ndr_mappings; i++) {
1185                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1186                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1187                 struct resource *res;
1188
1189                 for_each_dpa_resource(ndd, res)
1190                         if (strcmp(res->name, label_id.id) == 0)
1191                                 count++;
1192         }
1193  out:
1194         nvdimm_bus_unlock(dev);
1195
1196         return sprintf(buf, "%d\n", count);
1197 }
1198 static DEVICE_ATTR_RO(dpa_extents);
1199
1200 static ssize_t holder_show(struct device *dev,
1201                 struct device_attribute *attr, char *buf)
1202 {
1203         struct nd_namespace_common *ndns = to_ndns(dev);
1204         ssize_t rc;
1205
1206         device_lock(dev);
1207         rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
1208         device_unlock(dev);
1209
1210         return rc;
1211 }
1212 static DEVICE_ATTR_RO(holder);
1213
1214 static ssize_t force_raw_store(struct device *dev,
1215                 struct device_attribute *attr, const char *buf, size_t len)
1216 {
1217         bool force_raw;
1218         int rc = strtobool(buf, &force_raw);
1219
1220         if (rc)
1221                 return rc;
1222
1223         to_ndns(dev)->force_raw = force_raw;
1224         return len;
1225 }
1226
1227 static ssize_t force_raw_show(struct device *dev,
1228                 struct device_attribute *attr, char *buf)
1229 {
1230         return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
1231 }
1232 static DEVICE_ATTR_RW(force_raw);
1233
1234 static struct attribute *nd_namespace_attributes[] = {
1235         &dev_attr_nstype.attr,
1236         &dev_attr_size.attr,
1237         &dev_attr_uuid.attr,
1238         &dev_attr_holder.attr,
1239         &dev_attr_resource.attr,
1240         &dev_attr_alt_name.attr,
1241         &dev_attr_force_raw.attr,
1242         &dev_attr_sector_size.attr,
1243         &dev_attr_dpa_extents.attr,
1244         NULL,
1245 };
1246
1247 static umode_t namespace_visible(struct kobject *kobj,
1248                 struct attribute *a, int n)
1249 {
1250         struct device *dev = container_of(kobj, struct device, kobj);
1251
1252         if (a == &dev_attr_resource.attr) {
1253                 if (is_namespace_blk(dev))
1254                         return 0;
1255                 return a->mode;
1256         }
1257
1258         if (is_namespace_pmem(dev) || is_namespace_blk(dev)) {
1259                 if (a == &dev_attr_size.attr)
1260                         return S_IWUSR | S_IRUGO;
1261
1262                 if (is_namespace_pmem(dev) && a == &dev_attr_sector_size.attr)
1263                         return 0;
1264
1265                 return a->mode;
1266         }
1267
1268         if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr
1269                         || a == &dev_attr_holder.attr
1270                         || a == &dev_attr_force_raw.attr)
1271                 return a->mode;
1272
1273         return 0;
1274 }
1275
1276 static struct attribute_group nd_namespace_attribute_group = {
1277         .attrs = nd_namespace_attributes,
1278         .is_visible = namespace_visible,
1279 };
1280
1281 static const struct attribute_group *nd_namespace_attribute_groups[] = {
1282         &nd_device_attribute_group,
1283         &nd_namespace_attribute_group,
1284         &nd_numa_attribute_group,
1285         NULL,
1286 };
1287
1288 struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
1289 {
1290         struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
1291         struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
1292         struct nd_namespace_common *ndns;
1293         resource_size_t size;
1294
1295         if (nd_btt || nd_pfn) {
1296                 struct device *host = NULL;
1297
1298                 if (nd_btt) {
1299                         host = &nd_btt->dev;
1300                         ndns = nd_btt->ndns;
1301                 } else if (nd_pfn) {
1302                         host = &nd_pfn->dev;
1303                         ndns = nd_pfn->ndns;
1304                 }
1305
1306                 if (!ndns || !host)
1307                         return ERR_PTR(-ENODEV);
1308
1309                 /*
1310                  * Flush any in-progess probes / removals in the driver
1311                  * for the raw personality of this namespace.
