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/sort.h> 16 #include <linux/slab.h> 17 #include <linux/pmem.h> 18 #include <linux/list.h> 19 #include <linux/nd.h> 20 #include "nd-core.h" 21 #include "nd.h" 22 23 static void namespace_io_release(struct device *dev) 24 { 25 struct nd_namespace_io *nsio = to_nd_namespace_io(dev); 26 27 kfree(nsio); 28 } 29 30 static void namespace_pmem_release(struct device *dev) 31 { 32 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 33 struct nd_region *nd_region = to_nd_region(dev->parent); 34 35 if (nspm->id >= 0) 36 ida_simple_remove(&nd_region->ns_ida, nspm->id); 37 kfree(nspm->alt_name); 38 kfree(nspm->uuid); 39 kfree(nspm); 40 } 41 42 static void namespace_blk_release(struct device *dev) 43 { 44 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); 45 struct nd_region *nd_region = to_nd_region(dev->parent); 46 47 if (nsblk->id >= 0) 48 ida_simple_remove(&nd_region->ns_ida, nsblk->id); 49 kfree(nsblk->alt_name); 50 kfree(nsblk->uuid); 51 kfree(nsblk->res); 52 kfree(nsblk); 53 } 54 55 static struct device_type namespace_io_device_type = { 56 .name = "nd_namespace_io", 57 .release = namespace_io_release, 58 }; 59 60 static struct device_type namespace_pmem_device_type = { 61 .name = "nd_namespace_pmem", 62 .release = namespace_pmem_release, 63 }; 64 65 static struct device_type namespace_blk_device_type = { 66 .name = "nd_namespace_blk", 67 .release = namespace_blk_release, 68 }; 69 70 static bool is_namespace_pmem(const struct device *dev) 71 { 72 return dev ? dev->type == &namespace_pmem_device_type : false; 73 } 74 75 static bool is_namespace_blk(const struct device *dev) 76 { 77 return dev ? dev->type == &namespace_blk_device_type : false; 78 } 79 80 static bool is_namespace_io(const struct device *dev) 81 { 82 return dev ? dev->type == &namespace_io_device_type : false; 83 } 84 85 static int is_uuid_busy(struct device *dev, void *data) 86 { 87 u8 *uuid1 = data, *uuid2 = NULL; 88 89 if (is_namespace_pmem(dev)) { 90 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 91 92 uuid2 = nspm->uuid; 93 } else if (is_namespace_blk(dev)) { 94 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); 95 96 uuid2 = nsblk->uuid; 97 } else if (is_nd_btt(dev)) { 98 struct nd_btt *nd_btt = to_nd_btt(dev); 99 100 uuid2 = nd_btt->uuid; 101 } else if (is_nd_pfn(dev)) { 102 struct nd_pfn *nd_pfn = to_nd_pfn(dev); 103 104 uuid2 = nd_pfn->uuid; 105 } 106 107 if (uuid2 && memcmp(uuid1, uuid2, NSLABEL_UUID_LEN) == 0) 108 return -EBUSY; 109 110 return 0; 111 } 112 113 static int is_namespace_uuid_busy(struct device *dev, void *data) 114 { 115 if (is_nd_pmem(dev) || is_nd_blk(dev)) 116 return device_for_each_child(dev, data, is_uuid_busy); 117 return 0; 118 } 119 120 /** 121 * nd_is_uuid_unique - verify that no other namespace has @uuid 122 * @dev: any device on a nvdimm_bus 123 * @uuid: uuid to check 124 */ 125 bool nd_is_uuid_unique(struct device *dev, u8 *uuid) 126 { 127 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev); 128 129 if (!nvdimm_bus) 130 return false; 131 WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm_bus->dev)); 132 if (device_for_each_child(&nvdimm_bus->dev, uuid, 133 is_namespace_uuid_busy) != 0) 134 return false; 135 return true; 136 } 137 138 bool pmem_should_map_pages(struct device *dev) 139 { 140 struct nd_region *nd_region = to_nd_region(dev->parent); 141 struct nd_namespace_io *nsio; 142 143 if (!IS_ENABLED(CONFIG_ZONE_DEVICE)) 144 return false; 145 146 if (!test_bit(ND_REGION_PAGEMAP, &nd_region->flags)) 147 return false; 148 149 if (is_nd_pfn(dev) || is_nd_btt(dev)) 150 return false; 151 152 nsio = to_nd_namespace_io(dev); 153 if (region_intersects(nsio->res.start, resource_size(&nsio->res), 154 IORESOURCE_SYSTEM_RAM, 155 IORES_DESC_NONE) == REGION_MIXED) 156 return false; 157 158 #ifdef ARCH_MEMREMAP_PMEM 159 return ARCH_MEMREMAP_PMEM == MEMREMAP_WB; 160 #else 161 return false; 162 #endif 163 } 164 EXPORT_SYMBOL(pmem_should_map_pages); 165 166 const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns, 167 char *name) 168 { 169 struct nd_region *nd_region = to_nd_region(ndns->dev.parent); 170 const char *suffix = NULL; 171 172 if (ndns->claim && is_nd_btt(ndns->claim)) 173 suffix = "s"; 174 175 if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev)) { 176 int nsidx = 0; 177 178 if (is_namespace_pmem(&ndns->dev)) { 179 struct nd_namespace_pmem *nspm; 180 181 nspm = to_nd_namespace_pmem(&ndns->dev); 182 nsidx = nspm->id; 183 } 184 185 if (nsidx) 186 sprintf(name, "pmem%d.%d%s", nd_region->id, nsidx, 187 suffix ? suffix : ""); 188 else 189 sprintf(name, "pmem%d%s", nd_region->id, 190 suffix ? suffix : ""); 191 } else if (is_namespace_blk(&ndns->dev)) { 192 struct nd_namespace_blk *nsblk; 193 194 nsblk = to_nd_namespace_blk(&ndns->dev); 195 sprintf(name, "ndblk%d.%d%s", nd_region->id, nsblk->id, 196 suffix ? suffix : ""); 197 } else { 198 return NULL; 199 } 200 201 return name; 202 } 203 EXPORT_SYMBOL(nvdimm_namespace_disk_name); 204 205 const u8 *nd_dev_to_uuid(struct device *dev) 206 { 207 static const u8 null_uuid[16]; 208 209 if (!dev) 210 return null_uuid; 211 212 if (is_namespace_pmem(dev)) { 213 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 214 215 return nspm->uuid; 216 } else if (is_namespace_blk(dev)) { 217 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); 218 219 return nsblk->uuid; 220 } else 221 return null_uuid; 222 } 223 EXPORT_SYMBOL(nd_dev_to_uuid); 224 225 static ssize_t nstype_show(struct device *dev, 226 struct device_attribute *attr, char *buf) 227 { 228 struct nd_region *nd_region = to_nd_region(dev->parent); 229 230 return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region)); 231 } 232 static DEVICE_ATTR_RO(nstype); 233 234 static ssize_t __alt_name_store(struct device *dev, const char *buf, 235 const size_t len) 236 { 237 char *input, *pos, *alt_name, **ns_altname; 238 ssize_t rc; 239 240 if (is_namespace_pmem(dev)) { 241 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 242 243 ns_altname = &nspm->alt_name; 244 } else if (is_namespace_blk(dev)) { 245 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); 246 247 ns_altname = &nsblk->alt_name; 248 } else 249 return -ENXIO; 250 251 if (dev->driver || to_ndns(dev)->claim) 252 return -EBUSY; 253 254 input = kmemdup(buf, len + 1, GFP_KERNEL); 255 if (!input) 256 return -ENOMEM; 257 258 input[len] = '\0'; 259 pos = strim(input); 260 if (strlen(pos) + 1 > NSLABEL_NAME_LEN) { 261 rc = -EINVAL; 262 goto out; 263 } 264 265 alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL); 266 if (!alt_name) { 267 rc = -ENOMEM; 268 goto out; 269 } 270 kfree(*ns_altname); 271 *ns_altname = alt_name; 272 sprintf(*ns_altname, "%s", pos); 273 rc = len; 274 275 out: 276 kfree(input); 277 return rc; 278 } 279 280 static resource_size_t nd_namespace_blk_size(struct nd_namespace_blk *nsblk) 281 { 282 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent); 283 struct nd_mapping *nd_mapping = &nd_region->mapping[0]; 284 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 285 struct nd_label_id label_id; 286 resource_size_t size = 0; 287 struct resource *res; 288 289 if (!nsblk->uuid) 290 return 0; 291 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL); 292 for_each_dpa_resource(ndd, res) 293 if (strcmp(res->name, label_id.id) == 0) 294 size += resource_size(res); 295 return size; 296 } 297 298 static bool __nd_namespace_blk_validate(struct nd_namespace_blk *nsblk) 299 { 300 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent); 301 struct nd_mapping *nd_mapping = &nd_region->mapping[0]; 302 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 303 struct nd_label_id label_id; 304 struct resource *res; 305 int count, i; 306 307 if (!nsblk->uuid || !nsblk->lbasize || !ndd) 308 return false; 309 310 count = 0; 311 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL); 312 for_each_dpa_resource(ndd, res) { 313 if (strcmp(res->name, label_id.id) != 0) 314 continue; 315 /* 316 * Resources with unacknowledged adjustments indicate a 317 * failure to update labels 318 */ 319 if (res->flags & DPA_RESOURCE_ADJUSTED) 320 return false; 321 count++; 322 } 323 324 /* These values match after a successful label update */ 325 if (count != nsblk->num_resources) 326 return false; 327 328 for (i = 0; i < nsblk->num_resources; i++) { 329 struct resource *found = NULL; 330 331 for_each_dpa_resource(ndd, res) 332 if (res == nsblk->res[i]) { 333 found = res; 334 break; 335 } 336 /* stale resource */ 337 if (!found) 338 return false; 339 } 340 341 return true; 342 } 343 344 resource_size_t nd_namespace_blk_validate(struct nd_namespace_blk *nsblk) 345 { 346 resource_size_t size; 347 348 nvdimm_bus_lock(&nsblk->common.dev); 349 size = __nd_namespace_blk_validate(nsblk); 350 nvdimm_bus_unlock(&nsblk->common.dev); 351 352 return size; 353 } 354 EXPORT_SYMBOL(nd_namespace_blk_validate); 355 356 357 static int nd_namespace_label_update(struct nd_region *nd_region, 358 struct device *dev) 359 { 360 dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim, 361 "namespace must be idle during label update\n"); 362 if (dev->driver || to_ndns(dev)->claim) 363 return 0; 364 365 /* 366 * Only allow label writes that will result in a valid namespace 367 * or deletion of an existing namespace. 