Merge branch 'x86-pti-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[sfrench/cifs-2.6.git] / drivers / nvdimm / namespace_devs.c
1 /*
2  * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
3  *
4  * This program is free software; you can redistribute it and/or modify
5  * it under the terms of version 2 of the GNU General Public License as
6  * published by the Free Software Foundation.
7  *
8  * This program is distributed in the hope that it will be useful, but
9  * WITHOUT ANY WARRANTY; without even the implied warranty of
10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
11  * General Public License for more details.
12  */
13 #include <linux/module.h>
14 #include <linux/device.h>
15 #include <linux/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
23 static void namespace_io_release(struct device *dev)
24 {
25         struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
26
27         kfree(nsio);
28 }
29
30 static void namespace_pmem_release(struct device *dev)
31 {
32         struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
33         struct nd_region *nd_region = to_nd_region(dev->parent);
34
35         if (nspm->id >= 0)
36                 ida_simple_remove(&nd_region->ns_ida, nspm->id);
37         kfree(nspm->alt_name);
38         kfree(nspm->uuid);
39         kfree(nspm);
40 }
41
42 static void namespace_blk_release(struct device *dev)
43 {
44         struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
45         struct nd_region *nd_region = to_nd_region(dev->parent);
46
47         if (nsblk->id >= 0)
48                 ida_simple_remove(&nd_region->ns_ida, nsblk->id);
49         kfree(nsblk->alt_name);
50         kfree(nsblk->uuid);
51         kfree(nsblk->res);
52         kfree(nsblk);
53 }
54
55 static const struct device_type namespace_io_device_type = {
56         .name = "nd_namespace_io",
57         .release = namespace_io_release,
58 };
59
60 static const struct device_type namespace_pmem_device_type = {
61         .name = "nd_namespace_pmem",
62         .release = namespace_pmem_release,
63 };
64
65 static const struct device_type namespace_blk_device_type = {
66         .name = "nd_namespace_blk",
67         .release = namespace_blk_release,
68 };
69
70 static bool is_namespace_pmem(const struct device *dev)
71 {
72         return dev ? dev->type == &namespace_pmem_device_type : false;
73 }
74
75 static bool is_namespace_blk(const struct device *dev)
76 {
77         return dev ? dev->type == &namespace_blk_device_type : false;
78 }
79
80 static bool is_namespace_io(const struct device *dev)
81 {
82         return dev ? dev->type == &namespace_io_device_type : false;
83 }
84
85 static int is_uuid_busy(struct device *dev, void *data)
86 {
87         u8 *uuid1 = data, *uuid2 = NULL;
88
89         if (is_namespace_pmem(dev)) {
90                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
91
92                 uuid2 = nspm->uuid;
93         } else if (is_namespace_blk(dev)) {
94                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
95
96                 uuid2 = nsblk->uuid;
97         } else if (is_nd_btt(dev)) {
98                 struct nd_btt *nd_btt = to_nd_btt(dev);
99
100                 uuid2 = nd_btt->uuid;
101         } else if (is_nd_pfn(dev)) {
102                 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
103
104                 uuid2 = nd_pfn->uuid;
105         }
106
107         if (uuid2 && memcmp(uuid1, uuid2, NSLABEL_UUID_LEN) == 0)
108                 return -EBUSY;
109
110         return 0;
111 }
112
113 static int is_namespace_uuid_busy(struct device *dev, void *data)
114 {
115         if (is_nd_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  */
125 bool nd_is_uuid_unique(struct device *dev, u8 *uuid)
126 {
127         struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
128
129         if (!nvdimm_bus)
130                 return false;
131         WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm_bus->dev));
132         if (device_for_each_child(&nvdimm_bus->dev, uuid,
133                                 is_namespace_uuid_busy) != 0)
134                 return false;
135         return true;
136 }
137
138 bool pmem_should_map_pages(struct device *dev)
139 {
140         struct nd_region *nd_region = to_nd_region(dev->parent);
141         struct nd_namespace_io *nsio;
142
143         if (!IS_ENABLED(CONFIG_ZONE_DEVICE))
144                 return false;
145
146         if (!test_bit(ND_REGION_PAGEMAP, &nd_region->flags))
147                 return false;
148
149         if (is_nd_pfn(dev) || is_nd_btt(dev))
150                 return false;
151
152         nsio = to_nd_namespace_io(dev);
153         if (region_intersects(nsio->res.start, resource_size(&nsio->res),
154                                 IORESOURCE_SYSTEM_RAM,
155                                 IORES_DESC_NONE) == REGION_MIXED)
156                 return false;
157
158         return ARCH_MEMREMAP_PMEM == MEMREMAP_WB;
159 }
160 EXPORT_SYMBOL(pmem_should_map_pages);
161
162 unsigned 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 }
183 EXPORT_SYMBOL(pmem_sector_size);
184
185 const 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 }
222 EXPORT_SYMBOL(nvdimm_namespace_disk_name);
223
224 const 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 }
242 EXPORT_SYMBOL(nd_dev_to_uuid);
243
244 static 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 }
251 static DEVICE_ATTR_RO(nstype);
252
253 static 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
294 out:
295         kfree(input);
296         return rc;
297 }
298
299 static 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
317 static 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
363 resource_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 }
373 EXPORT_SYMBOL(nd_namespace_blk_validate);
374
375
376 static 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
412 static 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: %s(%zd)\n", __func__, rc < 0 ? "fail " : "", rc);
425         nvdimm_bus_unlock(dev);
426         device_unlock(dev);
427
428         return rc < 0 ? rc : len;
429 }
430
431 static 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 }
449 static DEVICE_ATTR_RW(alt_name);
450
451 static 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  */
509 static 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
526 static 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  */
568 static 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
615 enum alloc_loc {
616         ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
617 };
618
619 static 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
771 static 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
802 static 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
839 static 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
850 static 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  */
876 static 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
925 static 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
966 static 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
975 static 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, "%s: not associated with dimm(s)\n", __func__);
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
1073 static 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, "%s: %llx %s (%d)\n", __func__, val, rc < 0
1109                         ? "fail" : "success", rc);
1110
1111         nvdimm_bus_unlock(dev);
1112         device_unlock(dev);
1113
1114         return rc < 0 ? rc : len;
1115 }
1116
1117 resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1118 {
1119         struct device *dev = &ndns->dev;
1120
1121         if (is_namespace_pmem(dev)) {
1122                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1123
1124                 return resource_size(&nspm->nsio.res);
1125         } else if (is_namespace_blk(dev)) {
1126                 return nd_namespace_blk_size(to_nd_namespace_blk(dev));
1127         } else if (is_namespace_io(dev)) {
1128                 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1129
1130                 return resource_size(&nsio->res);
1131         } else
1132                 WARN_ONCE(1, "unknown namespace type\n");
1133         return 0;
1134 }
1135
1136 resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1137 {
1138         resource_size_t size;
1139
1140         nvdimm_bus_lock(&ndns->dev);
1141         size = __nvdimm_namespace_capacity(ndns);
1142         nvdimm_bus_unlock(&ndns->dev);
1143
1144         return size;
1145 }
1146 EXPORT_SYMBOL(nvdimm_namespace_capacity);
1147
1148 static ssize_t size_show(struct device *dev,
1149                 struct device_attribute *attr, char *buf)
1150 {
1151         return sprintf(buf, "%llu\n", (unsigned long long)
1152                         nvdimm_namespace_capacity(to_ndns(dev)));
1153 }
1154 static DEVICE_ATTR(size, 0444, size_show, size_store);
1155
1156 static u8 *namespace_to_uuid(struct device *dev)
1157 {
1158         if (is_namespace_pmem(dev)) {
1159                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1160
1161                 return nspm->uuid;
1162         } else if (is_namespace_blk(dev)) {
1163                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1164
1165                 return nsblk->uuid;
1166         } else
1167                 return ERR_PTR(-ENXIO);
1168 }
1169
1170 static ssize_t uuid_show(struct device *dev,
1171                 struct device_attribute *attr, char *buf)
1172 {
1173         u8 *uuid = namespace_to_uuid(dev);
1174
1175         if (IS_ERR(uuid))
1176                 return PTR_ERR(uuid);
1177         if (uuid)
1178                 return sprintf(buf, "%pUb\n", uuid);
1179         return sprintf(buf, "\n");
1180 }
1181
1182 /**
1183  * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
1184  * @nd_region: parent region so we can updates all dimms in the set
1185  * @dev: namespace type for generating label_id
1186  * @new_uuid: incoming uuid
1187  * @old_uuid: reference to the uuid storage location in the namespace object
1188  */
1189 static int namespace_update_uuid(struct nd_region *nd_region,
1190                 struct device *dev, u8 *new_uuid, u8 **old_uuid)
1191 {
1192         u32 flags = is_namespace_blk(dev) ? NSLABEL_FLAG_LOCAL : 0;
1193         struct nd_label_id old_label_id;
1194         struct nd_label_id new_label_id;
1195         int i;
1196
1197         if (!nd_is_uuid_unique(dev, new_uuid))
1198                 return -EINVAL;
1199
1200         if (*old_uuid == NULL)
1201                 goto out;
1202
1203         /*
1204          * If we've already written a label with this uuid, then it's
1205          * too late to rename because we can't reliably update the uuid
1206          * without losing the old namespace.  Userspace must delete this
1207          * namespace to abandon the old uuid.
