Loading...
Note: File does not exist in v3.1.
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/list.h>
18#include <linux/nd.h>
19#include "nd-core.h"
20#include "pmem.h"
21#include "nd.h"
22
23static 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
30static 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
42static 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
55static const struct device_type namespace_io_device_type = {
56 .name = "nd_namespace_io",
57 .release = namespace_io_release,
58};
59
60static const struct device_type namespace_pmem_device_type = {
61 .name = "nd_namespace_pmem",
62 .release = namespace_pmem_release,
63};
64
65static const struct device_type namespace_blk_device_type = {
66 .name = "nd_namespace_blk",
67 .release = namespace_blk_release,
68};
69
70static bool is_namespace_pmem(const struct device *dev)
71{
72 return dev ? dev->type == &namespace_pmem_device_type : false;
73}
74
75static bool is_namespace_blk(const struct device *dev)
76{
77 return dev ? dev->type == &namespace_blk_device_type : false;
78}
79
80static bool is_namespace_io(const struct device *dev)
81{
82 return dev ? dev->type == &namespace_io_device_type : false;
83}
84
85static 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
113static int is_namespace_uuid_busy(struct device *dev, void *data)
114{
115 if (is_nd_region(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 */
125bool 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
138bool 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 return ARCH_MEMREMAP_PMEM == MEMREMAP_WB;
159}
160EXPORT_SYMBOL(pmem_should_map_pages);
161
162unsigned int pmem_sector_size(struct nd_namespace_common *ndns)
163{
164 if (is_namespace_pmem(&ndns->dev)) {
165 struct nd_namespace_pmem *nspm;
166
167 nspm = to_nd_namespace_pmem(&ndns->dev);
168 if (nspm->lbasize == 0 || nspm->lbasize == 512)
169 /* default */;
170 else if (nspm->lbasize == 4096)
171 return 4096;
172 else
173 dev_WARN(&ndns->dev, "unsupported sector size: %ld\n",
174 nspm->lbasize);
175 }
176
177 /*
178 * There is no namespace label (is_namespace_io()), or the label
179 * indicates the default sector size.
180 */
181 return 512;
182}
183EXPORT_SYMBOL(pmem_sector_size);
184
185const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,
186 char *name)
187{
188 struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
189 const char *suffix = NULL;
190
191 if (ndns->claim && is_nd_btt(ndns->claim))
192 suffix = "s";
193
194 if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev)) {
195 int nsidx = 0;
196
197 if (is_namespace_pmem(&ndns->dev)) {
198 struct nd_namespace_pmem *nspm;
199
200 nspm = to_nd_namespace_pmem(&ndns->dev);
201 nsidx = nspm->id;
202 }
203
204 if (nsidx)
205 sprintf(name, "pmem%d.%d%s", nd_region->id, nsidx,
206 suffix ? suffix : "");
207 else
208 sprintf(name, "pmem%d%s", nd_region->id,
209 suffix ? suffix : "");
210 } else if (is_namespace_blk(&ndns->dev)) {
211 struct nd_namespace_blk *nsblk;
212
213 nsblk = to_nd_namespace_blk(&ndns->dev);
214 sprintf(name, "ndblk%d.%d%s", nd_region->id, nsblk->id,
215 suffix ? suffix : "");
216 } else {
217 return NULL;
218 }
219
220 return name;
221}
222EXPORT_SYMBOL(nvdimm_namespace_disk_name);
223
224const u8 *nd_dev_to_uuid(struct device *dev)
225{
226 static const u8 null_uuid[16];
227
228 if (!dev)
229 return null_uuid;
230
231 if (is_namespace_pmem(dev)) {
232 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
233
234 return nspm->uuid;
235 } else if (is_namespace_blk(dev)) {
236 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
237
238 return nsblk->uuid;
239 } else
240 return null_uuid;
241}
242EXPORT_SYMBOL(nd_dev_to_uuid);
243
244static ssize_t nstype_show(struct device *dev,
245 struct device_attribute *attr, char *buf)
246{
247 struct nd_region *nd_region = to_nd_region(dev->parent);
248
249 return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
250}
251static DEVICE_ATTR_RO(nstype);
252
253static ssize_t __alt_name_store(struct device *dev, const char *buf,
254 const size_t len)
255{
256 char *input, *pos, *alt_name, **ns_altname;
257 ssize_t rc;
258
259 if (is_namespace_pmem(dev)) {
260 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
261
262 ns_altname = &nspm->alt_name;
263 } else if (is_namespace_blk(dev)) {
264 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
265
266 ns_altname = &nsblk->alt_name;
267 } else
268 return -ENXIO;
269
270 if (dev->driver || to_ndns(dev)->claim)
271 return -EBUSY;
272
273 input = kmemdup(buf, len + 1, GFP_KERNEL);
274 if (!input)
275 return -ENOMEM;
276
277 input[len] = '\0';
278 pos = strim(input);
279 if (strlen(pos) + 1 > NSLABEL_NAME_LEN) {
280 rc = -EINVAL;
281 goto out;
282 }
283
284 alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL);
285 if (!alt_name) {
286 rc = -ENOMEM;
287 goto out;
288 }
289 kfree(*ns_altname);
290 *ns_altname = alt_name;
291 sprintf(*ns_altname, "%s", pos);
292 rc = len;
293
294out:
295 kfree(input);
296 return rc;
297}
298
299static resource_size_t nd_namespace_blk_size(struct nd_namespace_blk *nsblk)
300{
301 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
302 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
303 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
304 struct nd_label_id label_id;
305 resource_size_t size = 0;
306 struct resource *res;
307
308 if (!nsblk->uuid)
309 return 0;
310 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
311 for_each_dpa_resource(ndd, res)
312 if (strcmp(res->name, label_id.id) == 0)
313 size += resource_size(res);
314 return size;
315}
316
317static bool __nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
318{
319 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
320 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
321 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
322 struct nd_label_id label_id;
323 struct resource *res;
324 int count, i;
325
326 if (!nsblk->uuid || !nsblk->lbasize || !ndd)
327 return false;
328
329 count = 0;
330 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
331 for_each_dpa_resource(ndd, res) {
332 if (strcmp(res->name, label_id.id) != 0)
333 continue;
334 /*
335 * Resources with unacknowledged adjustments indicate a
336 * failure to update labels
337 */
338 if (res->flags & DPA_RESOURCE_ADJUSTED)
339 return false;
340 count++;
341 }
342
343 /* These values match after a successful label update */
344 if (count != nsblk->num_resources)
345 return false;
346
347 for (i = 0; i < nsblk->num_resources; i++) {
348 struct resource *found = NULL;
349
350 for_each_dpa_resource(ndd, res)
351 if (res == nsblk->res[i]) {
352 found = res;
353 break;
354 }
355 /* stale resource */
356 if (!found)
357 return false;
358 }
359
360 return true;
361}
362
363resource_size_t nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
364{
365 resource_size_t size;
366
367 nvdimm_bus_lock(&nsblk->common.dev);
368 size = __nd_namespace_blk_validate(nsblk);
369 nvdimm_bus_unlock(&nsblk->common.dev);
370
371 return size;
372}
373EXPORT_SYMBOL(nd_namespace_blk_validate);
374
375
376static int nd_namespace_label_update(struct nd_region *nd_region,
377 struct device *dev)
378{
379 dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim,
380 "namespace must be idle during label update\n");
381 if (dev->driver || to_ndns(dev)->claim)
382 return 0;
383
384 /*
385 * Only allow label writes that will result in a valid namespace
386 * or deletion of an existing namespace.
387 */
388 if (is_namespace_pmem(dev)) {
389 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
390 resource_size_t size = resource_size(&nspm->nsio.res);
391
392 if (size == 0 && nspm->uuid)
393 /* delete allocation */;
394 else if (!nspm->uuid)
395 return 0;
396
397 return nd_pmem_namespace_label_update(nd_region, nspm, size);
398 } else if (is_namespace_blk(dev)) {
399 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
400 resource_size_t size = nd_namespace_blk_size(nsblk);
401
402 if (size == 0 && nsblk->uuid)
403 /* delete allocation */;
404 else if (!nsblk->uuid || !nsblk->lbasize)
405 return 0;
406
407 return nd_blk_namespace_label_update(nd_region, nsblk, size);
408 } else
409 return -ENXIO;
410}
411
412static ssize_t alt_name_store(struct device *dev,
413 struct device_attribute *attr, const char *buf, size_t len)
414{
415 struct nd_region *nd_region = to_nd_region(dev->parent);
416 ssize_t rc;
417
418 device_lock(dev);
419 nvdimm_bus_lock(dev);
420 wait_nvdimm_bus_probe_idle(dev);
421 rc = __alt_name_store(dev, buf, len);
422 if (rc >= 0)
423 rc = nd_namespace_label_update(nd_region, dev);
424 dev_dbg(dev, "%s(%zd)\n", rc < 0 ? "fail " : "", rc);
425 nvdimm_bus_unlock(dev);
426 device_unlock(dev);
427
428 return rc < 0 ? rc : len;
429}
430
431static ssize_t alt_name_show(struct device *dev,
432 struct device_attribute *attr, char *buf)
433{
434 char *ns_altname;
435
436 if (is_namespace_pmem(dev)) {
437 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
438
439 ns_altname = nspm->alt_name;
440 } else if (is_namespace_blk(dev)) {
441 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
442
443 ns_altname = nsblk->alt_name;
444 } else
445 return -ENXIO;
446
447 return sprintf(buf, "%s\n", ns_altname ? ns_altname : "");
448}
449static DEVICE_ATTR_RW(alt_name);
450
451static int scan_free(struct nd_region *nd_region,
452 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
453 resource_size_t n)
454{
455 bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
456 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
457 int rc = 0;
458
459 while (n) {
460 struct resource *res, *last;
461 resource_size_t new_start;
462
463 last = NULL;
464 for_each_dpa_resource(ndd, res)
465 if (strcmp(res->name, label_id->id) == 0)
466 last = res;
467 res = last;
468 if (!res)
469 return 0;
470
471 if (n >= resource_size(res)) {
472 n -= resource_size(res);
473 nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc);
474 nvdimm_free_dpa(ndd, res);
475 /* retry with last resource deleted */
476 continue;
477 }
478
479 /*
480 * Keep BLK allocations relegated to high DPA as much as
481 * possible
482 */
483 if (is_blk)
484 new_start = res->start + n;
485 else
486 new_start = res->start;
487
488 rc = adjust_resource(res, new_start, resource_size(res) - n);
489 if (rc == 0)
490 res->flags |= DPA_RESOURCE_ADJUSTED;
491 nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc);
492 break;
493 }
494
495 return rc;
496}
497
498/**
499 * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
500 * @nd_region: the set of dimms to reclaim @n bytes from
501 * @label_id: unique identifier for the namespace consuming this dpa range
502 * @n: number of bytes per-dimm to release
503 *
504 * Assumes resources are ordered. Starting from the end try to
505 * adjust_resource() the allocation to @n, but if @n is larger than the
506 * allocation delete it and find the 'new' last allocation in the label
507 * set.