1312                  */
1313                 device_lock(&ndns->dev);
1314                 device_unlock(&ndns->dev);
1315                 if (ndns->dev.driver) {
1316                         dev_dbg(&ndns->dev, "is active, can't bind %s\n",
1317                                         dev_name(host));
1318                         return ERR_PTR(-EBUSY);
1319                 }
1320                 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != host,
1321                                         "host (%s) vs claim (%s) mismatch\n",
1322                                         dev_name(host),
1323                                         dev_name(ndns->claim)))
1324                         return ERR_PTR(-ENXIO);
1325         } else {
1326                 ndns = to_ndns(dev);
1327                 if (ndns->claim) {
1328                         dev_dbg(dev, "claimed by %s, failing probe\n",
1329                                 dev_name(ndns->claim));
1330
1331                         return ERR_PTR(-ENXIO);
1332                 }
1333         }
1334
1335         size = nvdimm_namespace_capacity(ndns);
1336         if (size < ND_MIN_NAMESPACE_SIZE) {
1337                 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
1338                                 &size, ND_MIN_NAMESPACE_SIZE);
1339                 return ERR_PTR(-ENODEV);
1340         }
1341
1342         if (is_namespace_pmem(&ndns->dev)) {
1343                 struct nd_namespace_pmem *nspm;
1344
1345                 nspm = to_nd_namespace_pmem(&ndns->dev);
1346                 if (!nspm->uuid) {
1347                         dev_dbg(&ndns->dev, "%s: uuid not set\n", __func__);
1348                         return ERR_PTR(-ENODEV);
1349                 }
1350         } else if (is_namespace_blk(&ndns->dev)) {
1351                 struct nd_namespace_blk *nsblk;
1352
1353                 nsblk = to_nd_namespace_blk(&ndns->dev);
1354                 if (!nd_namespace_blk_validate(nsblk))
1355                         return ERR_PTR(-ENODEV);
1356         }
1357
1358         return ndns;
1359 }
1360 EXPORT_SYMBOL(nvdimm_namespace_common_probe);
1361
1362 static struct device **create_namespace_io(struct nd_region *nd_region)
1363 {
1364         struct nd_namespace_io *nsio;
1365         struct device *dev, **devs;
1366         struct resource *res;
1367
1368         nsio = kzalloc(sizeof(*nsio), GFP_KERNEL);
1369         if (!nsio)
1370                 return NULL;
1371
1372         devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1373         if (!devs) {
1374                 kfree(nsio);
1375                 return NULL;
1376         }
1377
1378         dev = &nsio->common.dev;
1379         dev->type = &namespace_io_device_type;
1380         dev->parent = &nd_region->dev;
1381         res = &nsio->res;
1382         res->name = dev_name(&nd_region->dev);
1383         res->flags = IORESOURCE_MEM;
1384         res->start = nd_region->ndr_start;
1385         res->end = res->start + nd_region->ndr_size - 1;
1386
1387         devs[0] = dev;
1388         return devs;
1389 }
1390
1391 static bool has_uuid_at_pos(struct nd_region *nd_region, u8 *uuid,
1392                 u64 cookie, u16 pos)
1393 {
1394         struct nd_namespace_label *found = NULL;
1395         int i;
1396
1397         for (i = 0; i < nd_region->ndr_mappings; i++) {
1398                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1399                 struct nd_namespace_label *nd_label;
1400                 bool found_uuid = false;
1401                 int l;
1402
1403                 for_each_label(l, nd_label, nd_mapping->labels) {
1404                         u64 isetcookie = __le64_to_cpu(nd_label->isetcookie);
1405                         u16 position = __le16_to_cpu(nd_label->position);
1406                         u16 nlabel = __le16_to_cpu(nd_label->nlabel);
1407
1408                         if (isetcookie != cookie)
1409                                 continue;
1410
1411                         if (memcmp(nd_label->uuid, uuid, NSLABEL_UUID_LEN) != 0)
1412                                 continue;
1413
1414                         if (found_uuid) {
1415                                 dev_dbg(to_ndd(nd_mapping)->dev,
1416                                                 "%s duplicate entry for uuid\n",
1417                                                 __func__);
1418                                 return false;
1419                         }
1420                         found_uuid = true;
1421                         if (nlabel != nd_region->ndr_mappings)
1422                                 continue;
1423                         if (position != pos)
