Loading...
Note: File does not exist in v3.1.
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
4 */
5#include <linux/module.h>
6#include <linux/device.h>
7#include <linux/sort.h>
8#include <linux/slab.h>
9#include <linux/list.h>
10#include <linux/nd.h>
11#include "nd-core.h"
12#include "pmem.h"
13#include "nd.h"
14
15static void namespace_io_release(struct device *dev)
16{
17 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
18
19 kfree(nsio);
20}
21
22static void namespace_pmem_release(struct device *dev)
23{
24 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
25 struct nd_region *nd_region = to_nd_region(dev->parent);
26
27 if (nspm->id >= 0)
28 ida_simple_remove(&nd_region->ns_ida, nspm->id);
29 kfree(nspm->alt_name);
30 kfree(nspm->uuid);
31 kfree(nspm);
32}
33
34static void namespace_blk_release(struct device *dev)
35{
36 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
37 struct nd_region *nd_region = to_nd_region(dev->parent);
38
39 if (nsblk->id >= 0)
40 ida_simple_remove(&nd_region->ns_ida, nsblk->id);
41 kfree(nsblk->alt_name);
42 kfree(nsblk->uuid);
43 kfree(nsblk->res);
44 kfree(nsblk);
45}
46
47static const struct device_type namespace_io_device_type = {
48 .name = "nd_namespace_io",
49 .release = namespace_io_release,
50};
51
52static const struct device_type namespace_pmem_device_type = {
53 .name = "nd_namespace_pmem",
54 .release = namespace_pmem_release,
55};
56
57static const struct device_type namespace_blk_device_type = {
58 .name = "nd_namespace_blk",
59 .release = namespace_blk_release,
60};
61
62static bool is_namespace_pmem(const struct device *dev)
63{
64 return dev ? dev->type == &namespace_pmem_device_type : false;
65}
66
67static bool is_namespace_blk(const struct device *dev)
68{
69 return dev ? dev->type == &namespace_blk_device_type : false;
70}
71
72static bool is_namespace_io(const struct device *dev)
73{
74 return dev ? dev->type == &namespace_io_device_type : false;
75}
76
77static int is_uuid_busy(struct device *dev, void *data)
78{
79 u8 *uuid1 = data, *uuid2 = NULL;
80
81 if (is_namespace_pmem(dev)) {
82 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
83
84 uuid2 = nspm->uuid;
85 } else if (is_namespace_blk(dev)) {
86 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
87
88 uuid2 = nsblk->uuid;
89 } else if (is_nd_btt(dev)) {
90 struct nd_btt *nd_btt = to_nd_btt(dev);
91
92 uuid2 = nd_btt->uuid;
93 } else if (is_nd_pfn(dev)) {
94 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
95
96 uuid2 = nd_pfn->uuid;
97 }
98
99 if (uuid2 && memcmp(uuid1, uuid2, NSLABEL_UUID_LEN) == 0)
100 return -EBUSY;
101
102 return 0;
103}
104
105static int is_namespace_uuid_busy(struct device *dev, void *data)
106{
107 if (is_nd_region(dev))
108 return device_for_each_child(dev, data, is_uuid_busy);
109 return 0;
110}
111
112/**
113 * nd_is_uuid_unique - verify that no other namespace has @uuid
114 * @dev: any device on a nvdimm_bus
115 * @uuid: uuid to check
116 */
117bool nd_is_uuid_unique(struct device *dev, u8 *uuid)
118{
119 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
120
121 if (!nvdimm_bus)
122 return false;
123 WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm_bus->dev));
124 if (device_for_each_child(&nvdimm_bus->dev, uuid,
125 is_namespace_uuid_busy) != 0)
126 return false;
127 return true;
128}
129
130bool pmem_should_map_pages(struct device *dev)
131{
132 struct nd_region *nd_region = to_nd_region(dev->parent);
133 struct nd_namespace_common *ndns = to_ndns(dev);
134 struct nd_namespace_io *nsio;
135
136 if (!IS_ENABLED(CONFIG_ZONE_DEVICE))
137 return false;
138
139 if (!test_bit(ND_REGION_PAGEMAP, &nd_region->flags))
140 return false;
141
142 if (is_nd_pfn(dev) || is_nd_btt(dev))
143 return false;
144
145 if (ndns->force_raw)
146 return false;
147
148 nsio = to_nd_namespace_io(dev);
149 if (region_intersects(nsio->res.start, resource_size(&nsio->res),
150 IORESOURCE_SYSTEM_RAM,
151 IORES_DESC_NONE) == REGION_MIXED)
152 return false;
153
154 return ARCH_MEMREMAP_PMEM == MEMREMAP_WB;
155}
156EXPORT_SYMBOL(pmem_should_map_pages);
157
158unsigned int pmem_sector_size(struct nd_namespace_common *ndns)
159{
160 if (is_namespace_pmem(&ndns->dev)) {
161 struct nd_namespace_pmem *nspm;
162
163 nspm = to_nd_namespace_pmem(&ndns->dev);
164 if (nspm->lbasize == 0 || nspm->lbasize == 512)
165 /* default */;
166 else if (nspm->lbasize == 4096)
167 return 4096;
168 else
169 dev_WARN(&ndns->dev, "unsupported sector size: %ld\n",
170 nspm->lbasize);
171 }
172
173 /*
174 * There is no namespace label (is_namespace_io()), or the label
175 * indicates the default sector size.
176 */
177 return 512;
178}
179EXPORT_SYMBOL(pmem_sector_size);
180
181const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,
182 char *name)
183{
184 struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
185 const char *suffix = NULL;
186
187 if (ndns->claim && is_nd_btt(ndns->claim))
188 suffix = "s";
189
190 if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev)) {
191 int nsidx = 0;
192
193 if (is_namespace_pmem(&ndns->dev)) {
194 struct nd_namespace_pmem *nspm;
195
196 nspm = to_nd_namespace_pmem(&ndns->dev);
197 nsidx = nspm->id;
198 }
199
200 if (nsidx)
201 sprintf(name, "pmem%d.%d%s", nd_region->id, nsidx,
202 suffix ? suffix : "");
203 else
204 sprintf(name, "pmem%d%s", nd_region->id,
205 suffix ? suffix : "");
206 } else if (is_namespace_blk(&ndns->dev)) {
207 struct nd_namespace_blk *nsblk;
208
209 nsblk = to_nd_namespace_blk(&ndns->dev);
210 sprintf(name, "ndblk%d.%d%s", nd_region->id, nsblk->id,
211 suffix ? suffix : "");
212 } else {
213 return NULL;
214 }
215
216 return name;
217}
218EXPORT_SYMBOL(nvdimm_namespace_disk_name);
219
220const u8 *nd_dev_to_uuid(struct device *dev)
221{
222 static const u8 null_uuid[16];
223
224 if (!dev)
225 return null_uuid;
226
227 if (is_namespace_pmem(dev)) {
228 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
229
230 return nspm->uuid;
231 } else if (is_namespace_blk(dev)) {
232 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
233
234 return nsblk->uuid;
235 } else
236 return null_uuid;
237}
238EXPORT_SYMBOL(nd_dev_to_uuid);
239
240static ssize_t nstype_show(struct device *dev,
241 struct device_attribute *attr, char *buf)
242{
243 struct nd_region *nd_region = to_nd_region(dev->parent);
244
245 return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
246}
247static DEVICE_ATTR_RO(nstype);
248
249static ssize_t __alt_name_store(struct device *dev, const char *buf,
250 const size_t len)
251{
252 char *input, *pos, *alt_name, **ns_altname;
253 ssize_t rc;
254
255 if (is_namespace_pmem(dev)) {
256 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
257
258 ns_altname = &nspm->alt_name;
259 } else if (is_namespace_blk(dev)) {
260 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
261
262 ns_altname = &nsblk->alt_name;
263 } else
264 return -ENXIO;
265
266 if (dev->driver || to_ndns(dev)->claim)
267 return -EBUSY;
268
269 input = kstrndup(buf, len, GFP_KERNEL);
270 if (!input)
271 return -ENOMEM;
272
273 pos = strim(input);
274 if (strlen(pos) + 1 > NSLABEL_NAME_LEN) {
275 rc = -EINVAL;
276 goto out;
277 }
278
279 alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL);
280 if (!alt_name) {
281 rc = -ENOMEM;
282 goto out;
283 }
284 kfree(*ns_altname);
285 *ns_altname = alt_name;
286 sprintf(*ns_altname, "%s", pos);
287 rc = len;
288
289out:
290 kfree(input);
291 return rc;
292}
293
294static resource_size_t nd_namespace_blk_size(struct nd_namespace_blk *nsblk)
295{
296 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
297 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
298 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
299 struct nd_label_id label_id;
300 resource_size_t size = 0;
301 struct resource *res;
302
303 if (!nsblk->uuid)
304 return 0;
305 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
306 for_each_dpa_resource(ndd, res)
307 if (strcmp(res->name, label_id.id) == 0)
308 size += resource_size(res);
309 return size;
310}
311
312static bool __nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
313{
314 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
315 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
316 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
317 struct nd_label_id label_id;
318 struct resource *res;
319 int count, i;
320
321 if (!nsblk->uuid || !nsblk->lbasize || !ndd)
322 return false;
323
324 count = 0;
325 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
326 for_each_dpa_resource(ndd, res) {
327 if (strcmp(res->name, label_id.id) != 0)
328 continue;
329 /*
330 * Resources with unacknowledged adjustments indicate a
331 * failure to update labels
332 */
333 if (res->flags & DPA_RESOURCE_ADJUSTED)
334 return false;
335 count++;
336 }
337
338 /* These values match after a successful label update */
339 if (count != nsblk->num_resources)
340 return false;
341
342 for (i = 0; i < nsblk->num_resources; i++) {
343 struct resource *found = NULL;
344
345 for_each_dpa_resource(ndd, res)
346 if (res == nsblk->res[i]) {
347 found = res;
348 break;
349 }
350 /* stale resource */
351 if (!found)
352 return false;
353 }
354
355 return true;
356}
357
358resource_size_t nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
359{
360 resource_size_t size;
361
362 nvdimm_bus_lock(&nsblk->common.dev);
363 size = __nd_namespace_blk_validate(nsblk);
364 nvdimm_bus_unlock(&nsblk->common.dev);
365
366 return size;
367}
368EXPORT_SYMBOL(nd_namespace_blk_validate);
369
370
371static int nd_namespace_label_update(struct nd_region *nd_region,
372 struct device *dev)
373{
374 dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim,
375 "namespace must be idle during label update\n");
376 if (dev->driver || to_ndns(dev)->claim)
377 return 0;
378
379 /*
380 * Only allow label writes that will result in a valid namespace
381 * or deletion of an existing namespace.