1208          */
1209         for (i = 0; i < nd_region->ndr_mappings; i++) {
1210                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1211
1212                 /*
1213                  * This check by itself is sufficient because old_uuid
1214                  * would be NULL above if this uuid did not exist in the
1215                  * currently written set.
1216                  *
1217                  * FIXME: can we delete uuid with zero dpa allocated?
1218                  */
1219                 if (list_empty(&nd_mapping->labels))
1220                         return -EBUSY;
1221         }
1222
1223         nd_label_gen_id(&old_label_id, *old_uuid, flags);
1224         nd_label_gen_id(&new_label_id, new_uuid, flags);
1225         for (i = 0; i < nd_region->ndr_mappings; i++) {
1226                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1227                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1228                 struct resource *res;
1229
1230                 for_each_dpa_resource(ndd, res)
1231                         if (strcmp(res->name, old_label_id.id) == 0)
1232                                 sprintf((void *) res->name, "%s",
1233                                                 new_label_id.id);
1234         }
1235         kfree(*old_uuid);
1236  out:
1237         *old_uuid = new_uuid;
1238         return 0;
1239 }
1240
1241 static ssize_t uuid_store(struct device *dev,
1242                 struct device_attribute *attr, const char *buf, size_t len)
1243 {
1244         struct nd_region *nd_region = to_nd_region(dev->parent);
1245         u8 *uuid = NULL;
1246         ssize_t rc = 0;
1247         u8 **ns_uuid;
1248
1249         if (is_namespace_pmem(dev)) {
1250                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1251
1252                 ns_uuid = &nspm->uuid;
1253         } else if (is_namespace_blk(dev)) {
1254                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1255
1256                 ns_uuid = &nsblk->uuid;
1257         } else
1258                 return -ENXIO;
1259
1260         device_lock(dev);
1261         nvdimm_bus_lock(dev);
1262         wait_nvdimm_bus_probe_idle(dev);
1263         if (to_ndns(dev)->claim)
1264                 rc = -EBUSY;
1265         if (rc >= 0)
1266                 rc = nd_uuid_store(dev, &uuid, buf, len);
1267         if (rc >= 0)
1268                 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
1269         if (rc >= 0)
1270                 rc = nd_namespace_label_update(nd_region, dev);
1271         else
1272                 kfree(uuid);
1273         dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
1274                         rc, buf, buf[len - 1] == '\n' ? "" : "\n");
1275         nvdimm_bus_unlock(dev);
1276         device_unlock(dev);
1277
1278         return rc < 0 ? rc : len;
1279 }
1280 static DEVICE_ATTR_RW(uuid);
1281
1282 static ssize_t resource_show(struct device *dev,
1283                 struct device_attribute *attr, char *buf)
1284 {
1285         struct resource *res;
1286
1287         if (is_namespace_pmem(dev)) {
1288                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1289
1290                 res = &nspm->nsio.res;
1291         } else if (is_namespace_io(dev)) {
1292                 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1293
1294                 res = &nsio->res;
1295         } else
1296                 return -ENXIO;
1297
1298         /* no address to convey if the namespace has no allocation */
1299         if (resource_size(res) == 0)
1300                 return -ENXIO;
1301         return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
1302 }
1303 static DEVICE_ATTR_RO(resource);
1304
1305 static const unsigned long blk_lbasize_supported[] = { 512, 520, 528,
1306         4096, 4104, 4160, 4224, 0 };
1307
1308 static const unsigned long pmem_lbasize_supported[] = { 512, 4096, 0 };
1309
1310 static ssize_t sector_size_show(struct device *dev,
1311                 struct device_attribute *attr, char *buf)
1312 {
1313         if (is_namespace_blk(dev)) {
1314                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1315
1316                 return nd_size_select_show(nsblk->lbasize,
1317                                 blk_lbasize_supported, buf);
1318         }
1319
1320         if (is_namespace_pmem(dev)) {
1321                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1322
1323                 return nd_size_select_show(nspm->lbasize,
1324                                 pmem_lbasize_supported, buf);
1325         }
1326         return -ENXIO;
1327 }
1328
1329 static ssize_t sector_size_store(struct device *dev,
1330                 struct device_attribute *attr, const char *buf, size_t len)
1331 {
1332         struct nd_region *nd_region = to_nd_region(dev->parent);
1333         const unsigned long *supported;
1334         unsigned long *lbasize;
1335         ssize_t rc = 0;
1336
1337         if (is_namespace_blk(dev)) {
1338                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1339
1340                 lbasize = &nsblk->lbasize;
1341                 supported = blk_lbasize_supported;
1342         } else if (is_namespace_pmem(dev)) {
1343                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1344
1345                 lbasize = &nspm->lbasize;
1346                 supported = pmem_lbasize_supported;
1347         } else
1348                 return -ENXIO;
1349
1350         device_lock(dev);
1351         nvdimm_bus_lock(dev);
1352         if (to_ndns(dev)->claim)
1353                 rc = -EBUSY;
1354         if (rc >= 0)
1355                 rc = nd_size_select_store(dev, buf, lbasize, supported);
1356         if (rc >= 0)
1357                 rc = nd_namespace_label_update(nd_region, dev);
1358         dev_dbg(dev, "%s: result: %zd %s: %s%s", __func__,
1359                         rc, rc < 0 ? "tried" : "wrote", buf,
1360                         buf[len - 1] == '\n' ? "" : "\n");
1361         nvdimm_bus_unlock(dev);
1362         device_unlock(dev);
1363
1364         return rc ? rc : len;
1365 }
1366 static DEVICE_ATTR_RW(sector_size);
1367
1368 static ssize_t dpa_extents_show(struct device *dev,
1369                 struct device_attribute *attr, char *buf)
1370 {
1371         struct nd_region *nd_region = to_nd_region(dev->parent);
1372         struct nd_label_id label_id;
1373         int count = 0, i;
1374         u8 *uuid = NULL;
1375         u32 flags = 0;
1376
1377         nvdimm_bus_lock(dev);
1378         if (is_namespace_pmem(dev)) {
1379                 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1380
1381                 uuid = nspm->uuid;
1382                 flags = 0;
1383         } else if (is_namespace_blk(dev)) {
1384                 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1385
1386                 uuid = nsblk->uuid;
1387                 flags = NSLABEL_FLAG_LOCAL;
1388         }
1389
1390         if (!uuid)
1391                 goto out;
1392
1393         nd_label_gen_id(&label_id, uuid, flags);
1394         for (i = 0; i < nd_region->ndr_mappings; i++) {
1395                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1396                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1397                 struct resource *res;
1398
1399                 for_each_dpa_resource(ndd, res)
1400                         if (strcmp(res->name, label_id.id) == 0)
1401                                 count++;
1402         }
1403  out:
1404         nvdimm_bus_unlock(dev);
1405
1406         return sprintf(buf, "%d\n", count);
1407 }
1408 static DEVICE_ATTR_RO(dpa_extents);
1409
1410 static int btt_claim_class(struct device *dev)
1411 {
1412         struct nd_region *nd_region = to_nd_region(dev->parent);
1413         int i, loop_bitmask = 0;
1414
1415         for (i = 0; i < nd_region->ndr_mappings; i++) {
1416                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1417                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1418                 struct nd_namespace_index *nsindex;
1419
1420                 /*
1421                  * If any of the DIMMs do not support labels the only
1422                  * possible BTT format is v1.
