Loading...
1/*
2 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 */
13#include <linux/module.h>
14#include <linux/device.h>
15#include <linux/sort.h>
16#include <linux/slab.h>
17#include <linux/pmem.h>
18#include <linux/list.h>
19#include <linux/nd.h>
20#include "nd-core.h"
21#include "nd.h"
22
23static void namespace_io_release(struct device *dev)
24{
25 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
26
27 kfree(nsio);
28}
29
30static void namespace_pmem_release(struct device *dev)
31{
32 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
33 struct nd_region *nd_region = to_nd_region(dev->parent);
34
35 if (nspm->id >= 0)
36 ida_simple_remove(&nd_region->ns_ida, nspm->id);
37 kfree(nspm->alt_name);
38 kfree(nspm->uuid);
39 kfree(nspm);
40}
41
42static void namespace_blk_release(struct device *dev)
43{
44 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
45 struct nd_region *nd_region = to_nd_region(dev->parent);
46
47 if (nsblk->id >= 0)
48 ida_simple_remove(&nd_region->ns_ida, nsblk->id);
49 kfree(nsblk->alt_name);
50 kfree(nsblk->uuid);
51 kfree(nsblk->res);
52 kfree(nsblk);
53}
54
55static const struct device_type namespace_io_device_type = {
56 .name = "nd_namespace_io",
57 .release = namespace_io_release,
58};
59
60static const struct device_type namespace_pmem_device_type = {
61 .name = "nd_namespace_pmem",
62 .release = namespace_pmem_release,
63};
64
65static const struct device_type namespace_blk_device_type = {
66 .name = "nd_namespace_blk",
67 .release = namespace_blk_release,
68};
69
70static bool is_namespace_pmem(const struct device *dev)
71{
72 return dev ? dev->type == &namespace_pmem_device_type : false;
73}
74
75static bool is_namespace_blk(const struct device *dev)
76{
77 return dev ? dev->type == &namespace_blk_device_type : false;
78}
79
80static bool is_namespace_io(const struct device *dev)
81{
82 return dev ? dev->type == &namespace_io_device_type : false;
83}
84
85static int is_uuid_busy(struct device *dev, void *data)
86{
87 u8 *uuid1 = data, *uuid2 = NULL;
88
89 if (is_namespace_pmem(dev)) {
90 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
91
92 uuid2 = nspm->uuid;
93 } else if (is_namespace_blk(dev)) {
94 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
95
96 uuid2 = nsblk->uuid;
97 } else if (is_nd_btt(dev)) {
98 struct nd_btt *nd_btt = to_nd_btt(dev);
99
100 uuid2 = nd_btt->uuid;
101 } else if (is_nd_pfn(dev)) {
102 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
103
104 uuid2 = nd_pfn->uuid;
105 }
106
107 if (uuid2 && memcmp(uuid1, uuid2, NSLABEL_UUID_LEN) == 0)
108 return -EBUSY;
109
110 return 0;
111}
112
113static int is_namespace_uuid_busy(struct device *dev, void *data)
114{
115 if (is_nd_pmem(dev) || is_nd_blk(dev))
116 return device_for_each_child(dev, data, is_uuid_busy);
117 return 0;
118}
119
120/**
121 * nd_is_uuid_unique - verify that no other namespace has @uuid
122 * @dev: any device on a nvdimm_bus
123 * @uuid: uuid to check
124 */
125bool nd_is_uuid_unique(struct device *dev, u8 *uuid)
126{
127 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
128
129 if (!nvdimm_bus)
130 return false;
131 WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm_bus->dev));
132 if (device_for_each_child(&nvdimm_bus->dev, uuid,
133 is_namespace_uuid_busy) != 0)
134 return false;
135 return true;
136}
137
138bool pmem_should_map_pages(struct device *dev)
139{
140 struct nd_region *nd_region = to_nd_region(dev->parent);
141 struct nd_namespace_io *nsio;
142
143 if (!IS_ENABLED(CONFIG_ZONE_DEVICE))
144 return false;
145
146 if (!test_bit(ND_REGION_PAGEMAP, &nd_region->flags))
147 return false;
148
149 if (is_nd_pfn(dev) || is_nd_btt(dev))
150 return false;
151
152 nsio = to_nd_namespace_io(dev);
153 if (region_intersects(nsio->res.start, resource_size(&nsio->res),
154 IORESOURCE_SYSTEM_RAM,
155 IORES_DESC_NONE) == REGION_MIXED)
156 return false;
157
158#ifdef ARCH_MEMREMAP_PMEM
159 return ARCH_MEMREMAP_PMEM == MEMREMAP_WB;
160#else
161 return false;
162#endif
163}
164EXPORT_SYMBOL(pmem_should_map_pages);
165
166const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,
167 char *name)
168{
169 struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
170 const char *suffix = NULL;
171
172 if (ndns->claim && is_nd_btt(ndns->claim))
173 suffix = "s";
174
175 if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev)) {
176 int nsidx = 0;
177
178 if (is_namespace_pmem(&ndns->dev)) {
179 struct nd_namespace_pmem *nspm;
180
181 nspm = to_nd_namespace_pmem(&ndns->dev);
182 nsidx = nspm->id;
183 }
184
185 if (nsidx)
186 sprintf(name, "pmem%d.%d%s", nd_region->id, nsidx,
187 suffix ? suffix : "");
188 else
189 sprintf(name, "pmem%d%s", nd_region->id,
190 suffix ? suffix : "");
191 } else if (is_namespace_blk(&ndns->dev)) {
192 struct nd_namespace_blk *nsblk;
193
194 nsblk = to_nd_namespace_blk(&ndns->dev);
195 sprintf(name, "ndblk%d.%d%s", nd_region->id, nsblk->id,
196 suffix ? suffix : "");
197 } else {
198 return NULL;
199 }
200
201 return name;
202}
203EXPORT_SYMBOL(nvdimm_namespace_disk_name);
204
205const u8 *nd_dev_to_uuid(struct device *dev)
206{
207 static const u8 null_uuid[16];
208
209 if (!dev)
210 return null_uuid;
211
212 if (is_namespace_pmem(dev)) {
213 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
214
215 return nspm->uuid;
216 } else if (is_namespace_blk(dev)) {
217 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
218
219 return nsblk->uuid;
220 } else
221 return null_uuid;
222}
223EXPORT_SYMBOL(nd_dev_to_uuid);
224
225static ssize_t nstype_show(struct device *dev,
226 struct device_attribute *attr, char *buf)
227{
228 struct nd_region *nd_region = to_nd_region(dev->parent);
229
230 return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
231}
232static DEVICE_ATTR_RO(nstype);
233
234static ssize_t __alt_name_store(struct device *dev, const char *buf,
235 const size_t len)
236{
237 char *input, *pos, *alt_name, **ns_altname;
238 ssize_t rc;
239
240 if (is_namespace_pmem(dev)) {
241 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
242
243 ns_altname = &nspm->alt_name;
244 } else if (is_namespace_blk(dev)) {
245 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
246
247 ns_altname = &nsblk->alt_name;
248 } else
249 return -ENXIO;
250
251 if (dev->driver || to_ndns(dev)->claim)
252 return -EBUSY;
253
254 input = kmemdup(buf, len + 1, GFP_KERNEL);
255 if (!input)
256 return -ENOMEM;
257
258 input[len] = '\0';
259 pos = strim(input);
260 if (strlen(pos) + 1 > NSLABEL_NAME_LEN) {
261 rc = -EINVAL;
262 goto out;
263 }
264
265 alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL);
266 if (!alt_name) {
267 rc = -ENOMEM;
268 goto out;
269 }
270 kfree(*ns_altname);
271 *ns_altname = alt_name;
272 sprintf(*ns_altname, "%s", pos);
273 rc = len;
274
275out:
276 kfree(input);
277 return rc;
278}
279
280static resource_size_t nd_namespace_blk_size(struct nd_namespace_blk *nsblk)
281{
282 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
283 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
284 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
285 struct nd_label_id label_id;
286 resource_size_t size = 0;
287 struct resource *res;
288
289 if (!nsblk->uuid)
290 return 0;
291 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
292 for_each_dpa_resource(ndd, res)
293 if (strcmp(res->name, label_id.id) == 0)
294 size += resource_size(res);
295 return size;
296}
297
298static bool __nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
299{
300 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
301 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
302 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
303 struct nd_label_id label_id;
304 struct resource *res;
305 int count, i;
306
307 if (!nsblk->uuid || !nsblk->lbasize || !ndd)
308 return false;
309
310 count = 0;
311 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
312 for_each_dpa_resource(ndd, res) {
313 if (strcmp(res->name, label_id.id) != 0)
314 continue;
315 /*
316 * Resources with unacknowledged adjustments indicate a
317 * failure to update labels
318 */
319 if (res->flags & DPA_RESOURCE_ADJUSTED)
320 return false;
321 count++;
322 }
323
324 /* These values match after a successful label update */
325 if (count != nsblk->num_resources)
326 return false;
327
328 for (i = 0; i < nsblk->num_resources; i++) {
329 struct resource *found = NULL;
330
331 for_each_dpa_resource(ndd, res)
332 if (res == nsblk->res[i]) {
333 found = res;
334 break;
335 }
336 /* stale resource */
337 if (!found)
338 return false;
339 }
340
341 return true;
342}
343
344resource_size_t nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
345{
346 resource_size_t size;
347
348 nvdimm_bus_lock(&nsblk->common.dev);
349 size = __nd_namespace_blk_validate(nsblk);
350 nvdimm_bus_unlock(&nsblk->common.dev);
351
352 return size;
353}
354EXPORT_SYMBOL(nd_namespace_blk_validate);
355
356
357static int nd_namespace_label_update(struct nd_region *nd_region,
358 struct device *dev)
359{
360 dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim,
361 "namespace must be idle during label update\n");
362 if (dev->driver || to_ndns(dev)->claim)
363 return 0;
364
365 /*
366 * Only allow label writes that will result in a valid namespace
367 * or deletion of an existing namespace.
368 */
369 if (is_namespace_pmem(dev)) {
370 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
371 resource_size_t size = resource_size(&nspm->nsio.res);
372
373 if (size == 0 && nspm->uuid)
374 /* delete allocation */;
375 else if (!nspm->uuid)
376 return 0;
377
378 return nd_pmem_namespace_label_update(nd_region, nspm, size);
379 } else if (is_namespace_blk(dev)) {
380 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
381 resource_size_t size = nd_namespace_blk_size(nsblk);
382
383 if (size == 0 && nsblk->uuid)
384 /* delete allocation */;
385 else if (!nsblk->uuid || !nsblk->lbasize)
386 return 0;
387
388 return nd_blk_namespace_label_update(nd_region, nsblk, size);
389 } else
390 return -ENXIO;
391}
392
393static ssize_t alt_name_store(struct device *dev,
394 struct device_attribute *attr, const char *buf, size_t len)
395{
396 struct nd_region *nd_region = to_nd_region(dev->parent);
397 ssize_t rc;
398
399 device_lock(dev);
400 nvdimm_bus_lock(dev);
401 wait_nvdimm_bus_probe_idle(dev);
402 rc = __alt_name_store(dev, buf, len);
403 if (rc >= 0)
404 rc = nd_namespace_label_update(nd_region, dev);
405 dev_dbg(dev, "%s: %s(%zd)\n", __func__, rc < 0 ? "fail " : "", rc);
406 nvdimm_bus_unlock(dev);
407 device_unlock(dev);
408
409 return rc < 0 ? rc : len;
410}
411
412static ssize_t alt_name_show(struct device *dev,
413 struct device_attribute *attr, char *buf)
414{
415 char *ns_altname;
416
417 if (is_namespace_pmem(dev)) {
418 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
419
420 ns_altname = nspm->alt_name;
421 } else if (is_namespace_blk(dev)) {
422 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
423
424 ns_altname = nsblk->alt_name;
425 } else
426 return -ENXIO;
427
428 return sprintf(buf, "%s\n", ns_altname ? ns_altname : "");
429}
430static DEVICE_ATTR_RW(alt_name);
431
432static int scan_free(struct nd_region *nd_region,
433 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
434 resource_size_t n)
435{
436 bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
437 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
438 int rc = 0;
439
440 while (n) {
441 struct resource *res, *last;
442 resource_size_t new_start;
443
444 last = NULL;
445 for_each_dpa_resource(ndd, res)
446 if (strcmp(res->name, label_id->id) == 0)
447 last = res;
448 res = last;
449 if (!res)
450 return 0;
451
452 if (n >= resource_size(res)) {
453 n -= resource_size(res);
454 nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc);
455 nvdimm_free_dpa(ndd, res);
456 /* retry with last resource deleted */
457 continue;
458 }
459
460 /*
461 * Keep BLK allocations relegated to high DPA as much as
462 * possible
463 */
464 if (is_blk)
465 new_start = res->start + n;
466 else
467 new_start = res->start;
468
469 rc = adjust_resource(res, new_start, resource_size(res) - n);
470 if (rc == 0)
471 res->flags |= DPA_RESOURCE_ADJUSTED;
472 nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc);
473 break;
474 }
475
476 return rc;
477}
478
479/**
480 * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
481 * @nd_region: the set of dimms to reclaim @n bytes from
482 * @label_id: unique identifier for the namespace consuming this dpa range
483 * @n: number of bytes per-dimm to release
484 *
485 * Assumes resources are ordered. Starting from the end try to
486 * adjust_resource() the allocation to @n, but if @n is larger than the
487 * allocation delete it and find the 'new' last allocation in the label
488 * set.