508 */
509static int shrink_dpa_allocation(struct nd_region *nd_region,
510 struct nd_label_id *label_id, resource_size_t n)
511{
512 int i;
513
514 for (i = 0; i < nd_region->ndr_mappings; i++) {
515 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
516 int rc;
517
518 rc = scan_free(nd_region, nd_mapping, label_id, n);
519 if (rc)
520 return rc;
521 }
522
523 return 0;
524}
525
526static resource_size_t init_dpa_allocation(struct nd_label_id *label_id,
527 struct nd_region *nd_region, struct nd_mapping *nd_mapping,
528 resource_size_t n)
529{
530 bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
531 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
532 resource_size_t first_dpa;
533 struct resource *res;
534 int rc = 0;
535
536 /* allocate blk from highest dpa first */
537 if (is_blk)
538 first_dpa = nd_mapping->start + nd_mapping->size - n;
539 else
540 first_dpa = nd_mapping->start;
541
542 /* first resource allocation for this label-id or dimm */
543 res = nvdimm_allocate_dpa(ndd, label_id, first_dpa, n);
544 if (!res)
545 rc = -EBUSY;
546
547 nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc);
548 return rc ? n : 0;
549}
550
551
552/**
553 * space_valid() - validate free dpa space against constraints
554 * @nd_region: hosting region of the free space
555 * @ndd: dimm device data for debug
556 * @label_id: namespace id to allocate space
557 * @prev: potential allocation that precedes free space
558 * @next: allocation that follows the given free space range
559 * @exist: first allocation with same id in the mapping
560 * @n: range that must satisfied for pmem allocations
561 * @valid: free space range to validate
562 *
563 * BLK-space is valid as long as it does not precede a PMEM
564 * allocation in a given region. PMEM-space must be contiguous
565 * and adjacent to an existing existing allocation (if one
566 * exists). If reserving PMEM any space is valid.
567 */
568static void space_valid(struct nd_region *nd_region, struct nvdimm_drvdata *ndd,
569 struct nd_label_id *label_id, struct resource *prev,
570 struct resource *next, struct resource *exist,
571 resource_size_t n, struct resource *valid)
572{
573 bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0;
574 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
575
576 if (valid->start >= valid->end)
577 goto invalid;
578
579 if (is_reserve)
580 return;
581
582 if (!is_pmem) {
583 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
584 struct nvdimm_bus *nvdimm_bus;
585 struct blk_alloc_info info = {
586 .nd_mapping = nd_mapping,
587 .available = nd_mapping->size,
588 .res = valid,
589 };
590
591 WARN_ON(!is_nd_blk(&nd_region->dev));
592 nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
593 device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
594 return;
595 }
596
597 /* allocation needs to be contiguous, so this is all or nothing */
598 if (resource_size(valid) < n)
599 goto invalid;
600
601 /* we've got all the space we need and no existing allocation */
602 if (!exist)
603 return;
604
605 /* allocation needs to be contiguous with the existing namespace */
606 if (valid->start == exist->end + 1
607 || valid->end == exist->start - 1)
608 return;
609
610 invalid:
611 /* truncate @valid size to 0 */
612 valid->end = valid->start - 1;
613}
614
615enum alloc_loc {
616 ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
617};
618
619static resource_size_t scan_allocate(struct nd_region *nd_region,
620 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
621 resource_size_t n)
622{
623 resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1;
624 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
625 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
626 struct resource *res, *exist = NULL, valid;
627 const resource_size_t to_allocate = n;
628 int first;
629
630 for_each_dpa_resource(ndd, res)
631 if (strcmp(label_id->id, res->name) == 0)
632 exist = res;
633
634 valid.start = nd_mapping->start;
635 valid.end = mapping_end;
636 valid.name = "free space";
637 retry:
638 first = 0;
639 for_each_dpa_resource(ndd, res) {
640 struct resource *next = res->sibling, *new_res = NULL;
641 resource_size_t allocate, available = 0;
642 enum alloc_loc loc = ALLOC_ERR;
643 const char *action;
644 int rc = 0;
645
646 /* ignore resources outside this nd_mapping */
647 if (res->start > mapping_end)
648 continue;
649 if (res->end < nd_mapping->start)
650 continue;
651
652 /* space at the beginning of the mapping */
653 if (!first++ && res->start > nd_mapping->start) {
654 valid.start = nd_mapping->start;
655 valid.end = res->start - 1;
656 space_valid(nd_region, ndd, label_id, NULL, next, exist,
657 to_allocate, &valid);
658 available = resource_size(&valid);
659 if (available)
660 loc = ALLOC_BEFORE;
661 }
662
663 /* space between allocations */
664 if (!loc && next) {
665 valid.start = res->start + resource_size(res);
666 valid.end = min(mapping_end, next->start - 1);
667 space_valid(nd_region, ndd, label_id, res, next, exist,
668 to_allocate, &valid);
669 available = resource_size(&valid);
670 if (available)
671 loc = ALLOC_MID;
672 }
673
674 /* space at the end of the mapping */
675 if (!loc && !next) {
676 valid.start = res->start + resource_size(res);
677 valid.end = mapping_end;
678 space_valid(nd_region, ndd, label_id, res, next, exist,
679 to_allocate, &valid);
680 available = resource_size(&valid);
681 if (available)
682 loc = ALLOC_AFTER;
683 }
684
685 if (!loc || !available)
686 continue;
687 allocate = min(available, n);
688 switch (loc) {
689 case ALLOC_BEFORE:
690 if (strcmp(res->name, label_id->id) == 0) {
691 /* adjust current resource up */
692 rc = adjust_resource(res, res->start - allocate,
693 resource_size(res) + allocate);
694 action = "cur grow up";
695 } else
696 action = "allocate";
697 break;
698 case ALLOC_MID:
699 if (strcmp(next->name, label_id->id) == 0) {
700 /* adjust next resource up */
701 rc = adjust_resource(next, next->start
702 - allocate, resource_size(next)
703 + allocate);
704 new_res = next;
705 action = "next grow up";
706 } else if (strcmp(res->name, label_id->id) == 0) {
707 action = "grow down";
708 } else
709 action = "allocate";
710 break;
711 case ALLOC_AFTER:
712 if (strcmp(res->name, label_id->id) == 0)
713 action = "grow down";
714 else
715 action = "allocate";
716 break;
717 default:
718 return n;
719 }
720
721 if (strcmp(action, "allocate") == 0) {
722 /* BLK allocate bottom up */
723 if (!is_pmem)
724 valid.start += available - allocate;
725
726 new_res = nvdimm_allocate_dpa(ndd, label_id,
727 valid.start, allocate);
728 if (!new_res)
729 rc = -EBUSY;
730 } else if (strcmp(action, "grow down") == 0) {
731 /* adjust current resource down */
732 rc = adjust_resource(res, res->start, resource_size(res)
733 + allocate);
734 if (rc == 0)
735 res->flags |= DPA_RESOURCE_ADJUSTED;
736 }
737
738 if (!new_res)
739 new_res = res;
740
741 nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n",
742 action, loc, rc);
743
744 if (rc)
745 return n;
746
747 n -= allocate;
748 if (n) {
749 /*
750 * Retry scan with newly inserted resources.
751 * For example, if we did an ALLOC_BEFORE
752 * insertion there may also have been space
753 * available for an ALLOC_AFTER insertion, so we
754 * need to check this same resource again
755 */
756 goto retry;
757 } else
758 return 0;
759 }
760
761 /*
762 * If we allocated nothing in the BLK case it may be because we are in
763 * an initial "pmem-reserve pass". Only do an initial BLK allocation
764 * when none of the DPA space is reserved.