1424                                 continue;
1425                         found = nd_label;
1426                         break;
1427                 }
1428                 if (found)
1429                         break;
1430         }
1431         return found != NULL;
1432 }
1433
1434 static int select_pmem_id(struct nd_region *nd_region, u8 *pmem_id)
1435 {
1436         struct nd_namespace_label *select = NULL;
1437         int i;
1438
1439         if (!pmem_id)
1440                 return -ENODEV;
1441
1442         for (i = 0; i < nd_region->ndr_mappings; i++) {
1443                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1444                 struct nd_namespace_label *nd_label;
1445                 u64 hw_start, hw_end, pmem_start, pmem_end;
1446                 int l;
1447
1448                 for_each_label(l, nd_label, nd_mapping->labels)
1449                         if (memcmp(nd_label->uuid, pmem_id, NSLABEL_UUID_LEN) == 0)
1450                                 break;
1451
1452                 if (!nd_label) {
1453                         WARN_ON(1);
1454                         return -EINVAL;
1455                 }
1456
1457                 select = nd_label;
1458                 /*
1459                  * Check that this label is compliant with the dpa
1460                  * range published in NFIT
1461                  */
1462                 hw_start = nd_mapping->start;
1463                 hw_end = hw_start + nd_mapping->size;
1464                 pmem_start = __le64_to_cpu(select->dpa);
1465                 pmem_end = pmem_start + __le64_to_cpu(select->rawsize);
1466                 if (pmem_start == hw_start && pmem_end <= hw_end)
1467                         /* pass */;
1468                 else
1469                         return -EINVAL;
1470
1471                 nd_mapping->labels[0] = select;
1472                 nd_mapping->labels[1] = NULL;
1473         }
1474         return 0;
1475 }
1476
1477 /**
1478  * find_pmem_label_set - validate interleave set labelling, retrieve label0
1479  * @nd_region: region with mappings to validate
1480  */
1481 static int find_pmem_label_set(struct nd_region *nd_region,
1482                 struct nd_namespace_pmem *nspm)
1483 {
1484         u64 cookie = nd_region_interleave_set_cookie(nd_region);
1485         struct nd_namespace_label *nd_label;
1486         u8 select_id[NSLABEL_UUID_LEN];
1487         resource_size_t size = 0;
1488         u8 *pmem_id = NULL;
1489         int rc = -ENODEV, l;
1490         u16 i;
1491
1492         if (cookie == 0)
1493                 return -ENXIO;
1494
1495         /*
1496          * Find a complete set of labels by uuid.  By definition we can start
1497          * with any mapping as the reference label
1498          */
1499         for_each_label(l, nd_label, nd_region->mapping[0].labels) {
1500                 u64 isetcookie = __le64_to_cpu(nd_label->isetcookie);
1501
1502                 if (isetcookie != cookie)
1503                         continue;
1504
1505                 for (i = 0; nd_region->ndr_mappings; i++)
1506                         if (!has_uuid_at_pos(nd_region, nd_label->uuid,
1507                                                 cookie, i))
1508                                 break;
1509                 if (i < nd_region->ndr_mappings) {
1510                         /*
1511                          * Give up if we don't find an instance of a
1512                          * uuid at each position (from 0 to
1513                          * nd_region->ndr_mappings - 1), or if we find a
1514                          * dimm with two instances of the same uuid.
1515                          */
1516                         rc = -EINVAL;
1517                         goto err;
1518                 } else if (pmem_id) {
1519                         /*
1520                          * If there is more than one valid uuid set, we
1521                          * need userspace to clean this up.
1522                          */
1523                         rc = -EBUSY;
1524                         goto err;
1525                 }
1526                 memcpy(select_id, nd_label->uuid, NSLABEL_UUID_LEN);
1527                 pmem_id = select_id;
1528         }
1529
1530         /*
1531          * Fix up each mapping's 'labels' to have the validated pmem label for
1532          * that position at labels[0], and NULL at labels[1].  In the process,
1533          * check that the namespace aligns with interleave-set.  We know
1534          * that it does not overlap with any blk namespaces by virtue of
1535          * the dimm being enabled (i.e. nd_label_reserve_dpa()
1536          * succeeded).