368 */ 369 if (is_namespace_pmem(dev)) { 370 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 371 resource_size_t size = resource_size(&nspm->nsio.res); 372 373 if (size == 0 && nspm->uuid) 374 /* delete allocation */; 375 else if (!nspm->uuid) 376 return 0; 377 378 return nd_pmem_namespace_label_update(nd_region, nspm, size); 379 } else if (is_namespace_blk(dev)) { 380 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); 381 resource_size_t size = nd_namespace_blk_size(nsblk); 382 383 if (size == 0 && nsblk->uuid) 384 /* delete allocation */; 385 else if (!nsblk->uuid || !nsblk->lbasize) 386 return 0; 387 388 return nd_blk_namespace_label_update(nd_region, nsblk, size); 389 } else 390 return -ENXIO; 391 } 392 393 static ssize_t alt_name_store(struct device *dev, 394 struct device_attribute *attr, const char *buf, size_t len) 395 { 396 struct nd_region *nd_region = to_nd_region(dev->parent); 397 ssize_t rc; 398 399 device_lock(dev); 400 nvdimm_bus_lock(dev); 401 wait_nvdimm_bus_probe_idle(dev); 402 rc = __alt_name_store(dev, buf, len); 403 if (rc >= 0) 404 rc = nd_namespace_label_update(nd_region, dev); 405 dev_dbg(dev, "%s: %s(%zd)\n", __func__, rc < 0 ? "fail " : "", rc); 406 nvdimm_bus_unlock(dev); 407 device_unlock(dev); 408 409 return rc < 0 ? rc : len; 410 } 411 412 static ssize_t alt_name_show(struct device *dev, 413 struct device_attribute *attr, char *buf) 414 { 415 char *ns_altname; 416 417 if (is_namespace_pmem(dev)) { 418 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 419 420 ns_altname = nspm->alt_name; 421 } else if (is_namespace_blk(dev)) { 422 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); 423 424 ns_altname = nsblk->alt_name; 425 } else 426 return -ENXIO; 427 428 return sprintf(buf, "%s\n", ns_altname ? ns_altname : ""); 429 } 430 static DEVICE_ATTR_RW(alt_name); 431 432 static int scan_free(struct nd_region *nd_region, 433 struct nd_mapping *nd_mapping, struct nd_label_id *label_id, 434 resource_size_t n) 435 { 436 bool is_blk = strncmp(label_id->id, "blk", 3) == 0; 437 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 438 int rc = 0; 439 440 while (n) { 441 struct resource *res, *last; 442 resource_size_t new_start; 443 444 last = NULL; 445 for_each_dpa_resource(ndd, res) 446 if (strcmp(res->name, label_id->id) == 0) 447 last = res; 448 res = last; 449 if (!res) 450 return 0; 451 452 if (n >= resource_size(res)) { 453 n -= resource_size(res); 454 nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc); 455 nvdimm_free_dpa(ndd, res); 456 /* retry with last resource deleted */ 457 continue; 458 } 459 460 /* 461 * Keep BLK allocations relegated to high DPA as much as 462 * possible 463 */ 464 if (is_blk) 465 new_start = res->start + n; 466 else 467 new_start = res->start; 468 469 rc = adjust_resource(res, new_start, resource_size(res) - n); 470 if (rc == 0) 471 res->flags |= DPA_RESOURCE_ADJUSTED; 472 nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc); 473 break; 474 } 475 476 return rc; 477 } 478 479 /** 480 * shrink_dpa_allocation - for each dimm in region free n bytes for label_id 481 * @nd_region: the set of dimms to reclaim @n bytes from 482 * @label_id: unique identifier for the namespace consuming this dpa range 483 * @n: number of bytes per-dimm to release 484 * 485 * Assumes resources are ordered. Starting from the end try to 486 * adjust_resource() the allocation to @n, but if @n is larger than the 487 * allocation delete it and find the 'new' last allocation in the label 488 * set. 489 */ 490 static int shrink_dpa_allocation(struct nd_region *nd_region, 491 struct nd_label_id *label_id, resource_size_t n) 492 { 493 int i; 494 495 for (i = 0; i < nd_region->ndr_mappings; i++) { 496 struct nd_mapping *nd_mapping = &nd_region->mapping[i]; 497 int rc; 498 499 rc = scan_free(nd_region, nd_mapping, label_id, n); 500 if (rc) 501 return rc; 502 } 503 504 return 0; 505 } 506 507 static resource_size_t init_dpa_allocation(struct nd_label_id *label_id, 508 struct nd_region *nd_region, struct nd_mapping *nd_mapping, 509 resource_size_t n) 510 { 511 bool is_blk = strncmp(label_id->id, "blk", 3) == 0; 512 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 513 resource_size_t first_dpa; 514 struct resource *res; 515 int rc = 0; 516 517 /* allocate blk from highest dpa first */ 518 if (is_blk) 519 first_dpa = nd_mapping->start + nd_mapping->size - n; 520 else 521 first_dpa = nd_mapping->start; 522 523 /* first resource allocation for this label-id or dimm */ 524 res = nvdimm_allocate_dpa(ndd, label_id, first_dpa, n); 525 if (!res) 526 rc = -EBUSY; 527 528 nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc); 529 return rc ? n : 0; 530 } 531 532 533 /** 534 * space_valid() - validate free dpa space against constraints 535 * @nd_region: hosting region of the free space 536 * @ndd: dimm device data for debug 537 * @label_id: namespace id to allocate space 538 * @prev: potential allocation that precedes free space 539 * @next: allocation that follows the given free space range 540 * @exist: first allocation with same id in the mapping 541 * @n: range that must satisfied for pmem allocations 542 * @valid: free space range to validate 543 * 544 * BLK-space is valid as long as it does not precede a PMEM 545 * allocation in a given region. PMEM-space must be contiguous 546 * and adjacent to an existing existing allocation (if one 547 * exists). If reserving PMEM any space is valid. 548 */ 549 static void space_valid(struct nd_region *nd_region, struct nvdimm_drvdata *ndd, 550 struct nd_label_id *label_id, struct resource *prev, 551 struct resource *next, struct resource *exist, 552 resource_size_t n, struct resource *valid) 553 { 554 bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0; 555 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0; 556 557 if (valid->start >= valid->end) 558 goto invalid; 559 560 if (is_reserve) 561 return; 562 563 if (!is_pmem) { 564 struct nd_mapping *nd_mapping = &nd_region->mapping[0]; 565 struct nvdimm_bus *nvdimm_bus; 566 struct blk_alloc_info info = { 567 .nd_mapping = nd_mapping, 568 .available = nd_mapping->size, 569 .res = valid, 570 }; 571 572 WARN_ON(!is_nd_blk(&nd_region->dev)); 573 nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev); 574 device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy); 575 return; 576 } 577 578 /* allocation needs to be contiguous, so this is all or nothing */ 579 if (resource_size(valid) < n) 580 goto invalid; 581 582 /* we've got all the space we need and no existing allocation */ 583 if (!exist) 584 return; 585 586 /* allocation needs to be contiguous with the existing namespace */ 587 if (valid->start == exist->end + 1 588 || valid->end == exist->start - 1) 589 return; 590 591 invalid: 592 /* truncate @valid size to 0 */ 593 valid->end = valid->start - 1; 594 } 595 596 enum alloc_loc { 597 ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER, 598 }; 599 600 static resource_size_t scan_allocate(struct nd_region *nd_region, 601 struct nd_mapping *nd_mapping, struct nd_label_id *label_id, 602 resource_size_t n) 603 { 604 resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1; 605 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0; 606 