382 */
383 if (is_namespace_pmem(dev)) {
384 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
385 resource_size_t size = resource_size(&nspm->nsio.res);
386
387 if (size == 0 && nspm->uuid)
388 /* delete allocation */;
389 else if (!nspm->uuid)
390 return 0;
391
392 return nd_pmem_namespace_label_update(nd_region, nspm, size);
393 } else if (is_namespace_blk(dev)) {
394 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
395 resource_size_t size = nd_namespace_blk_size(nsblk);
396
397 if (size == 0 && nsblk->uuid)
398 /* delete allocation */;
399 else if (!nsblk->uuid || !nsblk->lbasize)
400 return 0;
401
402 return nd_blk_namespace_label_update(nd_region, nsblk, size);
403 } else
404 return -ENXIO;
405}
406
407static ssize_t alt_name_store(struct device *dev,
408 struct device_attribute *attr, const char *buf, size_t len)
409{
410 struct nd_region *nd_region = to_nd_region(dev->parent);
411 ssize_t rc;
412
413 nd_device_lock(dev);
414 nvdimm_bus_lock(dev);
415 wait_nvdimm_bus_probe_idle(dev);
416 rc = __alt_name_store(dev, buf, len);
417 if (rc >= 0)
418 rc = nd_namespace_label_update(nd_region, dev);
419 dev_dbg(dev, "%s(%zd)\n", rc < 0 ? "fail " : "", rc);
420 nvdimm_bus_unlock(dev);
421 nd_device_unlock(dev);
422
423 return rc < 0 ? rc : len;
424}
425
426static ssize_t alt_name_show(struct device *dev,
427 struct device_attribute *attr, char *buf)
428{
429 char *ns_altname;
430
431 if (is_namespace_pmem(dev)) {
432 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
433
434 ns_altname = nspm->alt_name;
435 } else if (is_namespace_blk(dev)) {
436 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
437
438 ns_altname = nsblk->alt_name;
439 } else
440 return -ENXIO;
441
442 return sprintf(buf, "%s\n", ns_altname ? ns_altname : "");
443}
444static DEVICE_ATTR_RW(alt_name);
445
446static int scan_free(struct nd_region *nd_region,
447 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
448 resource_size_t n)
449{
450 bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
451 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
452 int rc = 0;
453
454 while (n) {
455 struct resource *res, *last;
456 resource_size_t new_start;
457
458 last = NULL;
459 for_each_dpa_resource(ndd, res)
460 if (strcmp(res->name, label_id->id) == 0)
461 last = res;
462 res = last;
463 if (!res)
464 return 0;
465
466 if (n >= resource_size(res)) {
467 n -= resource_size(res);
468 nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc);
469 nvdimm_free_dpa(ndd, res);
470 /* retry with last resource deleted */
471 continue;
472 }
473
474 /*
475 * Keep BLK allocations relegated to high DPA as much as
476 * possible
477 */
478 if (is_blk)
479 new_start = res->start + n;
480 else
481 new_start = res->start;
482
483 rc = adjust_resource(res, new_start, resource_size(res) - n);
484 if (rc == 0)
485 res->flags |= DPA_RESOURCE_ADJUSTED;
486 nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc);
487 break;
488 }
489
490 return rc;
491}
492
493/**
494 * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
495 * @nd_region: the set of dimms to reclaim @n bytes from
496 * @label_id: unique identifier for the namespace consuming this dpa range
497 * @n: number of bytes per-dimm to release
498 *
499 * Assumes resources are ordered. Starting from the end try to
500 * adjust_resource() the allocation to @n, but if @n is larger than the
501 * allocation delete it and find the 'new' last allocation in the label
502 * set.
503 */
504static int shrink_dpa_allocation(struct nd_region *nd_region,
505 struct nd_label_id *label_id, resource_size_t n)
506{
507 int i;
508
509 for (i = 0; i < nd_region->ndr_mappings; i++) {
510 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
511 int rc;
512
513 rc = scan_free(nd_region, nd_mapping, label_id, n);
514 if (rc)
515 return rc;
516 }
517
518 return 0;
519}
520
521static resource_size_t init_dpa_allocation(struct nd_label_id *label_id,
522 struct nd_region *nd_region, struct nd_mapping *nd_mapping,
523 resource_size_t n)
524{
525 bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
526 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
527 resource_size_t first_dpa;
528 struct resource *res;
529 int rc = 0;
530
531 /* allocate blk from highest dpa first */
532 if (is_blk)
533 first_dpa = nd_mapping->start + nd_mapping->size - n;
534 else
535 first_dpa = nd_mapping->start;
536
537 /* first resource allocation for this label-id or dimm */
538 res = nvdimm_allocate_dpa(ndd, label_id, first_dpa, n);
539 if (!res)
540 rc = -EBUSY;
541
542 nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc);
543 return rc ? n : 0;
544}
545
546
547/**
548 * space_valid() - validate free dpa space against constraints
549 * @nd_region: hosting region of the free space
550 * @ndd: dimm device data for debug
551 * @label_id: namespace id to allocate space
552 * @prev: potential allocation that precedes free space
553 * @next: allocation that follows the given free space range
554 * @exist: first allocation with same id in the mapping
555 * @n: range that must satisfied for pmem allocations
556 * @valid: free space range to validate
557 *
558 * BLK-space is valid as long as it does not precede a PMEM
559 * allocation in a given region. PMEM-space must be contiguous
560 * and adjacent to an existing existing allocation (if one
561 * exists). If reserving PMEM any space is valid.
562 */
563static void space_valid(struct nd_region *nd_region, struct nvdimm_drvdata *ndd,
564 struct nd_label_id *label_id, struct resource *prev,
565 struct resource *next, struct resource *exist,
566 resource_size_t n, struct resource *valid)
567{
568 bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0;
569 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
570
571 if (valid->start >= valid->end)
572 goto invalid;
573
574 if (is_reserve)
575 return;
576
577 if (!is_pmem) {
578 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
579 struct nvdimm_bus *nvdimm_bus;
580 struct blk_alloc_info info = {
581 .nd_mapping = nd_mapping,
582 .available = nd_mapping->size,
583 .res = valid,
584 };
585
586 WARN_ON(!is_nd_blk(&nd_region->dev));
587 nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
588 device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
589 return;
590 }
591
592 /* allocation needs to be contiguous, so this is all or nothing */
593 if (resource_size(valid) < n)
594 goto invalid;
595
596 /* we've got all the space we need and no existing allocation */
597 if (!exist)
598 return;
599
600 /* allocation needs to be contiguous with the existing namespace */
601 if (valid->start == exist->end + 1
602 || valid->end == exist->start - 1)
603 return;
604
605 invalid:
606 /* truncate @valid size to 0 */
607 valid->end = valid->start - 1;
608}
609
610enum alloc_loc {
611 ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
612};
613
614static resource_size_t scan_allocate(struct nd_region *nd_region,
615 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
616 resource_size_t n)
617{
618 resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1;
619 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
620 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
621 struct resource *res, *exist = NULL, valid;
622 const resource_size_t to_allocate = n;
623 int first;
624
625 for_each_dpa_resource(ndd, res)
626 if (strcmp(label_id->id, res->name) == 0)
627 exist = res;
628
629 valid.start = nd_mapping->start;
630 valid.end = mapping_end;
631 valid.name = "free space";
632 retry:
633 first = 0;
634 for_each_dpa_resource(ndd, res) {
635 struct resource *next = res->sibling, *new_res = NULL;
636 resource_size_t allocate, available = 0;
637 enum alloc_loc loc = ALLOC_ERR;
638 const char *action;
639 int rc = 0;
640
641 /* ignore resources outside this nd_mapping */
642 if (res->start > mapping_end)
643 continue;
644 if (res->end < nd_mapping->start)
645 continue;
646
647 /* space at the beginning of the mapping */
648 if (!first++ && res->start > nd_mapping->start) {
649 valid.start = nd_mapping->start;
650 valid.end = res->start - 1;
651 space_valid(nd_region, ndd, label_id, NULL, next, exist,
652 to_allocate, &valid);
653 available = resource_size(&valid);
654 if (available)
655 loc = ALLOC_BEFORE;
656 }
657
658 /* space between allocations */
659 if (!loc && next) {
660 valid.start = res->start + resource_size(res);
661 valid.end = min(mapping_end, next->start - 1);
662 space_valid(nd_region, ndd, label_id, res, next, exist,
663 to_allocate, &valid);
664 available = resource_size(&valid);
665 if (available)
666 loc = ALLOC_MID;
667 }
668
669 /* space at the end of the mapping */
670 if (!loc && !next) {
671 valid.start = res->start + resource_size(res);
672 valid.end = mapping_end;
673 space_valid(nd_region, ndd, label_id, res, next, exist,
674 to_allocate, &valid);
675 available = resource_size(&valid);
676 if (available)
677 loc = ALLOC_AFTER;
678 }
679
680 if (!loc || !available)
681 continue;
682 allocate = min(available, n);
683 switch (loc) {
684 case ALLOC_BEFORE:
685 if (strcmp(res->name, label_id->id) == 0) {
686 /* adjust current resource up */
687 rc = adjust_resource(res, res->start - allocate,
688 resource_size(res) + allocate);
689 action = "cur grow up";
690 } else
691 action = "allocate";
692 break;
693 case ALLOC_MID:
694 if (strcmp(next->name, label_id->id) == 0) {
695 /* adjust next resource up */
696 rc = adjust_resource(next, next->start
697 - allocate, resource_size(next)
698 + allocate);
699 new_res = next;
700 action = "next grow up";
701 } else if (strcmp(res->name, label_id->id) == 0) {
702 action = "grow down";
703 } else
704 action = "allocate";
705 break;
706 case ALLOC_AFTER:
707 if (strcmp(res->name, label_id->id) == 0)
708 action = "grow down";
709 else
710 action = "allocate";
711 break;
712 default:
713 return n;
714 }
715
716 if (strcmp(action, "allocate") == 0) {
717 /* BLK allocate bottom up */
718 if (!is_pmem)
719 valid.start += available - allocate;
720
721 new_res = nvdimm_allocate_dpa(ndd, label_id,
722 valid.start, allocate);
723 if (!new_res)
724 rc = -EBUSY;
725 } else if (strcmp(action, "grow down") == 0) {
726 /* adjust current resource down */
727 rc = adjust_resource(res, res->start, resource_size(res)
728 + allocate);
729 if (rc == 0)
730 res->flags |= DPA_RESOURCE_ADJUSTED;
731 }
732
733 if (!new_res)
734 new_res = res;
735
736 nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n",
737 action, loc, rc);
738
739 if (rc)
740 return n;
741
742 n -= allocate;
743 if (n) {
744 /*
745 * Retry scan with newly inserted resources.
746 * For example, if we did an ALLOC_BEFORE
747 * insertion there may also have been space
748 * available for an ALLOC_AFTER insertion, so we
749 * need to check this same resource again
750 */
751 goto retry;
752 } else
753 return 0;
754 }
755
756 /*
757 * If we allocated nothing in the BLK case it may be because we are in
758 * an initial "pmem-reserve pass". Only do an initial BLK allocation
759 * when none of the DPA space is reserved.