1423                  */
1424                 if (!ndd) {
1425                         loop_bitmask = 0;
1426                         break;
1427                 }
1428
1429                 nsindex = to_namespace_index(ndd, ndd->ns_current);
1430                 if (nsindex == NULL)
1431                         loop_bitmask |= 1;
1432                 else {
1433                         /* check whether existing labels are v1.1 or v1.2 */
1434                         if (__le16_to_cpu(nsindex->major) == 1
1435                                         && __le16_to_cpu(nsindex->minor) == 1)
1436                                 loop_bitmask |= 2;
1437                         else
1438                                 loop_bitmask |= 4;
1439                 }
1440         }
1441         /*
1442          * If nsindex is null loop_bitmask's bit 0 will be set, and if an index
1443          * block is found, a v1.1 label for any mapping will set bit 1, and a
1444          * v1.2 label will set bit 2.
1445          *
1446          * At the end of the loop, at most one of the three bits must be set.
1447          * If multiple bits were set, it means the different mappings disagree
1448          * about their labels, and this must be cleaned up first.
1449          *
1450          * If all the label index blocks are found to agree, nsindex of NULL
1451          * implies labels haven't been initialized yet, and when they will,
1452          * they will be of the 1.2 format, so we can assume BTT2.0
1453          *
1454          * If 1.1 labels are found, we enforce BTT1.1, and if 1.2 labels are
1455          * found, we enforce BTT2.0
1456          *
1457          * If the loop was never entered, default to BTT1.1 (legacy namespaces)
1458          */
1459         switch (loop_bitmask) {
1460         case 0:
1461         case 2:
1462                 return NVDIMM_CCLASS_BTT;
1463         case 1:
1464         case 4:
1465                 return NVDIMM_CCLASS_BTT2;
1466         default:
1467                 return -ENXIO;
1468         }
1469 }
1470
1471 static ssize_t holder_show(struct device *dev,
1472                 struct device_attribute *attr, char *buf)
1473 {
1474         struct nd_namespace_common *ndns = to_ndns(dev);
1475         ssize_t rc;
1476
1477         device_lock(dev);
1478         rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
1479         device_unlock(dev);
1480
1481         return rc;
1482 }
1483 static DEVICE_ATTR_RO(holder);
1484
1485 static ssize_t __holder_class_store(struct device *dev, const char *buf)
1486 {
1487         struct nd_namespace_common *ndns = to_ndns(dev);
1488
1489         if (dev->driver || ndns->claim)
1490                 return -EBUSY;
1491
1492         if (strcmp(buf, "btt") == 0 || strcmp(buf, "btt\n") == 0)
1493                 ndns->claim_class = btt_claim_class(dev);
1494         else if (strcmp(buf, "pfn") == 0 || strcmp(buf, "pfn\n") == 0)
1495                 ndns->claim_class = NVDIMM_CCLASS_PFN;
1496         else if (strcmp(buf, "dax") == 0 || strcmp(buf, "dax\n") == 0)
1497                 ndns->claim_class = NVDIMM_CCLASS_DAX;
1498         else if (strcmp(buf, "") == 0 || strcmp(buf, "\n") == 0)
1499                 ndns->claim_class = NVDIMM_CCLASS_NONE;
1500         else
1501                 return -EINVAL;
1502
1503         /* btt_claim_class() could've returned an error */
1504         if (ndns->claim_class < 0)
1505                 return ndns->claim_class;
1506
1507         return 0;
1508 }
1509
1510 static ssize_t holder_class_store(struct device *dev,
1511                 struct device_attribute *attr, const char *buf, size_t len)
1512 {
1513         struct nd_region *nd_region = to_nd_region(dev->parent);
1514         ssize_t rc;
1515
1516         device_lock(dev);
1517         nvdimm_bus_lock(dev);
1518         wait_nvdimm_bus_probe_idle(dev);
1519         rc = __holder_class_store(dev, buf);
1520         if (rc >= 0)
1521                 rc = nd_namespace_label_update(nd_region, dev);
1522         dev_dbg(dev, "%s: %s(%zd)\n", __func__, rc < 0 ? "fail " : "", rc);
1523         nvdimm_bus_unlock(dev);
1524         device_unlock(dev);
1525
1526         return rc < 0 ? rc : len;
1527 }
1528
1529 static ssize_t holder_class_show(struct device *dev,
1530                 struct device_attribute *attr, char *buf)
1531 {
1532         struct nd_namespace_common *ndns = to_ndns(dev);
1533         ssize_t rc;
1534
1535         device_lock(dev);
1536         if (ndns->claim_class == NVDIMM_CCLASS_NONE)
1537                 rc = sprintf(buf, "\n");
1538         else if ((ndns->claim_class == NVDIMM_CCLASS_BTT) ||
1539                         (ndns->claim_class == NVDIMM_CCLASS_BTT2))
1540                 rc = sprintf(buf, "btt\n");
1541         else if (ndns->claim_class == NVDIMM_CCLASS_PFN)
1542                 rc = sprintf(buf, "pfn\n");
1543         else if (ndns->claim_class == NVDIMM_CCLASS_DAX)
1544                 rc = sprintf(buf, "dax\n");
1545         else
1546                 rc = sprintf(buf, "<unknown>\n");
1547         device_unlock(dev);
1548
1549         return rc;
1550 }
1551 static DEVICE_ATTR_RW(holder_class);
1552
1553 static ssize_t mode_show(struct device *dev,
1554                 struct device_attribute *attr, char *buf)
1555 {
1556         struct nd_namespace_common *ndns = to_ndns(dev);
1557         struct device *claim;
1558         char *mode;
1559         ssize_t rc;
1560
1561         device_lock(dev);
1562         claim = ndns->claim;
1563         if (claim && is_nd_btt(claim))
1564                 mode = "safe";
1565         else if (claim && is_nd_pfn(claim))
1566                 mode = "memory";
1567         else if (claim && is_nd_dax(claim))
1568                 mode = "dax";
1569         else if (!claim && pmem_should_map_pages(dev))
1570                 mode = "memory";
1571         else
1572                 mode = "raw";
1573         rc = sprintf(buf, "%s\n", mode);
1574         device_unlock(dev);
1575
1576         return rc;
1577 }
1578 static DEVICE_ATTR_RO(mode);
1579
1580 static ssize_t force_raw_store(struct device *dev,
1581                 struct device_attribute *attr, const char *buf, size_t len)
1582 {
1583         bool force_raw;
1584         int rc = strtobool(buf, &force_raw);
1585
1586         if (rc)
1587                 return rc;
1588
1589         to_ndns(dev)->force_raw = force_raw;
1590         return len;
1591 }
1592
1593 static ssize_t force_raw_show(struct device *dev,
1594                 struct device_attribute *attr, char *buf)
1595 {
1596         return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
1597 }
1598 static DEVICE_ATTR_RW(force_raw);
1599
1600 static struct attribute *nd_namespace_attributes[] = {
1601         &dev_attr_nstype.attr,
1602         &dev_attr_size.attr,
1603         &dev_attr_mode.attr,
1604         &dev_attr_uuid.attr,
1605         &dev_attr_holder.attr,
1606         &dev_attr_resource.attr,
1607         &dev_attr_alt_name.attr,
1608         &dev_attr_force_raw.attr,
1609         &dev_attr_sector_size.attr,
1610         &dev_attr_dpa_extents.attr,
1611         &dev_attr_holder_class.attr,