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 607 struct resource *res, *exist = NULL, valid; 608 const resource_size_t to_allocate = n; 609 int first; 610 611 for_each_dpa_resource(ndd, res) 612 if (strcmp(label_id->id, res->name) == 0) 613 exist = res; 614 615 valid.start = nd_mapping->start; 616 valid.end = mapping_end; 617 valid.name = "free space"; 618 retry: 619 first = 0; 620 for_each_dpa_resource(ndd, res) { 621 struct resource *next = res->sibling, *new_res = NULL; 622 resource_size_t allocate, available = 0; 623 enum alloc_loc loc = ALLOC_ERR; 624 const char *action; 625 int rc = 0; 626 627 /* ignore resources outside this nd_mapping */ 628 if (res->start > mapping_end) 629 continue; 630 if (res->end < nd_mapping->start) 631 continue; 632 633 /* space at the beginning of the mapping */ 634 if (!first++ && res->start > nd_mapping->start) { 635 valid.start = nd_mapping->start; 636 valid.end = res->start - 1; 637 space_valid(nd_region, ndd, label_id, NULL, next, exist, 638 to_allocate, &valid); 639 available = resource_size(&valid); 640 if (available) 641 loc = ALLOC_BEFORE; 642 } 643 644 /* space between allocations */ 645 if (!loc && next) { 646 valid.start = res->start + resource_size(res); 647 valid.end = min(mapping_end, next->start - 1); 648 space_valid(nd_region, ndd, label_id, res, next, exist, 649 to_allocate, &valid); 650 available = resource_size(&valid); 651 if (available) 652 loc = ALLOC_MID; 653 } 654 655 /* space at the end of the mapping */ 656 if (!loc && !next) { 657 valid.start = res->start + resource_size(res); 658 valid.end = mapping_end; 659 space_valid(nd_region, ndd, label_id, res, next, exist, 660 to_allocate, &valid); 661 available = resource_size(&valid); 662 if (available) 663 loc = ALLOC_AFTER; 664 } 665 666 if (!loc || !available) 667 continue; 668 allocate = min(available, n); 669 switch (loc) { 670 case ALLOC_BEFORE: 671 if (strcmp(res->name, label_id->id) == 0) { 672 /* adjust current resource up */ 673 rc = adjust_resource(res, res->start - allocate, 674 resource_size(res) + allocate); 675 action = "cur grow up"; 676 } else 677 action = "allocate"; 678 break; 679 case ALLOC_MID: 680 if (strcmp(next->name, label_id->id) == 0) { 681 /* adjust next resource up */ 682 rc = adjust_resource(next, next->start 683 - allocate, resource_size(next) 684 + allocate); 685 new_res = next; 686 action = "next grow up"; 687 } else if (strcmp(res->name, label_id->id) == 0) { 688 action = "grow down"; 689 } else 690 action = "allocate"; 691 break; 692 case ALLOC_AFTER: 693 if (strcmp(res->name, label_id->id) == 0) 694 action = "grow down"; 695 else 696 action = "allocate"; 697 break; 698 default: 699 return n; 700 } 701 702 if (strcmp(action, "allocate") == 0) { 703 /* BLK allocate bottom up */ 704 if (!is_pmem) 705 valid.start += available - allocate; 706 707 new_res = nvdimm_allocate_dpa(ndd, label_id, 708 valid.start, allocate); 709 if (!new_res) 710 rc = -EBUSY; 711 } else if (strcmp(action, "grow down") == 0) { 712 /* adjust current resource down */ 713 rc = adjust_resource(res, res->start, resource_size(res) 714 + allocate); 715 if (rc == 0) 716 res->flags |= DPA_RESOURCE_ADJUSTED; 717 } 718 719 if (!new_res) 720 new_res = res; 721 722 nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n", 723 action, loc, rc); 724 725 if (rc) 726 return n; 727 728 n -= allocate; 729 if (n) { 730 /* 731 * Retry scan with newly inserted resources. 732 * For example, if we did an ALLOC_BEFORE 733 * insertion there may also have been space 734 * available for an ALLOC_AFTER insertion, so we 735 * need to check this same resource again 736 */ 737 goto retry; 738 } else 739 return 0; 740 } 741 742 /* 743 * If we allocated nothing in the BLK case it may be because we are in 744 * an initial "pmem-reserve pass". Only do an initial BLK allocation 745 * when none of the DPA space is reserved. 746 */ 747 if ((is_pmem || !ndd->dpa.child) && n == to_allocate) 748 return init_dpa_allocation(label_id, nd_region, nd_mapping, n); 749 return n; 750 } 751 752 static int merge_dpa(struct nd_region *nd_region, 753 struct nd_mapping *nd_mapping, struct nd_label_id *label_id) 754 { 755 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 756 struct resource *res; 757 758 if (strncmp("pmem", label_id->id, 4) == 0) 759 return 0; 760 retry: 761 for_each_dpa_resource(ndd, res) { 762 int rc; 763 struct resource *next = res->sibling; 764 resource_size_t end = res->start + resource_size(res); 765 766 if (!next || strcmp(res->name, label_id->id) != 0 767 || strcmp(next->name, label_id->id) != 0 768 || end != next->start) 769 continue; 770 end += resource_size(next); 771 nvdimm_free_dpa(ndd, next); 772 rc = adjust_resource(res, res->start, end - res->start); 773 nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc); 774 if (rc) 775 return rc; 776 res->flags |= DPA_RESOURCE_ADJUSTED; 777 goto retry; 778 } 779 780 return 0; 781 } 782 783 static int __reserve_free_pmem(struct device *dev, void *data) 784 { 785 struct nvdimm *nvdimm = data; 786 struct nd_region *nd_region; 787 struct nd_label_id label_id; 788 int i; 789 790 if (!is_nd_pmem(dev)) 791 return 0; 792 793 nd_region = to_nd_region(dev); 794 if (nd_region->ndr_mappings == 0) 795 return 0; 796 797 memset(&label_id, 0, sizeof(label_id)); 798 strcat(label_id.id, "pmem-reserve"); 799 for (i = 0; i < nd_region->ndr_mappings; i++) { 800 struct nd_mapping *nd_mapping = &nd_region->mapping[i]; 801 resource_size_t n, rem = 0; 802 803 if (nd_mapping->nvdimm != nvdimm) 804 continue; 805 806 n = nd_pmem_available_dpa(nd_region, nd_mapping, &rem); 807 if (n == 0) 808 return 0; 809 rem = scan_allocate(nd_region, nd_mapping, &label_id, n); 810 dev_WARN_ONCE(&nd_region->dev, rem, 811 "pmem reserve underrun: %#llx of %#llx bytes\n", 812 (unsigned long long) n - rem, 813 (unsigned long long) n); 814 return rem ? -ENXIO : 0; 815 } 816 817 return 0; 818 } 819 820 static void release_free_pmem(struct nvdimm_bus *nvdimm_bus, 821 struct nd_mapping *nd_mapping) 822 { 823 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 824 struct resource *res, *_res; 825 826 for_each_dpa_resource_safe(ndd, res, _res) 827 if (strcmp(res->name, "pmem-reserve") == 0) 828 nvdimm_free_dpa(ndd, res); 829 } 830 831 static int reserve_free_pmem(struct nvdimm_bus *nvdimm_bus, 832 struct nd_mapping *nd_mapping) 833 { 834 struct nvdimm *nvdimm = nd_mapping->nvdimm; 835 int rc; 836 837 rc = device_for_each_child(&nvdimm_bus->dev, nvdimm, 838 __reserve_free_pmem); 839 if (rc) 840 release_free_pmem(nvdimm_bus, nd_mapping); 841 return rc; 842 } 843 844 /** 845 * grow_dpa_allocation - for each dimm allocate n bytes for @label_id 846 * @nd_region: the set of dimms to allocate @n more bytes from 847 * @label_id: unique identifier for the namespace consuming this dpa range 848 * @n: number of bytes per-dimm to add to the existing allocation 849 * 850 * Assumes resources are ordered. For BLK regions, first consume 851 * BLK-only available DPA free space, then consume PMEM-aliased DPA 852 * space starting at the highest DPA. For PMEM regions start 853 * allocations from the start of an interleave set and end at the first 854 * BLK allocation or the end of the interleave set, whichever comes 855 * first. 