765 */
766 if ((is_pmem || !ndd->dpa.child) && n == to_allocate)
767 return init_dpa_allocation(label_id, nd_region, nd_mapping, n);
768 return n;
769}
770
771static int merge_dpa(struct nd_region *nd_region,
772 struct nd_mapping *nd_mapping, struct nd_label_id *label_id)
773{
774 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
775 struct resource *res;
776
777 if (strncmp("pmem", label_id->id, 4) == 0)
778 return 0;
779 retry:
780 for_each_dpa_resource(ndd, res) {
781 int rc;
782 struct resource *next = res->sibling;
783 resource_size_t end = res->start + resource_size(res);
784
785 if (!next || strcmp(res->name, label_id->id) != 0
786 || strcmp(next->name, label_id->id) != 0
787 || end != next->start)
788 continue;
789 end += resource_size(next);
790 nvdimm_free_dpa(ndd, next);
791 rc = adjust_resource(res, res->start, end - res->start);
792 nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc);
793 if (rc)
794 return rc;
795 res->flags |= DPA_RESOURCE_ADJUSTED;
796 goto retry;
797 }
798
799 return 0;
800}
801
802static int __reserve_free_pmem(struct device *dev, void *data)
803{
804 struct nvdimm *nvdimm = data;
805 struct nd_region *nd_region;
806 struct nd_label_id label_id;
807 int i;
808
809 if (!is_memory(dev))
810 return 0;
811
812 nd_region = to_nd_region(dev);
813 if (nd_region->ndr_mappings == 0)
814 return 0;
815
816 memset(&label_id, 0, sizeof(label_id));
817 strcat(label_id.id, "pmem-reserve");
818 for (i = 0; i < nd_region->ndr_mappings; i++) {
819 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
820 resource_size_t n, rem = 0;
821
822 if (nd_mapping->nvdimm != nvdimm)
823 continue;
824
825 n = nd_pmem_available_dpa(nd_region, nd_mapping, &rem);
826 if (n == 0)
827 return 0;
828 rem = scan_allocate(nd_region, nd_mapping, &label_id, n);
829 dev_WARN_ONCE(&nd_region->dev, rem,
830 "pmem reserve underrun: %#llx of %#llx bytes\n",
831 (unsigned long long) n - rem,
832 (unsigned long long) n);
833 return rem ? -ENXIO : 0;
834 }
835
836 return 0;
837}
838
839static void release_free_pmem(struct nvdimm_bus *nvdimm_bus,
840 struct nd_mapping *nd_mapping)
841{
842 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
843 struct resource *res, *_res;
844
845 for_each_dpa_resource_safe(ndd, res, _res)
846 if (strcmp(res->name, "pmem-reserve") == 0)
847 nvdimm_free_dpa(ndd, res);
848}
849
850static int reserve_free_pmem(struct nvdimm_bus *nvdimm_bus,
851 struct nd_mapping *nd_mapping)
852{
853 struct nvdimm *nvdimm = nd_mapping->nvdimm;
854 int rc;
855
856 rc = device_for_each_child(&nvdimm_bus->dev, nvdimm,
857 __reserve_free_pmem);
858 if (rc)
859 release_free_pmem(nvdimm_bus, nd_mapping);
860 return rc;
861}
862
863/**
864 * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
865 * @nd_region: the set of dimms to allocate @n more bytes from
866 * @label_id: unique identifier for the namespace consuming this dpa range
867 * @n: number of bytes per-dimm to add to the existing allocation
868 *
869 * Assumes resources are ordered. For BLK regions, first consume
870 * BLK-only available DPA free space, then consume PMEM-aliased DPA
871 * space starting at the highest DPA. For PMEM regions start
872 * allocations from the start of an interleave set and end at the first
873 * BLK allocation or the end of the interleave set, whichever comes
874 * first.
875 */
876static int grow_dpa_allocation(struct nd_region *nd_region,
877 struct nd_label_id *label_id, resource_size_t n)
878{
879 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
880 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
881 int i;
882
883 for (i = 0; i < nd_region->ndr_mappings; i++) {
884 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
885 resource_size_t rem = n;
886 int rc, j;
887
888 /*
889 * In the BLK case try once with all unallocated PMEM
890 * reserved, and once without
891 */
892 for (j = is_pmem; j < 2; j++) {
893 bool blk_only = j == 0;
894
895 if (blk_only) {
896 rc = reserve_free_pmem(nvdimm_bus, nd_mapping);
897 if (rc)
898 return rc;
899 }
900 rem = scan_allocate(nd_region, nd_mapping,
901 label_id, rem);
902 if (blk_only)
903 release_free_pmem(nvdimm_bus, nd_mapping);
904
905 /* try again and allow encroachments into PMEM */
906 if (rem == 0)
907 break;
908 }
909
910 dev_WARN_ONCE(&nd_region->dev, rem,
911 "allocation underrun: %#llx of %#llx bytes\n",
912 (unsigned long long) n - rem,
913 (unsigned long long) n);
914 if (rem)
915 return -ENXIO;
916
917 rc = merge_dpa(nd_region, nd_mapping, label_id);
918 if (rc)
919 return rc;
920 }
921
922 return 0;
923}
924
925static void nd_namespace_pmem_set_resource(struct nd_region *nd_region,
926 struct nd_namespace_pmem *nspm, resource_size_t size)
927{
928 struct resource *res = &nspm->nsio.res;
929 resource_size_t offset = 0;
930
931 if (size && !nspm->uuid) {
932 WARN_ON_ONCE(1);
933 size = 0;
934 }
935
936 if (size && nspm->uuid) {
937 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
938 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
939 struct nd_label_id label_id;
940 struct resource *res;
941
942 if (!ndd) {
943 size = 0;
944 goto out;
945 }
946
947 nd_label_gen_id(&label_id, nspm->uuid, 0);
948
949 /* calculate a spa offset from the dpa allocation offset */
950 for_each_dpa_resource(ndd, res)
951 if (strcmp(res->name, label_id.id) == 0) {
952 offset = (res->start - nd_mapping->start)
953 * nd_region->ndr_mappings;
954 goto out;
955 }
956
957 WARN_ON_ONCE(1);
958 size = 0;
959 }
960
961 out:
962 res->start = nd_region->ndr_start + offset;
963 res->end = res->start + size - 1;
964}
965
966static bool uuid_not_set(const u8 *uuid, struct device *dev, const char *where)
967{
968 if (!uuid) {
969 dev_dbg(dev, "%s: uuid not set\n", where);
970 return true;
971 }
972 return false;
973}
974
975static ssize_t __size_store(struct device *dev, unsigned long long val)
976{
977 resource_size_t allocated = 0, available = 0;
978 struct nd_region *nd_region = to_nd_region(dev->parent);
979 struct nd_namespace_common *ndns = to_ndns(dev);
980 struct nd_mapping *nd_mapping;
981 struct nvdimm_drvdata *ndd;
982 struct nd_label_id label_id;
983 u32 flags = 0, remainder;
984 int rc, i, id = -1;
985 u8 *uuid = NULL;
986
987 if (dev->driver || ndns->claim)
988 return -EBUSY;
989
990 if (is_namespace_pmem(dev)) {
991 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
992
993 uuid = nspm->uuid;
994 id = nspm->id;
995 } else if (is_namespace_blk(dev)) {
996 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
997
998 uuid = nsblk->uuid;
999 flags = NSLABEL_FLAG_LOCAL;
1000 id = nsblk->id;
1001 }
1002
1003 /*
1004 * We need a uuid for the allocation-label and dimm(s) on which
1005 * to store the label.
1006 */
1007 if (uuid_not_set(uuid, dev, __func__))
1008 return -ENXIO;
1009 if (nd_region->ndr_mappings == 0) {
1010 dev_dbg(dev, "not associated with dimm(s)\n");
1011 return -ENXIO;
1012 }
1013
1014 div_u64_rem(val, SZ_4K * nd_region->ndr_mappings, &remainder);
1015 if (remainder) {
1016 dev_dbg(dev, "%llu is not %dK aligned\n", val,
1017 (SZ_4K * nd_region->ndr_mappings) / SZ_1K);
1018 return -EINVAL;
1019 }
1020
1021 nd_label_gen_id(&label_id, uuid, flags);
1022 for (i = 0; i < nd_region->ndr_mappings; i++) {
1023 nd_mapping = &nd_region->mapping[i];
1024 ndd = to_ndd(nd_mapping);
1025
1026 /*
1027 * All dimms in an interleave set, or the base dimm for a blk
1028 * region, need to be enabled for the size to be changed.
1029 */
1030 if (!ndd)
1031 return -ENXIO;
1032
1033 allocated += nvdimm_allocated_dpa(ndd, &label_id);
1034 }
1035 available = nd_region_available_dpa(nd_region);
1036
1037 if (val > available + allocated)
1038 return -ENOSPC;
1039
1040 if (val == allocated)
1041 return 0;
1042
1043 val = div_u64(val, nd_region->ndr_mappings);
1044 allocated = div_u64(allocated, nd_region->ndr_mappings);
1045 if (val < allocated)
1046 rc = shrink_dpa_allocation(nd_region, &label_id,
1047 allocated - val);
1048 else
1049 rc = grow_dpa_allocation(nd_region, &label_id, val - allocated);
1050
1051 if (rc)
1052 return rc;
1053
1054 if (is_namespace_pmem(dev)) {
1055 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1056
1057 nd_namespace_pmem_set_resource(nd_region, nspm,
1058 val * nd_region->ndr_mappings);
1059 }
1060
1061 /*
1062 * Try to delete the namespace if we deleted all of its
1063 * allocation, this is not the seed or 0th device for the
1064 * region, and it is not actively claimed by a btt, pfn, or dax
1065 * instance.