1612         NULL,
1613 };
1614
1615 static umode_t namespace_visible(struct kobject *kobj,
1616                 struct attribute *a, int n)
1617 {
1618         struct device *dev = container_of(kobj, struct device, kobj);
1619
1620         if (a == &dev_attr_resource.attr) {
1621                 if (is_namespace_blk(dev))
1622                         return 0;
1623                 return 0400;
1624         }
1625
1626         if (is_namespace_pmem(dev) || is_namespace_blk(dev)) {
1627                 if (a == &dev_attr_size.attr)
1628                         return 0644;
1629
1630                 return a->mode;
1631         }
1632
1633         if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr
1634                         || a == &dev_attr_holder.attr
1635                         || a == &dev_attr_holder_class.attr
1636                         || a == &dev_attr_force_raw.attr
1637                         || a == &dev_attr_mode.attr)
1638                 return a->mode;
1639
1640         return 0;
1641 }
1642
1643 static struct attribute_group nd_namespace_attribute_group = {
1644         .attrs = nd_namespace_attributes,
1645         .is_visible = namespace_visible,
1646 };
1647
1648 static const struct attribute_group *nd_namespace_attribute_groups[] = {
1649         &nd_device_attribute_group,
1650         &nd_namespace_attribute_group,
1651         &nd_numa_attribute_group,
1652         NULL,
1653 };
1654
1655 struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
1656 {
1657         struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
1658         struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
1659         struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
1660         struct nd_namespace_common *ndns = NULL;
1661         resource_size_t size;
1662
1663         if (nd_btt || nd_pfn || nd_dax) {
1664                 if (nd_btt)
1665                         ndns = nd_btt->ndns;
1666                 else if (nd_pfn)
1667                         ndns = nd_pfn->ndns;
1668                 else if (nd_dax)
1669                         ndns = nd_dax->nd_pfn.ndns;
1670
1671                 if (!ndns)
1672                         return ERR_PTR(-ENODEV);
1673
1674                 /*
1675                  * Flush any in-progess probes / removals in the driver
1676                  * for the raw personality of this namespace.
1677                  */
1678                 device_lock(&ndns->dev);
1679                 device_unlock(&ndns->dev);
1680                 if (ndns->dev.driver) {
1681                         dev_dbg(&ndns->dev, "is active, can't bind %s\n",
1682                                         dev_name(dev));
1683                         return ERR_PTR(-EBUSY);
1684                 }
1685                 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != dev,
1686                                         "host (%s) vs claim (%s) mismatch\n",
1687                                         dev_name(dev),
1688                                         dev_name(ndns->claim)))
1689                         return ERR_PTR(-ENXIO);
1690         } else {
1691                 ndns = to_ndns(dev);
1692                 if (ndns->claim) {
1693                         dev_dbg(dev, "claimed by %s, failing probe\n",
1694                                 dev_name(ndns->claim));
1695
1696                         return ERR_PTR(-ENXIO);
1697                 }
1698         }
1699
1700         size = nvdimm_namespace_capacity(ndns);
1701         if (size < ND_MIN_NAMESPACE_SIZE) {
1702                 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
1703                                 &size, ND_MIN_NAMESPACE_SIZE);
1704                 return ERR_PTR(-ENODEV);
1705         }
1706
1707         if (is_namespace_pmem(&ndns->dev)) {
1708                 struct nd_namespace_pmem *nspm;
1709
1710                 nspm = to_nd_namespace_pmem(&ndns->dev);
1711                 if (uuid_not_set(nspm->uuid, &ndns->dev, __func__))
1712                         return ERR_PTR(-ENODEV);
1713         } else if (is_namespace_blk(&ndns->dev)) {
1714                 struct nd_namespace_blk *nsblk;
1715
1716                 nsblk = to_nd_namespace_blk(&ndns->dev);
1717                 if (uuid_not_set(nsblk->uuid, &ndns->dev, __func__))
1718                         return ERR_PTR(-ENODEV);
1719                 if (!nsblk->lbasize) {
1720                         dev_dbg(&ndns->dev, "%s: sector size not set\n",
1721                                 __func__);
1722                         return ERR_PTR(-ENODEV);
1723                 }
1724                 if (!nd_namespace_blk_validate(nsblk))
1725                         return ERR_PTR(-ENODEV);
1726         }
1727
1728         return ndns;
1729 }
1730 EXPORT_SYMBOL(nvdimm_namespace_common_probe);
1731
1732 static struct device **create_namespace_io(struct nd_region *nd_region)
1733 {
1734         struct nd_namespace_io *nsio;
1735         struct device *dev, **devs;
1736         struct resource *res;
1737
1738         nsio = kzalloc(sizeof(*nsio), GFP_KERNEL);
1739         if (!nsio)
1740                 return NULL;
1741
1742         devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1743         if (!devs) {
1744                 kfree(nsio);
1745                 return NULL;
1746         }
1747
1748         dev = &nsio->common.dev;
1749         dev->type = &namespace_io_device_type;
1750         dev->parent = &nd_region->dev;
1751         res = &nsio->res;
1752         res->name = dev_name(&nd_region->dev);
1753         res->flags = IORESOURCE_MEM;
1754         res->start = nd_region->ndr_start;
1755         res->end = res->start + nd_region->ndr_size - 1;
1756
1757         devs[0] = dev;
1758         return devs;
1759 }
1760
1761 static bool has_uuid_at_pos(struct nd_region *nd_region, u8 *uuid,
1762                 u64 cookie, u16 pos)
1763 {
1764         struct nd_namespace_label *found = NULL;
1765         int i;
1766
1767         for (i = 0; i < nd_region->ndr_mappings; i++) {
1768                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1769                 struct nd_interleave_set *nd_set = nd_region->nd_set;
1770                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1771                 struct nd_label_ent *label_ent;
1772                 bool found_uuid = false;
1773
1774                 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1775                         struct nd_namespace_label *nd_label = label_ent->label;
1776                         u16 position, nlabel;
1777                         u64 isetcookie;
1778
1779                         if (!nd_label)
1780                                 continue;
1781                         isetcookie = __le64_to_cpu(nd_label->isetcookie);
1782                         position = __le16_to_cpu(nd_label->position);
1783                         nlabel = __le16_to_cpu(nd_label->nlabel);
1784
1785                         if (isetcookie != cookie)
1786                                 continue;
1787
1788                         if (memcmp(nd_label->uuid, uuid, NSLABEL_UUID_LEN) != 0)