856 */ 857 static int grow_dpa_allocation(struct nd_region *nd_region, 858 struct nd_label_id *label_id, resource_size_t n) 859 { 860 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev); 861 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0; 862 int i; 863 864 for (i = 0; i < nd_region->ndr_mappings; i++) { 865 struct nd_mapping *nd_mapping = &nd_region->mapping[i]; 866 resource_size_t rem = n; 867 int rc, j; 868 869 /* 870 * In the BLK case try once with all unallocated PMEM 871 * reserved, and once without 872 */ 873 for (j = is_pmem; j < 2; j++) { 874 bool blk_only = j == 0; 875 876 if (blk_only) { 877 rc = reserve_free_pmem(nvdimm_bus, nd_mapping); 878 if (rc) 879 return rc; 880 } 881 rem = scan_allocate(nd_region, nd_mapping, 882 label_id, rem); 883 if (blk_only) 884 release_free_pmem(nvdimm_bus, nd_mapping); 885 886 /* try again and allow encroachments into PMEM */ 887 if (rem == 0) 888 break; 889 } 890 891 dev_WARN_ONCE(&nd_region->dev, rem, 892 "allocation underrun: %#llx of %#llx bytes\n", 893 (unsigned long long) n - rem, 894 (unsigned long long) n); 895 if (rem) 896 return -ENXIO; 897 898 rc = merge_dpa(nd_region, nd_mapping, label_id); 899 if (rc) 900 return rc; 901 } 902 903 return 0; 904 } 905 906 static void nd_namespace_pmem_set_resource(struct nd_region *nd_region, 907 struct nd_namespace_pmem *nspm, resource_size_t size) 908 { 909 struct resource *res = &nspm->nsio.res; 910 resource_size_t offset = 0; 911 912 if (size && !nspm->uuid) { 913 WARN_ON_ONCE(1); 914 size = 0; 915 } 916 917 if (size && nspm->uuid) { 918 struct nd_mapping *nd_mapping = &nd_region->mapping[0]; 919 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 920 struct nd_label_id label_id; 921 struct resource *res; 922 923 if (!ndd) { 924 size = 0; 925 goto out; 926 } 927 928 nd_label_gen_id(&label_id, nspm->uuid, 0); 929 930 /* calculate a spa offset from the dpa allocation offset */ 931 for_each_dpa_resource(ndd, res) 932 if (strcmp(res->name, label_id.id) == 0) { 933 offset = (res->start - nd_mapping->start) 934 * nd_region->ndr_mappings; 935 goto out; 936 } 937 938 WARN_ON_ONCE(1); 939 size = 0; 940 } 941 942 out: 943 res->start = nd_region->ndr_start + offset; 944 res->end = res->start + size - 1; 945 } 946 947 static bool uuid_not_set(const u8 *uuid, struct device *dev, const char *where) 948 { 949 if (!uuid) { 950 dev_dbg(dev, "%s: uuid not set\n", where); 951 return true; 952 } 953 return false; 954 } 955 956 static ssize_t __size_store(struct device *dev, unsigned long long val) 957 { 958 resource_size_t allocated = 0, available = 0; 959 struct nd_region *nd_region = to_nd_region(dev->parent); 960 struct nd_namespace_common *ndns = to_ndns(dev); 961 struct nd_mapping *nd_mapping; 962 struct nvdimm_drvdata *ndd; 963 struct nd_label_id label_id; 964 u32 flags = 0, remainder; 965 u8 *uuid = NULL; 966 int rc, i; 967 968 if (dev->driver || ndns->claim) 969 return -EBUSY; 970 971 if (is_namespace_pmem(dev)) { 972 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 973 974 uuid = nspm->uuid; 975 } else if (is_namespace_blk(dev)) { 976 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); 977 978 uuid = nsblk->uuid; 979 flags = NSLABEL_FLAG_LOCAL; 980 } 981 982 /* 983 * We need a uuid for the allocation-label and dimm(s) on which 984 * to store the label. 985 */ 986 if (uuid_not_set(uuid, dev, __func__)) 987 return -ENXIO; 988 if (nd_region->ndr_mappings == 0) { 989 dev_dbg(dev, "%s: not associated with dimm(s)\n", __func__); 990 return -ENXIO; 991 } 992 993 div_u64_rem(val, SZ_4K * nd_region->ndr_mappings, &remainder); 994 if (remainder) { 995 dev_dbg(dev, "%llu is not %dK aligned\n", val, 996 (SZ_4K * nd_region->ndr_mappings) / SZ_1K); 997 return -EINVAL; 998 } 999 1000 nd_label_gen_id(&label_id, uuid, flags); 1001 for (i = 0; i < nd_region->ndr_mappings; i++) { 1002 nd_mapping = &nd_region->mapping[i]; 1003 ndd = to_ndd(nd_mapping); 1004 1005 /* 1006 * All dimms in an interleave set, or the base dimm for a blk 1007 * region, need to be enabled for the size to be changed. 1008 */ 1009 if (!ndd) 1010 return -ENXIO; 1011 1012 allocated += nvdimm_allocated_dpa(ndd, &label_id); 1013 } 1014 available = nd_region_available_dpa(nd_region); 1015 1016 if (val > available + allocated) 1017 return -ENOSPC; 1018 1019 if (val == allocated) 1020 return 0; 1021 1022 val = div_u64(val, nd_region->ndr_mappings); 1023 allocated = div_u64(allocated, nd_region->ndr_mappings); 1024 if (val < allocated) 1025 rc = shrink_dpa_allocation(nd_region, &label_id, 1026 allocated - val); 1027 else 1028 rc = grow_dpa_allocation(nd_region, &label_id, val - allocated); 1029 1030 if (rc) 1031 return rc; 1032 1033 if (is_namespace_pmem(dev)) { 1034 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 1035 1036 nd_namespace_pmem_set_resource(nd_region, nspm, 1037 val * nd_region->ndr_mappings); 1038 } 1039 1040 /* 1041 * Try to delete the namespace if we deleted all of its 1042 * allocation, this is not the seed device for the region, and 1043 * it is not actively claimed by a btt instance. 1044 */ 1045 if (val == 0 && nd_region->ns_seed != dev && !ndns->claim) 1046 nd_device_unregister(dev, ND_ASYNC); 1047 1048 return rc; 1049 } 1050 1051 static ssize_t size_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 unsigned long long val; 1056 u8 **uuid = NULL; 1057 int rc; 1058 1059 rc = kstrtoull(buf, 0, &val); 1060 if (rc) 1061 return rc; 1062 1063 device_lock(dev); 1064 nvdimm_bus_lock(dev); 1065 wait_nvdimm_bus_probe_idle(dev); 1066 rc = __size_store(dev, val); 1067 if (rc >= 0) 1068 rc = nd_namespace_label_update(nd_region, dev); 1069 1070 if (is_namespace_pmem(dev)) { 1071 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 1072 1073 uuid = &nspm->uuid; 1074 } else if (is_namespace_blk(dev)) { 1075 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); 1076 1077 uuid = &nsblk->uuid; 1078 } 1079 1080 if (rc == 0 && val == 0 && uuid) { 1081 /* setting size zero == 'delete namespace' */ 1082 kfree(*uuid); 1083 *uuid = NULL; 1084 } 1085 1086 dev_dbg(dev, "%s: %llx %s (%d)\n", __func__, val, rc < 0 1087 ? "fail" : "success", rc); 1088 1089 nvdimm_bus_unlock(dev); 1090 device_unlock(dev); 1091 1092 return rc < 0 ? rc : len; 1093 } 1094 1095 resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns) 1096 { 1097 struct device *dev = &ndns->dev; 1098 1099 if (is_namespace_pmem(dev)) { 1100 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 1101 1102 return resource_size(&nspm->nsio.res); 1103 } else if (is_namespace_blk(dev)) { 1104 return nd_namespace_blk_size(to_nd_namespace_blk(dev)); 1105 } else if (is_namespace_io(dev)) { 1106 struct nd_namespace_io *nsio = to_nd_namespace_io(dev); 1107 1108 return resource_size(&nsio->res); 1109 } else 1110 WARN_ONCE(1, "unknown namespace type\n"); 1111 return 0; 1112 } 1113 1114 resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns) 1115 { 1116 resource_size_t size; 1117 1118 nvdimm_bus_lock(&ndns->dev); 1119 size = __nvdimm_namespace_capacity(ndns); 1120 nvdimm_bus_unlock(&ndns->dev); 1121 1122 return size; 1123 } 1124 EXPORT_SYMBOL(nvdimm_namespace_capacity); 1125 1126 static ssize_t size_show(struct device *dev, 1127 struct device_attribute *attr, char *buf) 1128 { 1129 return sprintf(buf, "%llu\n", (unsigned long long) 1130 nvdimm_namespace_capacity(to_ndns(dev))); 1131 } 1132 static DEVICE_ATTR(size, 0444, size_show, size_store); 1133 1134 static u8 *namespace_to_uuid(struct device *dev) 1135 { 1136 if (is_namespace_pmem(dev)) { 1137 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 1138 1139 return nspm->uuid; 1140 } else if (is_namespace_blk(dev)) { 1141 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); 1142 1143 return nsblk->uuid; 1144 } else 1145 return ERR_PTR(-ENXIO); 1146 } 1147 1148 static ssize_t uuid_show(struct device *dev, 1149 struct device_attribute *attr, char *buf) 1150 { 1151 u8 *uuid = namespace_to_uuid(dev); 1152 1153 if (IS_ERR(uuid)) 1154 return PTR_ERR(uuid); 1155 if (uuid) 1156 return sprintf(buf, "%pUb\n", uuid); 1157 return sprintf(buf, "\n"); 1158 } 1159 1160 /** 1161 * namespace_update_uuid - check for a unique uuid and whether we're "renaming" 1162 * @nd_region: parent region so we can updates all dimms in the set 1163 * @dev: namespace type for generating label_id 1164 * @new_uuid: incoming uuid 1165 * @old_uuid: reference to the uuid storage location in the namespace object 1166 */ 1167 static int namespace_update_uuid(struct nd_region *nd_region, 1168 struct device *dev, u8 *new_uuid, u8 **old_uuid) 1169 { 1170 u32 flags = is_namespace_blk(dev) ? NSLABEL_FLAG_LOCAL : 0; 1171 struct nd_label_id old_label_id; 1172 struct nd_label_id new_label_id; 1173 int i; 1174 1175 if (!nd_is_uuid_unique(dev, new_uuid)) 1176 return -EINVAL; 1177 1178 if (*old_uuid == NULL) 1179 goto out; 1180 1181 /* 1182 * If we've already written a label with this uuid, then it's 1183 * too late to rename because we can't reliably update the uuid 1184 * without losing the old namespace. Userspace must delete this 1185 * namespace to abandon the old uuid. 1186 */ 1187 for (i = 0; i < nd_region->ndr_mappings; i++) { 1188 struct nd_mapping *nd_mapping = &nd_region->mapping[i]; 1189 1190 /* 1191 * This check by itself is sufficient because old_uuid 1192 * would be NULL above if this uuid did not exist in the 1193 * currently written set. 1194 * 1195 * FIXME: can we delete uuid with zero dpa allocated? 