760 */
761 if ((is_pmem || !ndd->dpa.child) && n == to_allocate)
762 return init_dpa_allocation(label_id, nd_region, nd_mapping, n);
763 return n;
764}
765
766static int merge_dpa(struct nd_region *nd_region,
767 struct nd_mapping *nd_mapping, struct nd_label_id *label_id)
768{
769 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
770 struct resource *res;
771
772 if (strncmp("pmem", label_id->id, 4) == 0)
773 return 0;
774 retry:
775 for_each_dpa_resource(ndd, res) {
776 int rc;
777 struct resource *next = res->sibling;
778 resource_size_t end = res->start + resource_size(res);
779
780 if (!next || strcmp(res->name, label_id->id) != 0
781 || strcmp(next->name, label_id->id) != 0
782 || end != next->start)
783 continue;
784 end += resource_size(next);
785 nvdimm_free_dpa(ndd, next);
786 rc = adjust_resource(res, res->start, end - res->start);
787 nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc);
788 if (rc)
789 return rc;
790 res->flags |= DPA_RESOURCE_ADJUSTED;
791 goto retry;
792 }
793
794 return 0;
795}
796
797int __reserve_free_pmem(struct device *dev, void *data)
798{
799 struct nvdimm *nvdimm = data;
800 struct nd_region *nd_region;
801 struct nd_label_id label_id;
802 int i;
803
804 if (!is_memory(dev))
805 return 0;
806
807 nd_region = to_nd_region(dev);
808 if (nd_region->ndr_mappings == 0)
809 return 0;
810
811 memset(&label_id, 0, sizeof(label_id));
812 strcat(label_id.id, "pmem-reserve");
813 for (i = 0; i < nd_region->ndr_mappings; i++) {
814 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
815 resource_size_t n, rem = 0;
816
817 if (nd_mapping->nvdimm != nvdimm)
818 continue;
819
820 n = nd_pmem_available_dpa(nd_region, nd_mapping, &rem);
821 if (n == 0)
822 return 0;
823 rem = scan_allocate(nd_region, nd_mapping, &label_id, n);
824 dev_WARN_ONCE(&nd_region->dev, rem,
825 "pmem reserve underrun: %#llx of %#llx bytes\n",
826 (unsigned long long) n - rem,
827 (unsigned long long) n);
828 return rem ? -ENXIO : 0;
829 }
830
831 return 0;
832}
833
834void release_free_pmem(struct nvdimm_bus *nvdimm_bus,
835 struct nd_mapping *nd_mapping)
836{
837 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
838 struct resource *res, *_res;
839
840 for_each_dpa_resource_safe(ndd, res, _res)
841 if (strcmp(res->name, "pmem-reserve") == 0)
842 nvdimm_free_dpa(ndd, res);
843}
844
845static int reserve_free_pmem(struct nvdimm_bus *nvdimm_bus,
846 struct nd_mapping *nd_mapping)
847{
848 struct nvdimm *nvdimm = nd_mapping->nvdimm;
849 int rc;
850
851 rc = device_for_each_child(&nvdimm_bus->dev, nvdimm,
852 __reserve_free_pmem);
853 if (rc)
854 release_free_pmem(nvdimm_bus, nd_mapping);
855 return rc;
856}
857
858/**
859 * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
860 * @nd_region: the set of dimms to allocate @n more bytes from
861 * @label_id: unique identifier for the namespace consuming this dpa range
862 * @n: number of bytes per-dimm to add to the existing allocation
863 *
864 * Assumes resources are ordered. For BLK regions, first consume
865 * BLK-only available DPA free space, then consume PMEM-aliased DPA
866 * space starting at the highest DPA. For PMEM regions start
867 * allocations from the start of an interleave set and end at the first
868 * BLK allocation or the end of the interleave set, whichever comes
869 * first.
870 */
871static int grow_dpa_allocation(struct nd_region *nd_region,
872 struct nd_label_id *label_id, resource_size_t n)
873{
874 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
875 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
876 int i;
877
878 for (i = 0; i < nd_region->ndr_mappings; i++) {
879 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
880 resource_size_t rem = n;
881 int rc, j;
882
883 /*
884 * In the BLK case try once with all unallocated PMEM
885 * reserved, and once without
886 */
887 for (j = is_pmem; j < 2; j++) {
888 bool blk_only = j == 0;
889
890 if (blk_only) {
891 rc = reserve_free_pmem(nvdimm_bus, nd_mapping);
892 if (rc)
893 return rc;
894 }
895 rem = scan_allocate(nd_region, nd_mapping,
896 label_id, rem);
897 if (blk_only)
898 release_free_pmem(nvdimm_bus, nd_mapping);
899
900 /* try again and allow encroachments into PMEM */
901 if (rem == 0)
902 break;
903 }
904
905 dev_WARN_ONCE(&nd_region->dev, rem,
906 "allocation underrun: %#llx of %#llx bytes\n",
907 (unsigned long long) n - rem,
908 (unsigned long long) n);
909 if (rem)
910 return -ENXIO;
911
912 rc = merge_dpa(nd_region, nd_mapping, label_id);
913 if (rc)
914 return rc;
915 }
916
917 return 0;
918}
919
920static void nd_namespace_pmem_set_resource(struct nd_region *nd_region,
921 struct nd_namespace_pmem *nspm, resource_size_t size)
922{
923 struct resource *res = &nspm->nsio.res;
924 resource_size_t offset = 0;
925
926 if (size && !nspm->uuid) {
927 WARN_ON_ONCE(1);
928 size = 0;
929 }
930
931 if (size && nspm->uuid) {
932 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
933 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
934 struct nd_label_id label_id;
935 struct resource *res;
936
937 if (!ndd) {
938 size = 0;
939 goto out;
940 }
941
942 nd_label_gen_id(&label_id, nspm->uuid, 0);
943
944 /* calculate a spa offset from the dpa allocation offset */
945 for_each_dpa_resource(ndd, res)
946 if (strcmp(res->name, label_id.id) == 0) {
947 offset = (res->start - nd_mapping->start)
948 * nd_region->ndr_mappings;
949 goto out;
950 }
951
952 WARN_ON_ONCE(1);
953 size = 0;
954 }
955
956 out:
957 res->start = nd_region->ndr_start + offset;
958 res->end = res->start + size - 1;
959}
960
961static bool uuid_not_set(const u8 *uuid, struct device *dev, const char *where)
962{
963 if (!uuid) {
964 dev_dbg(dev, "%s: uuid not set\n", where);
965 return true;
966 }
967 return false;
968}
969
970static ssize_t __size_store(struct device *dev, unsigned long long val)
971{
972 resource_size_t allocated = 0, available = 0;
973 struct nd_region *nd_region = to_nd_region(dev->parent);
974 struct nd_namespace_common *ndns = to_ndns(dev);
975 struct nd_mapping *nd_mapping;
976 struct nvdimm_drvdata *ndd;
977 struct nd_label_id label_id;
978 u32 flags = 0, remainder;
979 int rc, i, id = -1;
980 u8 *uuid = NULL;
981
982 if (dev->driver || ndns->claim)
983 return -EBUSY;
984
985 if (is_namespace_pmem(dev)) {
986 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
987
988 uuid = nspm->uuid;
989 id = nspm->id;
990 } else if (is_namespace_blk(dev)) {
991 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
992
993 uuid = nsblk->uuid;
994 flags = NSLABEL_FLAG_LOCAL;
995 id = nsblk->id;
996 }
997
998 /*
999 * We need a uuid for the allocation-label and dimm(s) on which
1000 * to store the label.
1001 */
1002 if (uuid_not_set(uuid, dev, __func__))
1003 return -ENXIO;
1004 if (nd_region->ndr_mappings == 0) {
1005 dev_dbg(dev, "not associated with dimm(s)\n");
1006 return -ENXIO;
1007 }
1008
1009 div_u64_rem(val, PAGE_SIZE * nd_region->ndr_mappings, &remainder);
1010 if (remainder) {
1011 dev_dbg(dev, "%llu is not %ldK aligned\n", val,
1012 (PAGE_SIZE * nd_region->ndr_mappings) / SZ_1K);
1013 return -EINVAL;
1014 }
1015
1016 nd_label_gen_id(&label_id, uuid, flags);
1017 for (i = 0; i < nd_region->ndr_mappings; i++) {
1018 nd_mapping = &nd_region->mapping[i];
1019 ndd = to_ndd(nd_mapping);
1020
1021 /*
1022 * All dimms in an interleave set, or the base dimm for a blk
1023 * region, need to be enabled for the size to be changed.
1024 */
1025 if (!ndd)
1026 return -ENXIO;
1027
1028 allocated += nvdimm_allocated_dpa(ndd, &label_id);
1029 }
1030 available = nd_region_allocatable_dpa(nd_region);
1031
1032 if (val > available + allocated)
1033 return -ENOSPC;
1034
1035 if (val == allocated)
1036 return 0;
1037
1038 val = div_u64(val, nd_region->ndr_mappings);
1039 allocated = div_u64(allocated, nd_region->ndr_mappings);
1040 if (val < allocated)
1041 rc = shrink_dpa_allocation(nd_region, &label_id,
1042 allocated - val);
1043 else
1044 rc = grow_dpa_allocation(nd_region, &label_id, val - allocated);
1045
1046 if (rc)
1047 return rc;
1048
1049 if (is_namespace_pmem(dev)) {
1050 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1051
1052 nd_namespace_pmem_set_resource(nd_region, nspm,
1053 val * nd_region->ndr_mappings);
1054 }
1055
1056 /*
1057 * Try to delete the namespace if we deleted all of its
1058 * allocation, this is not the seed or 0th device for the
1059 * region, and it is not actively claimed by a btt, pfn, or dax
1060 * instance.