1789                                 continue;
1790
1791                         if (namespace_label_has(ndd, type_guid)
1792                                         && !guid_equal(&nd_set->type_guid,
1793                                                 &nd_label->type_guid)) {
1794                                 dev_dbg(ndd->dev, "expect type_guid %pUb got %pUb\n",
1795                                                 nd_set->type_guid.b,
1796                                                 nd_label->type_guid.b);
1797                                 continue;
1798                         }
1799
1800                         if (found_uuid) {
1801                                 dev_dbg(ndd->dev,
1802                                                 "%s duplicate entry for uuid\n",
1803                                                 __func__);
1804                                 return false;
1805                         }
1806                         found_uuid = true;
1807                         if (nlabel != nd_region->ndr_mappings)
1808                                 continue;
1809                         if (position != pos)
1810                                 continue;
1811                         found = nd_label;
1812                         break;
1813                 }
1814                 if (found)
1815                         break;
1816         }
1817         return found != NULL;
1818 }
1819
1820 static int select_pmem_id(struct nd_region *nd_region, u8 *pmem_id)
1821 {
1822         int i;
1823
1824         if (!pmem_id)
1825                 return -ENODEV;
1826
1827         for (i = 0; i < nd_region->ndr_mappings; i++) {
1828                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1829                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1830                 struct nd_namespace_label *nd_label = NULL;
1831                 u64 hw_start, hw_end, pmem_start, pmem_end;
1832                 struct nd_label_ent *label_ent;
1833
1834                 lockdep_assert_held(&nd_mapping->lock);
1835                 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1836                         nd_label = label_ent->label;
1837                         if (!nd_label)
1838                                 continue;
1839                         if (memcmp(nd_label->uuid, pmem_id, NSLABEL_UUID_LEN) == 0)
1840                                 break;
1841                         nd_label = NULL;
1842                 }
1843
1844                 if (!nd_label) {
1845                         WARN_ON(1);
1846                         return -EINVAL;
1847                 }
1848
1849                 /*
1850                  * Check that this label is compliant with the dpa
1851                  * range published in NFIT
1852                  */
1853                 hw_start = nd_mapping->start;
1854                 hw_end = hw_start + nd_mapping->size;
1855                 pmem_start = __le64_to_cpu(nd_label->dpa);
1856                 pmem_end = pmem_start + __le64_to_cpu(nd_label->rawsize);
1857                 if (pmem_start >= hw_start && pmem_start < hw_end
1858                                 && pmem_end <= hw_end && pmem_end > hw_start)
1859                         /* pass */;
1860                 else {
1861                         dev_dbg(&nd_region->dev, "%s invalid label for %pUb\n",
1862                                         dev_name(ndd->dev), nd_label->uuid);
1863                         return -EINVAL;
1864                 }
1865
1866                 /* move recently validated label to the front of the list */
1867                 list_move(&label_ent->list, &nd_mapping->labels);
1868         }
1869         return 0;
1870 }
1871
1872 /**
1873  * create_namespace_pmem - validate interleave set labelling, retrieve label0
1874  * @nd_region: region with mappings to validate
1875  * @nspm: target namespace to create
1876  * @nd_label: target pmem namespace label to evaluate
1877  */
1878 static struct device *create_namespace_pmem(struct nd_region *nd_region,
1879                 struct nd_namespace_index *nsindex,
1880                 struct nd_namespace_label *nd_label)
1881 {
1882         u64 cookie = nd_region_interleave_set_cookie(nd_region, nsindex);
1883         u64 altcookie = nd_region_interleave_set_altcookie(nd_region);
1884         struct nd_label_ent *label_ent;
1885         struct nd_namespace_pmem *nspm;
1886         struct nd_mapping *nd_mapping;
1887         resource_size_t size = 0;
1888         struct resource *res;
1889         struct device *dev;
1890         int rc = 0;
1891         u16 i;
1892
1893         if (cookie == 0) {
1894                 dev_dbg(&nd_region->dev, "invalid interleave-set-cookie\n");
1895                 return ERR_PTR(-ENXIO);
1896         }
1897
1898         if (__le64_to_cpu(nd_label->isetcookie) != cookie) {
1899                 dev_dbg(&nd_region->dev, "invalid cookie in label: %pUb\n",
1900                                 nd_label->uuid);
1901                 if (__le64_to_cpu(nd_label->isetcookie) != altcookie)
1902                         return ERR_PTR(-EAGAIN);
1903
1904                 dev_dbg(&nd_region->dev, "valid altcookie in label: %pUb\n",
1905                                 nd_label->uuid);
1906         }
1907
1908         nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1909         if (!nspm)
1910                 return ERR_PTR(-ENOMEM);
1911
1912         nspm->id = -1;
1913         dev = &nspm->nsio.common.dev;
1914         dev->type = &namespace_pmem_device_type;
1915         dev->parent = &nd_region->dev;
1916         res = &nspm->nsio.res;
1917         res->name = dev_name(&nd_region->dev);
1918         res->flags = IORESOURCE_MEM;
1919
1920         for (i = 0; i < nd_region->ndr_mappings; i++) {
1921                 if (has_uuid_at_pos(nd_region, nd_label->uuid, cookie, i))
1922                         continue;
1923                 if (has_uuid_at_pos(nd_region, nd_label->uuid, altcookie, i))
1924                         continue;
1925                 break;
1926         }
1927
1928         if (i < nd_region->ndr_mappings) {
1929                 struct nvdimm_drvdata *ndd = to_ndd(&nd_region->mapping[i]);
1930
1931                 /*
1932                  * Give up if we don't find an instance of a uuid at each
1933                  * position (from 0 to nd_region->ndr_mappings - 1), or if we
1934                  * find a dimm with two instances of the same uuid.
1935                  */
1936                 dev_err(&nd_region->dev, "%s missing label for %pUb\n",
1937                                 dev_name(ndd->dev), nd_label->uuid);
1938                 rc = -EINVAL;
1939                 goto err;
1940         }
1941
1942         /*
1943          * Fix up each mapping's 'labels' to have the validated pmem label for
1944          * that position at labels[0], and NULL at labels[1].  In the process,
1945          * check that the namespace aligns with interleave-set.  We know
1946          * that it does not overlap with any blk namespaces by virtue of
1947          * the dimm being enabled (i.e. nd_label_reserve_dpa()
1948          * succeeded).