1537          */
1538         rc = select_pmem_id(nd_region, pmem_id);
1539         if (rc)
1540                 goto err;
1541
1542         /* Calculate total size and populate namespace properties from label0 */
1543         for (i = 0; i < nd_region->ndr_mappings; i++) {
1544                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1545                 struct nd_namespace_label *label0 = nd_mapping->labels[0];
1546
1547                 size += __le64_to_cpu(label0->rawsize);
1548                 if (__le16_to_cpu(label0->position) != 0)
1549                         continue;
1550                 WARN_ON(nspm->alt_name || nspm->uuid);
1551                 nspm->alt_name = kmemdup((void __force *) label0->name,
1552                                 NSLABEL_NAME_LEN, GFP_KERNEL);
1553                 nspm->uuid = kmemdup((void __force *) label0->uuid,
1554                                 NSLABEL_UUID_LEN, GFP_KERNEL);
1555         }
1556
1557         if (!nspm->alt_name || !nspm->uuid) {
1558                 rc = -ENOMEM;
1559                 goto err;
1560         }
1561
1562         nd_namespace_pmem_set_size(nd_region, nspm, size);
1563
1564         return 0;
1565  err:
1566         switch (rc) {
1567         case -EINVAL:
1568                 dev_dbg(&nd_region->dev, "%s: invalid label(s)\n", __func__);
1569                 break;
1570         case -ENODEV:
1571                 dev_dbg(&nd_region->dev, "%s: label not found\n", __func__);
1572                 break;
1573         default:
1574                 dev_dbg(&nd_region->dev, "%s: unexpected err: %d\n",
1575                                 __func__, rc);
1576                 break;
1577         }
1578         return rc;
1579 }
1580
1581 static struct device **create_namespace_pmem(struct nd_region *nd_region)
1582 {
1583         struct nd_namespace_pmem *nspm;
1584         struct device *dev, **devs;
1585         struct resource *res;
1586         int rc;
1587
1588         nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1589         if (!nspm)
1590                 return NULL;
1591
1592         dev = &nspm->nsio.common.dev;
1593         dev->type = &namespace_pmem_device_type;
1594         dev->parent = &nd_region->dev;
1595         res = &nspm->nsio.res;
1596         res->name = dev_name(&nd_region->dev);
1597         res->flags = IORESOURCE_MEM;
1598         rc = find_pmem_label_set(nd_region, nspm);
1599         if (rc == -ENODEV) {
1600                 int i;
1601
1602                 /* Pass, try to permit namespace creation... */
1603                 for (i = 0; i < nd_region->ndr_mappings; i++) {
1604                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1605
1606                         kfree(nd_mapping->labels);
1607                         nd_mapping->labels = NULL;
1608                 }
1609
1610                 /* Publish a zero-sized namespace for userspace to configure. */
1611                 nd_namespace_pmem_set_size(nd_region, nspm, 0);
1612
1613                 rc = 0;
1614         } else if (rc)
1615                 goto err;
1616
1617         devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1618         if (!devs)
1619                 goto err;
1620
1621         devs[0] = dev;
1622         return devs;
1623
1624  err:
1625         namespace_pmem_release(&nspm->nsio.common.dev);
1626         return NULL;
1627 }
1628
1629 struct resource *nsblk_add_resource(struct nd_region *nd_region,
1630                 struct nvdimm_drvdata *ndd, struct nd_namespace_blk *nsblk,
1631                 resource_size_t start)
1632 {
1633         struct nd_label_id label_id;
1634         struct resource *res;
1635
1636         nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
1637         res = krealloc(nsblk->res,
1638                         sizeof(void *) * (nsblk->num_resources + 1),
1639                         GFP_KERNEL);
1640         if (!res)
1641                 return NULL;
1642         nsblk->res = (struct resource **) res;
1643         for_each_dpa_resource(ndd, res)
1644                 if (strcmp(res->name, label_id.id) == 0
1645                                 && res->start == start) {
1646                         nsblk->res[nsblk->num_resources++] = res;
1647                         return res;
1648                 }
1649         return NULL;
1650 }
1651
1652 static struct device *nd_namespace_blk_create(struct nd_region *nd_region)
1653 {
1654         struct nd_namespace_blk *nsblk;
1655         struct device *dev;
1656
1657         if (!is_nd_blk(&nd_region->dev))
1658                 return NULL;
1659
1660         nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
1661         if (!nsblk)
1662                 return NULL;
1663
1664         dev = &nsblk->common.dev;
1665         dev->type = &namespace_blk_device_type;
1666         nsblk->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
1667         if (nsblk->id < 0) {
1668                 kfree(nsblk);
1669                 return NULL;
1670         }
1671         dev_set_name(dev, "namespace%d.%d", nd_region->id, nsblk->id);
1672         dev->parent = &nd_region->dev;
1673         dev->groups = nd_namespace_attribute_groups;
1674
1675         return &nsblk->common.dev;
1676 }
1677
1678 void nd_region_create_blk_seed(struct nd_region *nd_region)
1679 {
1680         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1681         nd_region->ns_seed = nd_namespace_blk_create(nd_region);
1682         /*
1683          * Seed creation failures are not fatal, provisioning is simply
1684          * disabled until memory becomes available
1685          */
1686         if (!nd_region->ns_seed)
1687                 dev_err(&nd_region->dev, "failed to create blk namespace\n");
1688         else
1689                 nd_device_register(nd_region->ns_seed);
1690 }
1691