1061 */
1062 if (val == 0 && id != 0 && nd_region->ns_seed != dev && !ndns->claim)
1063 nd_device_unregister(dev, ND_ASYNC);
1064
1065 return rc;
1066}
1067
1068static ssize_t size_store(struct device *dev,
1069 struct device_attribute *attr, const char *buf, size_t len)
1070{
1071 struct nd_region *nd_region = to_nd_region(dev->parent);
1072 unsigned long long val;
1073 u8 **uuid = NULL;
1074 int rc;
1075
1076 rc = kstrtoull(buf, 0, &val);
1077 if (rc)
1078 return rc;
1079
1080 nd_device_lock(dev);
1081 nvdimm_bus_lock(dev);
1082 wait_nvdimm_bus_probe_idle(dev);
1083 rc = __size_store(dev, val);
1084 if (rc >= 0)
1085 rc = nd_namespace_label_update(nd_region, dev);
1086
1087 if (is_namespace_pmem(dev)) {
1088 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1089
1090 uuid = &nspm->uuid;
1091 } else if (is_namespace_blk(dev)) {
1092 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1093
1094 uuid = &nsblk->uuid;
1095 }
1096
1097 if (rc == 0 && val == 0 && uuid) {
1098 /* setting size zero == 'delete namespace' */
1099 kfree(*uuid);
1100 *uuid = NULL;
1101 }
1102
1103 dev_dbg(dev, "%llx %s (%d)\n", val, rc < 0 ? "fail" : "success", rc);
1104
1105 nvdimm_bus_unlock(dev);
1106 nd_device_unlock(dev);
1107
1108 return rc < 0 ? rc : len;
1109}
1110
1111resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1112{
1113 struct device *dev = &ndns->dev;
1114
1115 if (is_namespace_pmem(dev)) {
1116 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1117
1118 return resource_size(&nspm->nsio.res);
1119 } else if (is_namespace_blk(dev)) {
1120 return nd_namespace_blk_size(to_nd_namespace_blk(dev));
1121 } else if (is_namespace_io(dev)) {
1122 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1123
1124 return resource_size(&nsio->res);
1125 } else
1126 WARN_ONCE(1, "unknown namespace type\n");
1127 return 0;
1128}
1129
1130resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1131{
1132 resource_size_t size;
1133
1134 nvdimm_bus_lock(&ndns->dev);
1135 size = __nvdimm_namespace_capacity(ndns);
1136 nvdimm_bus_unlock(&ndns->dev);
1137
1138 return size;
1139}
1140EXPORT_SYMBOL(nvdimm_namespace_capacity);
1141
1142bool nvdimm_namespace_locked(struct nd_namespace_common *ndns)
1143{
1144 int i;
1145 bool locked = false;
1146 struct device *dev = &ndns->dev;
1147 struct nd_region *nd_region = to_nd_region(dev->parent);
1148
1149 for (i = 0; i < nd_region->ndr_mappings; i++) {
1150 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1151 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1152
1153 if (test_bit(NDD_LOCKED, &nvdimm->flags)) {
1154 dev_dbg(dev, "%s locked\n", nvdimm_name(nvdimm));
1155 locked = true;
1156 }
1157 }
1158 return locked;
1159}
1160EXPORT_SYMBOL(nvdimm_namespace_locked);
1161
1162static ssize_t size_show(struct device *dev,
1163 struct device_attribute *attr, char *buf)
1164{
1165 return sprintf(buf, "%llu\n", (unsigned long long)
1166 nvdimm_namespace_capacity(to_ndns(dev)));
1167}
1168static DEVICE_ATTR(size, 0444, size_show, size_store);
1169
1170static u8 *namespace_to_uuid(struct device *dev)
1171{
1172 if (is_namespace_pmem(dev)) {
1173 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1174
1175 return nspm->uuid;
1176 } else if (is_namespace_blk(dev)) {
1177 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1178
1179 return nsblk->uuid;
1180 } else
1181 return ERR_PTR(-ENXIO);
1182}
1183
1184static ssize_t uuid_show(struct device *dev,
1185 struct device_attribute *attr, char *buf)
1186{
1187 u8 *uuid = namespace_to_uuid(dev);
1188
1189 if (IS_ERR(uuid))
1190 return PTR_ERR(uuid);
1191 if (uuid)
1192 return sprintf(buf, "%pUb\n", uuid);
1193 return sprintf(buf, "\n");
1194}
1195
1196/**
1197 * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
1198 * @nd_region: parent region so we can updates all dimms in the set
1199 * @dev: namespace type for generating label_id
1200 * @new_uuid: incoming uuid
1201 * @old_uuid: reference to the uuid storage location in the namespace object
1202 */
1203static int namespace_update_uuid(struct nd_region *nd_region,
1204 struct device *dev, u8 *new_uuid, u8 **old_uuid)
1205{
1206 u32 flags = is_namespace_blk(dev) ? NSLABEL_FLAG_LOCAL : 0;
1207 struct nd_label_id old_label_id;
1208 struct nd_label_id new_label_id;
1209 int i;
1210
1211 if (!nd_is_uuid_unique(dev, new_uuid))
1212 return -EINVAL;
1213
1214 if (*old_uuid == NULL)
1215 goto out;
1216
1217 /*
1218 * If we've already written a label with this uuid, then it's
1219 * too late to rename because we can't reliably update the uuid
1220 * without losing the old namespace. Userspace must delete this
1221 * namespace to abandon the old uuid.
1222 */
1223 for (i = 0; i < nd_region->ndr_mappings; i++) {
1224 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1225
1226 /*
1227 * This check by itself is sufficient because old_uuid
1228 * would be NULL above if this uuid did not exist in the
1229 * currently written set.
1230 *
1231 * FIXME: can we delete uuid with zero dpa allocated?
1232 */
1233 if (list_empty(&nd_mapping->labels))
1234 return -EBUSY;
1235 }
1236
1237 nd_label_gen_id(&old_label_id, *old_uuid, flags);
1238 nd_label_gen_id(&new_label_id, new_uuid, flags);
1239 for (i = 0; i < nd_region->ndr_mappings; i++) {
1240 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1241 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1242 struct nd_label_ent *label_ent;
1243 struct resource *res;
1244
1245 for_each_dpa_resource(ndd, res)
1246 if (strcmp(res->name, old_label_id.id) == 0)
1247 sprintf((void *) res->name, "%s",
1248 new_label_id.id);
1249
1250 mutex_lock(&nd_mapping->lock);
1251 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1252 struct nd_namespace_label *nd_label = label_ent->label;
1253 struct nd_label_id label_id;
1254
1255 if (!nd_label)
1256 continue;
1257 nd_label_gen_id(&label_id, nd_label->uuid,
1258 __le32_to_cpu(nd_label->flags));
1259 if (strcmp(old_label_id.id, label_id.id) == 0)
1260 set_bit(ND_LABEL_REAP, &label_ent->flags);
1261 }
1262 mutex_unlock(&nd_mapping->lock);
1263 }
1264 kfree(*old_uuid);
1265 out:
1266 *old_uuid = new_uuid;
1267 return 0;
1268}
1269
1270static ssize_t uuid_store(struct device *dev,
1271 struct device_attribute *attr, const char *buf, size_t len)
1272{
1273 struct nd_region *nd_region = to_nd_region(dev->parent);
1274 u8 *uuid = NULL;
1275 ssize_t rc = 0;
1276 u8 **ns_uuid;
1277
1278 if (is_namespace_pmem(dev)) {
1279 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1280
1281 ns_uuid = &nspm->uuid;
1282 } else if (is_namespace_blk(dev)) {
1283 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1284
1285 ns_uuid = &nsblk->uuid;
1286 } else
1287 return -ENXIO;
1288
1289 nd_device_lock(dev);
1290 nvdimm_bus_lock(dev);
1291 wait_nvdimm_bus_probe_idle(dev);
1292 if (to_ndns(dev)->claim)
1293 rc = -EBUSY;
1294 if (rc >= 0)
1295 rc = nd_uuid_store(dev, &uuid, buf, len);
1296 if (rc >= 0)
1297 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
1298 if (rc >= 0)
1299 rc = nd_namespace_label_update(nd_region, dev);
1300 else
1301 kfree(uuid);
1302 dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
1303 buf[len - 1] == '\n' ? "" : "\n");
1304 nvdimm_bus_unlock(dev);
1305 nd_device_unlock(dev);
1306
1307 return rc < 0 ? rc : len;
1308}
1309static DEVICE_ATTR_RW(uuid);
1310
1311static ssize_t resource_show(struct device *dev,
1312 struct device_attribute *attr, char *buf)
1313{
1314 struct resource *res;
1315
1316 if (is_namespace_pmem(dev)) {
1317 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1318
1319 res = &nspm->nsio.res;
1320 } else if (is_namespace_io(dev)) {
1321 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1322
1323 res = &nsio->res;
1324 } else
1325 return -ENXIO;
1326
1327 /* no address to convey if the namespace has no allocation */
1328 if (resource_size(res) == 0)
1329 return -ENXIO;
1330 return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
1331}
1332static DEVICE_ATTR_RO(resource);
1333
1334static const unsigned long blk_lbasize_supported[] = { 512, 520, 528,
1335 4096, 4104, 4160, 4224, 0 };
1336
1337static const unsigned long pmem_lbasize_supported[] = { 512, 4096, 0 };
1338
1339static ssize_t sector_size_show(struct device *dev,
1340 struct device_attribute *attr, char *buf)
1341{
1342 if (is_namespace_blk(dev)) {
1343 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1344
1345 return nd_size_select_show(nsblk->lbasize,
1346 blk_lbasize_supported, buf);
1347 }
1348
1349 if (is_namespace_pmem(dev)) {
1350 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1351
1352 return nd_size_select_show(nspm->lbasize,
1353 pmem_lbasize_supported, buf);
1354 }
1355 return -ENXIO;
1356}
1357
1358static ssize_t sector_size_store(struct device *dev,
1359 struct device_attribute *attr, const char *buf, size_t len)
1360{
1361 struct nd_region *nd_region = to_nd_region(dev->parent);
1362 const unsigned long *supported;
1363 unsigned long *lbasize;
1364 ssize_t rc = 0;
1365
1366 if (is_namespace_blk(dev)) {
1367 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1368
1369 lbasize = &nsblk->lbasize;
1370 supported = blk_lbasize_supported;
1371 } else if (is_namespace_pmem(dev)) {
1372 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1373
1374 lbasize = &nspm->lbasize;
1375 supported = pmem_lbasize_supported;
1376 } else
1377 return -ENXIO;
1378
1379 nd_device_lock(dev);
1380 nvdimm_bus_lock(dev);
1381 if (to_ndns(dev)->claim)
1382 rc = -EBUSY;
1383 if (rc >= 0)
1384 rc = nd_size_select_store(dev, buf, lbasize, supported);
1385 if (rc >= 0)
1386 rc = nd_namespace_label_update(nd_region, dev);
1387 dev_dbg(dev, "result: %zd %s: %s%s", rc, rc < 0 ? "tried" : "wrote",
1388 buf, buf[len - 1] == '\n' ? "" : "\n");
1389 nvdimm_bus_unlock(dev);
1390 nd_device_unlock(dev);
1391
1392 return rc ? rc : len;
1393}
1394static DEVICE_ATTR_RW(sector_size);
1395
1396static ssize_t dpa_extents_show(struct device *dev,
1397 struct device_attribute *attr, char *buf)
1398{
1399 struct nd_region *nd_region = to_nd_region(dev->parent);
1400 struct nd_label_id label_id;
1401 int count = 0, i;
1402 u8 *uuid = NULL;
1403 u32 flags = 0;
1404
1405 nvdimm_bus_lock(dev);
1406 if (is_namespace_pmem(dev)) {
1407 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1408
1409 uuid = nspm->uuid;
1410 flags = 0;
1411 } else if (is_namespace_blk(dev)) {
1412 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1413
1414 uuid = nsblk->uuid;
1415 flags = NSLABEL_FLAG_LOCAL;
1416 }
1417
1418 if (!uuid)
1419 goto out;
1420
1421 nd_label_gen_id(&label_id, uuid, flags);
1422 for (i = 0; i < nd_region->ndr_mappings; i++) {
1423 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1424 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1425 struct resource *res;
1426
1427 for_each_dpa_resource(ndd, res)
1428 if (strcmp(res->name, label_id.id) == 0)
1429 count++;
1430 }
1431 out:
1432 nvdimm_bus_unlock(dev);
1433
1434 return sprintf(buf, "%d\n", count);
1435}
1436static DEVICE_ATTR_RO(dpa_extents);
1437
1438static int btt_claim_class(struct device *dev)
1439{
1440 struct nd_region *nd_region = to_nd_region(dev->parent);
1441 int i, loop_bitmask = 0;
1442
1443 for (i = 0; i < nd_region->ndr_mappings; i++) {
1444 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1445 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1446 struct nd_namespace_index *nsindex;
1447
1448 /*
1449 * If any of the DIMMs do not support labels the only
1450 * possible BTT format is v1.