489 */
490static int shrink_dpa_allocation(struct nd_region *nd_region,
491 struct nd_label_id *label_id, resource_size_t n)
492{
493 int i;
494
495 for (i = 0; i < nd_region->ndr_mappings; i++) {
496 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
497 int rc;
498
499 rc = scan_free(nd_region, nd_mapping, label_id, n);
500 if (rc)
501 return rc;
502 }
503
504 return 0;
505}
506
507static resource_size_t init_dpa_allocation(struct nd_label_id *label_id,
508 struct nd_region *nd_region, struct nd_mapping *nd_mapping,
509 resource_size_t n)
510{
511 bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
512 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
513 resource_size_t first_dpa;
514 struct resource *res;
515 int rc = 0;
516
517 /* allocate blk from highest dpa first */
518 if (is_blk)
519 first_dpa = nd_mapping->start + nd_mapping->size - n;
520 else
521 first_dpa = nd_mapping->start;
522
523 /* first resource allocation for this label-id or dimm */
524 res = nvdimm_allocate_dpa(ndd, label_id, first_dpa, n);
525 if (!res)
526 rc = -EBUSY;
527
528 nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc);
529 return rc ? n : 0;
530}
531
532
533/**
534 * space_valid() - validate free dpa space against constraints
535 * @nd_region: hosting region of the free space
536 * @ndd: dimm device data for debug
537 * @label_id: namespace id to allocate space
538 * @prev: potential allocation that precedes free space
539 * @next: allocation that follows the given free space range
540 * @exist: first allocation with same id in the mapping
541 * @n: range that must satisfied for pmem allocations
542 * @valid: free space range to validate
543 *
544 * BLK-space is valid as long as it does not precede a PMEM
545 * allocation in a given region. PMEM-space must be contiguous
546 * and adjacent to an existing existing allocation (if one
547 * exists). If reserving PMEM any space is valid.
548 */
549static void space_valid(struct nd_region *nd_region, struct nvdimm_drvdata *ndd,
550 struct nd_label_id *label_id, struct resource *prev,
551 struct resource *next, struct resource *exist,
552 resource_size_t n, struct resource *valid)
553{
554 bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0;
555 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
556
557 if (valid->start >= valid->end)
558 goto invalid;
559
560 if (is_reserve)
561 return;
562
563 if (!is_pmem) {
564 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
565 struct nvdimm_bus *nvdimm_bus;
566 struct blk_alloc_info info = {
567 .nd_mapping = nd_mapping,
568 .available = nd_mapping->size,
569 .res = valid,
570 };
571
572 WARN_ON(!is_nd_blk(&nd_region->dev));
573 nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
574 device_for_each_child(&nvdimm_bus->dev, &info, alias_dpa_busy);
575 return;
576 }
577
578 /* allocation needs to be contiguous, so this is all or nothing */
579 if (resource_size(valid) < n)
580 goto invalid;
581
582 /* we've got all the space we need and no existing allocation */
583 if (!exist)
584 return;
585
586 /* allocation needs to be contiguous with the existing namespace */
587 if (valid->start == exist->end + 1
588 || valid->end == exist->start - 1)
589 return;
590
591 invalid:
592 /* truncate @valid size to 0 */
593 valid->end = valid->start - 1;
594}
595
596enum alloc_loc {
597 ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
598};
599
600static resource_size_t scan_allocate(struct nd_region *nd_region,
601 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
602 resource_size_t n)
603{
604 resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1;
605 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
606 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
607 struct resource *res, *exist = NULL, valid;
608 const resource_size_t to_allocate = n;
609 int first;
610
611 for_each_dpa_resource(ndd, res)
612 if (strcmp(label_id->id, res->name) == 0)
613 exist = res;
614
615 valid.start = nd_mapping->start;
616 valid.end = mapping_end;
617 valid.name = "free space";
618 retry:
619 first = 0;
620 for_each_dpa_resource(ndd, res) {
621 struct resource *next = res->sibling, *new_res = NULL;
622 resource_size_t allocate, available = 0;
623 enum alloc_loc loc = ALLOC_ERR;
624 const char *action;
625 int rc = 0;
626
627 /* ignore resources outside this nd_mapping */
628 if (res->start > mapping_end)
629 continue;
630 if (res->end < nd_mapping->start)
631 continue;
632
633 /* space at the beginning of the mapping */
634 if (!first++ && res->start > nd_mapping->start) {
635 valid.start = nd_mapping->start;
636 valid.end = res->start - 1;
637 space_valid(nd_region, ndd, label_id, NULL, next, exist,
638 to_allocate, &valid);
639 available = resource_size(&valid);
640 if (available)
641 loc = ALLOC_BEFORE;
642 }
643
644 /* space between allocations */
645 if (!loc && next) {
646 valid.start = res->start + resource_size(res);
647 valid.end = min(mapping_end, next->start - 1);
648 space_valid(nd_region, ndd, label_id, res, next, exist,
649 to_allocate, &valid);
650 available = resource_size(&valid);
651 if (available)
652 loc = ALLOC_MID;
653 }
654
655 /* space at the end of the mapping */
656 if (!loc && !next) {
657 valid.start = res->start + resource_size(res);
658 valid.end = mapping_end;
659 space_valid(nd_region, ndd, label_id, res, next, exist,
660 to_allocate, &valid);
661 available = resource_size(&valid);
662 if (available)
663 loc = ALLOC_AFTER;
664 }
665
666 if (!loc || !available)
667 continue;
668 allocate = min(available, n);
669 switch (loc) {
670 case ALLOC_BEFORE:
671 if (strcmp(res->name, label_id->id) == 0) {
672 /* adjust current resource up */
673 rc = adjust_resource(res, res->start - allocate,
674 resource_size(res) + allocate);
675 action = "cur grow up";
676 } else
677 action = "allocate";
678 break;
679 case ALLOC_MID:
680 if (strcmp(next->name, label_id->id) == 0) {
681 /* adjust next resource up */
682 rc = adjust_resource(next, next->start
683 - allocate, resource_size(next)
684 + allocate);
685 new_res = next;
686 action = "next grow up";
687 } else if (strcmp(res->name, label_id->id) == 0) {
688 action = "grow down";
689 } else
690 action = "allocate";
691 break;
692 case ALLOC_AFTER:
693 if (strcmp(res->name, label_id->id) == 0)
694 action = "grow down";
695 else
696 action = "allocate";
697 break;
698 default:
699 return n;
700 }
701
702 if (strcmp(action, "allocate") == 0) {
703 /* BLK allocate bottom up */
704 if (!is_pmem)
705 valid.start += available - allocate;
706
707 new_res = nvdimm_allocate_dpa(ndd, label_id,
708 valid.start, allocate);
709 if (!new_res)
710 rc = -EBUSY;
711 } else if (strcmp(action, "grow down") == 0) {
712 /* adjust current resource down */
713 rc = adjust_resource(res, res->start, resource_size(res)
714 + allocate);
715 if (rc == 0)
716 res->flags |= DPA_RESOURCE_ADJUSTED;
717 }
718
719 if (!new_res)
720 new_res = res;
721
722 nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n",
723 action, loc, rc);
724
725 if (rc)
726 return n;
727
728 n -= allocate;
729 if (n) {
730 /*
731 * Retry scan with newly inserted resources.
732 * For example, if we did an ALLOC_BEFORE
733 * insertion there may also have been space
734 * available for an ALLOC_AFTER insertion, so we
735 * need to check this same resource again
736 */
737 goto retry;
738 } else
739 return 0;
740 }
741
742 /*
743 * If we allocated nothing in the BLK case it may be because we are in
744 * an initial "pmem-reserve pass". Only do an initial BLK allocation
745 * when none of the DPA space is reserved.
746 */
747 if ((is_pmem || !ndd->dpa.child) && n == to_allocate)
748 return init_dpa_allocation(label_id, nd_region, nd_mapping, n);
749 return n;
750}
751
752static int merge_dpa(struct nd_region *nd_region,
753 struct nd_mapping *nd_mapping, struct nd_label_id *label_id)
754{
755 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
756 struct resource *res;
757
758 if (strncmp("pmem", label_id->id, 4) == 0)
759 return 0;
760 retry:
761 for_each_dpa_resource(ndd, res) {
762 int rc;
763 struct resource *next = res->sibling;
764 resource_size_t end = res->start + resource_size(res);
765
766 if (!next || strcmp(res->name, label_id->id) != 0
767 || strcmp(next->name, label_id->id) != 0
768 || end != next->start)
769 continue;
770 end += resource_size(next);
771 nvdimm_free_dpa(ndd, next);
772 rc = adjust_resource(res, res->start, end - res->start);
773 nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc);
774 if (rc)
775 return rc;
776 res->flags |= DPA_RESOURCE_ADJUSTED;
777 goto retry;
778 }
779
780 return 0;
781}
782
783static int __reserve_free_pmem(struct device *dev, void *data)
784{
785 struct nvdimm *nvdimm = data;
786 struct nd_region *nd_region;
787 struct nd_label_id label_id;
788 int i;
789
790 if (!is_nd_pmem(dev))
791 return 0;
792
793 nd_region = to_nd_region(dev);
794 if (nd_region->ndr_mappings == 0)
795 return 0;
796
797 memset(&label_id, 0, sizeof(label_id));
798 strcat(label_id.id, "pmem-reserve");
799 for (i = 0; i < nd_region->ndr_mappings; i++) {
800 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
801 resource_size_t n, rem = 0;
802
803 if (nd_mapping->nvdimm != nvdimm)
804 continue;
805
806 n = nd_pmem_available_dpa(nd_region, nd_mapping, &rem);
807 if (n == 0)
808 return 0;
809 rem = scan_allocate(nd_region, nd_mapping, &label_id, n);
810 dev_WARN_ONCE(&nd_region->dev, rem,
811 "pmem reserve underrun: %#llx of %#llx bytes\n",
812 (unsigned long long) n - rem,
813 (unsigned long long) n);
814 return rem ? -ENXIO : 0;
815 }
816
817 return 0;
818}
819
820static void release_free_pmem(struct nvdimm_bus *nvdimm_bus,
821 struct nd_mapping *nd_mapping)
822{
823 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
824 struct resource *res, *_res;
825
826 for_each_dpa_resource_safe(ndd, res, _res)
827 if (strcmp(res->name, "pmem-reserve") == 0)
828 nvdimm_free_dpa(ndd, res);
829}
830
831static int reserve_free_pmem(struct nvdimm_bus *nvdimm_bus,
832 struct nd_mapping *nd_mapping)
833{
834 struct nvdimm *nvdimm = nd_mapping->nvdimm;
835 int rc;
836
837 rc = device_for_each_child(&nvdimm_bus->dev, nvdimm,
838 __reserve_free_pmem);
839 if (rc)
840 release_free_pmem(nvdimm_bus, nd_mapping);
841 return rc;
842}
843
844/**
845 * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
846 * @nd_region: the set of dimms to allocate @n more bytes from
847 * @label_id: unique identifier for the namespace consuming this dpa range
848 * @n: number of bytes per-dimm to add to the existing allocation
849 *
850 * Assumes resources are ordered. For BLK regions, first consume
851 * BLK-only available DPA free space, then consume PMEM-aliased DPA
852 * space starting at the highest DPA. For PMEM regions start
853 * allocations from the start of an interleave set and end at the first
854 * BLK allocation or the end of the interleave set, whichever comes
855 * first.