1949          */
1950         rc = select_pmem_id(nd_region, nd_label->uuid);
1951         if (rc)
1952                 goto err;
1953
1954         /* Calculate total size and populate namespace properties from label0 */
1955         for (i = 0; i < nd_region->ndr_mappings; i++) {
1956                 struct nd_namespace_label *label0;
1957                 struct nvdimm_drvdata *ndd;
1958
1959                 nd_mapping = &nd_region->mapping[i];
1960                 label_ent = list_first_entry_or_null(&nd_mapping->labels,
1961                                 typeof(*label_ent), list);
1962                 label0 = label_ent ? label_ent->label : 0;
1963
1964                 if (!label0) {
1965                         WARN_ON(1);
1966                         continue;
1967                 }
1968
1969                 size += __le64_to_cpu(label0->rawsize);
1970                 if (__le16_to_cpu(label0->position) != 0)
1971                         continue;
1972                 WARN_ON(nspm->alt_name || nspm->uuid);
1973                 nspm->alt_name = kmemdup((void __force *) label0->name,
1974                                 NSLABEL_NAME_LEN, GFP_KERNEL);
1975                 nspm->uuid = kmemdup((void __force *) label0->uuid,
1976                                 NSLABEL_UUID_LEN, GFP_KERNEL);
1977                 nspm->lbasize = __le64_to_cpu(label0->lbasize);
1978                 ndd = to_ndd(nd_mapping);
1979                 if (namespace_label_has(ndd, abstraction_guid))
1980                         nspm->nsio.common.claim_class
1981                                 = to_nvdimm_cclass(&label0->abstraction_guid);
1982
1983         }
1984
1985         if (!nspm->alt_name || !nspm->uuid) {
1986                 rc = -ENOMEM;
1987                 goto err;
1988         }
1989
1990         nd_namespace_pmem_set_resource(nd_region, nspm, size);
1991
1992         return dev;
1993  err:
1994         namespace_pmem_release(dev);
1995         switch (rc) {
1996         case -EINVAL:
1997                 dev_dbg(&nd_region->dev, "%s: invalid label(s)\n", __func__);
1998                 break;
1999         case -ENODEV:
2000                 dev_dbg(&nd_region->dev, "%s: label not found\n", __func__);
2001                 break;
2002         default:
2003                 dev_dbg(&nd_region->dev, "%s: unexpected err: %d\n",
2004                                 __func__, rc);
2005                 break;
2006         }
2007         return ERR_PTR(rc);
2008 }
2009
2010 struct resource *nsblk_add_resource(struct nd_region *nd_region,
2011                 struct nvdimm_drvdata *ndd, struct nd_namespace_blk *nsblk,
2012                 resource_size_t start)
2013 {
2014         struct nd_label_id label_id;
2015         struct resource *res;
2016
2017         nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
2018         res = krealloc(nsblk->res,
2019                         sizeof(void *) * (nsblk->num_resources + 1),
2020                         GFP_KERNEL);
2021         if (!res)
2022                 return NULL;
2023         nsblk->res = (struct resource **) res;
2024         for_each_dpa_resource(ndd, res)
2025                 if (strcmp(res->name, label_id.id) == 0
2026                                 && res->start == start) {
2027                         nsblk->res[nsblk->num_resources++] = res;
2028                         return res;
2029                 }
2030         return NULL;
2031 }
2032
2033 static struct device *nd_namespace_blk_create(struct nd_region *nd_region)
2034 {
2035         struct nd_namespace_blk *nsblk;
2036         struct device *dev;
2037
2038         if (!is_nd_blk(&nd_region->dev))
2039                 return NULL;
2040
2041         nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2042         if (!nsblk)
2043                 return NULL;
2044
2045         dev = &nsblk->common.dev;
2046         dev->type = &namespace_blk_device_type;
2047         nsblk->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
2048         if (nsblk->id < 0) {
2049                 kfree(nsblk);
2050                 return NULL;
2051         }
2052         dev_set_name(dev, "namespace%d.%d", nd_region->id, nsblk->id);
2053         dev->parent = &nd_region->dev;
2054         dev->groups = nd_namespace_attribute_groups;
2055
2056         return &nsblk->common.dev;
2057 }
2058
2059 static struct device *nd_namespace_pmem_create(struct nd_region *nd_region)
2060 {
2061         struct nd_namespace_pmem *nspm;
2062         struct resource *res;
2063         struct device *dev;
2064
2065         if (!is_memory(&nd_region->dev))
2066                 return NULL;
2067
2068         nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2069         if (!nspm)
2070                 return NULL;
2071
2072         dev = &nspm->nsio.common.dev;
2073         dev->type = &namespace_pmem_device_type;
2074         dev->parent = &nd_region->dev;
2075         res = &nspm->nsio.res;
2076         res->name = dev_name(&nd_region->dev);
2077         res->flags = IORESOURCE_MEM;
2078
2079         nspm->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
2080         if (nspm->id < 0) {
2081                 kfree(nspm);
2082                 return NULL;
2083         }
2084         dev_set_name(dev, "namespace%d.%d", nd_region->id, nspm->id);
2085         dev->parent = &nd_region->dev;
2086         dev->groups = nd_namespace_attribute_groups;
2087         nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2088
2089         return dev;
2090 }
2091
2092 void nd_region_create_ns_seed(struct nd_region *nd_region)
2093 {
2094         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2095
2096         if (nd_region_to_nstype(nd_region) == ND_DEVICE_NAMESPACE_IO)
2097                 return;
2098
2099         if (is_nd_blk(&nd_region->dev))
2100                 nd_region->ns_seed = nd_namespace_blk_create(nd_region);
2101         else
2102                 nd_region->ns_seed = nd_namespace_pmem_create(nd_region);
2103
2104         /*
2105          * Seed creation failures are not fatal, provisioning is simply
2106          * disabled until memory becomes available
2107          */
2108         if (!nd_region->ns_seed)
2109                 dev_err(&nd_region->dev, "failed to create %s namespace\n",
2110                                 is_nd_blk(&nd_region->dev) ? "blk" : "pmem");
2111         else
2112                 nd_device_register(nd_region->ns_seed);
2113 }
2114
2115 void nd_region_create_dax_seed(struct nd_region *nd_region)
2116 {
2117         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2118         nd_region->dax_seed = nd_dax_create(nd_region);
2119         /*
2120          * Seed creation failures are not fatal, provisioning is simply
2121          * disabled until memory becomes available
2122          */
2123         if (!nd_region->dax_seed)
2124                 dev_err(&nd_region->dev, "failed to create dax namespace\n");
2125 }
2126
2127 void nd_region_create_pfn_seed(struct nd_region *nd_region)
2128 {
2129         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2130         nd_region->pfn_seed = nd_pfn_create(nd_region);
2131         /*
2132          * Seed creation failures are not fatal, provisioning is simply
2133          * disabled until memory becomes available
2134          */
2135         if (!nd_region->pfn_seed)
2136                 dev_err(&nd_region->dev, "failed to create pfn namespace\n");
2137 }
2138
2139 void nd_region_create_btt_seed(struct nd_region *nd_region)
2140 {
2141         WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
2142         nd_region->btt_seed = nd_btt_create(nd_region);
2143         /*
2144          * Seed creation failures are not fatal, provisioning is simply
2145          * disabled until memory becomes available
2146          */
2147         if (!nd_region->btt_seed)
2148                 dev_err(&nd_region->dev, "failed to create btt namespace\n");
2149 }
2150
2151 static int add_namespace_resource(struct nd_region *nd_region,
2152                 struct nd_namespace_label *nd_label, struct device **devs,
2153                 int count)
2154 {
2155         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2156         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2157         int i;
2158
2159         for (i = 0; i < count; i++) {
2160                 u8 *uuid = namespace_to_uuid(devs[i]);
2161                 struct resource *res;
2162
2163                 if (IS_ERR_OR_NULL(uuid)) {
2164                         WARN_ON(1);
2165                         continue;
2166                 }
2167
2168                 if (memcmp(uuid, nd_label->uuid, NSLABEL_UUID_LEN) != 0)