1066 */
1067 if (val == 0 && id != 0 && nd_region->ns_seed != dev && !ndns->claim)
1068 nd_device_unregister(dev, ND_ASYNC);
1069
1070 return rc;
1071}
1072
1073static ssize_t size_store(struct device *dev,
1074 struct device_attribute *attr, const char *buf, size_t len)
1075{
1076 struct nd_region *nd_region = to_nd_region(dev->parent);
1077 unsigned long long val;
1078 u8 **uuid = NULL;
1079 int rc;
1080
1081 rc = kstrtoull(buf, 0, &val);
1082 if (rc)
1083 return rc;
1084
1085 device_lock(dev);
1086 nvdimm_bus_lock(dev);
1087 wait_nvdimm_bus_probe_idle(dev);
1088 rc = __size_store(dev, val);
1089 if (rc >= 0)
1090 rc = nd_namespace_label_update(nd_region, dev);
1091
1092 if (is_namespace_pmem(dev)) {
1093 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1094
1095 uuid = &nspm->uuid;
1096 } else if (is_namespace_blk(dev)) {
1097 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1098
1099 uuid = &nsblk->uuid;
1100 }
1101
1102 if (rc == 0 && val == 0 && uuid) {
1103 /* setting size zero == 'delete namespace' */
1104 kfree(*uuid);
1105 *uuid = NULL;
1106 }
1107
1108 dev_dbg(dev, "%llx %s (%d)\n", val, rc < 0 ? "fail" : "success", rc);
1109
1110 nvdimm_bus_unlock(dev);
1111 device_unlock(dev);
1112
1113 return rc < 0 ? rc : len;
1114}
1115
1116resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1117{
1118 struct device *dev = &ndns->dev;
1119
1120 if (is_namespace_pmem(dev)) {
1121 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1122
1123 return resource_size(&nspm->nsio.res);
1124 } else if (is_namespace_blk(dev)) {
1125 return nd_namespace_blk_size(to_nd_namespace_blk(dev));
1126 } else if (is_namespace_io(dev)) {
1127 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1128
1129 return resource_size(&nsio->res);
1130 } else
1131 WARN_ONCE(1, "unknown namespace type\n");
1132 return 0;
1133}
1134
1135resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1136{
1137 resource_size_t size;
1138
1139 nvdimm_bus_lock(&ndns->dev);
1140 size = __nvdimm_namespace_capacity(ndns);
1141 nvdimm_bus_unlock(&ndns->dev);
1142
1143 return size;
1144}
1145EXPORT_SYMBOL(nvdimm_namespace_capacity);
1146
1147static ssize_t size_show(struct device *dev,
1148 struct device_attribute *attr, char *buf)
1149{
1150 return sprintf(buf, "%llu\n", (unsigned long long)
1151 nvdimm_namespace_capacity(to_ndns(dev)));
1152}
1153static DEVICE_ATTR(size, 0444, size_show, size_store);
1154
1155static u8 *namespace_to_uuid(struct device *dev)
1156{
1157 if (is_namespace_pmem(dev)) {
1158 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1159
1160 return nspm->uuid;
1161 } else if (is_namespace_blk(dev)) {
1162 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1163
1164 return nsblk->uuid;
1165 } else
1166 return ERR_PTR(-ENXIO);
1167}
1168
1169static ssize_t uuid_show(struct device *dev,
1170 struct device_attribute *attr, char *buf)
1171{
1172 u8 *uuid = namespace_to_uuid(dev);
1173
1174 if (IS_ERR(uuid))
1175 return PTR_ERR(uuid);
1176 if (uuid)
1177 return sprintf(buf, "%pUb\n", uuid);
1178 return sprintf(buf, "\n");
1179}
1180
1181/**
1182 * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
1183 * @nd_region: parent region so we can updates all dimms in the set
1184 * @dev: namespace type for generating label_id
1185 * @new_uuid: incoming uuid
1186 * @old_uuid: reference to the uuid storage location in the namespace object
1187 */
1188static int namespace_update_uuid(struct nd_region *nd_region,
1189 struct device *dev, u8 *new_uuid, u8 **old_uuid)
1190{
1191 u32 flags = is_namespace_blk(dev) ? NSLABEL_FLAG_LOCAL : 0;
1192 struct nd_label_id old_label_id;
1193 struct nd_label_id new_label_id;
1194 int i;
1195
1196 if (!nd_is_uuid_unique(dev, new_uuid))
1197 return -EINVAL;
1198
1199 if (*old_uuid == NULL)
1200 goto out;
1201
1202 /*
1203 * If we've already written a label with this uuid, then it's
1204 * too late to rename because we can't reliably update the uuid
1205 * without losing the old namespace. Userspace must delete this
1206 * namespace to abandon the old uuid.
1207 */
1208 for (i = 0; i < nd_region->ndr_mappings; i++) {
1209 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1210
1211 /*
1212 * This check by itself is sufficient because old_uuid
1213 * would be NULL above if this uuid did not exist in the
1214 * currently written set.
1215 *
1216 * FIXME: can we delete uuid with zero dpa allocated?
1217 */
1218 if (list_empty(&nd_mapping->labels))
1219 return -EBUSY;
1220 }
1221
1222 nd_label_gen_id(&old_label_id, *old_uuid, flags);
1223 nd_label_gen_id(&new_label_id, new_uuid, flags);
1224 for (i = 0; i < nd_region->ndr_mappings; i++) {
1225 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1226 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1227 struct resource *res;
1228
1229 for_each_dpa_resource(ndd, res)
1230 if (strcmp(res->name, old_label_id.id) == 0)
1231 sprintf((void *) res->name, "%s",
1232 new_label_id.id);
1233 }
1234 kfree(*old_uuid);
1235 out:
1236 *old_uuid = new_uuid;
1237 return 0;
1238}
1239
1240static ssize_t uuid_store(struct device *dev,
1241 struct device_attribute *attr, const char *buf, size_t len)
1242{
1243 struct nd_region *nd_region = to_nd_region(dev->parent);
1244 u8 *uuid = NULL;
1245 ssize_t rc = 0;
1246 u8 **ns_uuid;
1247
1248 if (is_namespace_pmem(dev)) {
1249 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1250
1251 ns_uuid = &nspm->uuid;
1252 } else if (is_namespace_blk(dev)) {
1253 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1254
1255 ns_uuid = &nsblk->uuid;
1256 } else
1257 return -ENXIO;
1258
1259 device_lock(dev);
1260 nvdimm_bus_lock(dev);
1261 wait_nvdimm_bus_probe_idle(dev);
1262 if (to_ndns(dev)->claim)
1263 rc = -EBUSY;
1264 if (rc >= 0)
1265 rc = nd_uuid_store(dev, &uuid, buf, len);
1266 if (rc >= 0)
1267 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
1268 if (rc >= 0)
1269 rc = nd_namespace_label_update(nd_region, dev);
1270 else
1271 kfree(uuid);
1272 dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
1273 buf[len - 1] == '\n' ? "" : "\n");
1274 nvdimm_bus_unlock(dev);
1275 device_unlock(dev);
1276
1277 return rc < 0 ? rc : len;
1278}
1279static DEVICE_ATTR_RW(uuid);
1280
1281static ssize_t resource_show(struct device *dev,
1282 struct device_attribute *attr, char *buf)
1283{
1284 struct resource *res;
1285
1286 if (is_namespace_pmem(dev)) {
1287 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1288
1289 res = &nspm->nsio.res;
1290 } else if (is_namespace_io(dev)) {
1291 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1292
1293 res = &nsio->res;
1294 } else
1295 return -ENXIO;
1296
1297 /* no address to convey if the namespace has no allocation */
1298 if (resource_size(res) == 0)
1299 return -ENXIO;
1300 return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
1301}
1302static DEVICE_ATTR_RO(resource);
1303
1304static const unsigned long blk_lbasize_supported[] = { 512, 520, 528,
1305 4096, 4104, 4160, 4224, 0 };
1306
1307static const unsigned long pmem_lbasize_supported[] = { 512, 4096, 0 };
1308
1309static ssize_t sector_size_show(struct device *dev,
1310 struct device_attribute *attr, char *buf)
1311{
1312 if (is_namespace_blk(dev)) {
1313 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1314
1315 return nd_size_select_show(nsblk->lbasize,
1316 blk_lbasize_supported, buf);
1317 }
1318
1319 if (is_namespace_pmem(dev)) {
1320 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1321
1322 return nd_size_select_show(nspm->lbasize,
1323 pmem_lbasize_supported, buf);
1324 }
1325 return -ENXIO;
1326}
1327
1328static ssize_t sector_size_store(struct device *dev,
1329 struct device_attribute *attr, const char *buf, size_t len)
1330{
1331 struct nd_region *nd_region = to_nd_region(dev->parent);
1332 const unsigned long *supported;
1333 unsigned long *lbasize;
1334 ssize_t rc = 0;
1335
1336 if (is_namespace_blk(dev)) {
1337 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1338
1339 lbasize = &nsblk->lbasize;
1340 supported = blk_lbasize_supported;
1341 } else if (is_namespace_pmem(dev)) {
1342 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1343
1344 lbasize = &nspm->lbasize;
1345 supported = pmem_lbasize_supported;
1346 } else
1347 return -ENXIO;
1348
1349 device_lock(dev);
1350 nvdimm_bus_lock(dev);
1351 if (to_ndns(dev)->claim)
1352 rc = -EBUSY;
1353 if (rc >= 0)
1354 rc = nd_size_select_store(dev, buf, lbasize, supported);
1355 if (rc >= 0)
1356 rc = nd_namespace_label_update(nd_region, dev);
1357 dev_dbg(dev, "result: %zd %s: %s%s", rc, rc < 0 ? "tried" : "wrote",
1358 buf, buf[len - 1] == '\n' ? "" : "\n");
1359 nvdimm_bus_unlock(dev);
1360 device_unlock(dev);
1361
1362 return rc ? rc : len;
1363}
1364static DEVICE_ATTR_RW(sector_size);
1365
1366static ssize_t dpa_extents_show(struct device *dev,
1367 struct device_attribute *attr, char *buf)
1368{
1369 struct nd_region *nd_region = to_nd_region(dev->parent);
1370 struct nd_label_id label_id;
1371 int count = 0, i;
1372 u8 *uuid = NULL;
1373 u32 flags = 0;
1374
1375 nvdimm_bus_lock(dev);
1376 if (is_namespace_pmem(dev)) {
1377 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1378
1379 uuid = nspm->uuid;
1380 flags = 0;
1381 } else if (is_namespace_blk(dev)) {
1382 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1383
1384 uuid = nsblk->uuid;
1385 flags = NSLABEL_FLAG_LOCAL;
1386 }
1387
1388 if (!uuid)
1389 goto out;
1390
1391 nd_label_gen_id(&label_id, uuid, flags);
1392 for (i = 0; i < nd_region->ndr_mappings; i++) {
1393 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1394 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1395 struct resource *res;
1396
1397 for_each_dpa_resource(ndd, res)
1398 if (strcmp(res->name, label_id.id) == 0)
1399 count++;
1400 }
1401 out:
1402 nvdimm_bus_unlock(dev);
1403
1404 return sprintf(buf, "%d\n", count);
1405}
1406static DEVICE_ATTR_RO(dpa_extents);
1407
1408static int btt_claim_class(struct device *dev)
1409{
1410 struct nd_region *nd_region = to_nd_region(dev->parent);
1411 int i, loop_bitmask = 0;
1412
1413 for (i = 0; i < nd_region->ndr_mappings; i++) {
1414 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1415 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1416 struct nd_namespace_index *nsindex;
1417
1418 /*
1419 * If any of the DIMMs do not support labels the only
1420 * possible BTT format is v1.
1421 */
1422 if (!ndd) {
1423 loop_bitmask = 0;
1424 break;
1425 }
1426
1427 nsindex = to_namespace_index(ndd, ndd->ns_current);
1428 if (nsindex == NULL)
1429 loop_bitmask |= 1;
1430 else {
1431 /* check whether existing labels are v1.1 or v1.2 */
1432 if (__le16_to_cpu(nsindex->major) == 1
1433 && __le16_to_cpu(nsindex->minor) == 1)
1434 loop_bitmask |= 2;
1435 else
1436 loop_bitmask |= 4;
1437 }
1438 }
1439 /*
1440 * If nsindex is null loop_bitmask's bit 0 will be set, and if an index
1441 * block is found, a v1.1 label for any mapping will set bit 1, and a
1442 * v1.2 label will set bit 2.