856 */
857static int grow_dpa_allocation(struct nd_region *nd_region,
858 struct nd_label_id *label_id, resource_size_t n)
859{
860 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
861 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
862 int i;
863
864 for (i = 0; i < nd_region->ndr_mappings; i++) {
865 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
866 resource_size_t rem = n;
867 int rc, j;
868
869 /*
870 * In the BLK case try once with all unallocated PMEM
871 * reserved, and once without
872 */
873 for (j = is_pmem; j < 2; j++) {
874 bool blk_only = j == 0;
875
876 if (blk_only) {
877 rc = reserve_free_pmem(nvdimm_bus, nd_mapping);
878 if (rc)
879 return rc;
880 }
881 rem = scan_allocate(nd_region, nd_mapping,
882 label_id, rem);
883 if (blk_only)
884 release_free_pmem(nvdimm_bus, nd_mapping);
885
886 /* try again and allow encroachments into PMEM */
887 if (rem == 0)
888 break;
889 }
890
891 dev_WARN_ONCE(&nd_region->dev, rem,
892 "allocation underrun: %#llx of %#llx bytes\n",
893 (unsigned long long) n - rem,
894 (unsigned long long) n);
895 if (rem)
896 return -ENXIO;
897
898 rc = merge_dpa(nd_region, nd_mapping, label_id);
899 if (rc)
900 return rc;
901 }
902
903 return 0;
904}
905
906static void nd_namespace_pmem_set_resource(struct nd_region *nd_region,
907 struct nd_namespace_pmem *nspm, resource_size_t size)
908{
909 struct resource *res = &nspm->nsio.res;
910 resource_size_t offset = 0;
911
912 if (size && !nspm->uuid) {
913 WARN_ON_ONCE(1);
914 size = 0;
915 }
916
917 if (size && nspm->uuid) {
918 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
919 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
920 struct nd_label_id label_id;
921 struct resource *res;
922
923 if (!ndd) {
924 size = 0;
925 goto out;
926 }
927
928 nd_label_gen_id(&label_id, nspm->uuid, 0);
929
930 /* calculate a spa offset from the dpa allocation offset */
931 for_each_dpa_resource(ndd, res)
932 if (strcmp(res->name, label_id.id) == 0) {
933 offset = (res->start - nd_mapping->start)
934 * nd_region->ndr_mappings;
935 goto out;
936 }
937
938 WARN_ON_ONCE(1);
939 size = 0;
940 }
941
942 out:
943 res->start = nd_region->ndr_start + offset;
944 res->end = res->start + size - 1;
945}
946
947static bool uuid_not_set(const u8 *uuid, struct device *dev, const char *where)
948{
949 if (!uuid) {
950 dev_dbg(dev, "%s: uuid not set\n", where);
951 return true;
952 }
953 return false;
954}
955
956static ssize_t __size_store(struct device *dev, unsigned long long val)
957{
958 resource_size_t allocated = 0, available = 0;
959 struct nd_region *nd_region = to_nd_region(dev->parent);
960 struct nd_namespace_common *ndns = to_ndns(dev);
961 struct nd_mapping *nd_mapping;
962 struct nvdimm_drvdata *ndd;
963 struct nd_label_id label_id;
964 u32 flags = 0, remainder;
965 int rc, i, id = -1;
966 u8 *uuid = NULL;
967
968 if (dev->driver || ndns->claim)
969 return -EBUSY;
970
971 if (is_namespace_pmem(dev)) {
972 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
973
974 uuid = nspm->uuid;
975 id = nspm->id;
976 } else if (is_namespace_blk(dev)) {
977 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
978
979 uuid = nsblk->uuid;
980 flags = NSLABEL_FLAG_LOCAL;
981 id = nsblk->id;
982 }
983
984 /*
985 * We need a uuid for the allocation-label and dimm(s) on which
986 * to store the label.
987 */
988 if (uuid_not_set(uuid, dev, __func__))
989 return -ENXIO;
990 if (nd_region->ndr_mappings == 0) {
991 dev_dbg(dev, "%s: not associated with dimm(s)\n", __func__);
992 return -ENXIO;
993 }
994
995 div_u64_rem(val, SZ_4K * nd_region->ndr_mappings, &remainder);
996 if (remainder) {
997 dev_dbg(dev, "%llu is not %dK aligned\n", val,
998 (SZ_4K * nd_region->ndr_mappings) / SZ_1K);
999 return -EINVAL;
1000 }
1001
1002 nd_label_gen_id(&label_id, uuid, flags);
1003 for (i = 0; i < nd_region->ndr_mappings; i++) {
1004 nd_mapping = &nd_region->mapping[i];
1005 ndd = to_ndd(nd_mapping);
1006
1007 /*
1008 * All dimms in an interleave set, or the base dimm for a blk
1009 * region, need to be enabled for the size to be changed.
1010 */
1011 if (!ndd)
1012 return -ENXIO;
1013
1014 allocated += nvdimm_allocated_dpa(ndd, &label_id);
1015 }
1016 available = nd_region_available_dpa(nd_region);
1017
1018 if (val > available + allocated)
1019 return -ENOSPC;
1020
1021 if (val == allocated)
1022 return 0;
1023
1024 val = div_u64(val, nd_region->ndr_mappings);
1025 allocated = div_u64(allocated, nd_region->ndr_mappings);
1026 if (val < allocated)
1027 rc = shrink_dpa_allocation(nd_region, &label_id,
1028 allocated - val);
1029 else
1030 rc = grow_dpa_allocation(nd_region, &label_id, val - allocated);
1031
1032 if (rc)
1033 return rc;
1034
1035 if (is_namespace_pmem(dev)) {
1036 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1037
1038 nd_namespace_pmem_set_resource(nd_region, nspm,
1039 val * nd_region->ndr_mappings);
1040 }
1041
1042 /*
1043 * Try to delete the namespace if we deleted all of its
1044 * allocation, this is not the seed or 0th device for the
1045 * region, and it is not actively claimed by a btt, pfn, or dax
1046 * instance.
1047 */
1048 if (val == 0 && id != 0 && nd_region->ns_seed != dev && !ndns->claim)
1049 nd_device_unregister(dev, ND_ASYNC);
1050
1051 return rc;
1052}
1053
1054static ssize_t size_store(struct device *dev,
1055 struct device_attribute *attr, const char *buf, size_t len)
1056{
1057 struct nd_region *nd_region = to_nd_region(dev->parent);
1058 unsigned long long val;
1059 u8 **uuid = NULL;
1060 int rc;
1061
1062 rc = kstrtoull(buf, 0, &val);
1063 if (rc)
1064 return rc;
1065
1066 device_lock(dev);
1067 nvdimm_bus_lock(dev);
1068 wait_nvdimm_bus_probe_idle(dev);
1069 rc = __size_store(dev, val);
1070 if (rc >= 0)
1071 rc = nd_namespace_label_update(nd_region, dev);
1072
1073 if (is_namespace_pmem(dev)) {
1074 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1075
1076 uuid = &nspm->uuid;
1077 } else if (is_namespace_blk(dev)) {
1078 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1079
1080 uuid = &nsblk->uuid;
1081 }
1082
1083 if (rc == 0 && val == 0 && uuid) {
1084 /* setting size zero == 'delete namespace' */
1085 kfree(*uuid);
1086 *uuid = NULL;
1087 }
1088
1089 dev_dbg(dev, "%s: %llx %s (%d)\n", __func__, val, rc < 0
1090 ? "fail" : "success", rc);
1091
1092 nvdimm_bus_unlock(dev);
1093 device_unlock(dev);
1094
1095 return rc < 0 ? rc : len;
1096}
1097
1098resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1099{
1100 struct device *dev = &ndns->dev;
1101
1102 if (is_namespace_pmem(dev)) {
1103 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1104
1105 return resource_size(&nspm->nsio.res);
1106 } else if (is_namespace_blk(dev)) {
1107 return nd_namespace_blk_size(to_nd_namespace_blk(dev));
1108 } else if (is_namespace_io(dev)) {
1109 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1110
1111 return resource_size(&nsio->res);
1112 } else
1113 WARN_ONCE(1, "unknown namespace type\n");
1114 return 0;
1115}
1116
1117resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1118{
1119 resource_size_t size;
1120
1121 nvdimm_bus_lock(&ndns->dev);
1122 size = __nvdimm_namespace_capacity(ndns);
1123 nvdimm_bus_unlock(&ndns->dev);
1124
1125 return size;
1126}
1127EXPORT_SYMBOL(nvdimm_namespace_capacity);
1128
1129static ssize_t size_show(struct device *dev,
1130 struct device_attribute *attr, char *buf)
1131{
1132 return sprintf(buf, "%llu\n", (unsigned long long)
1133 nvdimm_namespace_capacity(to_ndns(dev)));
1134}
1135static DEVICE_ATTR(size, 0444, size_show, size_store);
1136
1137static u8 *namespace_to_uuid(struct device *dev)
1138{
1139 if (is_namespace_pmem(dev)) {
1140 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1141
1142 return nspm->uuid;
1143 } else if (is_namespace_blk(dev)) {
1144 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1145
1146 return nsblk->uuid;
1147 } else
1148 return ERR_PTR(-ENXIO);
1149}
1150
1151static ssize_t uuid_show(struct device *dev,
1152 struct device_attribute *attr, char *buf)
1153{
1154 u8 *uuid = namespace_to_uuid(dev);
1155
1156 if (IS_ERR(uuid))
1157 return PTR_ERR(uuid);
1158 if (uuid)
1159 return sprintf(buf, "%pUb\n", uuid);
1160 return sprintf(buf, "\n");
1161}
1162
1163/**
1164 * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
1165 * @nd_region: parent region so we can updates all dimms in the set
1166 * @dev: namespace type for generating label_id
1167 * @new_uuid: incoming uuid
1168 * @old_uuid: reference to the uuid storage location in the namespace object
1169 */
1170static int namespace_update_uuid(struct nd_region *nd_region,
1171 struct device *dev, u8 *new_uuid, u8 **old_uuid)
1172{
1173 u32 flags = is_namespace_blk(dev) ? NSLABEL_FLAG_LOCAL : 0;
1174 struct nd_label_id old_label_id;
1175 struct nd_label_id new_label_id;
1176 int i;
1177
1178 if (!nd_is_uuid_unique(dev, new_uuid))
1179 return -EINVAL;
1180
1181 if (*old_uuid == NULL)
1182 goto out;
1183
1184 /*
1185 * If we've already written a label with this uuid, then it's
1186 * too late to rename because we can't reliably update the uuid
1187 * without losing the old namespace. Userspace must delete this
1188 * namespace to abandon the old uuid.
1189 */
1190 for (i = 0; i < nd_region->ndr_mappings; i++) {
1191 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1192
1193 /*
1194 * This check by itself is sufficient because old_uuid
1195 * would be NULL above if this uuid did not exist in the
1196 * currently written set.
1197 *
1198 * FIXME: can we delete uuid with zero dpa allocated?