1196 */ 1197 if (list_empty(&nd_mapping->labels)) 1198 return -EBUSY; 1199 } 1200 1201 nd_label_gen_id(&old_label_id, *old_uuid, flags); 1202 nd_label_gen_id(&new_label_id, new_uuid, flags); 1203 for (i = 0; i < nd_region->ndr_mappings; i++) { 1204 struct nd_mapping *nd_mapping = &nd_region->mapping[i]; 1205 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 1206 struct resource *res; 1207 1208 for_each_dpa_resource(ndd, res) 1209 if (strcmp(res->name, old_label_id.id) == 0) 1210 sprintf((void *) res->name, "%s", 1211 new_label_id.id); 1212 } 1213 kfree(*old_uuid); 1214 out: 1215 *old_uuid = new_uuid; 1216 return 0; 1217 } 1218 1219 static ssize_t uuid_store(struct device *dev, 1220 struct device_attribute *attr, const char *buf, size_t len) 1221 { 1222 struct nd_region *nd_region = to_nd_region(dev->parent); 1223 u8 *uuid = NULL; 1224 ssize_t rc = 0; 1225 u8 **ns_uuid; 1226 1227 if (is_namespace_pmem(dev)) { 1228 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 1229 1230 ns_uuid = &nspm->uuid; 1231 } else if (is_namespace_blk(dev)) { 1232 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); 1233 1234 ns_uuid = &nsblk->uuid; 1235 } else 1236 return -ENXIO; 1237 1238 device_lock(dev); 1239 nvdimm_bus_lock(dev); 1240 wait_nvdimm_bus_probe_idle(dev); 1241 if (to_ndns(dev)->claim) 1242 rc = -EBUSY; 1243 if (rc >= 0) 1244 rc = nd_uuid_store(dev, &uuid, buf, len); 1245 if (rc >= 0) 1246 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid); 1247 if (rc >= 0) 1248 rc = nd_namespace_label_update(nd_region, dev); 1249 else 1250 kfree(uuid); 1251 dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__, 1252 rc, buf, buf[len - 1] == '\n' ? "" : "\n"); 1253 nvdimm_bus_unlock(dev); 1254 device_unlock(dev); 1255 1256 return rc < 0 ? rc : len; 1257 } 1258 static DEVICE_ATTR_RW(uuid); 1259 1260 static ssize_t resource_show(struct device *dev, 1261 struct device_attribute *attr, char *buf) 1262 { 1263 struct resource *res; 1264 1265 if (is_namespace_pmem(dev)) { 1266 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 1267 1268 res = &nspm->nsio.res; 1269 } else if (is_namespace_io(dev)) { 1270 struct nd_namespace_io *nsio = to_nd_namespace_io(dev); 1271 1272 res = &nsio->res; 1273 } else 1274 return -ENXIO; 1275 1276 /* no address to convey if the namespace has no allocation */ 1277 if (resource_size(res) == 0) 1278 return -ENXIO; 1279 return sprintf(buf, "%#llx\n", (unsigned long long) res->start); 1280 } 1281 static DEVICE_ATTR_RO(resource); 1282 1283 static const unsigned long ns_lbasize_supported[] = { 512, 520, 528, 1284 4096, 4104, 4160, 4224, 0 }; 1285 1286 static ssize_t sector_size_show(struct device *dev, 1287 struct device_attribute *attr, char *buf) 1288 { 1289 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); 1290 1291 if (!is_namespace_blk(dev)) 1292 return -ENXIO; 1293 1294 return nd_sector_size_show(nsblk->lbasize, ns_lbasize_supported, buf); 1295 } 1296 1297 static ssize_t sector_size_store(struct device *dev, 1298 struct device_attribute *attr, const char *buf, size_t len) 1299 { 1300 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); 1301 struct nd_region *nd_region = to_nd_region(dev->parent); 1302 ssize_t rc = 0; 1303 1304 if (!is_namespace_blk(dev)) 1305 return -ENXIO; 1306 1307 device_lock(dev); 1308 nvdimm_bus_lock(dev); 1309 if (to_ndns(dev)->claim) 1310 rc = -EBUSY; 1311 if (rc >= 0) 1312 rc = nd_sector_size_store(dev, buf, &nsblk->lbasize, 1313 ns_lbasize_supported); 1314 if (rc >= 0) 1315 rc = nd_namespace_label_update(nd_region, dev); 1316 dev_dbg(dev, "%s: result: %zd %s: %s%s", __func__, 1317 rc, rc < 0 ? "tried" : "wrote", buf, 1318 buf[len - 1] == '\n' ? "" : "\n"); 1319 nvdimm_bus_unlock(dev); 1320 device_unlock(dev); 1321 1322 return rc ? rc : len; 1323 } 1324 static DEVICE_ATTR_RW(sector_size); 1325 1326 static ssize_t dpa_extents_show(struct device *dev, 1327 struct device_attribute *attr, char *buf) 1328 { 1329 struct nd_region *nd_region = to_nd_region(dev->parent); 1330 struct nd_label_id label_id; 1331 int count = 0, i; 1332 u8 *uuid = NULL; 1333 u32 flags = 0; 1334 1335 nvdimm_bus_lock(dev); 1336 if (is_namespace_pmem(dev)) { 1337 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); 1338 1339 uuid = nspm->uuid; 1340 flags = 0; 1341 } else if (is_namespace_blk(dev)) { 1342 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev); 1343 1344 uuid = nsblk->uuid; 1345 flags = NSLABEL_FLAG_LOCAL; 1346 } 1347 1348 if (!uuid) 1349 goto out; 1350 1351 nd_label_gen_id(&label_id, uuid, flags); 1352 for (i = 0; i < nd_region->ndr_mappings; i++) { 1353 struct nd_mapping *nd_mapping = &nd_region->mapping[i]; 1354 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 1355 struct resource *res; 1356 1357 for_each_dpa_resource(ndd, res) 1358 if (strcmp(res->name, label_id.id) == 0) 1359 count++; 1360 } 1361 out: 1362 nvdimm_bus_unlock(dev); 1363 1364 return sprintf(buf, "%d\n", count); 1365 } 1366 static DEVICE_ATTR_RO(dpa_extents); 1367 1368 static ssize_t holder_show(struct device *dev, 1369 struct device_attribute *attr, char *buf) 1370 { 1371 struct nd_namespace_common *ndns = to_ndns(dev); 1372 ssize_t rc; 1373 1374 device_lock(dev); 1375 rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : ""); 1376 device_unlock(dev); 1377 1378 return rc; 1379 } 1380 static DEVICE_ATTR_RO(holder); 1381 1382 static ssize_t mode_show(struct device *dev, 1383 struct device_attribute *attr, char *buf) 1384 { 1385 struct nd_namespace_common *ndns = to_ndns(dev); 1386 struct device *claim; 1387 char *mode; 1388 ssize_t rc; 1389 1390 device_lock(dev); 1391 claim = ndns->claim; 1392 if (claim && is_nd_btt(claim)) 1393 mode = "safe"; 1394 else if (claim && is_nd_pfn(claim)) 1395 mode = "memory"; 1396 else if (claim && is_nd_dax(claim)) 1397 mode = "dax"; 1398 else if (!claim && pmem_should_map_pages(dev)) 1399 mode = "memory"; 1400 else 1401 mode = "raw"; 1402 rc = sprintf(buf, "%s\n", mode); 1403 device_unlock(dev); 1404 1405 return rc; 1406 } 1407 static DEVICE_ATTR_RO(mode); 1408 1409 static ssize_t force_raw_store(struct device *dev, 1410 struct device_attribute *attr, const char *buf, size_t len) 1411 { 1412 bool force_raw; 1413 int rc = strtobool(buf, &force_raw); 1414 1415 if (rc) 1416 return rc; 1417 1418 to_ndns(dev)->force_raw = force_raw; 1419 return len; 1420 } 1421 1422 static ssize_t force_raw_show(struct device *dev, 1423 struct device_attribute *attr, char *buf) 1424 { 1425 return sprintf(buf, "%d\n", to_ndns(dev)->force_raw); 1426 } 1427 static DEVICE_ATTR_RW(force_raw); 1428 1429 static struct attribute *nd_namespace_attributes[] = { 1430 &dev_attr_nstype.attr, 1431 &dev_attr_size.attr, 1432 &dev_attr_mode.attr, 1433 &dev_attr_uuid.attr, 1434 &dev_attr_holder.attr, 1435 &dev_attr_resource.attr, 1436 &dev_attr_alt_name.attr, 1437 &dev_attr_force_raw.attr, 1438 &dev_attr_sector_size.attr, 1439 &dev_attr_dpa_extents.attr, 1440 NULL, 1441 }; 1442 1443 static umode_t namespace_visible(struct kobject *kobj, 1444 struct attribute *a, int n) 1445 { 1446 struct device *dev = container_of(kobj, struct device, kobj); 1447 1448 if (a == &dev_attr_resource.attr) { 1449 if (is_namespace_blk(dev)) 1450 return 0; 1451 return a->mode; 1452 } 1453 1454 if (is_namespace_pmem(dev) || is_namespace_blk(dev)) { 1455 if (a == &dev_attr_size.attr) 1456 return 0644; 1457 1458 if (is_namespace_pmem(dev) && a == &dev_attr_sector_size.attr) 1459 return 0; 1460 1461 return a->mode; 1462 } 1463 1464 if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr 1465 || a == &dev_attr_holder.attr 1466 || a == &dev_attr_force_raw.attr 1467 || a == &dev_attr_mode.attr) 1468 return a->mode; 1469 1470 return 0; 1471 } 1472 1473 static struct attribute_group nd_namespace_attribute_group = { 1474 .attrs = nd_namespace_attributes, 1475 .is_visible = namespace_visible, 1476 }; 1477 1478 static const struct attribute_group *nd_namespace_attribute_groups[] = { 1479 &nd_device_attribute_group, 1480 &nd_namespace_attribute_group, 1481 &nd_numa_attribute_group, 1482 NULL, 1483 }; 1484 1485 struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev) 1486 { 1487 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL; 1488 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL; 1489 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL; 1490 struct nd_namespace_common *ndns = NULL; 1491 resource_size_t size; 1492 1493 if (nd_btt || nd_pfn || nd_dax) { 1494 if (nd_btt) 1495 ndns = nd_btt->ndns; 1496 else if (nd_pfn) 1497 ndns = nd_pfn->ndns; 1498 else if (nd_dax) 1499 ndns = nd_dax->nd_pfn.ndns; 1500 1501 if (!ndns) 1502 return ERR_PTR(-ENODEV); 1503 1504 /* 1505 * Flush any in-progess probes / removals in the driver 1506 * for the raw personality of this namespace. 