1451 */
1452 if (!ndd) {
1453 loop_bitmask = 0;
1454 break;
1455 }
1456
1457 nsindex = to_namespace_index(ndd, ndd->ns_current);
1458 if (nsindex == NULL)
1459 loop_bitmask |= 1;
1460 else {
1461 /* check whether existing labels are v1.1 or v1.2 */
1462 if (__le16_to_cpu(nsindex->major) == 1
1463 && __le16_to_cpu(nsindex->minor) == 1)
1464 loop_bitmask |= 2;
1465 else
1466 loop_bitmask |= 4;
1467 }
1468 }
1469 /*
1470 * If nsindex is null loop_bitmask's bit 0 will be set, and if an index
1471 * block is found, a v1.1 label for any mapping will set bit 1, and a
1472 * v1.2 label will set bit 2.
1473 *
1474 * At the end of the loop, at most one of the three bits must be set.
1475 * If multiple bits were set, it means the different mappings disagree
1476 * about their labels, and this must be cleaned up first.
1477 *
1478 * If all the label index blocks are found to agree, nsindex of NULL
1479 * implies labels haven't been initialized yet, and when they will,
1480 * they will be of the 1.2 format, so we can assume BTT2.0
1481 *
1482 * If 1.1 labels are found, we enforce BTT1.1, and if 1.2 labels are
1483 * found, we enforce BTT2.0
1484 *
1485 * If the loop was never entered, default to BTT1.1 (legacy namespaces)
1486 */
1487 switch (loop_bitmask) {
1488 case 0:
1489 case 2:
1490 return NVDIMM_CCLASS_BTT;
1491 case 1:
1492 case 4:
1493 return NVDIMM_CCLASS_BTT2;
1494 default:
1495 return -ENXIO;
1496 }
1497}
1498
1499static ssize_t holder_show(struct device *dev,
1500 struct device_attribute *attr, char *buf)
1501{
1502 struct nd_namespace_common *ndns = to_ndns(dev);
1503 ssize_t rc;
1504
1505 nd_device_lock(dev);
1506 rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
1507 nd_device_unlock(dev);
1508
1509 return rc;
1510}
1511static DEVICE_ATTR_RO(holder);
1512
1513static ssize_t __holder_class_store(struct device *dev, const char *buf)
1514{
1515 struct nd_namespace_common *ndns = to_ndns(dev);
1516
1517 if (dev->driver || ndns->claim)
1518 return -EBUSY;
1519
1520 if (sysfs_streq(buf, "btt"))
1521 ndns->claim_class = btt_claim_class(dev);
1522 else if (sysfs_streq(buf, "pfn"))
1523 ndns->claim_class = NVDIMM_CCLASS_PFN;
1524 else if (sysfs_streq(buf, "dax"))
1525 ndns->claim_class = NVDIMM_CCLASS_DAX;
1526 else if (sysfs_streq(buf, ""))
1527 ndns->claim_class = NVDIMM_CCLASS_NONE;
1528 else
1529 return -EINVAL;
1530
1531 /* btt_claim_class() could've returned an error */
1532 if (ndns->claim_class < 0)
1533 return ndns->claim_class;
1534
1535 return 0;
1536}
1537
1538static ssize_t holder_class_store(struct device *dev,
1539 struct device_attribute *attr, const char *buf, size_t len)
1540{
1541 struct nd_region *nd_region = to_nd_region(dev->parent);
1542 ssize_t rc;
1543
1544 nd_device_lock(dev);
1545 nvdimm_bus_lock(dev);
1546 wait_nvdimm_bus_probe_idle(dev);
1547 rc = __holder_class_store(dev, buf);
1548 if (rc >= 0)
1549 rc = nd_namespace_label_update(nd_region, dev);
1550 dev_dbg(dev, "%s(%zd)\n", rc < 0 ? "fail " : "", rc);
1551 nvdimm_bus_unlock(dev);
1552 nd_device_unlock(dev);
1553
1554 return rc < 0 ? rc : len;
1555}
1556
1557static ssize_t holder_class_show(struct device *dev,
1558 struct device_attribute *attr, char *buf)
1559{
1560 struct nd_namespace_common *ndns = to_ndns(dev);
1561 ssize_t rc;
1562
1563 nd_device_lock(dev);
1564 if (ndns->claim_class == NVDIMM_CCLASS_NONE)
1565 rc = sprintf(buf, "\n");
1566 else if ((ndns->claim_class == NVDIMM_CCLASS_BTT) ||
1567 (ndns->claim_class == NVDIMM_CCLASS_BTT2))
1568 rc = sprintf(buf, "btt\n");
1569 else if (ndns->claim_class == NVDIMM_CCLASS_PFN)
1570 rc = sprintf(buf, "pfn\n");
1571 else if (ndns->claim_class == NVDIMM_CCLASS_DAX)
1572 rc = sprintf(buf, "dax\n");
1573 else
1574 rc = sprintf(buf, "<unknown>\n");
1575 nd_device_unlock(dev);
1576
1577 return rc;
1578}
1579static DEVICE_ATTR_RW(holder_class);
1580
1581static ssize_t mode_show(struct device *dev,
1582 struct device_attribute *attr, char *buf)
1583{
1584 struct nd_namespace_common *ndns = to_ndns(dev);
1585 struct device *claim;
1586 char *mode;
1587 ssize_t rc;
1588
1589 nd_device_lock(dev);
1590 claim = ndns->claim;
1591 if (claim && is_nd_btt(claim))
1592 mode = "safe";
1593 else if (claim && is_nd_pfn(claim))
1594 mode = "memory";
1595 else if (claim && is_nd_dax(claim))
1596 mode = "dax";
1597 else if (!claim && pmem_should_map_pages(dev))
1598 mode = "memory";
1599 else
1600 mode = "raw";
1601 rc = sprintf(buf, "%s\n", mode);
1602 nd_device_unlock(dev);
1603
1604 return rc;
1605}
1606static DEVICE_ATTR_RO(mode);
1607
1608static ssize_t force_raw_store(struct device *dev,
1609 struct device_attribute *attr, const char *buf, size_t len)
1610{
1611 bool force_raw;
1612 int rc = strtobool(buf, &force_raw);
1613
1614 if (rc)
1615 return rc;
1616
1617 to_ndns(dev)->force_raw = force_raw;
1618 return len;
1619}
1620
1621static ssize_t force_raw_show(struct device *dev,
1622 struct device_attribute *attr, char *buf)
1623{
1624 return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
1625}
1626static DEVICE_ATTR_RW(force_raw);
1627
1628static struct attribute *nd_namespace_attributes[] = {
1629 &dev_attr_nstype.attr,
1630 &dev_attr_size.attr,
1631 &dev_attr_mode.attr,
1632 &dev_attr_uuid.attr,
1633 &dev_attr_holder.attr,
1634 &dev_attr_resource.attr,
1635 &dev_attr_alt_name.attr,
1636 &dev_attr_force_raw.attr,
1637 &dev_attr_sector_size.attr,
1638 &dev_attr_dpa_extents.attr,
1639 &dev_attr_holder_class.attr,
1640 NULL,
1641};
1642
1643static umode_t namespace_visible(struct kobject *kobj,
1644 struct attribute *a, int n)
1645{
1646 struct device *dev = container_of(kobj, struct device, kobj);
1647
1648 if (a == &dev_attr_resource.attr) {
1649 if (is_namespace_blk(dev))
1650 return 0;
1651 return 0400;
1652 }
1653
1654 if (is_namespace_pmem(dev) || is_namespace_blk(dev)) {
1655 if (a == &dev_attr_size.attr)
1656 return 0644;
1657
1658 return a->mode;
1659 }
1660
1661 if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr
1662 || a == &dev_attr_holder.attr
1663 || a == &dev_attr_holder_class.attr
1664 || a == &dev_attr_force_raw.attr
1665 || a == &dev_attr_mode.attr)
1666 return a->mode;
1667
1668 return 0;
1669}
1670
1671static struct attribute_group nd_namespace_attribute_group = {
1672 .attrs = nd_namespace_attributes,
1673 .is_visible = namespace_visible,
1674};
1675
1676static const struct attribute_group *nd_namespace_attribute_groups[] = {
1677 &nd_device_attribute_group,
1678 &nd_namespace_attribute_group,
1679 &nd_numa_attribute_group,
1680 NULL,
1681};
1682
1683struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
1684{
1685 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
1686 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
1687 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
1688 struct nd_namespace_common *ndns = NULL;
1689 resource_size_t size;
1690
1691 if (nd_btt || nd_pfn || nd_dax) {
1692 if (nd_btt)
1693 ndns = nd_btt->ndns;
1694 else if (nd_pfn)
1695 ndns = nd_pfn->ndns;
1696 else if (nd_dax)
1697 ndns = nd_dax->nd_pfn.ndns;
1698
1699 if (!ndns)
1700 return ERR_PTR(-ENODEV);
1701
1702 /*
1703 * Flush any in-progess probes / removals in the driver
1704 * for the raw personality of this namespace.