1199 */
1200 if (list_empty(&nd_mapping->labels))
1201 return -EBUSY;
1202 }
1203
1204 nd_label_gen_id(&old_label_id, *old_uuid, flags);
1205 nd_label_gen_id(&new_label_id, new_uuid, flags);
1206 for (i = 0; i < nd_region->ndr_mappings; i++) {
1207 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1208 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1209 struct resource *res;
1210
1211 for_each_dpa_resource(ndd, res)
1212 if (strcmp(res->name, old_label_id.id) == 0)
1213 sprintf((void *) res->name, "%s",
1214 new_label_id.id);
1215 }
1216 kfree(*old_uuid);
1217 out:
1218 *old_uuid = new_uuid;
1219 return 0;
1220}
1221
1222static ssize_t uuid_store(struct device *dev,
1223 struct device_attribute *attr, const char *buf, size_t len)
1224{
1225 struct nd_region *nd_region = to_nd_region(dev->parent);
1226 u8 *uuid = NULL;
1227 ssize_t rc = 0;
1228 u8 **ns_uuid;
1229
1230 if (is_namespace_pmem(dev)) {
1231 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1232
1233 ns_uuid = &nspm->uuid;
1234 } else if (is_namespace_blk(dev)) {
1235 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1236
1237 ns_uuid = &nsblk->uuid;
1238 } else
1239 return -ENXIO;
1240
1241 device_lock(dev);
1242 nvdimm_bus_lock(dev);
1243 wait_nvdimm_bus_probe_idle(dev);
1244 if (to_ndns(dev)->claim)
1245 rc = -EBUSY;
1246 if (rc >= 0)
1247 rc = nd_uuid_store(dev, &uuid, buf, len);
1248 if (rc >= 0)
1249 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
1250 if (rc >= 0)
1251 rc = nd_namespace_label_update(nd_region, dev);
1252 else
1253 kfree(uuid);
1254 dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
1255 rc, buf, buf[len - 1] == '\n' ? "" : "\n");
1256 nvdimm_bus_unlock(dev);
1257 device_unlock(dev);
1258
1259 return rc < 0 ? rc : len;
1260}
1261static DEVICE_ATTR_RW(uuid);
1262
1263static ssize_t resource_show(struct device *dev,
1264 struct device_attribute *attr, char *buf)
1265{
1266 struct resource *res;
1267
1268 if (is_namespace_pmem(dev)) {
1269 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1270
1271 res = &nspm->nsio.res;
1272 } else if (is_namespace_io(dev)) {
1273 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1274
1275 res = &nsio->res;
1276 } else
1277 return -ENXIO;
1278
1279 /* no address to convey if the namespace has no allocation */
1280 if (resource_size(res) == 0)
1281 return -ENXIO;
1282 return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
1283}
1284static DEVICE_ATTR_RO(resource);
1285
1286static const unsigned long ns_lbasize_supported[] = { 512, 520, 528,
1287 4096, 4104, 4160, 4224, 0 };
1288
1289static ssize_t sector_size_show(struct device *dev,
1290 struct device_attribute *attr, char *buf)
1291{
1292 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1293
1294 if (!is_namespace_blk(dev))
1295 return -ENXIO;
1296
1297 return nd_sector_size_show(nsblk->lbasize, ns_lbasize_supported, buf);
1298}
1299
1300static ssize_t sector_size_store(struct device *dev,
1301 struct device_attribute *attr, const char *buf, size_t len)
1302{
1303 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1304 struct nd_region *nd_region = to_nd_region(dev->parent);
1305 ssize_t rc = 0;
1306
1307 if (!is_namespace_blk(dev))
1308 return -ENXIO;
1309
1310 device_lock(dev);
1311 nvdimm_bus_lock(dev);
1312 if (to_ndns(dev)->claim)
1313 rc = -EBUSY;
1314 if (rc >= 0)
1315 rc = nd_sector_size_store(dev, buf, &nsblk->lbasize,
1316 ns_lbasize_supported);
1317 if (rc >= 0)
1318 rc = nd_namespace_label_update(nd_region, dev);
1319 dev_dbg(dev, "%s: result: %zd %s: %s%s", __func__,
1320 rc, rc < 0 ? "tried" : "wrote", buf,
1321 buf[len - 1] == '\n' ? "" : "\n");
1322 nvdimm_bus_unlock(dev);
1323 device_unlock(dev);
1324
1325 return rc ? rc : len;
1326}
1327static DEVICE_ATTR_RW(sector_size);
1328
1329static ssize_t dpa_extents_show(struct device *dev,
1330 struct device_attribute *attr, char *buf)
1331{
1332 struct nd_region *nd_region = to_nd_region(dev->parent);
1333 struct nd_label_id label_id;
1334 int count = 0, i;
1335 u8 *uuid = NULL;
1336 u32 flags = 0;
1337
1338 nvdimm_bus_lock(dev);
1339 if (is_namespace_pmem(dev)) {
1340 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1341
1342 uuid = nspm->uuid;
1343 flags = 0;
1344 } else if (is_namespace_blk(dev)) {
1345 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1346
1347 uuid = nsblk->uuid;
1348 flags = NSLABEL_FLAG_LOCAL;
1349 }
1350
1351 if (!uuid)
1352 goto out;
1353
1354 nd_label_gen_id(&label_id, uuid, flags);
1355 for (i = 0; i < nd_region->ndr_mappings; i++) {
1356 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1357 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1358 struct resource *res;
1359
1360 for_each_dpa_resource(ndd, res)
1361 if (strcmp(res->name, label_id.id) == 0)
1362 count++;
1363 }
1364 out:
1365 nvdimm_bus_unlock(dev);
1366
1367 return sprintf(buf, "%d\n", count);
1368}
1369static DEVICE_ATTR_RO(dpa_extents);
1370
1371static ssize_t holder_show(struct device *dev,
1372 struct device_attribute *attr, char *buf)
1373{
1374 struct nd_namespace_common *ndns = to_ndns(dev);
1375 ssize_t rc;
1376
1377 device_lock(dev);
1378 rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
1379 device_unlock(dev);
1380
1381 return rc;
1382}
1383static DEVICE_ATTR_RO(holder);
1384
1385static ssize_t mode_show(struct device *dev,
1386 struct device_attribute *attr, char *buf)
1387{
1388 struct nd_namespace_common *ndns = to_ndns(dev);
1389 struct device *claim;
1390 char *mode;
1391 ssize_t rc;
1392
1393 device_lock(dev);
1394 claim = ndns->claim;
1395 if (claim && is_nd_btt(claim))
1396 mode = "safe";
1397 else if (claim && is_nd_pfn(claim))
1398 mode = "memory";
1399 else if (claim && is_nd_dax(claim))
1400 mode = "dax";
1401 else if (!claim && pmem_should_map_pages(dev))
1402 mode = "memory";
1403 else
1404 mode = "raw";
1405 rc = sprintf(buf, "%s\n", mode);
1406 device_unlock(dev);
1407
1408 return rc;
1409}
1410static DEVICE_ATTR_RO(mode);
1411
1412static ssize_t force_raw_store(struct device *dev,
1413 struct device_attribute *attr, const char *buf, size_t len)
1414{
1415 bool force_raw;
1416 int rc = strtobool(buf, &force_raw);
1417
1418 if (rc)
1419 return rc;
1420
1421 to_ndns(dev)->force_raw = force_raw;
1422 return len;
1423}
1424
1425static ssize_t force_raw_show(struct device *dev,
1426 struct device_attribute *attr, char *buf)
1427{
1428 return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
1429}
1430static DEVICE_ATTR_RW(force_raw);
1431
1432static struct attribute *nd_namespace_attributes[] = {
1433 &dev_attr_nstype.attr,
1434 &dev_attr_size.attr,
1435 &dev_attr_mode.attr,
1436 &dev_attr_uuid.attr,
1437 &dev_attr_holder.attr,
1438 &dev_attr_resource.attr,
1439 &dev_attr_alt_name.attr,
1440 &dev_attr_force_raw.attr,
1441 &dev_attr_sector_size.attr,
1442 &dev_attr_dpa_extents.attr,
1443 NULL,
1444};
1445
1446static umode_t namespace_visible(struct kobject *kobj,
1447 struct attribute *a, int n)
1448{
1449 struct device *dev = container_of(kobj, struct device, kobj);
1450
1451 if (a == &dev_attr_resource.attr) {
1452 if (is_namespace_blk(dev))
1453 return 0;
1454 return a->mode;
1455 }
1456
1457 if (is_namespace_pmem(dev) || is_namespace_blk(dev)) {
1458 if (a == &dev_attr_size.attr)
1459 return 0644;
1460
1461 if (is_namespace_pmem(dev) && a == &dev_attr_sector_size.attr)
1462 return 0;
1463
1464 return a->mode;
1465 }
1466
1467 if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr
1468 || a == &dev_attr_holder.attr
1469 || a == &dev_attr_force_raw.attr
1470 || a == &dev_attr_mode.attr)
1471 return a->mode;
1472
1473 return 0;
1474}
1475
1476static struct attribute_group nd_namespace_attribute_group = {
1477 .attrs = nd_namespace_attributes,
1478 .is_visible = namespace_visible,
1479};
1480
1481static const struct attribute_group *nd_namespace_attribute_groups[] = {
1482 &nd_device_attribute_group,
1483 &nd_namespace_attribute_group,
1484 &nd_numa_attribute_group,
1485 NULL,
1486};
1487
1488struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
1489{
1490 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
1491 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
1492 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
1493 struct nd_namespace_common *ndns = NULL;
1494 resource_size_t size;
1495
1496 if (nd_btt || nd_pfn || nd_dax) {
1497 if (nd_btt)
1498 ndns = nd_btt->ndns;
1499 else if (nd_pfn)
1500 ndns = nd_pfn->ndns;
1501 else if (nd_dax)
1502 ndns = nd_dax->nd_pfn.ndns;
1503
1504 if (!ndns)
1505 return ERR_PTR(-ENODEV);
1506
1507 /*
1508 * Flush any in-progess probes / removals in the driver
1509 * for the raw personality of this namespace.
1510 */
1511 device_lock(&ndns->dev);
1512 device_unlock(&ndns->dev);
1513 if (ndns->dev.driver) {
1514 dev_dbg(&ndns->dev, "is active, can't bind %s\n",
1515 dev_name(dev));
1516 return ERR_PTR(-EBUSY);
1517 }
1518 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != dev,
1519 "host (%s) vs claim (%s) mismatch\n",
1520 dev_name(dev),
1521 dev_name(ndns->claim)))
1522 return ERR_PTR(-ENXIO);
1523 } else {
1524 ndns = to_ndns(dev);
1525 if (ndns->claim) {
1526 dev_dbg(dev, "claimed by %s, failing probe\n",
1527 dev_name(ndns->claim));
1528
1529 return ERR_PTR(-ENXIO);
1530 }
1531 }
1532
1533 size = nvdimm_namespace_capacity(ndns);
1534 if (size < ND_MIN_NAMESPACE_SIZE) {
1535 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
1536 &size, ND_MIN_NAMESPACE_SIZE);
1537 return ERR_PTR(-ENODEV);
1538 }
1539
1540 if (is_namespace_pmem(&ndns->dev)) {
1541 struct nd_namespace_pmem *nspm;
1542
1543 nspm = to_nd_namespace_pmem(&ndns->dev);
1544 if (uuid_not_set(nspm->uuid, &ndns->dev, __func__))
1545 return ERR_PTR(-ENODEV);
1546 } else if (is_namespace_blk(&ndns->dev)) {
1547 struct nd_namespace_blk *nsblk;
1548
1549 nsblk = to_nd_namespace_blk(&ndns->dev);
1550 if (uuid_not_set(nsblk->uuid, &ndns->dev, __func__))
1551 return ERR_PTR(-ENODEV);
1552 if (!nsblk->lbasize) {
1553 dev_dbg(&ndns->dev, "%s: sector size not set\n",
1554 __func__);
1555 return ERR_PTR(-ENODEV);
1556 }
1557 if (!nd_namespace_blk_validate(nsblk))
1558 return ERR_PTR(-ENODEV);
1559 }
1560
1561 return ndns;
1562}
1563EXPORT_SYMBOL(nvdimm_namespace_common_probe);
1564
1565static struct device **create_namespace_io(struct nd_region *nd_region)
1566{
1567 struct nd_namespace_io *nsio;
1568 struct device *dev, **devs;
1569 struct resource *res;
1570
1571 nsio = kzalloc(sizeof(*nsio), GFP_KERNEL);
1572 if (!nsio)
1573 return NULL;
1574
1575 devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1576 if (!devs) {
1577 kfree(nsio);
1578 return NULL;
1579 }
1580
1581 dev = &nsio->common.dev;
1582 dev->type = &namespace_io_device_type;
1583 dev->parent = &nd_region->dev;
1584 res = &nsio->res;
1585 res->name = dev_name(&nd_region->dev);
1586 res->flags = IORESOURCE_MEM;
1587 res->start = nd_region->ndr_start;
1588 res->end = res->start + nd_region->ndr_size - 1;
1589
1590 devs[0] = dev;
1591 return devs;
1592}
1593
1594static bool has_uuid_at_pos(struct nd_region *nd_region, u8 *uuid,
1595 u64 cookie, u16 pos)
1596{
1597 struct nd_namespace_label *found = NULL;
1598 int i;
1599
1600 for (i = 0; i < nd_region->ndr_mappings; i++) {
1601 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1602 struct nd_label_ent *label_ent;
1603 bool found_uuid = false;
1604
1605 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1606 struct nd_namespace_label *nd_label = label_ent->label;
1607 u16 position, nlabel;
1608 u64 isetcookie;
1609
1610 if (!nd_label)
1611 continue;
1612 isetcookie = __le64_to_cpu(nd_label->isetcookie);
1613 position = __le16_to_cpu(nd_label->position);
1614 nlabel = __le16_to_cpu(nd_label->nlabel);
1615
1616 if (isetcookie != cookie)
1617 continue;
1618
1619 if (memcmp(nd_label->uuid, uuid, NSLABEL_UUID_LEN) != 0)
1620 continue;
1621
1622 if (found_uuid) {
1623 dev_dbg(to_ndd(nd_mapping)->dev,
1624 "%s duplicate entry for uuid\n",
1625 __func__);
1626 return false;
1627 }
1628 found_uuid = true;
1629 if (nlabel != nd_region->ndr_mappings)
1630 continue;
1631 if (position != pos)
1632 continue;
1633 found = nd_label;
1634 break;
1635 }
1636 if (found)
1637 break;
1638 }
1639 return found != NULL;
1640}
1641
1642static int select_pmem_id(struct nd_region *nd_region, u8 *pmem_id)
1643{
1644 int i;
1645
1646 if (!pmem_id)
1647 return -ENODEV;
1648
1649 for (i = 0; i < nd_region->ndr_mappings; i++) {
1650 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1651 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1652 struct nd_namespace_label *nd_label = NULL;
1653 u64 hw_start, hw_end, pmem_start, pmem_end;
1654 struct nd_label_ent *label_ent;
1655
1656 lockdep_assert_held(&nd_mapping->lock);
1657 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1658 nd_label = label_ent->label;
1659 if (!nd_label)
1660 continue;
1661 if (memcmp(nd_label->uuid, pmem_id, NSLABEL_UUID_LEN) == 0)
1662 break;
1663 nd_label = NULL;
1664 }
1665
1666 if (!nd_label) {
1667 WARN_ON(1);
1668 return -EINVAL;
1669 }
1670
1671 /*
1672 * Check that this label is compliant with the dpa
1673 * range published in NFIT
1674 */
1675 hw_start = nd_mapping->start;
1676 hw_end = hw_start + nd_mapping->size;
1677 pmem_start = __le64_to_cpu(nd_label->dpa);
1678 pmem_end = pmem_start + __le64_to_cpu(nd_label->rawsize);
1679 if (pmem_start >= hw_start && pmem_start < hw_end
1680 && pmem_end <= hw_end && pmem_end > hw_start)
1681 /* pass */;
1682 else {
1683 dev_dbg(&nd_region->dev, "%s invalid label for %pUb\n",
1684 dev_name(ndd->dev), nd_label->uuid);
1685 return -EINVAL;
1686 }
1687
1688 /* move recently validated label to the front of the list */
1689 list_move(&label_ent->list, &nd_mapping->labels);
1690 }
1691 return 0;
1692}
1693
1694/**
1695 * create_namespace_pmem - validate interleave set labelling, retrieve label0
1696 * @nd_region: region with mappings to validate
1697 * @nspm: target namespace to create
1698 * @nd_label: target pmem namespace label to evaluate
1699 */
1700struct device *create_namespace_pmem(struct nd_region *nd_region,
1701 struct nd_namespace_label *nd_label)
1702{
1703 u64 altcookie = nd_region_interleave_set_altcookie(nd_region);
1704 u64 cookie = nd_region_interleave_set_cookie(nd_region);
1705 struct nd_label_ent *label_ent;
1706 struct nd_namespace_pmem *nspm;
1707 struct nd_mapping *nd_mapping;
1708 resource_size_t size = 0;
1709 struct resource *res;
1710 struct device *dev;
1711 int rc = 0;
1712 u16 i;
1713
1714 if (cookie == 0) {
1715 dev_dbg(&nd_region->dev, "invalid interleave-set-cookie\n");
1716 return ERR_PTR(-ENXIO);
1717 }
1718
1719 if (__le64_to_cpu(nd_label->isetcookie) != cookie) {
1720 dev_dbg(&nd_region->dev, "invalid cookie in label: %pUb\n",
1721 nd_label->uuid);
1722 if (__le64_to_cpu(nd_label->isetcookie) != altcookie)
1723 return ERR_PTR(-EAGAIN);
1724
1725 dev_dbg(&nd_region->dev, "valid altcookie in label: %pUb\n",
1726 nd_label->uuid);
1727 }
1728
1729 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1730 if (!nspm)
1731 return ERR_PTR(-ENOMEM);
1732
1733 nspm->id = -1;
1734 dev = &nspm->nsio.common.dev;
1735 dev->type = &namespace_pmem_device_type;
1736 dev->parent = &nd_region->dev;
1737 res = &nspm->nsio.res;
1738 res->name = dev_name(&nd_region->dev);
1739 res->flags = IORESOURCE_MEM;
1740
1741 for (i = 0; i < nd_region->ndr_mappings; i++) {
1742 if (has_uuid_at_pos(nd_region, nd_label->uuid, cookie, i))
1743 continue;
1744 if (has_uuid_at_pos(nd_region, nd_label->uuid, altcookie, i))
1745 continue;
1746 break;
1747 }
1748
1749 if (i < nd_region->ndr_mappings) {
1750 struct nvdimm_drvdata *ndd = to_ndd(&nd_region->mapping[i]);
1751
1752 /*
1753 * Give up if we don't find an instance of a uuid at each
1754 * position (from 0 to nd_region->ndr_mappings - 1), or if we
1755 * find a dimm with two instances of the same uuid.