1507 */ 1508 device_lock(&ndns->dev); 1509 device_unlock(&ndns->dev); 1510 if (ndns->dev.driver) { 1511 dev_dbg(&ndns->dev, "is active, can't bind %s\n", 1512 dev_name(dev)); 1513 return ERR_PTR(-EBUSY); 1514 } 1515 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != dev, 1516 "host (%s) vs claim (%s) mismatch\n", 1517 dev_name(dev), 1518 dev_name(ndns->claim))) 1519 return ERR_PTR(-ENXIO); 1520 } else { 1521 ndns = to_ndns(dev); 1522 if (ndns->claim) { 1523 dev_dbg(dev, "claimed by %s, failing probe\n", 1524 dev_name(ndns->claim)); 1525 1526 return ERR_PTR(-ENXIO); 1527 } 1528 } 1529 1530 size = nvdimm_namespace_capacity(ndns); 1531 if (size < ND_MIN_NAMESPACE_SIZE) { 1532 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n", 1533 &size, ND_MIN_NAMESPACE_SIZE); 1534 return ERR_PTR(-ENODEV); 1535 } 1536 1537 if (is_namespace_pmem(&ndns->dev)) { 1538 struct nd_namespace_pmem *nspm; 1539 1540 nspm = to_nd_namespace_pmem(&ndns->dev); 1541 if (uuid_not_set(nspm->uuid, &ndns->dev, __func__)) 1542 return ERR_PTR(-ENODEV); 1543 } else if (is_namespace_blk(&ndns->dev)) { 1544 struct nd_namespace_blk *nsblk; 1545 1546 nsblk = to_nd_namespace_blk(&ndns->dev); 1547 if (uuid_not_set(nsblk->uuid, &ndns->dev, __func__)) 1548 return ERR_PTR(-ENODEV); 1549 if (!nsblk->lbasize) { 1550 dev_dbg(&ndns->dev, "%s: sector size not set\n", 1551 __func__); 1552 return ERR_PTR(-ENODEV); 1553 } 1554 if (!nd_namespace_blk_validate(nsblk)) 1555 return ERR_PTR(-ENODEV); 1556 } 1557 1558 return ndns; 1559 } 1560 EXPORT_SYMBOL(nvdimm_namespace_common_probe); 1561 1562 static struct device **create_namespace_io(struct nd_region *nd_region) 1563 { 1564 struct nd_namespace_io *nsio; 1565 struct device *dev, **devs; 1566 struct resource *res; 1567 1568 nsio = kzalloc(sizeof(*nsio), GFP_KERNEL); 1569 if (!nsio) 1570 return NULL; 1571 1572 devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL); 1573 if (!devs) { 1574 kfree(nsio); 1575 return NULL; 1576 } 1577 1578 dev = &nsio->common.dev; 1579 dev->type = &namespace_io_device_type; 1580 dev->parent = &nd_region->dev; 1581 res = &nsio->res; 1582 res->name = dev_name(&nd_region->dev); 1583 res->flags = IORESOURCE_MEM; 1584 res->start = nd_region->ndr_start; 1585 res->end = res->start + nd_region->ndr_size - 1; 1586 1587 devs[0] = dev; 1588 return devs; 1589 } 1590 1591 static bool has_uuid_at_pos(struct nd_region *nd_region, u8 *uuid, 1592 u64 cookie, u16 pos) 1593 { 1594 struct nd_namespace_label *found = NULL; 1595 int i; 1596 1597 for (i = 0; i < nd_region->ndr_mappings; i++) { 1598 struct nd_mapping *nd_mapping = &nd_region->mapping[i]; 1599 struct nd_label_ent *label_ent; 1600 bool found_uuid = false; 1601 1602 list_for_each_entry(label_ent, &nd_mapping->labels, list) { 1603 struct nd_namespace_label *nd_label = label_ent->label; 1604 u16 position, nlabel; 1605 u64 isetcookie; 1606 1607 if (!nd_label) 1608 continue; 1609 isetcookie = __le64_to_cpu(nd_label->isetcookie); 1610 position = __le16_to_cpu(nd_label->position); 1611 nlabel = __le16_to_cpu(nd_label->nlabel); 1612 1613 if (isetcookie != cookie) 1614 continue; 1615 1616 if (memcmp(nd_label->uuid, uuid, NSLABEL_UUID_LEN) != 0) 1617 continue; 1618 1619 if (found_uuid) { 1620 dev_dbg(to_ndd(nd_mapping)->dev, 1621 "%s duplicate entry for uuid\n", 1622 __func__); 1623 return false; 1624 } 1625 found_uuid = true; 1626 if (nlabel != nd_region->ndr_mappings) 1627 continue; 1628 if (position != pos) 1629 continue; 1630 found = nd_label; 1631 break; 1632 } 1633 if (found) 1634 break; 1635 } 1636 return found != NULL; 1637 } 1638 1639 static int select_pmem_id(struct nd_region *nd_region, u8 *pmem_id) 1640 { 1641 int i; 1642 1643 if (!pmem_id) 1644 return -ENODEV; 1645 1646 for (i = 0; i < nd_region->ndr_mappings; i++) { 1647 struct nd_mapping *nd_mapping = &nd_region->mapping[i]; 1648 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 1649 struct nd_namespace_label *nd_label = NULL; 1650 u64 hw_start, hw_end, pmem_start, pmem_end; 1651 struct nd_label_ent *label_ent; 1652 1653 lockdep_assert_held(&nd_mapping->lock); 1654 list_for_each_entry(label_ent, &nd_mapping->labels, list) { 1655 nd_label = label_ent->label; 1656 if (!nd_label) 1657 continue; 1658 if (memcmp(nd_label->uuid, pmem_id, NSLABEL_UUID_LEN) == 0) 1659 break; 1660 nd_label = NULL; 1661 } 1662 1663 if (!nd_label) { 1664 WARN_ON(1); 1665 return -EINVAL; 1666 } 1667 1668 /* 1669 * Check that this label is compliant with the dpa 1670 * range published in NFIT 1671 */ 1672 hw_start = nd_mapping->start; 1673 hw_end = hw_start + nd_mapping->size; 1674 pmem_start = __le64_to_cpu(nd_label->dpa); 1675 pmem_end = pmem_start + __le64_to_cpu(nd_label->rawsize); 1676 if (pmem_start >= hw_start && pmem_start < hw_end 1677 && pmem_end <= hw_end && pmem_end > hw_start) 1678 /* pass */; 1679 else { 1680 dev_dbg(&nd_region->dev, "%s invalid label for %pUb\n", 1681 dev_name(ndd->dev), nd_label->uuid); 1682 return -EINVAL; 1683 } 1684 1685 /* move recently validated label to the front of the list */ 1686 list_move(&label_ent->list, &nd_mapping->labels); 1687 } 1688 return 0; 1689 } 1690 1691 /** 1692 * create_namespace_pmem - validate interleave set labelling, retrieve label0 1693 * @nd_region: region with mappings to validate 1694 * @nspm: target namespace to create 1695 * @nd_label: target pmem namespace label to evaluate 1696 */ 1697 struct device *create_namespace_pmem(struct nd_region *nd_region, 1698 struct nd_namespace_label *nd_label) 1699 { 1700 u64 cookie = nd_region_interleave_set_cookie(nd_region); 1701 struct nd_label_ent *label_ent; 1702 struct nd_namespace_pmem *nspm; 1703 struct nd_mapping *nd_mapping; 1704 resource_size_t size = 0; 1705 struct resource *res; 1706 struct device *dev; 1707 int rc = 0; 1708 u16 i; 1709 1710 if (cookie == 0) { 1711 dev_dbg(&nd_region->dev, "invalid interleave-set-cookie\n"); 1712 return ERR_PTR(-ENXIO); 1713 } 1714 1715 if (__le64_to_cpu(nd_label->isetcookie) != cookie) { 1716 dev_dbg(&nd_region->dev, "invalid cookie in label: %pUb\n", 1717 nd_label->uuid); 1718 return ERR_PTR(-EAGAIN); 1719 } 1720 1721 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL); 1722 if (!nspm) 1723 return ERR_PTR(-ENOMEM); 1724 1725 nspm->id = -1; 1726 dev = &nspm->nsio.common.dev; 1727 dev->type = &namespace_pmem_device_type; 1728 dev->parent = &nd_region->dev; 1729 res = &nspm->nsio.res; 1730 res->name = dev_name(&nd_region->dev); 1731 res->flags = IORESOURCE_MEM; 1732 1733 for (i = 0; i < nd_region->ndr_mappings; i++) 1734 if (!has_uuid_at_pos(nd_region, nd_label->uuid, cookie, i)) 1735 break; 1736 if (i < nd_region->ndr_mappings) { 1737 struct nvdimm_drvdata *ndd = to_ndd(&nd_region->mapping[i]); 1738 1739 /* 1740 * Give up if we don't find an instance of a uuid at each 1741 * position (from 0 to nd_region->ndr_mappings - 1), or if we 1742 * find a dimm with two instances of the same uuid. 1743 */ 1744 dev_err(&nd_region->dev, "%s missing label for %pUb\n", 1745 dev_name(ndd->dev), nd_label->uuid); 1746 rc = -EINVAL; 1747 goto err; 1748 } 1749 1750 /* 1751 * Fix up each mapping's 'labels' to have the validated pmem label for 1752 * that position at labels[0], and NULL at labels[1]. In the process, 1753 * check that the namespace aligns with interleave-set. We know 1754 * that it does not overlap with any blk namespaces by virtue of 1755 * the dimm being enabled (i.e. nd_label_reserve_dpa() 1756 * succeeded). 