2169                         continue;
2170                 if (is_namespace_blk(devs[i])) {
2171                         res = nsblk_add_resource(nd_region, ndd,
2172                                         to_nd_namespace_blk(devs[i]),
2173                                         __le64_to_cpu(nd_label->dpa));
2174                         if (!res)
2175                                 return -ENXIO;
2176                         nd_dbg_dpa(nd_region, ndd, res, "%d assign\n", count);
2177                 } else {
2178                         dev_err(&nd_region->dev,
2179                                         "error: conflicting extents for uuid: %pUb\n",
2180                                         nd_label->uuid);
2181                         return -ENXIO;
2182                 }
2183                 break;
2184         }
2185
2186         return i;
2187 }
2188
2189 static struct device *create_namespace_blk(struct nd_region *nd_region,
2190                 struct nd_namespace_label *nd_label, int count)
2191 {
2192
2193         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2194         struct nd_interleave_set *nd_set = nd_region->nd_set;
2195         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2196         struct nd_namespace_blk *nsblk;
2197         char name[NSLABEL_NAME_LEN];
2198         struct device *dev = NULL;
2199         struct resource *res;
2200
2201         if (namespace_label_has(ndd, type_guid)) {
2202                 if (!guid_equal(&nd_set->type_guid, &nd_label->type_guid)) {
2203                         dev_dbg(ndd->dev, "expect type_guid %pUb got %pUb\n",
2204                                         nd_set->type_guid.b,
2205                                         nd_label->type_guid.b);
2206                         return ERR_PTR(-EAGAIN);
2207                 }
2208
2209                 if (nd_label->isetcookie != __cpu_to_le64(nd_set->cookie2)) {
2210                         dev_dbg(ndd->dev, "expect cookie %#llx got %#llx\n",
2211                                         nd_set->cookie2,
2212                                         __le64_to_cpu(nd_label->isetcookie));
2213                         return ERR_PTR(-EAGAIN);
2214                 }
2215         }
2216
2217         nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2218         if (!nsblk)
2219                 return ERR_PTR(-ENOMEM);
2220         dev = &nsblk->common.dev;
2221         dev->type = &namespace_blk_device_type;
2222         dev->parent = &nd_region->dev;
2223         nsblk->id = -1;
2224         nsblk->lbasize = __le64_to_cpu(nd_label->lbasize);
2225         nsblk->uuid = kmemdup(nd_label->uuid, NSLABEL_UUID_LEN,
2226                         GFP_KERNEL);
2227         if (namespace_label_has(ndd, abstraction_guid))
2228                 nsblk->common.claim_class
2229                         = to_nvdimm_cclass(&nd_label->abstraction_guid);
2230         if (!nsblk->uuid)
2231                 goto blk_err;
2232         memcpy(name, nd_label->name, NSLABEL_NAME_LEN);
2233         if (name[0])
2234                 nsblk->alt_name = kmemdup(name, NSLABEL_NAME_LEN,
2235                                 GFP_KERNEL);
2236         res = nsblk_add_resource(nd_region, ndd, nsblk,
2237                         __le64_to_cpu(nd_label->dpa));
2238         if (!res)
2239                 goto blk_err;
2240         nd_dbg_dpa(nd_region, ndd, res, "%d: assign\n", count);
2241         return dev;
2242  blk_err:
2243         namespace_blk_release(dev);
2244         return ERR_PTR(-ENXIO);
2245 }
2246
2247 static int cmp_dpa(const void *a, const void *b)
2248 {
2249         const struct device *dev_a = *(const struct device **) a;
2250         const struct device *dev_b = *(const struct device **) b;
2251         struct nd_namespace_blk *nsblk_a, *nsblk_b;
2252         struct nd_namespace_pmem *nspm_a, *nspm_b;
2253
2254         if (is_namespace_io(dev_a))
2255                 return 0;
2256
2257         if (is_namespace_blk(dev_a)) {
2258                 nsblk_a = to_nd_namespace_blk(dev_a);
2259                 nsblk_b = to_nd_namespace_blk(dev_b);
2260
2261                 return memcmp(&nsblk_a->res[0]->start, &nsblk_b->res[0]->start,
2262                                 sizeof(resource_size_t));
2263         }
2264
2265         nspm_a = to_nd_namespace_pmem(dev_a);
2266         nspm_b = to_nd_namespace_pmem(dev_b);
2267
2268         return memcmp(&nspm_a->nsio.res.start, &nspm_b->nsio.res.start,
2269                         sizeof(resource_size_t));
2270 }
2271
2272 static struct device **scan_labels(struct nd_region *nd_region)
2273 {
2274         int i, count = 0;
2275         struct device *dev, **devs = NULL;
2276         struct nd_label_ent *label_ent, *e;
2277         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2278         resource_size_t map_end = nd_mapping->start + nd_mapping->size - 1;
2279
2280         /* "safe" because create_namespace_pmem() might list_move() label_ent */
2281         list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
2282                 struct nd_namespace_label *nd_label = label_ent->label;
2283                 struct device **__devs;
2284                 u32 flags;
2285
2286                 if (!nd_label)
2287                         continue;
2288                 flags = __le32_to_cpu(nd_label->flags);
2289                 if (is_nd_blk(&nd_region->dev)
2290                                 == !!(flags & NSLABEL_FLAG_LOCAL))
2291                         /* pass, region matches label type */;
2292                 else
2293                         continue;
2294
2295                 /* skip labels that describe extents outside of the region */
2296                 if (nd_label->dpa < nd_mapping->start || nd_label->dpa > map_end)
2297                         continue;
2298
2299                 i = add_namespace_resource(nd_region, nd_label, devs, count);
2300                 if (i < 0)
2301                         goto err;
2302                 if (i < count)
2303                         continue;
2304                 __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
2305                 if (!__devs)
2306                         goto err;
2307                 memcpy(__devs, devs, sizeof(dev) * count);
2308                 kfree(devs);
2309                 devs = __devs;
2310
2311                 if (is_nd_blk(&nd_region->dev))
2312                         dev = create_namespace_blk(nd_region, nd_label, count);
2313                 else {
2314                         struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2315                         struct nd_namespace_index *nsindex;
2316
2317                         nsindex = to_namespace_index(ndd, ndd->ns_current);
2318                         dev = create_namespace_pmem(nd_region, nsindex, nd_label);
2319                 }
2320
2321                 if (IS_ERR(dev)) {
2322                         switch (PTR_ERR(dev)) {
2323                         case -EAGAIN:
2324                                 /* skip invalid labels */
2325                                 continue;
2326                         case -ENODEV:
2327                                 /* fallthrough to seed creation */
2328                                 break;
2329                         default:
2330                                 goto err;
2331                         }
2332                 } else
2333                         devs[count++] = dev;
2334
2335         }
2336
2337         dev_dbg(&nd_region->dev, "%s: discovered %d %s namespace%s\n",
2338                         __func__, count, is_nd_blk(&nd_region->dev)
2339                         ? "blk" : "pmem", count == 1 ? "" : "s");
2340
2341         if (count == 0) {
2342                 /* Publish a zero-sized namespace for userspace to configure. */
2343                 nd_mapping_free_labels(nd_mapping);
2344
2345                 devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
2346                 if (!devs)
2347                         goto err;
2348                 if (is_nd_blk(&nd_region->dev)) {
2349                         struct nd_namespace_blk *nsblk;
2350
2351                         nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2352                         if (!nsblk)
2353                                 goto err;
2354                         dev = &nsblk->common.dev;
2355                         dev->type = &namespace_blk_device_type;
2356                 } else {
2357                         struct nd_namespace_pmem *nspm;
2358
2359                         nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2360                         if (!nspm)