1443 *
1444 * At the end of the loop, at most one of the three bits must be set.
1445 * If multiple bits were set, it means the different mappings disagree
1446 * about their labels, and this must be cleaned up first.
1447 *
1448 * If all the label index blocks are found to agree, nsindex of NULL
1449 * implies labels haven't been initialized yet, and when they will,
1450 * they will be of the 1.2 format, so we can assume BTT2.0
1451 *
1452 * If 1.1 labels are found, we enforce BTT1.1, and if 1.2 labels are
1453 * found, we enforce BTT2.0
1454 *
1455 * If the loop was never entered, default to BTT1.1 (legacy namespaces)
1456 */
1457 switch (loop_bitmask) {
1458 case 0:
1459 case 2:
1460 return NVDIMM_CCLASS_BTT;
1461 case 1:
1462 case 4:
1463 return NVDIMM_CCLASS_BTT2;
1464 default:
1465 return -ENXIO;
1466 }
1467}
1468
1469static ssize_t holder_show(struct device *dev,
1470 struct device_attribute *attr, char *buf)
1471{
1472 struct nd_namespace_common *ndns = to_ndns(dev);
1473 ssize_t rc;
1474
1475 device_lock(dev);
1476 rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
1477 device_unlock(dev);
1478
1479 return rc;
1480}
1481static DEVICE_ATTR_RO(holder);
1482
1483static ssize_t __holder_class_store(struct device *dev, const char *buf)
1484{
1485 struct nd_namespace_common *ndns = to_ndns(dev);
1486
1487 if (dev->driver || ndns->claim)
1488 return -EBUSY;
1489
1490 if (strcmp(buf, "btt") == 0 || strcmp(buf, "btt\n") == 0)
1491 ndns->claim_class = btt_claim_class(dev);
1492 else if (strcmp(buf, "pfn") == 0 || strcmp(buf, "pfn\n") == 0)
1493 ndns->claim_class = NVDIMM_CCLASS_PFN;
1494 else if (strcmp(buf, "dax") == 0 || strcmp(buf, "dax\n") == 0)
1495 ndns->claim_class = NVDIMM_CCLASS_DAX;
1496 else if (strcmp(buf, "") == 0 || strcmp(buf, "\n") == 0)
1497 ndns->claim_class = NVDIMM_CCLASS_NONE;
1498 else
1499 return -EINVAL;
1500
1501 /* btt_claim_class() could've returned an error */
1502 if (ndns->claim_class < 0)
1503 return ndns->claim_class;
1504
1505 return 0;
1506}
1507
1508static ssize_t holder_class_store(struct device *dev,
1509 struct device_attribute *attr, const char *buf, size_t len)
1510{
1511 struct nd_region *nd_region = to_nd_region(dev->parent);
1512 ssize_t rc;
1513
1514 device_lock(dev);
1515 nvdimm_bus_lock(dev);
1516 wait_nvdimm_bus_probe_idle(dev);
1517 rc = __holder_class_store(dev, buf);
1518 if (rc >= 0)
1519 rc = nd_namespace_label_update(nd_region, dev);
1520 dev_dbg(dev, "%s(%zd)\n", rc < 0 ? "fail " : "", rc);
1521 nvdimm_bus_unlock(dev);
1522 device_unlock(dev);
1523
1524 return rc < 0 ? rc : len;
1525}
1526
1527static ssize_t holder_class_show(struct device *dev,
1528 struct device_attribute *attr, char *buf)
1529{
1530 struct nd_namespace_common *ndns = to_ndns(dev);
1531 ssize_t rc;
1532
1533 device_lock(dev);
1534 if (ndns->claim_class == NVDIMM_CCLASS_NONE)
1535 rc = sprintf(buf, "\n");
1536 else if ((ndns->claim_class == NVDIMM_CCLASS_BTT) ||
1537 (ndns->claim_class == NVDIMM_CCLASS_BTT2))
1538 rc = sprintf(buf, "btt\n");
1539 else if (ndns->claim_class == NVDIMM_CCLASS_PFN)
1540 rc = sprintf(buf, "pfn\n");
1541 else if (ndns->claim_class == NVDIMM_CCLASS_DAX)
1542 rc = sprintf(buf, "dax\n");
1543 else
1544 rc = sprintf(buf, "<unknown>\n");
1545 device_unlock(dev);
1546
1547 return rc;
1548}
1549static DEVICE_ATTR_RW(holder_class);
1550
1551static ssize_t mode_show(struct device *dev,
1552 struct device_attribute *attr, char *buf)
1553{
1554 struct nd_namespace_common *ndns = to_ndns(dev);
1555 struct device *claim;
1556 char *mode;
1557 ssize_t rc;
1558
1559 device_lock(dev);
1560 claim = ndns->claim;
1561 if (claim && is_nd_btt(claim))
1562 mode = "safe";
1563 else if (claim && is_nd_pfn(claim))
1564 mode = "memory";
1565 else if (claim && is_nd_dax(claim))
1566 mode = "dax";
1567 else if (!claim && pmem_should_map_pages(dev))
1568 mode = "memory";
1569 else
1570 mode = "raw";
1571 rc = sprintf(buf, "%s\n", mode);
1572 device_unlock(dev);
1573
1574 return rc;
1575}
1576static DEVICE_ATTR_RO(mode);
1577
1578static ssize_t force_raw_store(struct device *dev,
1579 struct device_attribute *attr, const char *buf, size_t len)
1580{
1581 bool force_raw;
1582 int rc = strtobool(buf, &force_raw);
1583
1584 if (rc)
1585 return rc;
1586
1587 to_ndns(dev)->force_raw = force_raw;
1588 return len;
1589}
1590
1591static ssize_t force_raw_show(struct device *dev,
1592 struct device_attribute *attr, char *buf)
1593{
1594 return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
1595}
1596static DEVICE_ATTR_RW(force_raw);
1597
1598static struct attribute *nd_namespace_attributes[] = {
1599 &dev_attr_nstype.attr,
1600 &dev_attr_size.attr,
1601 &dev_attr_mode.attr,
1602 &dev_attr_uuid.attr,
1603 &dev_attr_holder.attr,
1604 &dev_attr_resource.attr,
1605 &dev_attr_alt_name.attr,
1606 &dev_attr_force_raw.attr,
1607 &dev_attr_sector_size.attr,
1608 &dev_attr_dpa_extents.attr,
1609 &dev_attr_holder_class.attr,
1610 NULL,
1611};
1612
1613static umode_t namespace_visible(struct kobject *kobj,
1614 struct attribute *a, int n)
1615{
1616 struct device *dev = container_of(kobj, struct device, kobj);
1617
1618 if (a == &dev_attr_resource.attr) {
1619 if (is_namespace_blk(dev))
1620 return 0;
1621 return 0400;
1622 }
1623
1624 if (is_namespace_pmem(dev) || is_namespace_blk(dev)) {
1625 if (a == &dev_attr_size.attr)
1626 return 0644;
1627
1628 return a->mode;
1629 }
1630
1631 if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr
1632 || a == &dev_attr_holder.attr
1633 || a == &dev_attr_holder_class.attr
1634 || a == &dev_attr_force_raw.attr
1635 || a == &dev_attr_mode.attr)
1636 return a->mode;
1637
1638 return 0;
1639}
1640
1641static struct attribute_group nd_namespace_attribute_group = {
1642 .attrs = nd_namespace_attributes,
1643 .is_visible = namespace_visible,
1644};
1645
1646static const struct attribute_group *nd_namespace_attribute_groups[] = {
1647 &nd_device_attribute_group,
1648 &nd_namespace_attribute_group,
1649 &nd_numa_attribute_group,
1650 NULL,
1651};
1652
1653struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
1654{
1655 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
1656 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
1657 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
1658 struct nd_namespace_common *ndns = NULL;
1659 resource_size_t size;
1660
1661 if (nd_btt || nd_pfn || nd_dax) {
1662 if (nd_btt)
1663 ndns = nd_btt->ndns;
1664 else if (nd_pfn)
1665 ndns = nd_pfn->ndns;
1666 else if (nd_dax)
1667 ndns = nd_dax->nd_pfn.ndns;
1668
1669 if (!ndns)
1670 return ERR_PTR(-ENODEV);
1671
1672 /*
1673 * Flush any in-progess probes / removals in the driver
1674 * for the raw personality of this namespace.