1705 */
1706 nd_device_lock(&ndns->dev);
1707 nd_device_unlock(&ndns->dev);
1708 if (ndns->dev.driver) {
1709 dev_dbg(&ndns->dev, "is active, can't bind %s\n",
1710 dev_name(dev));
1711 return ERR_PTR(-EBUSY);
1712 }
1713 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != dev,
1714 "host (%s) vs claim (%s) mismatch\n",
1715 dev_name(dev),
1716 dev_name(ndns->claim)))
1717 return ERR_PTR(-ENXIO);
1718 } else {
1719 ndns = to_ndns(dev);
1720 if (ndns->claim) {
1721 dev_dbg(dev, "claimed by %s, failing probe\n",
1722 dev_name(ndns->claim));
1723
1724 return ERR_PTR(-ENXIO);
1725 }
1726 }
1727
1728 if (nvdimm_namespace_locked(ndns))
1729 return ERR_PTR(-EACCES);
1730
1731 size = nvdimm_namespace_capacity(ndns);
1732 if (size < ND_MIN_NAMESPACE_SIZE) {
1733 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
1734 &size, ND_MIN_NAMESPACE_SIZE);
1735 return ERR_PTR(-ENODEV);
1736 }
1737
1738 if (is_namespace_pmem(&ndns->dev)) {
1739 struct nd_namespace_pmem *nspm;
1740
1741 nspm = to_nd_namespace_pmem(&ndns->dev);
1742 if (uuid_not_set(nspm->uuid, &ndns->dev, __func__))
1743 return ERR_PTR(-ENODEV);
1744 } else if (is_namespace_blk(&ndns->dev)) {
1745 struct nd_namespace_blk *nsblk;
1746
1747 nsblk = to_nd_namespace_blk(&ndns->dev);
1748 if (uuid_not_set(nsblk->uuid, &ndns->dev, __func__))
1749 return ERR_PTR(-ENODEV);
1750 if (!nsblk->lbasize) {
1751 dev_dbg(&ndns->dev, "sector size not set\n");
1752 return ERR_PTR(-ENODEV);
1753 }
1754 if (!nd_namespace_blk_validate(nsblk))
1755 return ERR_PTR(-ENODEV);
1756 }
1757
1758 return ndns;
1759}
1760EXPORT_SYMBOL(nvdimm_namespace_common_probe);
1761
1762static struct device **create_namespace_io(struct nd_region *nd_region)
1763{
1764 struct nd_namespace_io *nsio;
1765 struct device *dev, **devs;
1766 struct resource *res;
1767
1768 nsio = kzalloc(sizeof(*nsio), GFP_KERNEL);
1769 if (!nsio)
1770 return NULL;
1771
1772 devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1773 if (!devs) {
1774 kfree(nsio);
1775 return NULL;
1776 }
1777
1778 dev = &nsio->common.dev;
1779 dev->type = &namespace_io_device_type;
1780 dev->parent = &nd_region->dev;
1781 res = &nsio->res;
1782 res->name = dev_name(&nd_region->dev);
1783 res->flags = IORESOURCE_MEM;
1784 res->start = nd_region->ndr_start;
1785 res->end = res->start + nd_region->ndr_size - 1;
1786
1787 devs[0] = dev;
1788 return devs;
1789}
1790
1791static bool has_uuid_at_pos(struct nd_region *nd_region, u8 *uuid,
1792 u64 cookie, u16 pos)
1793{
1794 struct nd_namespace_label *found = NULL;
1795 int i;
1796
1797 for (i = 0; i < nd_region->ndr_mappings; i++) {
1798 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1799 struct nd_interleave_set *nd_set = nd_region->nd_set;
1800 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1801 struct nd_label_ent *label_ent;
1802 bool found_uuid = false;
1803
1804 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1805 struct nd_namespace_label *nd_label = label_ent->label;
1806 u16 position, nlabel;
1807 u64 isetcookie;
1808
1809 if (!nd_label)
1810 continue;
1811 isetcookie = __le64_to_cpu(nd_label->isetcookie);
1812 position = __le16_to_cpu(nd_label->position);
1813 nlabel = __le16_to_cpu(nd_label->nlabel);
1814
1815 if (isetcookie != cookie)
1816 continue;
1817
1818 if (memcmp(nd_label->uuid, uuid, NSLABEL_UUID_LEN) != 0)
1819 continue;
1820
1821 if (namespace_label_has(ndd, type_guid)
1822 && !guid_equal(&nd_set->type_guid,
1823 &nd_label->type_guid)) {
1824 dev_dbg(ndd->dev, "expect type_guid %pUb got %pUb\n",
1825 &nd_set->type_guid,
1826 &nd_label->type_guid);
1827 continue;
1828 }
1829
1830 if (found_uuid) {
1831 dev_dbg(ndd->dev, "duplicate entry for uuid\n");
1832 return false;
1833 }
1834 found_uuid = true;
1835 if (nlabel != nd_region->ndr_mappings)
1836 continue;
1837 if (position != pos)
1838 continue;
1839 found = nd_label;
1840 break;
1841 }
1842 if (found)
1843 break;
1844 }
1845 return found != NULL;
1846}
1847
1848static int select_pmem_id(struct nd_region *nd_region, u8 *pmem_id)
1849{
1850 int i;
1851
1852 if (!pmem_id)
1853 return -ENODEV;
1854
1855 for (i = 0; i < nd_region->ndr_mappings; i++) {
1856 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1857 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1858 struct nd_namespace_label *nd_label = NULL;
1859 u64 hw_start, hw_end, pmem_start, pmem_end;
1860 struct nd_label_ent *label_ent;
1861
1862 lockdep_assert_held(&nd_mapping->lock);
1863 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1864 nd_label = label_ent->label;
1865 if (!nd_label)
1866 continue;
1867 if (memcmp(nd_label->uuid, pmem_id, NSLABEL_UUID_LEN) == 0)
1868 break;
1869 nd_label = NULL;
1870 }
1871
1872 if (!nd_label) {
1873 WARN_ON(1);
1874 return -EINVAL;
1875 }
1876
1877 /*
1878 * Check that this label is compliant with the dpa
1879 * range published in NFIT
1880 */
1881 hw_start = nd_mapping->start;
1882 hw_end = hw_start + nd_mapping->size;
1883 pmem_start = __le64_to_cpu(nd_label->dpa);
1884 pmem_end = pmem_start + __le64_to_cpu(nd_label->rawsize);
1885 if (pmem_start >= hw_start && pmem_start < hw_end
1886 && pmem_end <= hw_end && pmem_end > hw_start)
1887 /* pass */;
1888 else {
1889 dev_dbg(&nd_region->dev, "%s invalid label for %pUb\n",
1890 dev_name(ndd->dev), nd_label->uuid);
1891 return -EINVAL;
1892 }
1893
1894 /* move recently validated label to the front of the list */
1895 list_move(&label_ent->list, &nd_mapping->labels);
1896 }
1897 return 0;
1898}
1899
1900/**
1901 * create_namespace_pmem - validate interleave set labelling, retrieve label0
1902 * @nd_region: region with mappings to validate
1903 * @nspm: target namespace to create
1904 * @nd_label: target pmem namespace label to evaluate
1905 */
1906static struct device *create_namespace_pmem(struct nd_region *nd_region,
1907 struct nd_namespace_index *nsindex,
1908 struct nd_namespace_label *nd_label)
1909{
1910 u64 cookie = nd_region_interleave_set_cookie(nd_region, nsindex);
1911 u64 altcookie = nd_region_interleave_set_altcookie(nd_region);
1912 struct nd_label_ent *label_ent;
1913 struct nd_namespace_pmem *nspm;
1914 struct nd_mapping *nd_mapping;
1915 resource_size_t size = 0;
1916 struct resource *res;
1917 struct device *dev;
1918 int rc = 0;
1919 u16 i;
1920
1921 if (cookie == 0) {
1922 dev_dbg(&nd_region->dev, "invalid interleave-set-cookie\n");
1923 return ERR_PTR(-ENXIO);
1924 }
1925
1926 if (__le64_to_cpu(nd_label->isetcookie) != cookie) {
1927 dev_dbg(&nd_region->dev, "invalid cookie in label: %pUb\n",
1928 nd_label->uuid);
1929 if (__le64_to_cpu(nd_label->isetcookie) != altcookie)
1930 return ERR_PTR(-EAGAIN);
1931
1932 dev_dbg(&nd_region->dev, "valid altcookie in label: %pUb\n",
1933 nd_label->uuid);
1934 }
1935
1936 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1937 if (!nspm)
1938 return ERR_PTR(-ENOMEM);
1939
1940 nspm->id = -1;
1941 dev = &nspm->nsio.common.dev;
1942 dev->type = &namespace_pmem_device_type;
1943 dev->parent = &nd_region->dev;
1944 res = &nspm->nsio.res;
1945 res->name = dev_name(&nd_region->dev);
1946 res->flags = IORESOURCE_MEM;
1947
1948 for (i = 0; i < nd_region->ndr_mappings; i++) {
1949 if (has_uuid_at_pos(nd_region, nd_label->uuid, cookie, i))
1950 continue;
1951 if (has_uuid_at_pos(nd_region, nd_label->uuid, altcookie, i))
1952 continue;
1953 break;
1954 }
1955
1956 if (i < nd_region->ndr_mappings) {
1957 struct nvdimm *nvdimm = nd_region->mapping[i].nvdimm;
1958
1959 /*
1960 * Give up if we don't find an instance of a uuid at each
1961 * position (from 0 to nd_region->ndr_mappings - 1), or if we
1962 * find a dimm with two instances of the same uuid.
1963 */
1964 dev_err(&nd_region->dev, "%s missing label for %pUb\n",
1965 nvdimm_name(nvdimm), nd_label->uuid);
1966 rc = -EINVAL;
1967 goto err;
1968 }
1969
1970 /*
1971 * Fix up each mapping's 'labels' to have the validated pmem label for
1972 * that position at labels[0], and NULL at labels[1]. In the process,
1973 * check that the namespace aligns with interleave-set. We know
1974 * that it does not overlap with any blk namespaces by virtue of
1975 * the dimm being enabled (i.e. nd_label_reserve_dpa()
1976 * succeeded).