1756 */
1757 dev_err(&nd_region->dev, "%s missing label for %pUb\n",
1758 dev_name(ndd->dev), nd_label->uuid);
1759 rc = -EINVAL;
1760 goto err;
1761 }
1762
1763 /*
1764 * Fix up each mapping's 'labels' to have the validated pmem label for
1765 * that position at labels[0], and NULL at labels[1]. In the process,
1766 * check that the namespace aligns with interleave-set. We know
1767 * that it does not overlap with any blk namespaces by virtue of
1768 * the dimm being enabled (i.e. nd_label_reserve_dpa()
1769 * succeeded).
1770 */
1771 rc = select_pmem_id(nd_region, nd_label->uuid);
1772 if (rc)
1773 goto err;
1774
1775 /* Calculate total size and populate namespace properties from label0 */
1776 for (i = 0; i < nd_region->ndr_mappings; i++) {
1777 struct nd_namespace_label *label0;
1778
1779 nd_mapping = &nd_region->mapping[i];
1780 label_ent = list_first_entry_or_null(&nd_mapping->labels,
1781 typeof(*label_ent), list);
1782 label0 = label_ent ? label_ent->label : 0;
1783
1784 if (!label0) {
1785 WARN_ON(1);
1786 continue;
1787 }
1788
1789 size += __le64_to_cpu(label0->rawsize);
1790 if (__le16_to_cpu(label0->position) != 0)
1791 continue;
1792 WARN_ON(nspm->alt_name || nspm->uuid);
1793 nspm->alt_name = kmemdup((void __force *) label0->name,
1794 NSLABEL_NAME_LEN, GFP_KERNEL);
1795 nspm->uuid = kmemdup((void __force *) label0->uuid,
1796 NSLABEL_UUID_LEN, GFP_KERNEL);
1797 }
1798
1799 if (!nspm->alt_name || !nspm->uuid) {
1800 rc = -ENOMEM;
1801 goto err;
1802 }
1803
1804 nd_namespace_pmem_set_resource(nd_region, nspm, size);
1805
1806 return dev;
1807 err:
1808 namespace_pmem_release(dev);
1809 switch (rc) {
1810 case -EINVAL:
1811 dev_dbg(&nd_region->dev, "%s: invalid label(s)\n", __func__);
1812 break;
1813 case -ENODEV:
1814 dev_dbg(&nd_region->dev, "%s: label not found\n", __func__);
1815 break;
1816 default:
1817 dev_dbg(&nd_region->dev, "%s: unexpected err: %d\n",
1818 __func__, rc);
1819 break;
1820 }
1821 return ERR_PTR(rc);
1822}
1823
1824struct resource *nsblk_add_resource(struct nd_region *nd_region,
1825 struct nvdimm_drvdata *ndd, struct nd_namespace_blk *nsblk,
1826 resource_size_t start)
1827{
1828 struct nd_label_id label_id;
1829 struct resource *res;
1830
1831 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
1832 res = krealloc(nsblk->res,
1833 sizeof(void *) * (nsblk->num_resources + 1),
1834 GFP_KERNEL);
1835 if (!res)
1836 return NULL;
1837 nsblk->res = (struct resource **) res;
1838 for_each_dpa_resource(ndd, res)
1839 if (strcmp(res->name, label_id.id) == 0
1840 && res->start == start) {
1841 nsblk->res[nsblk->num_resources++] = res;
1842 return res;
1843 }
1844 return NULL;
1845}
1846
1847static struct device *nd_namespace_blk_create(struct nd_region *nd_region)
1848{
1849 struct nd_namespace_blk *nsblk;
1850 struct device *dev;
1851
1852 if (!is_nd_blk(&nd_region->dev))
1853 return NULL;
1854
1855 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
1856 if (!nsblk)
1857 return NULL;
1858
1859 dev = &nsblk->common.dev;
1860 dev->type = &namespace_blk_device_type;
1861 nsblk->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
1862 if (nsblk->id < 0) {
1863 kfree(nsblk);
1864 return NULL;
1865 }
1866 dev_set_name(dev, "namespace%d.%d", nd_region->id, nsblk->id);
1867 dev->parent = &nd_region->dev;
1868 dev->groups = nd_namespace_attribute_groups;
1869
1870 return &nsblk->common.dev;
1871}
1872
1873static struct device *nd_namespace_pmem_create(struct nd_region *nd_region)
1874{
1875 struct nd_namespace_pmem *nspm;
1876 struct resource *res;
1877 struct device *dev;
1878
1879 if (!is_nd_pmem(&nd_region->dev))
1880 return NULL;
1881
1882 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1883 if (!nspm)
1884 return NULL;
1885
1886 dev = &nspm->nsio.common.dev;
1887 dev->type = &namespace_pmem_device_type;
1888 dev->parent = &nd_region->dev;
1889 res = &nspm->nsio.res;
1890 res->name = dev_name(&nd_region->dev);
1891 res->flags = IORESOURCE_MEM;
1892
1893 nspm->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
1894 if (nspm->id < 0) {
1895 kfree(nspm);
1896 return NULL;
1897 }
1898 dev_set_name(dev, "namespace%d.%d", nd_region->id, nspm->id);
1899 dev->parent = &nd_region->dev;
1900 dev->groups = nd_namespace_attribute_groups;
1901 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
1902
1903 return dev;
1904}
1905
1906void nd_region_create_ns_seed(struct nd_region *nd_region)
1907{
1908 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1909
1910 if (nd_region_to_nstype(nd_region) == ND_DEVICE_NAMESPACE_IO)
1911 return;
1912
1913 if (is_nd_blk(&nd_region->dev))
1914 nd_region->ns_seed = nd_namespace_blk_create(nd_region);
1915 else
1916 nd_region->ns_seed = nd_namespace_pmem_create(nd_region);
1917
1918 /*
1919 * Seed creation failures are not fatal, provisioning is simply
1920 * disabled until memory becomes available
1921 */
1922 if (!nd_region->ns_seed)
1923 dev_err(&nd_region->dev, "failed to create %s namespace\n",
1924 is_nd_blk(&nd_region->dev) ? "blk" : "pmem");
1925 else
1926 nd_device_register(nd_region->ns_seed);
1927}
1928
1929void nd_region_create_dax_seed(struct nd_region *nd_region)
1930{
1931 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1932 nd_region->dax_seed = nd_dax_create(nd_region);
1933 /*
1934 * Seed creation failures are not fatal, provisioning is simply
1935 * disabled until memory becomes available
1936 */
1937 if (!nd_region->dax_seed)
1938 dev_err(&nd_region->dev, "failed to create dax namespace\n");
1939}
1940
1941void nd_region_create_pfn_seed(struct nd_region *nd_region)
1942{
1943 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1944 nd_region->pfn_seed = nd_pfn_create(nd_region);
1945 /*
1946 * Seed creation failures are not fatal, provisioning is simply
1947 * disabled until memory becomes available
1948 */
1949 if (!nd_region->pfn_seed)
1950 dev_err(&nd_region->dev, "failed to create pfn namespace\n");
1951}
1952
1953void nd_region_create_btt_seed(struct nd_region *nd_region)
1954{
1955 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1956 nd_region->btt_seed = nd_btt_create(nd_region);
1957 /*
1958 * Seed creation failures are not fatal, provisioning is simply
1959 * disabled until memory becomes available
1960 */
1961 if (!nd_region->btt_seed)
1962 dev_err(&nd_region->dev, "failed to create btt namespace\n");
1963}
1964
1965static int add_namespace_resource(struct nd_region *nd_region,
1966 struct nd_namespace_label *nd_label, struct device **devs,
1967 int count)
1968{
1969 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
1970 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1971 int i;
1972
1973 for (i = 0; i < count; i++) {
1974 u8 *uuid = namespace_to_uuid(devs[i]);
1975 struct resource *res;
1976
1977 if (IS_ERR_OR_NULL(uuid)) {
1978 WARN_ON(1);
1979 continue;
1980 }
1981
1982 if (memcmp(uuid, nd_label->uuid, NSLABEL_UUID_LEN) != 0)
1983 continue;
1984 if (is_namespace_blk(devs[i])) {
1985 res = nsblk_add_resource(nd_region, ndd,
1986 to_nd_namespace_blk(devs[i]),
1987 __le64_to_cpu(nd_label->dpa));
1988 if (!res)
1989 return -ENXIO;
1990 nd_dbg_dpa(nd_region, ndd, res, "%d assign\n", count);
1991 } else {
1992 dev_err(&nd_region->dev,
1993 "error: conflicting extents for uuid: %pUb\n",
1994 nd_label->uuid);
1995 return -ENXIO;
1996 }
1997 break;
1998 }
1999
2000 return i;
2001}
2002
2003struct device *create_namespace_blk(struct nd_region *nd_region,
2004 struct nd_namespace_label *nd_label, int count)
2005{
2006
2007 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2008 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2009 struct nd_namespace_blk *nsblk;
2010 char name[NSLABEL_NAME_LEN];
2011 struct device *dev = NULL;
2012 struct resource *res;
2013
2014 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2015 if (!nsblk)
2016 return ERR_PTR(-ENOMEM);
2017 dev = &nsblk->common.dev;
2018 dev->type = &namespace_blk_device_type;
2019 dev->parent = &nd_region->dev;
2020 nsblk->id = -1;
2021 nsblk->lbasize = __le64_to_cpu(nd_label->lbasize);
2022 nsblk->uuid = kmemdup(nd_label->uuid, NSLABEL_UUID_LEN,
2023 GFP_KERNEL);
2024 if (!nsblk->uuid)
2025 goto blk_err;
2026 memcpy(name, nd_label->name, NSLABEL_NAME_LEN);
2027 if (name[0])
2028 nsblk->alt_name = kmemdup(name, NSLABEL_NAME_LEN,
2029 GFP_KERNEL);
2030 res = nsblk_add_resource(nd_region, ndd, nsblk,
2031 __le64_to_cpu(nd_label->dpa));
2032 if (!res)
2033 goto blk_err;
2034 nd_dbg_dpa(nd_region, ndd, res, "%d: assign\n", count);
2035 return dev;
2036 blk_err:
2037 namespace_blk_release(dev);
2038 return ERR_PTR(-ENXIO);
2039}
2040
2041static int cmp_dpa(const void *a, const void *b)
2042{
2043 const struct device *dev_a = *(const struct device **) a;
2044 const struct device *dev_b = *(const struct device **) b;
2045 struct nd_namespace_blk *nsblk_a, *nsblk_b;
2046 struct nd_namespace_pmem *nspm_a, *nspm_b;
2047
2048 if (is_namespace_io(dev_a))
2049 return 0;
2050
2051 if (is_namespace_blk(dev_a)) {
2052 nsblk_a = to_nd_namespace_blk(dev_a);
2053 nsblk_b = to_nd_namespace_blk(dev_b);
2054
2055 return memcmp(&nsblk_a->res[0]->start, &nsblk_b->res[0]->start,
2056 sizeof(resource_size_t));
2057 }
2058
2059 nspm_a = to_nd_namespace_pmem(dev_a);
2060 nspm_b = to_nd_namespace_pmem(dev_b);
2061
2062 return memcmp(&nspm_a->nsio.res.start, &nspm_b->nsio.res.start,
2063 sizeof(resource_size_t));
2064}
2065
2066static struct device **scan_labels(struct nd_region *nd_region)
2067{
2068 int i, count = 0;
2069 struct device *dev, **devs = NULL;
2070 struct nd_label_ent *label_ent, *e;
2071 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2072 resource_size_t map_end = nd_mapping->start + nd_mapping->size - 1;
2073
2074 /* "safe" because create_namespace_pmem() might list_move() label_ent */
2075 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
2076 struct nd_namespace_label *nd_label = label_ent->label;
2077 struct device **__devs;
2078 u32 flags;
2079
2080 if (!nd_label)
2081 continue;
2082 flags = __le32_to_cpu(nd_label->flags);
2083 if (is_nd_blk(&nd_region->dev)
2084 == !!(flags & NSLABEL_FLAG_LOCAL))
2085 /* pass, region matches label type */;
2086 else
2087 continue;
2088
2089 /* skip labels that describe extents outside of the region */
2090 if (nd_label->dpa < nd_mapping->start || nd_label->dpa > map_end)
2091 continue;
2092
2093 i = add_namespace_resource(nd_region, nd_label, devs, count);