1675 */
1676 device_lock(&ndns->dev);
1677 device_unlock(&ndns->dev);
1678 if (ndns->dev.driver) {
1679 dev_dbg(&ndns->dev, "is active, can't bind %s\n",
1680 dev_name(dev));
1681 return ERR_PTR(-EBUSY);
1682 }
1683 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != dev,
1684 "host (%s) vs claim (%s) mismatch\n",
1685 dev_name(dev),
1686 dev_name(ndns->claim)))
1687 return ERR_PTR(-ENXIO);
1688 } else {
1689 ndns = to_ndns(dev);
1690 if (ndns->claim) {
1691 dev_dbg(dev, "claimed by %s, failing probe\n",
1692 dev_name(ndns->claim));
1693
1694 return ERR_PTR(-ENXIO);
1695 }
1696 }
1697
1698 size = nvdimm_namespace_capacity(ndns);
1699 if (size < ND_MIN_NAMESPACE_SIZE) {
1700 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
1701 &size, ND_MIN_NAMESPACE_SIZE);
1702 return ERR_PTR(-ENODEV);
1703 }
1704
1705 if (is_namespace_pmem(&ndns->dev)) {
1706 struct nd_namespace_pmem *nspm;
1707
1708 nspm = to_nd_namespace_pmem(&ndns->dev);
1709 if (uuid_not_set(nspm->uuid, &ndns->dev, __func__))
1710 return ERR_PTR(-ENODEV);
1711 } else if (is_namespace_blk(&ndns->dev)) {
1712 struct nd_namespace_blk *nsblk;
1713
1714 nsblk = to_nd_namespace_blk(&ndns->dev);
1715 if (uuid_not_set(nsblk->uuid, &ndns->dev, __func__))
1716 return ERR_PTR(-ENODEV);
1717 if (!nsblk->lbasize) {
1718 dev_dbg(&ndns->dev, "sector size not set\n");
1719 return ERR_PTR(-ENODEV);
1720 }
1721 if (!nd_namespace_blk_validate(nsblk))
1722 return ERR_PTR(-ENODEV);
1723 }
1724
1725 return ndns;
1726}
1727EXPORT_SYMBOL(nvdimm_namespace_common_probe);
1728
1729static struct device **create_namespace_io(struct nd_region *nd_region)
1730{
1731 struct nd_namespace_io *nsio;
1732 struct device *dev, **devs;
1733 struct resource *res;
1734
1735 nsio = kzalloc(sizeof(*nsio), GFP_KERNEL);
1736 if (!nsio)
1737 return NULL;
1738
1739 devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1740 if (!devs) {
1741 kfree(nsio);
1742 return NULL;
1743 }
1744
1745 dev = &nsio->common.dev;
1746 dev->type = &namespace_io_device_type;
1747 dev->parent = &nd_region->dev;
1748 res = &nsio->res;
1749 res->name = dev_name(&nd_region->dev);
1750 res->flags = IORESOURCE_MEM;
1751 res->start = nd_region->ndr_start;
1752 res->end = res->start + nd_region->ndr_size - 1;
1753
1754 devs[0] = dev;
1755 return devs;
1756}
1757
1758static bool has_uuid_at_pos(struct nd_region *nd_region, u8 *uuid,
1759 u64 cookie, u16 pos)
1760{
1761 struct nd_namespace_label *found = NULL;
1762 int i;
1763
1764 for (i = 0; i < nd_region->ndr_mappings; i++) {
1765 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1766 struct nd_interleave_set *nd_set = nd_region->nd_set;
1767 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1768 struct nd_label_ent *label_ent;
1769 bool found_uuid = false;
1770
1771 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1772 struct nd_namespace_label *nd_label = label_ent->label;
1773 u16 position, nlabel;
1774 u64 isetcookie;
1775
1776 if (!nd_label)
1777 continue;
1778 isetcookie = __le64_to_cpu(nd_label->isetcookie);
1779 position = __le16_to_cpu(nd_label->position);
1780 nlabel = __le16_to_cpu(nd_label->nlabel);
1781
1782 if (isetcookie != cookie)
1783 continue;
1784
1785 if (memcmp(nd_label->uuid, uuid, NSLABEL_UUID_LEN) != 0)
1786 continue;
1787
1788 if (namespace_label_has(ndd, type_guid)
1789 && !guid_equal(&nd_set->type_guid,
1790 &nd_label->type_guid)) {
1791 dev_dbg(ndd->dev, "expect type_guid %pUb got %pUb\n",
1792 nd_set->type_guid.b,
1793 nd_label->type_guid.b);
1794 continue;
1795 }
1796
1797 if (found_uuid) {
1798 dev_dbg(ndd->dev, "duplicate entry for uuid\n");
1799 return false;
1800 }
1801 found_uuid = true;
1802 if (nlabel != nd_region->ndr_mappings)
1803 continue;
1804 if (position != pos)
1805 continue;
1806 found = nd_label;
1807 break;
1808 }
1809 if (found)
1810 break;
1811 }
1812 return found != NULL;
1813}
1814
1815static int select_pmem_id(struct nd_region *nd_region, u8 *pmem_id)
1816{
1817 int i;
1818
1819 if (!pmem_id)
1820 return -ENODEV;
1821
1822 for (i = 0; i < nd_region->ndr_mappings; i++) {
1823 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1824 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1825 struct nd_namespace_label *nd_label = NULL;
1826 u64 hw_start, hw_end, pmem_start, pmem_end;
1827 struct nd_label_ent *label_ent;
1828
1829 lockdep_assert_held(&nd_mapping->lock);
1830 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1831 nd_label = label_ent->label;
1832 if (!nd_label)
1833 continue;
1834 if (memcmp(nd_label->uuid, pmem_id, NSLABEL_UUID_LEN) == 0)
1835 break;
1836 nd_label = NULL;
1837 }
1838
1839 if (!nd_label) {
1840 WARN_ON(1);
1841 return -EINVAL;
1842 }
1843
1844 /*
1845 * Check that this label is compliant with the dpa
1846 * range published in NFIT
1847 */
1848 hw_start = nd_mapping->start;
1849 hw_end = hw_start + nd_mapping->size;
1850 pmem_start = __le64_to_cpu(nd_label->dpa);
1851 pmem_end = pmem_start + __le64_to_cpu(nd_label->rawsize);
1852 if (pmem_start >= hw_start && pmem_start < hw_end
1853 && pmem_end <= hw_end && pmem_end > hw_start)
1854 /* pass */;
1855 else {
1856 dev_dbg(&nd_region->dev, "%s invalid label for %pUb\n",
1857 dev_name(ndd->dev), nd_label->uuid);
1858 return -EINVAL;
1859 }
1860
1861 /* move recently validated label to the front of the list */
1862 list_move(&label_ent->list, &nd_mapping->labels);
1863 }
1864 return 0;
1865}
1866
1867/**
1868 * create_namespace_pmem - validate interleave set labelling, retrieve label0
1869 * @nd_region: region with mappings to validate
1870 * @nspm: target namespace to create
1871 * @nd_label: target pmem namespace label to evaluate
1872 */
1873static struct device *create_namespace_pmem(struct nd_region *nd_region,
1874 struct nd_namespace_index *nsindex,
1875 struct nd_namespace_label *nd_label)
1876{
1877 u64 cookie = nd_region_interleave_set_cookie(nd_region, nsindex);
1878 u64 altcookie = nd_region_interleave_set_altcookie(nd_region);
1879 struct nd_label_ent *label_ent;
1880 struct nd_namespace_pmem *nspm;
1881 struct nd_mapping *nd_mapping;
1882 resource_size_t size = 0;
1883 struct resource *res;
1884 struct device *dev;
1885 int rc = 0;
1886 u16 i;
1887
1888 if (cookie == 0) {
1889 dev_dbg(&nd_region->dev, "invalid interleave-set-cookie\n");
1890 return ERR_PTR(-ENXIO);
1891 }
1892
1893 if (__le64_to_cpu(nd_label->isetcookie) != cookie) {
1894 dev_dbg(&nd_region->dev, "invalid cookie in label: %pUb\n",
1895 nd_label->uuid);
1896 if (__le64_to_cpu(nd_label->isetcookie) != altcookie)
1897 return ERR_PTR(-EAGAIN);
1898
1899 dev_dbg(&nd_region->dev, "valid altcookie in label: %pUb\n",
1900 nd_label->uuid);
1901 }
1902
1903 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1904 if (!nspm)
1905 return ERR_PTR(-ENOMEM);
1906
1907 nspm->id = -1;
1908 dev = &nspm->nsio.common.dev;
1909 dev->type = &namespace_pmem_device_type;
1910 dev->parent = &nd_region->dev;
1911 res = &nspm->nsio.res;
1912 res->name = dev_name(&nd_region->dev);
1913 res->flags = IORESOURCE_MEM;
1914
1915 for (i = 0; i < nd_region->ndr_mappings; i++) {
1916 if (has_uuid_at_pos(nd_region, nd_label->uuid, cookie, i))
1917 continue;
1918 if (has_uuid_at_pos(nd_region, nd_label->uuid, altcookie, i))
1919 continue;
1920 break;
1921 }
1922
1923 if (i < nd_region->ndr_mappings) {
1924 struct nvdimm *nvdimm = nd_region->mapping[i].nvdimm;
1925
1926 /*
1927 * Give up if we don't find an instance of a uuid at each
1928 * position (from 0 to nd_region->ndr_mappings - 1), or if we
1929 * find a dimm with two instances of the same uuid.
1930 */
1931 dev_err(&nd_region->dev, "%s missing label for %pUb\n",
1932 nvdimm_name(nvdimm), nd_label->uuid);
1933 rc = -EINVAL;
1934 goto err;
1935 }
1936
1937 /*
1938 * Fix up each mapping's 'labels' to have the validated pmem label for
1939 * that position at labels[0], and NULL at labels[1]. In the process,
1940 * check that the namespace aligns with interleave-set. We know
1941 * that it does not overlap with any blk namespaces by virtue of
1942 * the dimm being enabled (i.e. nd_label_reserve_dpa()
1943 * succeeded).