1977 */
1978 rc = select_pmem_id(nd_region, nd_label->uuid);
1979 if (rc)
1980 goto err;
1981
1982 /* Calculate total size and populate namespace properties from label0 */
1983 for (i = 0; i < nd_region->ndr_mappings; i++) {
1984 struct nd_namespace_label *label0;
1985 struct nvdimm_drvdata *ndd;
1986
1987 nd_mapping = &nd_region->mapping[i];
1988 label_ent = list_first_entry_or_null(&nd_mapping->labels,
1989 typeof(*label_ent), list);
1990 label0 = label_ent ? label_ent->label : NULL;
1991
1992 if (!label0) {
1993 WARN_ON(1);
1994 continue;
1995 }
1996
1997 size += __le64_to_cpu(label0->rawsize);
1998 if (__le16_to_cpu(label0->position) != 0)
1999 continue;
2000 WARN_ON(nspm->alt_name || nspm->uuid);
2001 nspm->alt_name = kmemdup((void __force *) label0->name,
2002 NSLABEL_NAME_LEN, GFP_KERNEL);
2003 nspm->uuid = kmemdup((void __force *) label0->uuid,
2004 NSLABEL_UUID_LEN, GFP_KERNEL);
2005 nspm->lbasize = __le64_to_cpu(label0->lbasize);
2006 ndd = to_ndd(nd_mapping);
2007 if (namespace_label_has(ndd, abstraction_guid))
2008 nspm->nsio.common.claim_class
2009 = to_nvdimm_cclass(&label0->abstraction_guid);
2010
2011 }
2012
2013 if (!nspm->alt_name || !nspm->uuid) {
2014 rc = -ENOMEM;
2015 goto err;
2016 }
2017
2018 nd_namespace_pmem_set_resource(nd_region, nspm, size);
2019
2020 return dev;
2021 err:
2022 namespace_pmem_release(dev);
2023 switch (rc) {
2024 case -EINVAL:
2025 dev_dbg(&nd_region->dev, "invalid label(s)\n");
2026 break;
2027 case -ENODEV:
2028 dev_dbg(&nd_region->dev, "label not found\n");
2029 break;
2030 default:
2031 dev_dbg(&nd_region->dev, "unexpected err: %d\n", rc);
2032 break;
2033 }
2034 return ERR_PTR(rc);
2035}
2036
2037struct resource *nsblk_add_resource(struct nd_region *nd_region,
2038 struct nvdimm_drvdata *ndd, struct nd_namespace_blk *nsblk,
2039 resource_size_t start)
2040{
2041 struct nd_label_id label_id;
2042 struct resource *res;
2043
2044 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
2045 res = krealloc(nsblk->res,
2046 sizeof(void *) * (nsblk->num_resources + 1),
2047 GFP_KERNEL);
2048 if (!res)
2049 return NULL;
2050 nsblk->res = (struct resource **) res;
2051 for_each_dpa_resource(ndd, res)
2052 if (strcmp(res->name, label_id.id) == 0
2053 && res->start == start) {
2054 nsblk->res[nsblk->num_resources++] = res;
2055 return res;
2056 }
2057 return NULL;
2058}
2059
2060static struct device *nd_namespace_blk_create(struct nd_region *nd_region)
2061{
2062 struct nd_namespace_blk *nsblk;
2063 struct device *dev;
2064
2065 if (!is_nd_blk(&nd_region->dev))
2066 return NULL;
2067
2068 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2069 if (!nsblk)
2070 return NULL;
2071
2072 dev = &nsblk->common.dev;
2073 dev->type = &namespace_blk_device_type;
2074 nsblk->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
2075 if (nsblk->id < 0) {
2076 kfree(nsblk);
2077 return NULL;
2078 }
2079 dev_set_name(dev, "namespace%d.%d", nd_region->id, nsblk->id);
2080 dev->parent = &nd_region->dev;
2081 dev->groups = nd_namespace_attribute_groups;
2082
2083 return &nsblk->common.dev;
2084}
2085
2086static struct device *nd_namespace_pmem_create(struct nd_region *nd_region)
2087{
2088 struct nd_namespace_pmem *nspm;
2089 struct resource *res;
2090 struct device *dev;
2091
2092 if (!is_memory(&nd_region->dev))
2093 return NULL;
2094
2095 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2096 if (!nspm)
2097 return NULL;
2098
2099 dev = &nspm->nsio.common.dev;
2100 dev->type = &namespace_pmem_device_type;
2101 dev->parent = &nd_region->dev;
2102 res = &nspm->nsio.res;
2103 res->name = dev_name(&nd_region->dev);
2104 res->flags = IORESOURCE_MEM;
2105
2106 nspm->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
2107 if (nspm->id < 0) {
2108 kfree(nspm);
2109 return NULL;
2110 }
2111 dev_set_name(dev, "namespace%d.%d", nd_region->id, nspm->id);
2112 dev->groups = nd_namespace_attribute_groups;
2113 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2114
2115 return dev;
2116}
2117
2118void nd_region_create_ns_seed(struct nd_region *nd_region)
2119{
2120 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2121
2122 if (nd_region_to_nstype(nd_region) == ND_DEVICE_NAMESPACE_IO)
2123 return;
2124
2125 if (is_nd_blk(&nd_region->dev))
2126 nd_region->ns_seed = nd_namespace_blk_create(nd_region);
2127 else
2128 nd_region->ns_seed = nd_namespace_pmem_create(nd_region);
2129
2130 /*
2131 * Seed creation failures are not fatal, provisioning is simply
2132 * disabled until memory becomes available
2133 */
2134 if (!nd_region->ns_seed)
2135 dev_err(&nd_region->dev, "failed to create %s namespace\n",
2136 is_nd_blk(&nd_region->dev) ? "blk" : "pmem");
2137 else
2138 nd_device_register(nd_region->ns_seed);
2139}
2140
2141void nd_region_create_dax_seed(struct nd_region *nd_region)
2142{
2143 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2144 nd_region->dax_seed = nd_dax_create(nd_region);
2145 /*
2146 * Seed creation failures are not fatal, provisioning is simply
2147 * disabled until memory becomes available
2148 */
2149 if (!nd_region->dax_seed)
2150 dev_err(&nd_region->dev, "failed to create dax namespace\n");
2151}
2152
2153void nd_region_create_pfn_seed(struct nd_region *nd_region)
2154{
2155 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2156 nd_region->pfn_seed = nd_pfn_create(nd_region);
2157 /*
2158 * Seed creation failures are not fatal, provisioning is simply
2159 * disabled until memory becomes available
2160 */
2161 if (!nd_region->pfn_seed)
2162 dev_err(&nd_region->dev, "failed to create pfn namespace\n");
2163}
2164
2165void nd_region_create_btt_seed(struct nd_region *nd_region)
2166{
2167 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2168 nd_region->btt_seed = nd_btt_create(nd_region);
2169 /*
2170 * Seed creation failures are not fatal, provisioning is simply
2171 * disabled until memory becomes available
2172 */
2173 if (!nd_region->btt_seed)
2174 dev_err(&nd_region->dev, "failed to create btt namespace\n");
2175}
2176
2177static int add_namespace_resource(struct nd_region *nd_region,
2178 struct nd_namespace_label *nd_label, struct device **devs,
2179 int count)
2180{
2181 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2182 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2183 int i;
2184
2185 for (i = 0; i < count; i++) {
2186 u8 *uuid = namespace_to_uuid(devs[i]);
2187 struct resource *res;
2188
2189 if (IS_ERR_OR_NULL(uuid)) {
2190 WARN_ON(1);
2191 continue;
2192 }
2193
2194 if (memcmp(uuid, nd_label->uuid, NSLABEL_UUID_LEN) != 0)
2195 continue;
2196 if (is_namespace_blk(devs[i])) {
2197 res = nsblk_add_resource(nd_region, ndd,
2198 to_nd_namespace_blk(devs[i]),
2199 __le64_to_cpu(nd_label->dpa));
2200 if (!res)
2201 return -ENXIO;
2202 nd_dbg_dpa(nd_region, ndd, res, "%d assign\n", count);
2203 } else {
2204 dev_err(&nd_region->dev,
2205 "error: conflicting extents for uuid: %pUb\n",
2206 nd_label->uuid);
2207 return -ENXIO;
2208 }
2209 break;
2210 }
2211
2212 return i;
2213}
2214
2215static struct device *create_namespace_blk(struct nd_region *nd_region,
2216 struct nd_namespace_label *nd_label, int count)
2217{
2218
2219 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2220 struct nd_interleave_set *nd_set = nd_region->nd_set;
2221 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2222 struct nd_namespace_blk *nsblk;
2223 char name[NSLABEL_NAME_LEN];
2224 struct device *dev = NULL;
2225 struct resource *res;
2226
2227 if (namespace_label_has(ndd, type_guid)) {
2228 if (!guid_equal(&nd_set->type_guid, &nd_label->type_guid)) {
2229 dev_dbg(ndd->dev, "expect type_guid %pUb got %pUb\n",
2230 &nd_set->type_guid,
2231 &nd_label->type_guid);
2232 return ERR_PTR(-EAGAIN);
2233 }
2234
2235 if (nd_label->isetcookie != __cpu_to_le64(nd_set->cookie2)) {
2236 dev_dbg(ndd->dev, "expect cookie %#llx got %#llx\n",
2237 nd_set->cookie2,
2238 __le64_to_cpu(nd_label->isetcookie));
2239 return ERR_PTR(-EAGAIN);
2240 }
2241 }
2242
2243 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2244 if (!nsblk)
2245 return ERR_PTR(-ENOMEM);
2246 dev = &nsblk->common.dev;
2247 dev->type = &namespace_blk_device_type;
2248 dev->parent = &nd_region->dev;
2249 nsblk->id = -1;
2250 nsblk->lbasize = __le64_to_cpu(nd_label->lbasize);
2251 nsblk->uuid = kmemdup(nd_label->uuid, NSLABEL_UUID_LEN,
2252 GFP_KERNEL);
2253 if (namespace_label_has(ndd, abstraction_guid))
2254 nsblk->common.claim_class
2255 = to_nvdimm_cclass(&nd_label->abstraction_guid);
2256 if (!nsblk->uuid)
2257 goto blk_err;
2258 memcpy(name, nd_label->name, NSLABEL_NAME_LEN);
2259 if (name[0]) {
2260 nsblk->alt_name = kmemdup(name, NSLABEL_NAME_LEN,
2261 GFP_KERNEL);
2262 if (!nsblk->alt_name)
2263 goto blk_err;
2264 }
2265 res = nsblk_add_resource(nd_region, ndd, nsblk,
2266 __le64_to_cpu(nd_label->dpa));
2267 if (!res)
2268 goto blk_err;
2269 nd_dbg_dpa(nd_region, ndd, res, "%d: assign\n", count);
2270 return dev;
2271 blk_err:
2272 namespace_blk_release(dev);
2273 return ERR_PTR(-ENXIO);
2274}
2275
2276static int cmp_dpa(const void *a, const void *b)
2277{
2278 const struct device *dev_a = *(const struct device **) a;
2279 const struct device *dev_b = *(const struct device **) b;
2280 struct nd_namespace_blk *nsblk_a, *nsblk_b;
2281 struct nd_namespace_pmem *nspm_a, *nspm_b;
2282
2283 if (is_namespace_io(dev_a))
2284 return 0;
2285
2286 if (is_namespace_blk(dev_a)) {
2287 nsblk_a = to_nd_namespace_blk(dev_a);