1692 void nd_region_create_btt_seed(struct nd_region *nd_region)
1693 {
1694         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1695         nd_region->btt_seed = nd_btt_create(nd_region);
1696         /*
1697          * Seed creation failures are not fatal, provisioning is simply
1698          * disabled until memory becomes available
1699          */
1700         if (!nd_region->btt_seed)
1701                 dev_err(&nd_region->dev, "failed to create btt namespace\n");
1702 }
1703
1704 static struct device **create_namespace_blk(struct nd_region *nd_region)
1705 {
1706         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
1707         struct nd_namespace_label *nd_label;
1708         struct device *dev, **devs = NULL;
1709         struct nd_namespace_blk *nsblk;
1710         struct nvdimm_drvdata *ndd;
1711         int i, l, count = 0;
1712         struct resource *res;
1713
1714         if (nd_region->ndr_mappings == 0)
1715                 return NULL;
1716
1717         ndd = to_ndd(nd_mapping);
1718         for_each_label(l, nd_label, nd_mapping->labels) {
1719                 u32 flags = __le32_to_cpu(nd_label->flags);
1720                 char *name[NSLABEL_NAME_LEN];
1721                 struct device **__devs;
1722
1723                 if (flags & NSLABEL_FLAG_LOCAL)
1724                         /* pass */;
1725                 else
1726                         continue;
1727
1728                 for (i = 0; i < count; i++) {
1729                         nsblk = to_nd_namespace_blk(devs[i]);
1730                         if (memcmp(nsblk->uuid, nd_label->uuid,
1731                                                 NSLABEL_UUID_LEN) == 0) {
1732                                 res = nsblk_add_resource(nd_region, ndd, nsblk,
1733                                                 __le64_to_cpu(nd_label->dpa));
1734                                 if (!res)
1735                                         goto err;
1736                                 nd_dbg_dpa(nd_region, ndd, res, "%s assign\n",
1737                                         dev_name(&nsblk->common.dev));
1738                                 break;
1739                         }
1740                 }
1741                 if (i < count)
1742                         continue;
1743                 __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
1744                 if (!__devs)
1745                         goto err;
1746                 memcpy(__devs, devs, sizeof(dev) * count);
1747                 kfree(devs);
1748                 devs = __devs;
1749
1750                 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
1751                 if (!nsblk)
1752                         goto err;
1753                 dev = &nsblk->common.dev;
1754                 dev->type = &namespace_blk_device_type;
1755                 dev->parent = &nd_region->dev;
1756                 dev_set_name(dev, "namespace%d.%d", nd_region->id, count);
1757                 devs[count++] = dev;
1758                 nsblk->id = -1;
1759                 nsblk->lbasize = __le64_to_cpu(nd_label->lbasize);
1760                 nsblk->uuid = kmemdup(nd_label->uuid, NSLABEL_UUID_LEN,
1761                                 GFP_KERNEL);
1762                 if (!nsblk->uuid)
1763                         goto err;
1764                 memcpy(name, nd_label->name, NSLABEL_NAME_LEN);
1765                 if (name[0])
1766                         nsblk->alt_name = kmemdup(name, NSLABEL_NAME_LEN,
1767                                         GFP_KERNEL);
1768                 res = nsblk_add_resource(nd_region, ndd, nsblk,
1769                                 __le64_to_cpu(nd_label->dpa));
1770                 if (!res)
1771                         goto err;
1772                 nd_dbg_dpa(nd_region, ndd, res, "%s assign\n",
1773                                 dev_name(&nsblk->common.dev));
1774         }
1775
1776         dev_dbg(&nd_region->dev, "%s: discovered %d blk namespace%s\n",
1777                         __func__, count, count == 1 ? "" : "s");
1778
1779         if (count == 0) {
1780                 /* Publish a zero-sized namespace for userspace to configure. */
1781                 for (i = 0; i < nd_region->ndr_mappings; i++) {
1782                         struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1783
1784                         kfree(nd_mapping->labels);
1785                         nd_mapping->labels = NULL;
1786                 }
1787
1788                 devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
1789                 if (!devs)
1790                         goto err;
1791                 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
1792                 if (!nsblk)
1793                         goto err;
1794                 dev = &nsblk->common.dev;
1795                 dev->type = &namespace_blk_device_type;
1796                 dev->parent = &nd_region->dev;
1797                 devs[count++] = dev;
1798         }
1799
1800         return devs;
1801
1802 err:
1803         for (i = 0; i < count; i++) {
1804                 nsblk = to_nd_namespace_blk(devs[i]);
1805                 namespace_blk_release(&nsblk->common.dev);
1806         }
1807         kfree(devs);
1808         return NULL;
1809 }
1810
1811 static int init_active_labels(struct nd_region *nd_region)
1812 {
1813         int i;
1814
1815         for (i = 0; i < nd_region->ndr_mappings; i++) {
1816                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1817                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1818                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1819                 int count, j;
1820
1821                 /*
1822                  * If the dimm is disabled then prevent the region from
1823                  * being activated if it aliases DPA.