2094 if (i < 0)
2095 goto err;
2096 if (i < count)
2097 continue;
2098 __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
2099 if (!__devs)
2100 goto err;
2101 memcpy(__devs, devs, sizeof(dev) * count);
2102 kfree(devs);
2103 devs = __devs;
2104
2105 if (is_nd_blk(&nd_region->dev)) {
2106 dev = create_namespace_blk(nd_region, nd_label, count);
2107 if (IS_ERR(dev))
2108 goto err;
2109 devs[count++] = dev;
2110 } else {
2111 dev = create_namespace_pmem(nd_region, nd_label);
2112 if (IS_ERR(dev)) {
2113 switch (PTR_ERR(dev)) {
2114 case -EAGAIN:
2115 /* skip invalid labels */
2116 continue;
2117 case -ENODEV:
2118 /* fallthrough to seed creation */
2119 break;
2120 default:
2121 goto err;
2122 }
2123 } else
2124 devs[count++] = dev;
2125 }
2126 }
2127
2128 dev_dbg(&nd_region->dev, "%s: discovered %d %s namespace%s\n",
2129 __func__, count, is_nd_blk(&nd_region->dev)
2130 ? "blk" : "pmem", count == 1 ? "" : "s");
2131
2132 if (count == 0) {
2133 /* Publish a zero-sized namespace for userspace to configure. */
2134 nd_mapping_free_labels(nd_mapping);
2135
2136 devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
2137 if (!devs)
2138 goto err;
2139 if (is_nd_blk(&nd_region->dev)) {
2140 struct nd_namespace_blk *nsblk;
2141
2142 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
2143 if (!nsblk)
2144 goto err;
2145 dev = &nsblk->common.dev;
2146 dev->type = &namespace_blk_device_type;
2147 } else {
2148 struct nd_namespace_pmem *nspm;
2149
2150 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2151 if (!nspm)
2152 goto err;
2153 dev = &nspm->nsio.common.dev;
2154 dev->type = &namespace_pmem_device_type;
2155 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2156 }
2157 dev->parent = &nd_region->dev;
2158 devs[count++] = dev;
2159 } else if (is_nd_pmem(&nd_region->dev)) {
2160 /* clean unselected labels */
2161 for (i = 0; i < nd_region->ndr_mappings; i++) {
2162 struct list_head *l, *e;
2163 LIST_HEAD(list);
2164 int j;
2165
2166 nd_mapping = &nd_region->mapping[i];
2167 if (list_empty(&nd_mapping->labels)) {
2168 WARN_ON(1);
2169 continue;
2170 }
2171
2172 j = count;
2173 list_for_each_safe(l, e, &nd_mapping->labels) {
2174 if (!j--)
2175 break;
2176 list_move_tail(l, &list);
2177 }
2178 nd_mapping_free_labels(nd_mapping);
2179 list_splice_init(&list, &nd_mapping->labels);
2180 }
2181 }
2182
2183 if (count > 1)
2184 sort(devs, count, sizeof(struct device *), cmp_dpa, NULL);
2185
2186 return devs;
2187
2188 err:
2189 if (devs) {
2190 for (i = 0; devs[i]; i++)
2191 if (is_nd_blk(&nd_region->dev))
2192 namespace_blk_release(devs[i]);
2193 else
2194 namespace_pmem_release(devs[i]);
2195 kfree(devs);
2196 }
2197 return NULL;
2198}
2199
2200static struct device **create_namespaces(struct nd_region *nd_region)
2201{
2202 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
2203 struct device **devs;
2204 int i;
2205
2206 if (nd_region->ndr_mappings == 0)
2207 return NULL;
2208
2209 /* lock down all mappings while we scan labels */
2210 for (i = 0; i < nd_region->ndr_mappings; i++) {
2211 nd_mapping = &nd_region->mapping[i];
2212 mutex_lock_nested(&nd_mapping->lock, i);
2213 }
2214
2215 devs = scan_labels(nd_region);
2216
2217 for (i = 0; i < nd_region->ndr_mappings; i++) {
2218 int reverse = nd_region->ndr_mappings - 1 - i;
2219
2220 nd_mapping = &nd_region->mapping[reverse];
2221 mutex_unlock(&nd_mapping->lock);
2222 }
2223
2224 return devs;
2225}
2226
2227static int init_active_labels(struct nd_region *nd_region)
2228{
2229 int i;
2230
2231 for (i = 0; i < nd_region->ndr_mappings; i++) {
2232 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2233 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2234 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2235 struct nd_label_ent *label_ent;
2236 int count, j;
2237
2238 /*
2239 * If the dimm is disabled then prevent the region from
2240 * being activated if it aliases DPA.
2241 */
2242 if (!ndd) {
2243 if ((nvdimm->flags & NDD_ALIASING) == 0)
2244 return 0;
2245 dev_dbg(&nd_region->dev, "%s: is disabled, failing probe\n",
2246 dev_name(&nd_mapping->nvdimm->dev));
2247 return -ENXIO;
2248 }
2249 nd_mapping->ndd = ndd;
2250 atomic_inc(&nvdimm->busy);
2251 get_ndd(ndd);
2252
2253 count = nd_label_active_count(ndd);
2254 dev_dbg(ndd->dev, "%s: %d\n", __func__, count);
2255 if (!count)
2256 continue;
2257 for (j = 0; j < count; j++) {
2258 struct nd_namespace_label *label;
2259
2260 label_ent = kzalloc(sizeof(*label_ent), GFP_KERNEL);
2261 if (!label_ent)
2262 break;
2263 label = nd_label_active(ndd, j);
2264 label_ent->label = label;
2265
2266 mutex_lock(&nd_mapping->lock);
2267 list_add_tail(&label_ent->list, &nd_mapping->labels);
2268 mutex_unlock(&nd_mapping->lock);
2269 }
2270
2271 if (j >= count)
2272 continue;
2273
2274 mutex_lock(&nd_mapping->lock);
2275 nd_mapping_free_labels(nd_mapping);
2276 mutex_unlock(&nd_mapping->lock);
2277 return -ENOMEM;
2278 }
2279
2280 return 0;
2281}
2282
2283int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
2284{
2285 struct device **devs = NULL;
2286 int i, rc = 0, type;
2287
2288 *err = 0;
2289 nvdimm_bus_lock(&nd_region->dev);
2290 rc = init_active_labels(nd_region);
2291 if (rc) {
2292 nvdimm_bus_unlock(&nd_region->dev);
2293 return rc;
2294 }
2295
2296 type = nd_region_to_nstype(nd_region);
2297 switch (type) {
2298 case ND_DEVICE_NAMESPACE_IO:
2299 devs = create_namespace_io(nd_region);
2300 break;
2301 case ND_DEVICE_NAMESPACE_PMEM:
2302 case ND_DEVICE_NAMESPACE_BLK:
2303 devs = create_namespaces(nd_region);
2304 break;
2305 default:
2306 break;
2307 }
2308 nvdimm_bus_unlock(&nd_region->dev);
2309
2310 if (!devs)
2311 return -ENODEV;
2312
2313 for (i = 0; devs[i]; i++) {
2314 struct device *dev = devs[i];
2315 int id;
2316
2317 if (type == ND_DEVICE_NAMESPACE_BLK) {
2318 struct nd_namespace_blk *nsblk;
2319
2320 nsblk = to_nd_namespace_blk(dev);
2321 id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2322 GFP_KERNEL);
2323 nsblk->id = id;
2324 } else if (type == ND_DEVICE_NAMESPACE_PMEM) {
2325 struct nd_namespace_pmem *nspm;
2326
2327 nspm = to_nd_namespace_pmem(dev);
2328 id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2329 GFP_KERNEL);
2330 nspm->id = id;
2331 } else
2332 id = i;
2333
2334 if (id < 0)
2335 break;
2336 dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
2337 dev->groups = nd_namespace_attribute_groups;
2338 nd_device_register(dev);
2339 }
2340 if (i)
2341 nd_region->ns_seed = devs[0];
2342
2343 if (devs[i]) {
2344 int j;
2345
2346 for (j = i; devs[j]; j++) {
2347 struct device *dev = devs[j];
2348
2349 device_initialize(dev);
2350 put_device(dev);
2351 }
2352 *err = j - i;
2353 /*
2354 * All of the namespaces we tried to register failed, so
2355 * fail region activation.
2356 */
2357 if (*err == 0)
2358 rc = -ENODEV;
2359 }
2360 kfree(devs);
2361
2362 if (rc == -ENODEV)
2363 return rc;
2364
2365 return i;
2366}
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
4 */
5#include <linux/kstrtox.h>
6#include <linux/module.h>
7#include <linux/device.h>
8#include <linux/sort.h>
9#include <linux/slab.h>
10#include <linux/list.h>
11#include <linux/nd.h>
12#include "nd-core.h"
13#include "pmem.h"
14#include "pfn.h"
15#include "nd.h"
16
17static void namespace_io_release(struct device *dev)
18{
19 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
20
21 kfree(nsio);
22}
23
24static void namespace_pmem_release(struct device *dev)
25{
26 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
27 struct nd_region *nd_region = to_nd_region(dev->parent);
28
29 if (nspm->id >= 0)
30 ida_free(&nd_region->ns_ida, nspm->id);
31 kfree(nspm->alt_name);
32 kfree(nspm->uuid);
33 kfree(nspm);
34}
35
36static bool is_namespace_pmem(const struct device *dev);
37static bool is_namespace_io(const struct device *dev);
38
39static int is_uuid_busy(struct device *dev, void *data)
40{
41 uuid_t *uuid1 = data, *uuid2 = NULL;
42
43 if (is_namespace_pmem(dev)) {
44 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
45
46 uuid2 = nspm->uuid;
47 } else if (is_nd_btt(dev)) {
48 struct nd_btt *nd_btt = to_nd_btt(dev);
49
50 uuid2 = nd_btt->uuid;
51 } else if (is_nd_pfn(dev)) {
52 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
53
54 uuid2 = nd_pfn->uuid;
55 }
56
57 if (uuid2 && uuid_equal(uuid1, uuid2))
58 return -EBUSY;
59
60 return 0;
61}
62
63static int is_namespace_uuid_busy(struct device *dev, void *data)
64{
65 if (is_nd_region(dev))
66 return device_for_each_child(dev, data, is_uuid_busy);
67 return 0;
68}
69
70/**
71 * nd_is_uuid_unique - verify that no other namespace has @uuid
72 * @dev: any device on a nvdimm_bus
73 * @uuid: uuid to check
74 *
75 * Returns: %true if the uuid is unique, %false if not
76 */
77bool nd_is_uuid_unique(struct device *dev, uuid_t *uuid)
78{
79 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
80
81 if (!nvdimm_bus)
82 return false;
83 WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm_bus->dev));
84 if (device_for_each_child(&nvdimm_bus->dev, uuid,
85 is_namespace_uuid_busy) != 0)
86 return false;
87 return true;
88}
89
90bool pmem_should_map_pages(struct device *dev)
91{
92 struct nd_region *nd_region = to_nd_region(dev->parent);
93 struct nd_namespace_common *ndns = to_ndns(dev);
94 struct nd_namespace_io *nsio;
95
96 if (!IS_ENABLED(CONFIG_ZONE_DEVICE))
97 return false;
98
99 if (!test_bit(ND_REGION_PAGEMAP, &nd_region->flags))
100 return false;
101
102 if (is_nd_pfn(dev) || is_nd_btt(dev))
103 return false;
104
105 if (ndns->force_raw)
106 return false;
107
108 nsio = to_nd_namespace_io(dev);
109 if (region_intersects(nsio->res.start, resource_size(&nsio->res),
110 IORESOURCE_SYSTEM_RAM,
111 IORES_DESC_NONE) == REGION_MIXED)
112 return false;
113
114 return ARCH_MEMREMAP_PMEM == MEMREMAP_WB;
115}
116EXPORT_SYMBOL(pmem_should_map_pages);
117
118unsigned int pmem_sector_size(struct nd_namespace_common *ndns)
119{
120 if (is_namespace_pmem(&ndns->dev)) {
121 struct nd_namespace_pmem *nspm;
122
123 nspm = to_nd_namespace_pmem(&ndns->dev);
124 if (nspm->lbasize == 0 || nspm->lbasize == 512)
125 /* default */;
126 else if (nspm->lbasize == 4096)
127 return 4096;
128 else
129 dev_WARN(&ndns->dev, "unsupported sector size: %ld\n",
130 nspm->lbasize);
131 }
132
133 /*
134 * There is no namespace label (is_namespace_io()), or the label
135 * indicates the default sector size.