1757 */ 1758 rc = select_pmem_id(nd_region, nd_label->uuid); 1759 if (rc) 1760 goto err; 1761 1762 /* Calculate total size and populate namespace properties from label0 */ 1763 for (i = 0; i < nd_region->ndr_mappings; i++) { 1764 struct nd_namespace_label *label0; 1765 1766 nd_mapping = &nd_region->mapping[i]; 1767 label_ent = list_first_entry_or_null(&nd_mapping->labels, 1768 typeof(*label_ent), list); 1769 label0 = label_ent ? label_ent->label : 0; 1770 1771 if (!label0) { 1772 WARN_ON(1); 1773 continue; 1774 } 1775 1776 size += __le64_to_cpu(label0->rawsize); 1777 if (__le16_to_cpu(label0->position) != 0) 1778 continue; 1779 WARN_ON(nspm->alt_name || nspm->uuid); 1780 nspm->alt_name = kmemdup((void __force *) label0->name, 1781 NSLABEL_NAME_LEN, GFP_KERNEL); 1782 nspm->uuid = kmemdup((void __force *) label0->uuid, 1783 NSLABEL_UUID_LEN, GFP_KERNEL); 1784 } 1785 1786 if (!nspm->alt_name || !nspm->uuid) { 1787 rc = -ENOMEM; 1788 goto err; 1789 } 1790 1791 nd_namespace_pmem_set_resource(nd_region, nspm, size); 1792 1793 return dev; 1794 err: 1795 namespace_pmem_release(dev); 1796 switch (rc) { 1797 case -EINVAL: 1798 dev_dbg(&nd_region->dev, "%s: invalid label(s)\n", __func__); 1799 break; 1800 case -ENODEV: 1801 dev_dbg(&nd_region->dev, "%s: label not found\n", __func__); 1802 break; 1803 default: 1804 dev_dbg(&nd_region->dev, "%s: unexpected err: %d\n", 1805 __func__, rc); 1806 break; 1807 } 1808 return ERR_PTR(rc); 1809 } 1810 1811 struct resource *nsblk_add_resource(struct nd_region *nd_region, 1812 struct nvdimm_drvdata *ndd, struct nd_namespace_blk *nsblk, 1813 resource_size_t start) 1814 { 1815 struct nd_label_id label_id; 1816 struct resource *res; 1817 1818 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL); 1819 res = krealloc(nsblk->res, 1820 sizeof(void *) * (nsblk->num_resources + 1), 1821 GFP_KERNEL); 1822 if (!res) 1823 return NULL; 1824 nsblk->res = (struct resource **) res; 1825 for_each_dpa_resource(ndd, res) 1826 if (strcmp(res->name, label_id.id) == 0 1827 && res->start == start) { 1828 nsblk->res[nsblk->num_resources++] = res; 1829 return res; 1830 } 1831 return NULL; 1832 } 1833 1834 static struct device *nd_namespace_blk_create(struct nd_region *nd_region) 1835 { 1836 struct nd_namespace_blk *nsblk; 1837 struct device *dev; 1838 1839 if (!is_nd_blk(&nd_region->dev)) 1840 return NULL; 1841 1842 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL); 1843 if (!nsblk) 1844 return NULL; 1845 1846 dev = &nsblk->common.dev; 1847 dev->type = &namespace_blk_device_type; 1848 nsblk->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL); 1849 if (nsblk->id < 0) { 1850 kfree(nsblk); 1851 return NULL; 1852 } 1853 dev_set_name(dev, "namespace%d.%d", nd_region->id, nsblk->id); 1854 dev->parent = &nd_region->dev; 1855 dev->groups = nd_namespace_attribute_groups; 1856 1857 return &nsblk->common.dev; 1858 } 1859 1860 static struct device *nd_namespace_pmem_create(struct nd_region *nd_region) 1861 { 1862 struct nd_namespace_pmem *nspm; 1863 struct resource *res; 1864 struct device *dev; 1865 1866 if (!is_nd_pmem(&nd_region->dev)) 1867 return NULL; 1868 1869 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL); 1870 if (!nspm) 1871 return NULL; 1872 1873 dev = &nspm->nsio.common.dev; 1874 dev->type = &namespace_pmem_device_type; 1875 dev->parent = &nd_region->dev; 1876 res = &nspm->nsio.res; 1877 res->name = dev_name(&nd_region->dev); 1878 res->flags = IORESOURCE_MEM; 1879 1880 nspm->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL); 1881 if (nspm->id < 0) { 1882 kfree(nspm); 1883 return NULL; 1884 } 1885 dev_set_name(dev, "namespace%d.%d", nd_region->id, nspm->id); 1886 dev->parent = &nd_region->dev; 1887 dev->groups = nd_namespace_attribute_groups; 1888 nd_namespace_pmem_set_resource(nd_region, nspm, 0); 1889 1890 return dev; 1891 } 1892 1893 void nd_region_create_ns_seed(struct nd_region *nd_region) 1894 { 1895 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev)); 1896 1897 if (nd_region_to_nstype(nd_region) == ND_DEVICE_NAMESPACE_IO) 1898 return; 1899 1900 if (is_nd_blk(&nd_region->dev)) 1901 nd_region->ns_seed = nd_namespace_blk_create(nd_region); 1902 else 1903 nd_region->ns_seed = nd_namespace_pmem_create(nd_region); 1904 1905 /* 1906 * Seed creation failures are not fatal, provisioning is simply 1907 * disabled until memory becomes available 1908 */ 1909 if (!nd_region->ns_seed) 1910 dev_err(&nd_region->dev, "failed to create %s namespace\n", 1911 is_nd_blk(&nd_region->dev) ? "blk" : "pmem"); 1912 else 1913 nd_device_register(nd_region->ns_seed); 1914 } 1915 1916 void nd_region_create_dax_seed(struct nd_region *nd_region) 1917 { 1918 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev)); 1919 nd_region->dax_seed = nd_dax_create(nd_region); 1920 /* 1921 * Seed creation failures are not fatal, provisioning is simply 1922 * disabled until memory becomes available 1923 */ 1924 if (!nd_region->dax_seed) 1925 dev_err(&nd_region->dev, "failed to create dax namespace\n"); 1926 } 1927 1928 void nd_region_create_pfn_seed(struct nd_region *nd_region) 1929 { 1930 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev)); 1931 nd_region->pfn_seed = nd_pfn_create(nd_region); 1932 /* 1933 * Seed creation failures are not fatal, provisioning is simply 1934 * disabled until memory becomes available 1935 */ 1936 if (!nd_region->pfn_seed) 1937 dev_err(&nd_region->dev, "failed to create pfn namespace\n"); 1938 } 1939 1940 void nd_region_create_btt_seed(struct nd_region *nd_region) 1941 { 1942 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev)); 1943 nd_region->btt_seed = nd_btt_create(nd_region); 1944 /* 1945 * Seed creation failures are not fatal, provisioning is simply 1946 * disabled until memory becomes available 1947 */ 1948 if (!nd_region->btt_seed) 1949 dev_err(&nd_region->dev, "failed to create btt namespace\n"); 1950 } 1951 1952 static int add_namespace_resource(struct nd_region *nd_region, 1953 struct nd_namespace_label *nd_label, struct device **devs, 1954 int count) 1955 { 1956 struct nd_mapping *nd_mapping = &nd_region->mapping[0]; 1957 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 1958 int i; 1959 1960 for (i = 0; i < count; i++) { 1961 u8 *uuid = namespace_to_uuid(devs[i]); 1962 struct resource *res; 1963 1964 if (IS_ERR_OR_NULL(uuid)) { 1965 WARN_ON(1); 1966 continue; 1967 } 1968 1969 if (memcmp(uuid, nd_label->uuid, NSLABEL_UUID_LEN) != 0) 1970 continue; 1971 if (is_namespace_blk(devs[i])) { 1972 res = nsblk_add_resource(nd_region, ndd, 1973 to_nd_namespace_blk(devs[i]), 1974 __le64_to_cpu(nd_label->dpa)); 1975 if (!res) 1976 return -ENXIO; 1977 nd_dbg_dpa(nd_region, ndd, res, "%d assign\n", count); 1978 } else { 1979 dev_err(&nd_region->dev, 1980 "error: conflicting extents for uuid: %pUb\n", 1981 nd_label->uuid); 1982 return -ENXIO; 1983 } 1984 break; 1985 } 1986 1987 return i; 1988 } 1989 1990 struct device *create_namespace_blk(struct nd_region *nd_region, 1991 struct nd_namespace_label *nd_label, int count) 1992 { 1993 1994 struct nd_mapping *nd_mapping = &nd_region->mapping[0]; 1995 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 1996 struct nd_namespace_blk *nsblk; 1997 char name[NSLABEL_NAME_LEN]; 1998 struct device *dev = NULL; 1999 struct resource *res; 2000 2001 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL); 2002 if (!nsblk) 2003 return ERR_PTR(-ENOMEM); 2004 dev = &nsblk->common.dev; 2005 dev->type = &namespace_blk_device_type; 2006 dev->parent = &nd_region->dev; 2007 nsblk->id = -1; 2008 nsblk->lbasize = __le64_to_cpu(nd_label->lbasize); 2009 nsblk->uuid = kmemdup(nd_label->uuid, NSLABEL_UUID_LEN, 2010 GFP_KERNEL); 2011 if (!nsblk->uuid) 2012 goto blk_err; 2013 memcpy(name, nd_label->name, NSLABEL_NAME_LEN); 2014 if (name[0]) 2015 nsblk->alt_name = kmemdup(name, NSLABEL_NAME_LEN, 2016 GFP_KERNEL); 2017 res = nsblk_add_resource(nd_region, ndd, nsblk, 2018 __le64_to_cpu(nd_label->dpa)); 2019 if (!res) 2020 goto blk_err; 2021 nd_dbg_dpa(nd_region, ndd, res, "%d: assign\n", count); 2022 return dev; 2023 blk_err: 2024 namespace_blk_release(dev); 2025 return ERR_PTR(-ENXIO); 2026 } 2027 2028 static int cmp_dpa(const void *a, const void *b) 2029 { 2030 const struct device *dev_a = *(const struct device **) a; 2031 const struct device *dev_b = *(const struct device **) b; 2032 struct nd_namespace_blk *nsblk_a, *nsblk_b; 2033 struct nd_namespace_pmem *nspm_a, *nspm_b; 2034 2035 if (is_namespace_io(dev_a)) 2036 return 0; 2037 2038 if (is_namespace_blk(dev_a)) { 2039 nsblk_a = to_nd_namespace_blk(dev_a); 2040 nsblk_b = to_nd_namespace_blk(dev_b); 2041 2042 return memcmp(&nsblk_a->res[0]->start, &nsblk_b->res[0]->start, 2043 sizeof(resource_size_t)); 2044 } 2045 2046 nspm_a = to_nd_namespace_pmem(dev_a); 2047 nspm_b = to_nd_namespace_pmem(dev_b); 2048 2049 return memcmp(&nspm_a->nsio.res.start, &nspm_b->nsio.res.start, 2050 sizeof(resource_size_t)); 2051 } 2052 2053 static struct device **scan_labels(struct nd_region *nd_region) 2054 { 2055 int i, count = 0; 2056 struct device *dev, **devs = NULL; 2057 struct nd_label_ent *label_ent, *e; 2058 struct nd_mapping *nd_mapping = &nd_region->mapping[0]; 2059 resource_size_t map_end = nd_mapping->start + nd_mapping->size - 1; 2060 2061 /* "safe" because create_namespace_pmem() might list_move() label_ent */ 2062 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) { 2063 struct nd_namespace_label *nd_label = label_ent->label; 2064 struct device **__devs; 2065 u32 flags; 2066 2067 if (!nd_label) 2068 continue; 2069 flags = __le32_to_cpu(nd_label->flags); 2070 if (is_nd_blk(&nd_region->dev) 2071 == !!