1824                  */
1825                 if (!ndd) {
1826                         if ((nvdimm->flags & NDD_ALIASING) == 0)
1827                                 return 0;
1828                         dev_dbg(&nd_region->dev, "%s: is disabled, failing probe\n",
1829                                         dev_name(&nd_mapping->nvdimm->dev));
1830                         return -ENXIO;
1831                 }
1832                 nd_mapping->ndd = ndd;
1833                 atomic_inc(&nvdimm->busy);
1834                 get_ndd(ndd);
1835
1836                 count = nd_label_active_count(ndd);
1837                 dev_dbg(ndd->dev, "%s: %d\n", __func__, count);
1838                 if (!count)
1839                         continue;
1840                 nd_mapping->labels = kcalloc(count + 1, sizeof(void *),
1841                                 GFP_KERNEL);
1842                 if (!nd_mapping->labels)
1843                         return -ENOMEM;
1844                 for (j = 0; j < count; j++) {
1845                         struct nd_namespace_label *label;
1846
1847                         label = nd_label_active(ndd, j);
1848                         nd_mapping->labels[j] = label;
1849                 }
1850         }
1851
1852         return 0;
1853 }
1854
1855 int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
1856 {
1857         struct device **devs = NULL;
1858         int i, rc = 0, type;
1859
1860         *err = 0;
1861         nvdimm_bus_lock(&nd_region->dev);
1862         rc = init_active_labels(nd_region);
1863         if (rc) {
1864                 nvdimm_bus_unlock(&nd_region->dev);
1865                 return rc;
1866         }
1867
1868         type = nd_region_to_nstype(nd_region);
1869         switch (type) {
1870         case ND_DEVICE_NAMESPACE_IO:
1871                 devs = create_namespace_io(nd_region);
1872                 break;
1873         case ND_DEVICE_NAMESPACE_PMEM:
1874                 devs = create_namespace_pmem(nd_region);
1875                 break;
1876         case ND_DEVICE_NAMESPACE_BLK:
1877                 devs = create_namespace_blk(nd_region);
1878                 break;
1879         default:
1880                 break;
1881         }
1882         nvdimm_bus_unlock(&nd_region->dev);
1883
1884         if (!devs)
1885                 return -ENODEV;
1886
1887         for (i = 0; devs[i]; i++) {
1888                 struct device *dev = devs[i];
1889                 int id;
1890
1891                 if (type == ND_DEVICE_NAMESPACE_BLK) {
1892                         struct nd_namespace_blk *nsblk;
1893
1894                         nsblk = to_nd_namespace_blk(dev);
1895                         id = ida_simple_get(&nd_region->ns_ida, 0, 0,
1896                                         GFP_KERNEL);
1897                         nsblk->id = id;
1898                 } else
1899                         id = i;
1900
1901                 if (id < 0)
1902                         break;
1903                 dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
1904                 dev->groups = nd_namespace_attribute_groups;
1905                 nd_device_register(dev);
1906         }
1907         if (i)
1908                 nd_region->ns_seed = devs[0];
1909
1910         if (devs[i]) {
1911                 int j;
1912
1913                 for (j = i; devs[j]; j++) {
1914                         struct device *dev = devs[j];
1915
1916                         device_initialize(dev);
1917                         put_device(dev);
1918                 }
1919                 *err = j - i;
1920                 /*
1921                  * All of the namespaces we tried to register failed, so
1922                  * fail region activation.
1923                  */
1924                 if (*err == 0)
1925                         rc = -ENODEV;
1926         }
1927         kfree(devs);
1928
1929         if (rc == -ENODEV)
1930                 return rc;
1931
1932         return i;
1933 }