1944 */
1945 rc = select_pmem_id(nd_region, nd_label->uuid);
1946 if (rc)
1947 goto err;
1948
1949 /* Calculate total size and populate namespace properties from label0 */
1950 for (i = 0; i < nd_region->ndr_mappings; i++) {
1951 struct nd_namespace_label *label0;
1952 struct nvdimm_drvdata *ndd;
1953
1954 nd_mapping = &nd_region->mapping[i];
1955 label_ent = list_first_entry_or_null(&nd_mapping->labels,
1956 typeof(*label_ent), list);
1957 label0 = label_ent ? label_ent->label : 0;
1958
1959 if (!label0) {
1960 WARN_ON(1);
1961 continue;
1962 }
1963
1964 size += __le64_to_cpu(label0->rawsize);
1965 if (__le16_to_cpu(label0->position) != 0)
1966 continue;
1967 WARN_ON(nspm->alt_name || nspm->uuid);
1968 nspm->alt_name = kmemdup((void __force *) label0->name,
1969 NSLABEL_NAME_LEN, GFP_KERNEL);
1970 nspm->uuid = kmemdup((void __force *) label0->uuid,
1971 NSLABEL_UUID_LEN, GFP_KERNEL);
1972 nspm->lbasize = __le64_to_cpu(label0->lbasize);
1973 ndd = to_ndd(nd_mapping);
1974 if (namespace_label_has(ndd, abstraction_guid))
1975 nspm->nsio.common.claim_class
1976 = to_nvdimm_cclass(&label0->abstraction_guid);
1977
1978 }
1979
1980 if (!nspm->alt_name || !nspm->uuid) {
1981 rc = -ENOMEM;
1982 goto err;
1983 }
1984
1985 nd_namespace_pmem_set_resource(nd_region, nspm, size);
1986
1987 return dev;
1988 err:
1989 namespace_pmem_release(dev);
1990 switch (rc) {
1991 case -EINVAL:
1992 dev_dbg(&nd_region->dev, "invalid label(s)\n");
1993 break;
1994 case -ENODEV:
1995 dev_dbg(&nd_region->dev, "label not found\n");
1996 break;
1997 default:
1998 dev_dbg(&nd_region->dev, "unexpected err: %d\n", rc);
1999 break;
2000 }
2001 return ERR_PTR(rc);
2002}
2003
2004struct resource *nsblk_add_resource(struct nd_region *nd_region,
2005 struct nvdimm_drvdata *ndd, struct nd_namespace_blk *nsblk,
2006 resource_size_t start)
2007{
2008 struct nd_label_id label_id;
2009 struct resource *res;
2010
2011 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
2012 res = krealloc(nsblk->res,
2013 sizeof(void *) * (nsblk->num_resources + 1),
2014 GFP_KERNEL);
2015 if (!res)
2016 return NULL;
2017 nsblk->res = (struct resource **) res;
2018 for_each_dpa_resource(ndd, res)
2019 if (strcmp(res->name, label_id.id) == 0
2020 && res->start == start) {
2021 nsblk->res[nsblk->num_resources++] = res;
2022 return res;
2023 }
2024 return NULL;
2025}
2026
2027static struct device *nd_namespace_blk_create(struct nd_region *nd_region)
2028{
2029 struct nd_namespace_blk *nsblk;
2030 struct device *dev;
2031
2032 if (!is_nd_blk(&nd_region->dev))
2033 return NULL;
2034
2035 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2036 if (!nsblk)
2037 return NULL;
2038
2039 dev = &nsblk->common.dev;
2040 dev->type = &namespace_blk_device_type;
2041 nsblk->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
2042 if (nsblk->id < 0) {
2043 kfree(nsblk);
2044 return NULL;
2045 }
2046 dev_set_name(dev, "namespace%d.%d", nd_region->id, nsblk->id);
2047 dev->parent = &nd_region->dev;
2048 dev->groups = nd_namespace_attribute_groups;
2049
2050 return &nsblk->common.dev;
2051}
2052
2053static struct device *nd_namespace_pmem_create(struct nd_region *nd_region)
2054{
2055 struct nd_namespace_pmem *nspm;
2056 struct resource *res;
2057 struct device *dev;
2058
2059 if (!is_memory(&nd_region->dev))
2060 return NULL;
2061
2062 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2063 if (!nspm)
2064 return NULL;
2065
2066 dev = &nspm->nsio.common.dev;
2067 dev->type = &namespace_pmem_device_type;
2068 dev->parent = &nd_region->dev;
2069 res = &nspm->nsio.res;
2070 res->name = dev_name(&nd_region->dev);
2071 res->flags = IORESOURCE_MEM;
2072
2073 nspm->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
2074 if (nspm->id < 0) {
2075 kfree(nspm);
2076 return NULL;
2077 }
2078 dev_set_name(dev, "namespace%d.%d", nd_region->id, nspm->id);
2079 dev->parent = &nd_region->dev;
2080 dev->groups = nd_namespace_attribute_groups;
2081 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2082
2083 return dev;
2084}
2085
2086void nd_region_create_ns_seed(struct nd_region *nd_region)
2087{
2088 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2089
2090 if (nd_region_to_nstype(nd_region) == ND_DEVICE_NAMESPACE_IO)
2091 return;
2092
2093 if (is_nd_blk(&nd_region->dev))
2094 nd_region->ns_seed = nd_namespace_blk_create(nd_region);
2095 else
2096 nd_region->ns_seed = nd_namespace_pmem_create(nd_region);
2097
2098 /*
2099 * Seed creation failures are not fatal, provisioning is simply
2100 * disabled until memory becomes available
2101 */
2102 if (!nd_region->ns_seed)
2103 dev_err(&nd_region->dev, "failed to create %s namespace\n",
2104 is_nd_blk(&nd_region->dev) ? "blk" : "pmem");
2105 else
2106 nd_device_register(nd_region->ns_seed);
2107}
2108
2109void nd_region_create_dax_seed(struct nd_region *nd_region)
2110{
2111 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2112 nd_region->dax_seed = nd_dax_create(nd_region);
2113 /*
2114 * Seed creation failures are not fatal, provisioning is simply
2115 * disabled until memory becomes available
2116 */
2117 if (!nd_region->dax_seed)
2118 dev_err(&nd_region->dev, "failed to create dax namespace\n");
2119}
2120
2121void nd_region_create_pfn_seed(struct nd_region *nd_region)
2122{
2123 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2124 nd_region->pfn_seed = nd_pfn_create(nd_region);
2125 /*
2126 * Seed creation failures are not fatal, provisioning is simply
2127 * disabled until memory becomes available
2128 */
2129 if (!nd_region->pfn_seed)
2130 dev_err(&nd_region->dev, "failed to create pfn namespace\n");
2131}
2132
2133void nd_region_create_btt_seed(struct nd_region *nd_region)
2134{
2135 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2136 nd_region->btt_seed = nd_btt_create(nd_region);
2137 /*
2138 * Seed creation failures are not fatal, provisioning is simply
2139 * disabled until memory becomes available
2140 */
2141 if (!nd_region->btt_seed)
2142 dev_err(&nd_region->dev, "failed to create btt namespace\n");
2143}
2144
2145static int add_namespace_resource(struct nd_region *nd_region,
2146 struct nd_namespace_label *nd_label, struct device **devs,
2147 int count)
2148{
2149 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2150 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2151 int i;
2152
2153 for (i = 0; i < count; i++) {
2154 u8 *uuid = namespace_to_uuid(devs[i]);
2155 struct resource *res;
2156
2157 if (IS_ERR_OR_NULL(uuid)) {
2158 WARN_ON(1);
2159 continue;
2160 }
2161
2162 if (memcmp(uuid, nd_label->uuid, NSLABEL_UUID_LEN) != 0)
2163 continue;
2164 if (is_namespace_blk(devs[i])) {
2165 res = nsblk_add_resource(nd_region, ndd,
2166 to_nd_namespace_blk(devs[i]),
2167 __le64_to_cpu(nd_label->dpa));
2168 if (!res)
2169 return -ENXIO;
2170 nd_dbg_dpa(nd_region, ndd, res, "%d assign\n", count);
2171 } else {
2172 dev_err(&nd_region->dev,
2173 "error: conflicting extents for uuid: %pUb\n",
2174 nd_label->uuid);
2175 return -ENXIO;
2176 }
2177 break;
2178 }
2179
2180 return i;
2181}
2182
2183static struct device *create_namespace_blk(struct nd_region *nd_region,
2184 struct nd_namespace_label *nd_label, int count)
2185{
2186
2187 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2188 struct nd_interleave_set *nd_set = nd_region->nd_set;
2189 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2190 struct nd_namespace_blk *nsblk;
2191 char name[NSLABEL_NAME_LEN];
2192 struct device *dev = NULL;
2193 struct resource *res;
2194
2195 if (namespace_label_has(ndd, type_guid)) {
2196 if (!guid_equal(&nd_set->type_guid, &nd_label->type_guid)) {
2197 dev_dbg(ndd->dev, "expect type_guid %pUb got %pUb\n",
2198 nd_set->type_guid.b,
2199 nd_label->type_guid.b);
2200 return ERR_PTR(-EAGAIN);
2201 }
2202
2203 if (nd_label->isetcookie != __cpu_to_le64(nd_set->cookie2)) {
2204 dev_dbg(ndd->dev, "expect cookie %#llx got %#llx\n",
2205 nd_set->cookie2,
2206 __le64_to_cpu(nd_label->isetcookie));
2207 return ERR_PTR(-EAGAIN);
2208 }
2209 }
2210
2211 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2212 if (!nsblk)
2213 return ERR_PTR(-ENOMEM);
2214 dev = &nsblk->common.dev;
2215 dev->type = &namespace_blk_device_type;
2216 dev->parent = &nd_region->dev;
2217 nsblk->id = -1;
2218 nsblk->lbasize = __le64_to_cpu(nd_label->lbasize);
2219 nsblk->uuid = kmemdup(nd_label->uuid, NSLABEL_UUID_LEN,
2220 GFP_KERNEL);
2221 if (namespace_label_has(ndd, abstraction_guid))
2222 nsblk->common.claim_class
2223 = to_nvdimm_cclass(&nd_label->abstraction_guid);
2224 if (!nsblk->uuid)
2225 goto blk_err;
2226 memcpy(name, nd_label->name, NSLABEL_NAME_LEN);
2227 if (name[0])
2228 nsblk->alt_name = kmemdup(name, NSLABEL_NAME_LEN,
2229 GFP_KERNEL);
2230 res = nsblk_add_resource(nd_region, ndd, nsblk,
2231 __le64_to_cpu(nd_label->dpa));
2232 if (!res)
2233 goto blk_err;
2234 nd_dbg_dpa(nd_region, ndd, res, "%d: assign\n", count);
2235 return dev;
2236 blk_err:
2237 namespace_blk_release(dev);
2238 return ERR_PTR(-ENXIO);
2239}
2240