(flags & NSLABEL_FLAG_LOCAL)) 2072 /* pass, region matches label type */; 2073 else 2074 continue; 2075 2076 /* skip labels that describe extents outside of the region */ 2077 if (nd_label->dpa < nd_mapping->start || nd_label->dpa > map_end) 2078 continue; 2079 2080 i = add_namespace_resource(nd_region, nd_label, devs, count); 2081 if (i < 0) 2082 goto err; 2083 if (i < count) 2084 continue; 2085 __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL); 2086 if (!__devs) 2087 goto err; 2088 memcpy(__devs, devs, sizeof(dev) * count); 2089 kfree(devs); 2090 devs = __devs; 2091 2092 if (is_nd_blk(&nd_region->dev)) { 2093 dev = create_namespace_blk(nd_region, nd_label, count); 2094 if (IS_ERR(dev)) 2095 goto err; 2096 devs[count++] = dev; 2097 } else { 2098 dev = create_namespace_pmem(nd_region, nd_label); 2099 if (IS_ERR(dev)) { 2100 switch (PTR_ERR(dev)) { 2101 case -EAGAIN: 2102 /* skip invalid labels */ 2103 continue; 2104 case -ENODEV: 2105 /* fallthrough to seed creation */ 2106 break; 2107 default: 2108 goto err; 2109 } 2110 } else 2111 devs[count++] = dev; 2112 } 2113 } 2114 2115 dev_dbg(&nd_region->dev, "%s: discovered %d %s namespace%s\n", 2116 __func__, count, is_nd_blk(&nd_region->dev) 2117 ? "blk" : "pmem", count == 1 ? "" : "s"); 2118 2119 if (count == 0) { 2120 /* Publish a zero-sized namespace for userspace to configure. */ 2121 nd_mapping_free_labels(nd_mapping); 2122 2123 devs = kcalloc(2, sizeof(dev), GFP_KERNEL); 2124 if (!devs) 2125 goto err; 2126 if (is_nd_blk(&nd_region->dev)) { 2127 struct nd_namespace_blk *nsblk; 2128 2129 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL); 2130 if (!nsblk) 2131 goto err; 2132 dev = &nsblk->common.dev; 2133 dev->type = &namespace_blk_device_type; 2134 } else { 2135 struct nd_namespace_pmem *nspm; 2136 2137 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL); 2138 if (!nspm) 2139 goto err; 2140 dev = &nspm->nsio.common.dev; 2141 dev->type = &namespace_pmem_device_type; 2142 nd_namespace_pmem_set_resource(nd_region, nspm, 0); 2143 } 2144 dev->parent = &nd_region->dev; 2145 devs[count++] = dev; 2146 } else if (is_nd_pmem(&nd_region->dev)) { 2147 /* clean unselected labels */ 2148 for (i = 0; i < nd_region->ndr_mappings; i++) { 2149 struct list_head *l, *e; 2150 LIST_HEAD(list); 2151 int j; 2152 2153 nd_mapping = &nd_region->mapping[i]; 2154 if (list_empty(&nd_mapping->labels)) { 2155 WARN_ON(1); 2156 continue; 2157 } 2158 2159 j = count; 2160 list_for_each_safe(l, e, &nd_mapping->labels) { 2161 if (!j--) 2162 break; 2163 list_move_tail(l, &list); 2164 } 2165 nd_mapping_free_labels(nd_mapping); 2166 list_splice_init(&list, &nd_mapping->labels); 2167 } 2168 } 2169 2170 if (count > 1) 2171 sort(devs, count, sizeof(struct device *), cmp_dpa, NULL); 2172 2173 return devs; 2174 2175 err: 2176 if (devs) { 2177 for (i = 0; devs[i]; i++) 2178 if (is_nd_blk(&nd_region->dev)) 2179 namespace_blk_release(devs[i]); 2180 else 2181 namespace_pmem_release(devs[i]); 2182 kfree(devs); 2183 } 2184 return NULL; 2185 } 2186 2187 static struct device **create_namespaces(struct nd_region *nd_region) 2188 { 2189 struct nd_mapping *nd_mapping = &nd_region->mapping[0]; 2190 struct device **devs; 2191 int i; 2192 2193 if (nd_region->ndr_mappings == 0) 2194 return NULL; 2195 2196 /* lock down all mappings while we scan labels */ 2197 for (i = 0; i < nd_region->ndr_mappings; i++) { 2198 nd_mapping = &nd_region->mapping[i]; 2199 mutex_lock_nested(&nd_mapping->lock, i); 2200 } 2201 2202 devs = scan_labels(nd_region); 2203 2204 for (i = 0; i < nd_region->ndr_mappings; i++) { 2205 int reverse = nd_region->ndr_mappings - 1 - i; 2206 2207 nd_mapping = &nd_region->mapping[reverse]; 2208 mutex_unlock(&nd_mapping->lock); 2209 } 2210 2211 return devs; 2212 } 2213 2214 static int init_active_labels(struct nd_region *nd_region) 2215 { 2216 int i; 2217 2218 for (i = 0; i < nd_region->ndr_mappings; i++) { 2219 struct nd_mapping *nd_mapping = &nd_region->mapping[i]; 2220 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); 2221 struct nvdimm *nvdimm = nd_mapping->nvdimm; 2222 struct nd_label_ent *label_ent; 2223 int count, j; 2224 2225 /* 2226 * If the dimm is disabled then prevent the region from 2227 * being activated if it aliases DPA. 2228 */ 2229 if (!ndd) { 2230 if ((nvdimm->flags & NDD_ALIASING) == 0) 2231 return 0; 2232 dev_dbg(&nd_region->dev, "%s: is disabled, failing probe\n", 2233 dev_name(&nd_mapping->nvdimm->dev)); 2234 return -ENXIO; 2235 } 2236 nd_mapping->ndd = ndd; 2237 atomic_inc(&nvdimm->busy); 2238 get_ndd(ndd); 2239 2240 count = nd_label_active_count(ndd); 2241 dev_dbg(ndd->dev, "%s: %d\n", __func__, count); 2242 if (!count) 2243 continue; 2244 for (j = 0; j < count; j++) { 2245 struct nd_namespace_label *label; 2246 2247 label_ent = kzalloc(sizeof(*label_ent), GFP_KERNEL); 2248 if (!label_ent) 2249 break; 2250 label = nd_label_active(ndd, j); 2251 label_ent->label = label; 2252 2253 mutex_lock(&nd_mapping->lock); 2254 list_add_tail(&label_ent->list, &nd_mapping->labels); 2255 mutex_unlock(&nd_mapping->lock); 2256 } 2257 2258 if (j >= count) 2259 continue; 2260 2261 mutex_lock(&nd_mapping->lock); 2262 nd_mapping_free_labels(nd_mapping); 2263 mutex_unlock(&nd_mapping->lock); 2264 return -ENOMEM; 2265 } 2266 2267 return 0; 2268 } 2269 2270 int nd_region_register_namespaces(struct nd_region *nd_region, int *err) 2271 { 2272 struct device **devs = NULL; 2273 int i, rc = 0, type; 2274 2275 *err = 0; 2276 nvdimm_bus_lock(&nd_region->dev); 2277 rc = init_active_labels(nd_region); 2278 if (rc) { 2279 nvdimm_bus_unlock(&nd_region->dev); 2280 return rc; 2281 } 2282 2283 type = nd_region_to_nstype(nd_region); 2284 switch (type) { 2285 case ND_DEVICE_NAMESPACE_IO: 2286 devs = create_namespace_io(nd_region); 2287 break; 2288 case ND_DEVICE_NAMESPACE_PMEM: 2289 case ND_DEVICE_NAMESPACE_BLK: 2290 devs = create_namespaces(nd_region); 2291 break; 2292 default: 2293 break; 2294 } 2295 nvdimm_bus_unlock(&nd_region->dev); 2296 2297 if (!devs) 2298 return -ENODEV; 2299 2300 for (i = 0; devs[i]; i++) { 2301 struct device *dev = devs[i]; 2302 int id; 2303 2304 if (type == ND_DEVICE_NAMESPACE_BLK) { 2305 struct nd_namespace_blk *nsblk; 2306 2307 nsblk = to_nd_namespace_blk(dev); 2308 id = ida_simple_get(&nd_region->ns_ida, 0, 0, 2309 GFP_KERNEL); 2310 nsblk->id = id; 2311 } else if (type == ND_DEVICE_NAMESPACE_PMEM) { 2312 struct nd_namespace_pmem *nspm; 2313 2314 nspm = to_nd_namespace_pmem(dev); 2315 id = ida_simple_get(&nd_region->ns_ida, 0, 0, 2316 GFP_KERNEL); 2317 nspm->id = id; 2318 } else 2319 id = i; 2320 2321 if (id < 0) 2322 break; 2323 dev_set_name(dev, "namespace%d.%d", nd_region->id, id); 2324 dev->groups = nd_namespace_attribute_groups; 2325 nd_device_register(dev); 2326 } 2327 if (i) 2328 nd_region->ns_seed = devs[0]; 2329 2330 if (devs[i]) { 2331 int j; 2332 2333 for (j = i; devs[j]; j++) { 2334 struct device *dev = devs[j]; 2335 2336 device_initialize(dev); 2337 put_device(dev); 2338 } 2339 *err = j - i; 2340 /* 2341 * All of the namespaces we tried to register failed, so 2342 * fail region activation. 2343 */ 2344 if (*err == 0) 2345 rc = -ENODEV; 2346 } 2347 kfree(devs); 2348 2349 if (rc == -ENODEV) 2350 return rc; 2351 2352 return i; 2353 } 2354