2288 nsblk_b = to_nd_namespace_blk(dev_b);
2289
2290 return memcmp(&nsblk_a->res[0]->start, &nsblk_b->res[0]->start,
2291 sizeof(resource_size_t));
2292 }
2293
2294 nspm_a = to_nd_namespace_pmem(dev_a);
2295 nspm_b = to_nd_namespace_pmem(dev_b);
2296
2297 return memcmp(&nspm_a->nsio.res.start, &nspm_b->nsio.res.start,
2298 sizeof(resource_size_t));
2299}
2300
2301static struct device **scan_labels(struct nd_region *nd_region)
2302{
2303 int i, count = 0;
2304 struct device *dev, **devs = NULL;
2305 struct nd_label_ent *label_ent, *e;
2306 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2307 resource_size_t map_end = nd_mapping->start + nd_mapping->size - 1;
2308
2309 /* "safe" because create_namespace_pmem() might list_move() label_ent */
2310 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
2311 struct nd_namespace_label *nd_label = label_ent->label;
2312 struct device **__devs;
2313 u32 flags;
2314
2315 if (!nd_label)
2316 continue;
2317 flags = __le32_to_cpu(nd_label->flags);
2318 if (is_nd_blk(&nd_region->dev)
2319 == !!(flags & NSLABEL_FLAG_LOCAL))
2320 /* pass, region matches label type */;
2321 else
2322 continue;
2323
2324 /* skip labels that describe extents outside of the region */
2325 if (__le64_to_cpu(nd_label->dpa) < nd_mapping->start ||
2326 __le64_to_cpu(nd_label->dpa) > map_end)
2327 continue;
2328
2329 i = add_namespace_resource(nd_region, nd_label, devs, count);
2330 if (i < 0)
2331 goto err;
2332 if (i < count)
2333 continue;
2334 __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
2335 if (!__devs)
2336 goto err;
2337 memcpy(__devs, devs, sizeof(dev) * count);
2338 kfree(devs);
2339 devs = __devs;
2340
2341 if (is_nd_blk(&nd_region->dev))
2342 dev = create_namespace_blk(nd_region, nd_label, count);
2343 else {
2344 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2345 struct nd_namespace_index *nsindex;
2346
2347 nsindex = to_namespace_index(ndd, ndd->ns_current);
2348 dev = create_namespace_pmem(nd_region, nsindex, nd_label);
2349 }
2350
2351 if (IS_ERR(dev)) {
2352 switch (PTR_ERR(dev)) {
2353 case -EAGAIN:
2354 /* skip invalid labels */
2355 continue;
2356 case -ENODEV:
2357 /* fallthrough to seed creation */
2358 break;
2359 default:
2360 goto err;
2361 }
2362 } else
2363 devs[count++] = dev;
2364
2365 }
2366
2367 dev_dbg(&nd_region->dev, "discovered %d %s namespace%s\n",
2368 count, is_nd_blk(&nd_region->dev)
2369 ? "blk" : "pmem", count == 1 ? "" : "s");
2370
2371 if (count == 0) {
2372 /* Publish a zero-sized namespace for userspace to configure. */
2373 nd_mapping_free_labels(nd_mapping);
2374
2375 devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
2376 if (!devs)
2377 goto err;
2378 if (is_nd_blk(&nd_region->dev)) {
2379 struct nd_namespace_blk *nsblk;
2380
2381 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2382 if (!nsblk)
2383 goto err;
2384 dev = &nsblk->common.dev;
2385 dev->type = &namespace_blk_device_type;
2386 } else {
2387 struct nd_namespace_pmem *nspm;
2388
2389 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2390 if (!nspm)
2391 goto err;
2392 dev = &nspm->nsio.common.dev;
2393 dev->type = &namespace_pmem_device_type;
2394 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2395 }
2396 dev->parent = &nd_region->dev;
2397 devs[count++] = dev;
2398 } else if (is_memory(&nd_region->dev)) {
2399 /* clean unselected labels */
2400 for (i = 0; i < nd_region->ndr_mappings; i++) {
2401 struct list_head *l, *e;
2402 LIST_HEAD(list);
2403 int j;
2404
2405 nd_mapping = &nd_region->mapping[i];
2406 if (list_empty(&nd_mapping->labels)) {
2407 WARN_ON(1);
2408 continue;
2409 }
2410
2411 j = count;
2412 list_for_each_safe(l, e, &nd_mapping->labels) {
2413 if (!j--)
2414 break;
2415 list_move_tail(l, &list);
2416 }
2417 nd_mapping_free_labels(nd_mapping);
2418 list_splice_init(&list, &nd_mapping->labels);
2419 }
2420 }
2421
2422 if (count > 1)
2423 sort(devs, count, sizeof(struct device *), cmp_dpa, NULL);
2424
2425 return devs;
2426
2427 err:
2428 if (devs) {
2429 for (i = 0; devs[i]; i++)
2430 if (is_nd_blk(&nd_region->dev))
2431 namespace_blk_release(devs[i]);
2432 else
2433 namespace_pmem_release(devs[i]);
2434 kfree(devs);
2435 }
2436 return NULL;
2437}
2438
2439static struct device **create_namespaces(struct nd_region *nd_region)
2440{
2441 struct nd_mapping *nd_mapping;
2442 struct device **devs;
2443 int i;
2444
2445 if (nd_region->ndr_mappings == 0)
2446 return NULL;
2447
2448 /* lock down all mappings while we scan labels */
2449 for (i = 0; i < nd_region->ndr_mappings; i++) {
2450 nd_mapping = &nd_region->mapping[i];
2451 mutex_lock_nested(&nd_mapping->lock, i);
2452 }
2453
2454 devs = scan_labels(nd_region);
2455
2456 for (i = 0; i < nd_region->ndr_mappings; i++) {
2457 int reverse = nd_region->ndr_mappings - 1 - i;
2458
2459 nd_mapping = &nd_region->mapping[reverse];
2460 mutex_unlock(&nd_mapping->lock);
2461 }
2462
2463 return devs;
2464}
2465
2466static void deactivate_labels(void *region)
2467{
2468 struct nd_region *nd_region = region;
2469 int i;
2470
2471 for (i = 0; i < nd_region->ndr_mappings; i++) {
2472 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2473 struct nvdimm_drvdata *ndd = nd_mapping->ndd;
2474 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2475
2476 mutex_lock(&nd_mapping->lock);
2477 nd_mapping_free_labels(nd_mapping);
2478 mutex_unlock(&nd_mapping->lock);
2479
2480 put_ndd(ndd);
2481 nd_mapping->ndd = NULL;
2482 if (ndd)
2483 atomic_dec(&nvdimm->busy);
2484 }
2485}
2486
2487static int init_active_labels(struct nd_region *nd_region)
2488{
2489 int i;
2490
2491 for (i = 0; i < nd_region->ndr_mappings; i++) {
2492 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2493 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2494 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2495 struct nd_label_ent *label_ent;
2496 int count, j;
2497
2498 /*
2499 * If the dimm is disabled then we may need to prevent
2500 * the region from being activated.
2501 */
2502 if (!ndd) {
2503 if (test_bit(NDD_LOCKED, &nvdimm->flags))
2504 /* fail, label data may be unreadable */;
2505 else if (test_bit(NDD_ALIASING, &nvdimm->flags))
2506 /* fail, labels needed to disambiguate dpa */;
2507 else
2508 return 0;
2509
2510 dev_err(&nd_region->dev, "%s: is %s, failing probe\n",
2511 dev_name(&nd_mapping->nvdimm->dev),
2512 test_bit(NDD_LOCKED, &nvdimm->flags)
2513 ? "locked" : "disabled");
2514 return -ENXIO;
2515 }
2516 nd_mapping->ndd = ndd;
2517 atomic_inc(&nvdimm->busy);
2518 get_ndd(ndd);
2519
2520 count = nd_label_active_count(ndd);
2521 dev_dbg(ndd->dev, "count: %d\n", count);
2522 if (!count)
2523 continue;
2524 for (j = 0; j < count; j++) {
2525 struct nd_namespace_label *label;
2526
2527 label_ent = kzalloc(sizeof(*label_ent), GFP_KERNEL);
2528 if (!label_ent)
2529 break;
2530 label = nd_label_active(ndd, j);
2531 if (test_bit(NDD_NOBLK, &nvdimm->flags)) {
2532 u32 flags = __le32_to_cpu(label->flags);
2533
2534 flags &= ~NSLABEL_FLAG_LOCAL;
2535 label->flags = __cpu_to_le32(flags);
2536 }
2537 label_ent->label = label;
2538
2539 mutex_lock(&nd_mapping->lock);
2540 list_add_tail(&label_ent->list, &nd_mapping->labels);
2541 mutex_unlock(&nd_mapping->lock);
2542 }
2543
2544 if (j < count)
2545 break;
2546 }
2547
2548 if (i < nd_region->ndr_mappings) {
2549 deactivate_labels(nd_region);
2550 return -ENOMEM;
2551 }
2552
2553 return devm_add_action_or_reset(&nd_region->dev, deactivate_labels,
2554 nd_region);
2555}
2556
2557int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
2558{
2559 struct device **devs = NULL;
2560 int i, rc = 0, type;
2561
2562 *err = 0;
2563 nvdimm_bus_lock(&nd_region->dev);
2564 rc = init_active_labels(nd_region);
2565 if (rc) {
2566 nvdimm_bus_unlock(&nd_region->dev);
2567 return rc;
2568 }
2569
2570 type = nd_region_to_nstype(nd_region);
2571 switch (type) {
2572 case ND_DEVICE_NAMESPACE_IO:
2573 devs = create_namespace_io(nd_region);
2574 break;
2575 case ND_DEVICE_NAMESPACE_PMEM:
2576 case ND_DEVICE_NAMESPACE_BLK:
2577 devs = create_namespaces(nd_region);
2578 break;
2579 default:
2580 break;
2581 }
2582 nvdimm_bus_unlock(&nd_region->dev);
2583
2584 if (!devs)
2585 return -ENODEV;
2586
2587 for (i = 0; devs[i]; i++) {
2588 struct device *dev = devs[i];
2589 int id;
2590
2591 if (type == ND_DEVICE_NAMESPACE_BLK) {
2592 struct nd_namespace_blk *nsblk;
2593
2594 nsblk = to_nd_namespace_blk(dev);
2595 id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2596 GFP_KERNEL);
2597 nsblk->id = id;
2598 } else if (type == ND_DEVICE_NAMESPACE_PMEM) {
2599 struct nd_namespace_pmem *nspm;
2600
2601 nspm = to_nd_namespace_pmem(dev);
2602 id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2603 GFP_KERNEL);
2604 nspm->id = id;
2605 } else
2606 id = i;
2607
2608 if (id < 0)
2609 break;
2610 dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
2611 dev->groups = nd_namespace_attribute_groups;
2612 nd_device_register(dev);
2613 }
2614 if (i)
2615 nd_region->ns_seed = devs[0];
2616
2617 if (devs[i]) {
2618 int j;
2619
2620 for (j = i; devs[j]; j++) {
2621 struct device *dev = devs[j];
2622
2623 device_initialize(dev);
2624 put_device(dev);
2625 }
2626 *err = j - i;
2627 /*
2628 * All of the namespaces we tried to register failed, so
2629 * fail region activation.
2630 */
2631 if (*err == 0)
2632 rc = -ENODEV;
2633 }
2634 kfree(devs);
2635
2636 if (rc == -ENODEV)
2637 return rc;
2638
2639 return i;
2640}