136 */
137 return 512;
138}
139EXPORT_SYMBOL(pmem_sector_size);
140
141const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,
142 char *name)
143{
144 struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
145 const char *suffix = NULL;
146
147 if (ndns->claim && is_nd_btt(ndns->claim))
148 suffix = "s";
149
150 if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev)) {
151 int nsidx = 0;
152
153 if (is_namespace_pmem(&ndns->dev)) {
154 struct nd_namespace_pmem *nspm;
155
156 nspm = to_nd_namespace_pmem(&ndns->dev);
157 nsidx = nspm->id;
158 }
159
160 if (nsidx)
161 sprintf(name, "pmem%d.%d%s", nd_region->id, nsidx,
162 suffix ? suffix : "");
163 else
164 sprintf(name, "pmem%d%s", nd_region->id,
165 suffix ? suffix : "");
166 } else {
167 return NULL;
168 }
169
170 return name;
171}
172EXPORT_SYMBOL(nvdimm_namespace_disk_name);
173
174const uuid_t *nd_dev_to_uuid(struct device *dev)
175{
176 if (dev && is_namespace_pmem(dev)) {
177 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
178
179 return nspm->uuid;
180 }
181 return &uuid_null;
182}
183EXPORT_SYMBOL(nd_dev_to_uuid);
184
185static ssize_t nstype_show(struct device *dev,
186 struct device_attribute *attr, char *buf)
187{
188 struct nd_region *nd_region = to_nd_region(dev->parent);
189
190 return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
191}
192static DEVICE_ATTR_RO(nstype);
193
194static ssize_t __alt_name_store(struct device *dev, const char *buf,
195 const size_t len)
196{
197 char *input, *pos, *alt_name, **ns_altname;
198 ssize_t rc;
199
200 if (is_namespace_pmem(dev)) {
201 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
202
203 ns_altname = &nspm->alt_name;
204 } else
205 return -ENXIO;
206
207 if (dev->driver || to_ndns(dev)->claim)
208 return -EBUSY;
209
210 input = kstrndup(buf, len, GFP_KERNEL);
211 if (!input)
212 return -ENOMEM;
213
214 pos = strim(input);
215 if (strlen(pos) + 1 > NSLABEL_NAME_LEN) {
216 rc = -EINVAL;
217 goto out;
218 }
219
220 alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL);
221 if (!alt_name) {
222 rc = -ENOMEM;
223 goto out;
224 }
225 kfree(*ns_altname);
226 *ns_altname = alt_name;
227 sprintf(*ns_altname, "%s", pos);
228 rc = len;
229
230out:
231 kfree(input);
232 return rc;
233}
234
235static int nd_namespace_label_update(struct nd_region *nd_region,
236 struct device *dev)
237{
238 dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim,
239 "namespace must be idle during label update\n");
240 if (dev->driver || to_ndns(dev)->claim)
241 return 0;
242
243 /*
244 * Only allow label writes that will result in a valid namespace
245 * or deletion of an existing namespace.
246 */
247 if (is_namespace_pmem(dev)) {
248 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
249 resource_size_t size = resource_size(&nspm->nsio.res);
250
251 if (size == 0 && nspm->uuid)
252 /* delete allocation */;
253 else if (!nspm->uuid)
254 return 0;
255
256 return nd_pmem_namespace_label_update(nd_region, nspm, size);
257 } else
258 return -ENXIO;
259}
260
261static ssize_t alt_name_store(struct device *dev,
262 struct device_attribute *attr, const char *buf, size_t len)
263{
264 struct nd_region *nd_region = to_nd_region(dev->parent);
265 ssize_t rc;
266
267 device_lock(dev);
268 nvdimm_bus_lock(dev);
269 wait_nvdimm_bus_probe_idle(dev);
270 rc = __alt_name_store(dev, buf, len);
271 if (rc >= 0)
272 rc = nd_namespace_label_update(nd_region, dev);
273 dev_dbg(dev, "%s(%zd)\n", rc < 0 ? "fail " : "", rc);
274 nvdimm_bus_unlock(dev);
275 device_unlock(dev);
276
277 return rc < 0 ? rc : len;
278}
279
280static ssize_t alt_name_show(struct device *dev,
281 struct device_attribute *attr, char *buf)
282{
283 char *ns_altname;
284
285 if (is_namespace_pmem(dev)) {
286 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
287
288 ns_altname = nspm->alt_name;
289 } else
290 return -ENXIO;
291
292 return sprintf(buf, "%s\n", ns_altname ? ns_altname : "");
293}
294static DEVICE_ATTR_RW(alt_name);
295
296static int scan_free(struct nd_region *nd_region,
297 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
298 resource_size_t n)
299{
300 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
301 int rc = 0;
302
303 while (n) {
304 struct resource *res, *last;
305
306 last = NULL;
307 for_each_dpa_resource(ndd, res)
308 if (strcmp(res->name, label_id->id) == 0)
309 last = res;
310 res = last;
311 if (!res)
312 return 0;
313
314 if (n >= resource_size(res)) {
315 n -= resource_size(res);
316 nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc);
317 nvdimm_free_dpa(ndd, res);
318 /* retry with last resource deleted */
319 continue;
320 }
321
322 rc = adjust_resource(res, res->start, resource_size(res) - n);
323 if (rc == 0)
324 res->flags |= DPA_RESOURCE_ADJUSTED;
325 nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc);
326 break;
327 }
328
329 return rc;
330}
331
332/**
333 * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
334 * @nd_region: the set of dimms to reclaim @n bytes from
335 * @label_id: unique identifier for the namespace consuming this dpa range
336 * @n: number of bytes per-dimm to release
337 *
338 * Assumes resources are ordered. Starting from the end try to
339 * adjust_resource() the allocation to @n, but if @n is larger than the
340 * allocation delete it and find the 'new' last allocation in the label
341 * set.
342 *
343 * Returns: %0 on success on -errno on error
344 */
345static int shrink_dpa_allocation(struct nd_region *nd_region,
346 struct nd_label_id *label_id, resource_size_t n)
347{
348 int i;
349
350 for (i = 0; i < nd_region->ndr_mappings; i++) {
351 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
352 int rc;
353
354 rc = scan_free(nd_region, nd_mapping, label_id, n);
355 if (rc)
356 return rc;
357 }
358
359 return 0;
360}
361
362static resource_size_t init_dpa_allocation(struct nd_label_id *label_id,
363 struct nd_region *nd_region, struct nd_mapping *nd_mapping,
364 resource_size_t n)
365{
366 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
367 struct resource *res;
368 int rc = 0;
369
370 /* first resource allocation for this label-id or dimm */
371 res = nvdimm_allocate_dpa(ndd, label_id, nd_mapping->start, n);
372 if (!res)
373 rc = -EBUSY;
374
375 nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc);
376 return rc ? n : 0;
377}
378
379
380/**
381 * space_valid() - validate free dpa space against constraints
382 * @nd_region: hosting region of the free space
383 * @ndd: dimm device data for debug
384 * @label_id: namespace id to allocate space
385 * @prev: potential allocation that precedes free space
386 * @next: allocation that follows the given free space range
387 * @exist: first allocation with same id in the mapping
388 * @n: range that must satisfied for pmem allocations
389 * @valid: free space range to validate
390 *
391 * BLK-space is valid as long as it does not precede a PMEM
392 * allocation in a given region. PMEM-space must be contiguous
393 * and adjacent to an existing allocation (if one
394 * exists). If reserving PMEM any space is valid.
395 */
396static void space_valid(struct nd_region *nd_region, struct nvdimm_drvdata *ndd,
397 struct nd_label_id *label_id, struct resource *prev,
398 struct resource *next, struct resource *exist,
399 resource_size_t n, struct resource *valid)
400{
401 bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0;
402 unsigned long align;
403
404 align = nd_region->align / nd_region->ndr_mappings;
405 valid->start = ALIGN(valid->start, align);
406 valid->end = ALIGN_DOWN(valid->end + 1, align) - 1;
407
408 if (valid->start >= valid->end)
409 goto invalid;
410
411 if (is_reserve)
412 return;
413
414 /* allocation needs to be contiguous, so this is all or nothing */
415 if (resource_size(valid) < n)
416 goto invalid;
417
418 /* we've got all the space we need and no existing allocation */
419 if (!exist)
420 return;
421
422 /* allocation needs to be contiguous with the existing namespace */
423 if (valid->start == exist->end + 1
424 || valid->end == exist->start - 1)
425 return;
426
427 invalid:
428 /* truncate @valid size to 0 */
429 valid->end = valid->start - 1;
430}
431
432enum alloc_loc {
433 ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
434};
435
436static resource_size_t scan_allocate(struct nd_region *nd_region,
437 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
438 resource_size_t n)
439{
440 resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1;
441 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
442 struct resource *res, *exist = NULL, valid;
443 const resource_size_t to_allocate = n;
444 int first;
445
446 for_each_dpa_resource(ndd, res)
447 if (strcmp(label_id->id, res->name) == 0)
448 exist = res;
449
450 valid.start = nd_mapping->start;
451 valid.end = mapping_end;
452 valid.name = "free space";
453 retry:
454 first = 0;
455 for_each_dpa_resource(ndd, res) {
456 struct resource *next = res->sibling, *new_res = NULL;
457 resource_size_t allocate, available = 0;
458 enum alloc_loc loc = ALLOC_ERR;
459 const char *action;
460 int rc = 0;
461
462 /* ignore resources outside this nd_mapping */
463 if (res->start > mapping_end)
464 continue;
465 if (res->end < nd_mapping->start)
466 continue;
467
468 /* space at the beginning of the mapping */
469 if (!first++ && res->start > nd_mapping->start) {
470 valid.start = nd_mapping->start;
471 valid.end = res->start - 1;
472 space_valid(nd_region, ndd, label_id, NULL, next, exist,
473 to_allocate, &valid);
474 available = resource_size(&valid);
475 if (available)
476 loc = ALLOC_BEFORE;
477 }
478
479 /* space between allocations */
480 if (!loc && next) {
481 valid.start = res->start + resource_size(res);
482 valid.end = min(mapping_end, next->start - 1);
483 space_valid(nd_region, ndd, label_id, res, next, exist,
484 to_allocate, &valid);
485 available = resource_size(&valid);
486 if (available)
487 loc = ALLOC_MID;
488 }
489
490 /* space at the end of the mapping */
491 if (!loc && !next) {
492 valid.start = res->start + resource_size(res);
493 valid.end = mapping_end;
494 space_valid(nd_region, ndd, label_id, res, next, exist,
495 to_allocate, &valid);
496 available = resource_size(&valid);
497 if (available)
498 loc = ALLOC_AFTER;
499 }
500
501 if (!loc || !available)
502 continue;
503 allocate = min(available, n);
504 switch (loc) {
505 case ALLOC_BEFORE:
506 if (strcmp(res->name, label_id->id) == 0) {
507 /* adjust current resource up */
508 rc = adjust_resource(res, res->start - allocate,
509 resource_size(res) + allocate);
510 action = "cur grow up";
511 } else
512 action = "allocate";
513 break;
514 case ALLOC_MID:
515 if (strcmp(next->name, label_id->id) == 0) {
516 /* adjust next resource up */
517 rc = adjust_resource(next, next->start
518 - allocate, resource_size(next)
519 + allocate);
520 new_res = next;
521 action = "next grow up";
522 } else if (strcmp(res->name, label_id->id) == 0) {
523 action = "grow down";
524 } else
525 action = "allocate";
526 break;
527 case ALLOC_AFTER:
528 if (strcmp(res->name, label_id->id) == 0)
529 action = "grow down";
530 else
531 action = "allocate";
532 break;
533 default:
534 return n;
535 }
536
537 if (strcmp(action, "allocate") == 0) {
538 new_res = nvdimm_allocate_dpa(ndd, label_id,
539 valid.start, allocate);
540 if (!new_res)
541 rc = -EBUSY;
542 } else if (strcmp(action, "grow down") == 0) {
543 /* adjust current resource down */
544 rc = adjust_resource(res, res->start, resource_size(res)
545 + allocate);
546 if (rc == 0)
547 res->flags |= DPA_RESOURCE_ADJUSTED;
548 }
549
550 if (!new_res)
551 new_res = res;
552
553 nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n",
554 action, loc, rc);
555
556 if (rc)
557 return n;
558
559 n -= allocate;
560 if (n) {
561 /*
562 * Retry scan with newly inserted resources.