2361                                 goto err;
2362                         dev = &nspm->nsio.common.dev;
2363                         dev->type = &namespace_pmem_device_type;
2364                         nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2365                 }
2366                 dev->parent = &nd_region->dev;
2367                 devs[count++] = dev;
2368         } else if (is_memory(&nd_region->dev)) {
2369                 /* clean unselected labels */
2370                 for (i = 0; i < nd_region->ndr_mappings; i++) {
2371                         struct list_head *l, *e;
2372                         LIST_HEAD(list);
2373                         int j;
2374
2375                         nd_mapping = &nd_region->mapping[i];
2376                         if (list_empty(&nd_mapping->labels)) {
2377                                 WARN_ON(1);
2378                                 continue;
2379                         }
2380
2381                         j = count;
2382                         list_for_each_safe(l, e, &nd_mapping->labels) {
2383                                 if (!j--)
2384                                         break;
2385                                 list_move_tail(l, &list);
2386                         }
2387                         nd_mapping_free_labels(nd_mapping);
2388                         list_splice_init(&list, &nd_mapping->labels);
2389                 }
2390         }
2391
2392         if (count > 1)
2393                 sort(devs, count, sizeof(struct device *), cmp_dpa, NULL);
2394
2395         return devs;
2396
2397  err:
2398         if (devs) {
2399                 for (i = 0; devs[i]; i++)
2400                         if (is_nd_blk(&nd_region->dev))
2401                                 namespace_blk_release(devs[i]);
2402                         else
2403                                 namespace_pmem_release(devs[i]);
2404                 kfree(devs);
2405         }
2406         return NULL;
2407 }
2408
2409 static struct device **create_namespaces(struct nd_region *nd_region)
2410 {
2411         struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2412         struct device **devs;
2413         int i;
2414
2415         if (nd_region->ndr_mappings == 0)
2416                 return NULL;
2417
2418         /* lock down all mappings while we scan labels */
2419         for (i = 0; i < nd_region->ndr_mappings; i++) {
2420                 nd_mapping = &nd_region->mapping[i];
2421                 mutex_lock_nested(&nd_mapping->lock, i);
2422         }
2423
2424         devs = scan_labels(nd_region);
2425
2426         for (i = 0; i < nd_region->ndr_mappings; i++) {
2427                 int reverse = nd_region->ndr_mappings - 1 - i;
2428
2429                 nd_mapping = &nd_region->mapping[reverse];
2430                 mutex_unlock(&nd_mapping->lock);
2431         }
2432
2433         return devs;
2434 }
2435
2436 static int init_active_labels(struct nd_region *nd_region)
2437 {
2438         int i;
2439
2440         for (i = 0; i < nd_region->ndr_mappings; i++) {
2441                 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2442                 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2443                 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2444                 struct nd_label_ent *label_ent;
2445                 int count, j;
2446
2447                 /*
2448                  * If the dimm is disabled then we may need to prevent
2449                  * the region from being activated.
2450                  */
2451                 if (!ndd) {
2452                         if (test_bit(NDD_LOCKED, &nvdimm->flags))
2453                                 /* fail, label data may be unreadable */;
2454                         else if (test_bit(NDD_ALIASING, &nvdimm->flags))
2455                                 /* fail, labels needed to disambiguate dpa */;
2456                         else
2457                                 return 0;
2458
2459                         dev_err(&nd_region->dev, "%s: is %s, failing probe\n",
2460                                         dev_name(&nd_mapping->nvdimm->dev),
2461                                         test_bit(NDD_LOCKED, &nvdimm->flags)
2462                                         ? "locked" : "disabled");
2463                         return -ENXIO;
2464                 }
2465                 nd_mapping->ndd = ndd;
2466                 atomic_inc(&nvdimm->busy);
2467                 get_ndd(ndd);
2468
2469                 count = nd_label_active_count(ndd);
2470                 dev_dbg(ndd->dev, "%s: %d\n", __func__, count);
2471                 if (!count)
2472                         continue;
2473                 for (j = 0; j < count; j++) {
2474                         struct nd_namespace_label *label;
2475
2476                         label_ent = kzalloc(sizeof(*label_ent), GFP_KERNEL);
2477                         if (!label_ent)
2478                                 break;
2479                         label = nd_label_active(ndd, j);
2480                         label_ent->label = label;
2481
2482                         mutex_lock(&nd_mapping->lock);
2483                         list_add_tail(&label_ent->list, &nd_mapping->labels);
2484                         mutex_unlock(&nd_mapping->lock);
2485                 }
2486
2487                 if (j >= count)
2488                         continue;
2489
2490                 mutex_lock(&nd_mapping->lock);
2491                 nd_mapping_free_labels(nd_mapping);
2492                 mutex_unlock(&nd_mapping->lock);
2493                 return -ENOMEM;
2494         }
2495
2496         return 0;
2497 }
2498
2499 int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
2500 {
2501         struct device **devs = NULL;
2502         int i, rc = 0, type;
2503
2504         *err = 0;
2505         nvdimm_bus_lock(&nd_region->dev);
2506         rc = init_active_labels(nd_region);
2507         if (rc) {
2508                 nvdimm_bus_unlock(&nd_region->dev);
2509                 return rc;
2510         }
2511
2512         type = nd_region_to_nstype(nd_region);
2513         switch (type) {
2514         case ND_DEVICE_NAMESPACE_IO:
2515                 devs = create_namespace_io(nd_region);
2516                 break;
2517         case ND_DEVICE_NAMESPACE_PMEM:
2518         case ND_DEVICE_NAMESPACE_BLK:
2519                 devs = create_namespaces(nd_region);
2520                 break;
2521         default:
2522                 break;
2523         }
2524         nvdimm_bus_unlock(&nd_region->dev);
2525
2526         if (!devs)
2527                 return -ENODEV;
2528
2529         for (i = 0; devs[i]; i++) {
2530                 struct device *dev = devs[i];
2531                 int id;
2532
2533                 if (type == ND_DEVICE_NAMESPACE_BLK) {
2534                         struct nd_namespace_blk *nsblk;
2535
2536                         nsblk = to_nd_namespace_blk(dev);
2537                         id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2538                                         GFP_KERNEL);
2539                         nsblk->id = id;
2540                 } else if (type == ND_DEVICE_NAMESPACE_PMEM) {
2541                         struct nd_namespace_pmem *nspm;
2542
2543                         nspm = to_nd_namespace_pmem(dev);
2544                         id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2545                                         GFP_KERNEL);
2546                         nspm->id = id;
2547                 } else
2548                         id = i;
2549
2550                 if (id < 0)
2551                         break;
2552                 dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
2553                 dev->groups = nd_namespace_attribute_groups;
2554                 nd_device_register(dev);
2555         }
2556         if (i)
2557                 nd_region->ns_seed = devs[0];
2558
2559         if (devs[i]) {
2560                 int j;
2561
2562                 for (j = i; devs[j]; j++) {
2563                         struct device *dev = devs[j];
2564
2565                         device_initialize(dev);
2566                         put_device(dev);
2567                 }
2568                 *err = j - i;
2569                 /*
2570                  * All of the namespaces we tried to register failed, so
2571                  * fail region activation.
2572                  */
2573                 if (*err == 0)
2574                         rc = -ENODEV;
2575         }
2576         kfree(devs);
2577
2578         if (rc == -ENODEV)
2579                 return rc;
2580
2581         return i;
2582 }