2241static int cmp_dpa(const void *a, const void *b)
2242{
2243 const struct device *dev_a = *(const struct device **) a;
2244 const struct device *dev_b = *(const struct device **) b;
2245 struct nd_namespace_blk *nsblk_a, *nsblk_b;
2246 struct nd_namespace_pmem *nspm_a, *nspm_b;
2247
2248 if (is_namespace_io(dev_a))
2249 return 0;
2250
2251 if (is_namespace_blk(dev_a)) {
2252 nsblk_a = to_nd_namespace_blk(dev_a);
2253 nsblk_b = to_nd_namespace_blk(dev_b);
2254
2255 return memcmp(&nsblk_a->res[0]->start, &nsblk_b->res[0]->start,
2256 sizeof(resource_size_t));
2257 }
2258
2259 nspm_a = to_nd_namespace_pmem(dev_a);
2260 nspm_b = to_nd_namespace_pmem(dev_b);
2261
2262 return memcmp(&nspm_a->nsio.res.start, &nspm_b->nsio.res.start,
2263 sizeof(resource_size_t));
2264}
2265
2266static struct device **scan_labels(struct nd_region *nd_region)
2267{
2268 int i, count = 0;
2269 struct device *dev, **devs = NULL;
2270 struct nd_label_ent *label_ent, *e;
2271 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2272 resource_size_t map_end = nd_mapping->start + nd_mapping->size - 1;
2273
2274 /* "safe" because create_namespace_pmem() might list_move() label_ent */
2275 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
2276 struct nd_namespace_label *nd_label = label_ent->label;
2277 struct device **__devs;
2278 u32 flags;
2279
2280 if (!nd_label)
2281 continue;
2282 flags = __le32_to_cpu(nd_label->flags);
2283 if (is_nd_blk(&nd_region->dev)
2284 == !!(flags & NSLABEL_FLAG_LOCAL))
2285 /* pass, region matches label type */;
2286 else
2287 continue;
2288
2289 /* skip labels that describe extents outside of the region */
2290 if (nd_label->dpa < nd_mapping->start || nd_label->dpa > map_end)
2291 continue;
2292
2293 i = add_namespace_resource(nd_region, nd_label, devs, count);
2294 if (i < 0)
2295 goto err;
2296 if (i < count)
2297 continue;
2298 __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
2299 if (!__devs)
2300 goto err;
2301 memcpy(__devs, devs, sizeof(dev) * count);
2302 kfree(devs);
2303 devs = __devs;
2304
2305 if (is_nd_blk(&nd_region->dev))
2306 dev = create_namespace_blk(nd_region, nd_label, count);
2307 else {
2308 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2309 struct nd_namespace_index *nsindex;
2310
2311 nsindex = to_namespace_index(ndd, ndd->ns_current);
2312 dev = create_namespace_pmem(nd_region, nsindex, nd_label);
2313 }
2314
2315 if (IS_ERR(dev)) {
2316 switch (PTR_ERR(dev)) {
2317 case -EAGAIN:
2318 /* skip invalid labels */
2319 continue;
2320 case -ENODEV:
2321 /* fallthrough to seed creation */
2322 break;
2323 default:
2324 goto err;
2325 }
2326 } else
2327 devs[count++] = dev;
2328
2329 }
2330
2331 dev_dbg(&nd_region->dev, "discovered %d %s namespace%s\n",
2332 count, is_nd_blk(&nd_region->dev)
2333 ? "blk" : "pmem", count == 1 ? "" : "s");
2334
2335 if (count == 0) {
2336 /* Publish a zero-sized namespace for userspace to configure. */
2337 nd_mapping_free_labels(nd_mapping);
2338
2339 devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
2340 if (!devs)
2341 goto err;
2342 if (is_nd_blk(&nd_region->dev)) {
2343 struct nd_namespace_blk *nsblk;
2344
2345 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2346 if (!nsblk)
2347 goto err;
2348 dev = &nsblk->common.dev;
2349 dev->type = &namespace_blk_device_type;
2350 } else {
2351 struct nd_namespace_pmem *nspm;
2352
2353 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2354 if (!nspm)
2355 goto err;
2356 dev = &nspm->nsio.common.dev;
2357 dev->type = &namespace_pmem_device_type;
2358 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2359 }
2360 dev->parent = &nd_region->dev;
2361 devs[count++] = dev;
2362 } else if (is_memory(&nd_region->dev)) {
2363 /* clean unselected labels */
2364 for (i = 0; i < nd_region->ndr_mappings; i++) {
2365 struct list_head *l, *e;
2366 LIST_HEAD(list);
2367 int j;
2368
2369 nd_mapping = &nd_region->mapping[i];
2370 if (list_empty(&nd_mapping->labels)) {
2371 WARN_ON(1);
2372 continue;
2373 }
2374
2375 j = count;
2376 list_for_each_safe(l, e, &nd_mapping->labels) {
2377 if (!j--)
2378 break;
2379 list_move_tail(l, &list);
2380 }
2381 nd_mapping_free_labels(nd_mapping);
2382 list_splice_init(&list, &nd_mapping->labels);
2383 }
2384 }
2385
2386 if (count > 1)
2387 sort(devs, count, sizeof(struct device *), cmp_dpa, NULL);
2388
2389 return devs;
2390
2391 err:
2392 if (devs) {
2393 for (i = 0; devs[i]; i++)
2394 if (is_nd_blk(&nd_region->dev))
2395 namespace_blk_release(devs[i]);
2396 else
2397 namespace_pmem_release(devs[i]);
2398 kfree(devs);
2399 }
2400 return NULL;
2401}
2402
2403static struct device **create_namespaces(struct nd_region *nd_region)
2404{
2405 struct nd_mapping *nd_mapping;
2406 struct device **devs;
2407 int i;
2408
2409 if (nd_region->ndr_mappings == 0)
2410 return NULL;
2411
2412 /* lock down all mappings while we scan labels */
2413 for (i = 0; i < nd_region->ndr_mappings; i++) {
2414 nd_mapping = &nd_region->mapping[i];
2415 mutex_lock_nested(&nd_mapping->lock, i);
2416 }
2417
2418 devs = scan_labels(nd_region);
2419
2420 for (i = 0; i < nd_region->ndr_mappings; i++) {
2421 int reverse = nd_region->ndr_mappings - 1 - i;
2422
2423 nd_mapping = &nd_region->mapping[reverse];
2424 mutex_unlock(&nd_mapping->lock);
2425 }
2426
2427 return devs;
2428}
2429
2430static int init_active_labels(struct nd_region *nd_region)
2431{
2432 int i;
2433
2434 for (i = 0; i < nd_region->ndr_mappings; i++) {
2435 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2436 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2437 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2438 struct nd_label_ent *label_ent;
2439 int count, j;
2440
2441 /*
2442 * If the dimm is disabled then we may need to prevent
2443 * the region from being activated.
2444 */
2445 if (!ndd) {
2446 if (test_bit(NDD_LOCKED, &nvdimm->flags))
2447 /* fail, label data may be unreadable */;
2448 else if (test_bit(NDD_ALIASING, &nvdimm->flags))
2449 /* fail, labels needed to disambiguate dpa */;
2450 else
2451 return 0;
2452
2453 dev_err(&nd_region->dev, "%s: is %s, failing probe\n",
2454 dev_name(&nd_mapping->nvdimm->dev),
2455 test_bit(NDD_LOCKED, &nvdimm->flags)
2456 ? "locked" : "disabled");
2457 return -ENXIO;
2458 }
2459 nd_mapping->ndd = ndd;
2460 atomic_inc(&nvdimm->busy);
2461 get_ndd(ndd);
2462
2463 count = nd_label_active_count(ndd);
2464 dev_dbg(ndd->dev, "count: %d\n", count);
2465 if (!count)
2466 continue;
2467 for (j = 0; j < count; j++) {
2468 struct nd_namespace_label *label;
2469
2470 label_ent = kzalloc(sizeof(*label_ent), GFP_KERNEL);
2471 if (!label_ent)
2472 break;
2473 label = nd_label_active(ndd, j);
2474 label_ent->label = label;
2475
2476 mutex_lock(&nd_mapping->lock);
2477 list_add_tail(&label_ent->list, &nd_mapping->labels);
2478 mutex_unlock(&nd_mapping->lock);
2479 }
2480
2481 if (j >= count)
2482 continue;
2483
2484 mutex_lock(&nd_mapping->lock);
2485 nd_mapping_free_labels(nd_mapping);
2486 mutex_unlock(&nd_mapping->lock);
2487 return -ENOMEM;
2488 }
2489
2490 return 0;
2491}
2492
2493int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
2494{
2495 struct device **devs = NULL;
2496 int i, rc = 0, type;
2497
2498 *err = 0;
2499 nvdimm_bus_lock(&nd_region->dev);
2500 rc = init_active_labels(nd_region);
2501 if (rc) {
2502 nvdimm_bus_unlock(&nd_region->dev);
2503 return rc;
2504 }
2505
2506 type = nd_region_to_nstype(nd_region);
2507 switch (type) {
2508 case ND_DEVICE_NAMESPACE_IO:
2509 devs = create_namespace_io(nd_region);
2510 break;
2511 case ND_DEVICE_NAMESPACE_PMEM:
2512 case ND_DEVICE_NAMESPACE_BLK:
2513 devs = create_namespaces(nd_region);
2514 break;
2515 default:
2516 break;
2517 }
2518 nvdimm_bus_unlock(&nd_region->dev);
2519
2520 if (!devs)
2521 return -ENODEV;
2522
2523 for (i = 0; devs[i]; i++) {
2524 struct device *dev = devs[i];
2525 int id;
2526
2527 if (type == ND_DEVICE_NAMESPACE_BLK) {
2528 struct nd_namespace_blk *nsblk;
2529
2530 nsblk = to_nd_namespace_blk(dev);
2531 id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2532 GFP_KERNEL);
2533 nsblk->id = id;
2534 } else if (type == ND_DEVICE_NAMESPACE_PMEM) {
2535 struct nd_namespace_pmem *nspm;
2536
2537 nspm = to_nd_namespace_pmem(dev);
2538 id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2539 GFP_KERNEL);
2540 nspm->id = id;
2541 } else
2542 id = i;
2543
2544 if (id < 0)
2545 break;
2546 dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
2547 dev->groups = nd_namespace_attribute_groups;
2548 nd_device_register(dev);
2549 }
2550 if (i)
2551 nd_region->ns_seed = devs[0];
2552
2553 if (devs[i]) {
2554 int j;
2555
2556 for (j = i; devs[j]; j++) {
2557 struct device *dev = devs[j];
2558
2559 device_initialize(dev);
2560 put_device(dev);
2561 }
2562 *err = j - i;
2563 /*
2564 * All of the namespaces we tried to register failed, so
2565 * fail region activation.
2566 */
2567 if (*err == 0)
2568 rc = -ENODEV;
2569 }
2570 kfree(devs);
2571
2572 if (rc == -ENODEV)
2573 return rc;
2574
2575 return i;
2576}