563 * For example, if we did an ALLOC_BEFORE
564 * insertion there may also have been space
565 * available for an ALLOC_AFTER insertion, so we
566 * need to check this same resource again
567 */
568 goto retry;
569 } else
570 return 0;
571 }
572
573 if (n == to_allocate)
574 return init_dpa_allocation(label_id, nd_region, nd_mapping, n);
575 return n;
576}
577
578static int merge_dpa(struct nd_region *nd_region,
579 struct nd_mapping *nd_mapping, struct nd_label_id *label_id)
580{
581 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
582 struct resource *res;
583
584 if (strncmp("pmem", label_id->id, 4) == 0)
585 return 0;
586 retry:
587 for_each_dpa_resource(ndd, res) {
588 int rc;
589 struct resource *next = res->sibling;
590 resource_size_t end = res->start + resource_size(res);
591
592 if (!next || strcmp(res->name, label_id->id) != 0
593 || strcmp(next->name, label_id->id) != 0
594 || end != next->start)
595 continue;
596 end += resource_size(next);
597 nvdimm_free_dpa(ndd, next);
598 rc = adjust_resource(res, res->start, end - res->start);
599 nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc);
600 if (rc)
601 return rc;
602 res->flags |= DPA_RESOURCE_ADJUSTED;
603 goto retry;
604 }
605
606 return 0;
607}
608
609int __reserve_free_pmem(struct device *dev, void *data)
610{
611 struct nvdimm *nvdimm = data;
612 struct nd_region *nd_region;
613 struct nd_label_id label_id;
614 int i;
615
616 if (!is_memory(dev))
617 return 0;
618
619 nd_region = to_nd_region(dev);
620 if (nd_region->ndr_mappings == 0)
621 return 0;
622
623 memset(&label_id, 0, sizeof(label_id));
624 strcat(label_id.id, "pmem-reserve");
625 for (i = 0; i < nd_region->ndr_mappings; i++) {
626 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
627 resource_size_t n, rem = 0;
628
629 if (nd_mapping->nvdimm != nvdimm)
630 continue;
631
632 n = nd_pmem_available_dpa(nd_region, nd_mapping);
633 if (n == 0)
634 return 0;
635 rem = scan_allocate(nd_region, nd_mapping, &label_id, n);
636 dev_WARN_ONCE(&nd_region->dev, rem,
637 "pmem reserve underrun: %#llx of %#llx bytes\n",
638 (unsigned long long) n - rem,
639 (unsigned long long) n);
640 return rem ? -ENXIO : 0;
641 }
642
643 return 0;
644}
645
646void release_free_pmem(struct nvdimm_bus *nvdimm_bus,
647 struct nd_mapping *nd_mapping)
648{
649 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
650 struct resource *res, *_res;
651
652 for_each_dpa_resource_safe(ndd, res, _res)
653 if (strcmp(res->name, "pmem-reserve") == 0)
654 nvdimm_free_dpa(ndd, res);
655}
656
657/**
658 * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
659 * @nd_region: the set of dimms to allocate @n more bytes from
660 * @label_id: unique identifier for the namespace consuming this dpa range
661 * @n: number of bytes per-dimm to add to the existing allocation
662 *
663 * Assumes resources are ordered. For BLK regions, first consume
664 * BLK-only available DPA free space, then consume PMEM-aliased DPA
665 * space starting at the highest DPA. For PMEM regions start
666 * allocations from the start of an interleave set and end at the first
667 * BLK allocation or the end of the interleave set, whichever comes
668 * first.
669 *
670 * Returns: %0 on success on -errno on error
671 */
672static int grow_dpa_allocation(struct nd_region *nd_region,
673 struct nd_label_id *label_id, resource_size_t n)
674{
675 int i;
676
677 for (i = 0; i < nd_region->ndr_mappings; i++) {
678 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
679 resource_size_t rem = n;
680 int rc;
681
682 rem = scan_allocate(nd_region, nd_mapping, label_id, rem);
683 dev_WARN_ONCE(&nd_region->dev, rem,
684 "allocation underrun: %#llx of %#llx bytes\n",
685 (unsigned long long) n - rem,
686 (unsigned long long) n);
687 if (rem)
688 return -ENXIO;
689
690 rc = merge_dpa(nd_region, nd_mapping, label_id);
691 if (rc)
692 return rc;
693 }
694
695 return 0;
696}
697
698static void nd_namespace_pmem_set_resource(struct nd_region *nd_region,
699 struct nd_namespace_pmem *nspm, resource_size_t size)
700{
701 struct resource *res = &nspm->nsio.res;
702 resource_size_t offset = 0;
703
704 if (size && !nspm->uuid) {
705 WARN_ON_ONCE(1);
706 size = 0;
707 }
708
709 if (size && nspm->uuid) {
710 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
711 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
712 struct nd_label_id label_id;
713 struct resource *res;
714
715 if (!ndd) {
716 size = 0;
717 goto out;
718 }
719
720 nd_label_gen_id(&label_id, nspm->uuid, 0);
721
722 /* calculate a spa offset from the dpa allocation offset */
723 for_each_dpa_resource(ndd, res)
724 if (strcmp(res->name, label_id.id) == 0) {
725 offset = (res->start - nd_mapping->start)
726 * nd_region->ndr_mappings;
727 goto out;
728 }
729
730 WARN_ON_ONCE(1);
731 size = 0;
732 }
733
734 out:
735 res->start = nd_region->ndr_start + offset;
736 res->end = res->start + size - 1;
737}
738
739static bool uuid_not_set(const uuid_t *uuid, struct device *dev,
740 const char *where)
741{
742 if (!uuid) {
743 dev_dbg(dev, "%s: uuid not set\n", where);
744 return true;
745 }
746 return false;
747}
748
749static ssize_t __size_store(struct device *dev, unsigned long long val)
750{
751 resource_size_t allocated = 0, available = 0;
752 struct nd_region *nd_region = to_nd_region(dev->parent);
753 struct nd_namespace_common *ndns = to_ndns(dev);
754 struct nd_mapping *nd_mapping;
755 struct nvdimm_drvdata *ndd;
756 struct nd_label_id label_id;
757 u32 flags = 0, remainder;
758 int rc, i, id = -1;
759 uuid_t *uuid = NULL;
760
761 if (dev->driver || ndns->claim)
762 return -EBUSY;
763
764 if (is_namespace_pmem(dev)) {
765 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
766
767 uuid = nspm->uuid;
768 id = nspm->id;
769 }
770
771 /*
772 * We need a uuid for the allocation-label and dimm(s) on which
773 * to store the label.
774 */
775 if (uuid_not_set(uuid, dev, __func__))
776 return -ENXIO;
777 if (nd_region->ndr_mappings == 0) {
778 dev_dbg(dev, "not associated with dimm(s)\n");
779 return -ENXIO;
780 }
781
782 div_u64_rem(val, nd_region->align, &remainder);
783 if (remainder) {
784 dev_dbg(dev, "%llu is not %ldK aligned\n", val,
785 nd_region->align / SZ_1K);
786 return -EINVAL;
787 }
788
789 nd_label_gen_id(&label_id, uuid, flags);
790 for (i = 0; i < nd_region->ndr_mappings; i++) {
791 nd_mapping = &nd_region->mapping[i];
792 ndd = to_ndd(nd_mapping);
793
794 /*
795 * All dimms in an interleave set, need to be enabled
796 * for the size to be changed.
797 */
798 if (!ndd)
799 return -ENXIO;
800
801 allocated += nvdimm_allocated_dpa(ndd, &label_id);
802 }
803 available = nd_region_allocatable_dpa(nd_region);
804
805 if (val > available + allocated)
806 return -ENOSPC;
807
808 if (val == allocated)
809 return 0;
810
811 val = div_u64(val, nd_region->ndr_mappings);
812 allocated = div_u64(allocated, nd_region->ndr_mappings);
813 if (val < allocated)
814 rc = shrink_dpa_allocation(nd_region, &label_id,
815 allocated - val);
816 else
817 rc = grow_dpa_allocation(nd_region, &label_id, val - allocated);
818
819 if (rc)
820 return rc;
821
822 if (is_namespace_pmem(dev)) {
823 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
824
825 nd_namespace_pmem_set_resource(nd_region, nspm,
826 val * nd_region->ndr_mappings);
827 }
828
829 /*
830 * Try to delete the namespace if we deleted all of its
831 * allocation, this is not the seed or 0th device for the
832 * region, and it is not actively claimed by a btt, pfn, or dax
833 * instance.
834 */
835 if (val == 0 && id != 0 && nd_region->ns_seed != dev && !ndns->claim)
836 nd_device_unregister(dev, ND_ASYNC);
837
838 return rc;
839}
840
841static ssize_t size_store(struct device *dev,
842 struct device_attribute *attr, const char *buf, size_t len)
843{
844 struct nd_region *nd_region = to_nd_region(dev->parent);
845 unsigned long long val;
846 int rc;
847
848 rc = kstrtoull(buf, 0, &val);
849 if (rc)
850 return rc;
851
852 device_lock(dev);
853 nvdimm_bus_lock(dev);
854 wait_nvdimm_bus_probe_idle(dev);
855 rc = __size_store(dev, val);
856 if (rc >= 0)
857 rc = nd_namespace_label_update(nd_region, dev);
858
859 /* setting size zero == 'delete namespace' */
860 if (rc == 0 && val == 0 && is_namespace_pmem(dev)) {
861 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
862
863 kfree(nspm->uuid);
864 nspm->uuid = NULL;
865 }
866
867 dev_dbg(dev, "%llx %s (%d)\n", val, rc < 0 ? "fail" : "success", rc);
868
869 nvdimm_bus_unlock(dev);
870 device_unlock(dev);
871
872 return rc < 0 ? rc : len;
873}
874
875resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
876{
877 struct device *dev = &ndns->dev;
878
879 if (is_namespace_pmem(dev)) {
880 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
881
882 return resource_size(&nspm->nsio.res);
883 } else if (is_namespace_io(dev)) {
884 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
885
886 return resource_size(&nsio->res);
887 } else
888 WARN_ONCE(1, "unknown namespace type\n");
889 return 0;
890}
891
892resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
893{
894 resource_size_t size;
895
896 nvdimm_bus_lock(&ndns->dev);
897 size = __nvdimm_namespace_capacity(ndns);
898 nvdimm_bus_unlock(&ndns->dev);
899
900 return size;
901}
902EXPORT_SYMBOL(nvdimm_namespace_capacity);
903
904bool nvdimm_namespace_locked(struct nd_namespace_common *ndns)
905{
906 int i;
907 bool locked = false;
908 struct device *dev = &ndns->dev;
909 struct nd_region *nd_region = to_nd_region(dev->parent);
910
911 for (i = 0; i < nd_region->ndr_mappings; i++) {
912 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
913 struct nvdimm *nvdimm = nd_mapping->nvdimm;
914
915 if (test_bit(NDD_LOCKED, &nvdimm->flags)) {
916 dev_dbg(dev, "%s locked\n", nvdimm_name(nvdimm));
917 locked = true;
918 }
919 }
920 return locked;
921}
922EXPORT_SYMBOL(nvdimm_namespace_locked);
923
924static ssize_t size_show(struct device *dev,
925 struct device_attribute *attr, char *buf)
926{
927 return sprintf(buf, "%llu\n", (unsigned long long)
928 nvdimm_namespace_capacity(to_ndns(dev)));
929}
930static DEVICE_ATTR(size, 0444, size_show, size_store);
931
932static uuid_t *namespace_to_uuid(struct device *dev)
933{
934 if (is_namespace_pmem(dev)) {
935 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
936
937 return nspm->uuid;
938 }
939 return ERR_PTR(-ENXIO);
940}
941
942static ssize_t uuid_show(struct device *dev, struct device_attribute *attr,
943 char *buf)
944{
945 uuid_t *uuid = namespace_to_uuid(dev);
946
947 if (IS_ERR(uuid))
948 return PTR_ERR(uuid);
949 if (uuid)
950 return sprintf(buf, "%pUb\n", uuid);
951 return sprintf(buf, "\n");
952}
953
954/**
955 * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
956 * @nd_region: parent region so we can updates all dimms in the set
957 * @dev: namespace type for generating label_id
958 * @new_uuid: incoming uuid
959 * @old_uuid: reference to the uuid storage location in the namespace object
960 *
961 * Returns: %0 on success on -errno on error
962 */
963static int namespace_update_uuid(struct nd_region *nd_region,
964 struct device *dev, uuid_t *new_uuid,
965 uuid_t **old_uuid)
966{
967 struct nd_label_id old_label_id;
968 struct nd_label_id new_label_id;
969 int i;
970
971 if (!nd_is_uuid_unique(dev, new_uuid))
972 return -EINVAL;
973
974 if (*old_uuid == NULL)
975 goto out;
976
977 /*
978 * If we've already written a label with this uuid, then it's
979 * too late to rename because we can't reliably update the uuid
980 * without losing the old namespace. Userspace must delete this
981 * namespace to abandon the old uuid.
982 */
983 for (i = 0; i < nd_region->ndr_mappings; i++) {
984 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
985
986 /*
987 * This check by itself is sufficient because old_uuid
988 * would be NULL above if this uuid did not exist in the
989 * currently written set.
990 *
991 * FIXME: can we delete uuid with zero dpa allocated?
992 */
993 if (list_empty(&nd_mapping->labels))
994 return -EBUSY;
995 }
996
997 nd_label_gen_id(&old_label_id, *old_uuid, 0);
998 nd_label_gen_id(&new_label_id, new_uuid, 0);
999 for (i = 0; i < nd_region->ndr_mappings; i++) {
1000 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1001 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1002 struct nd_label_ent *label_ent;
1003 struct resource *res;
1004
1005 for_each_dpa_resource(ndd, res)
1006 if (strcmp(res->name, old_label_id.id) == 0)
1007 sprintf((void *) res->name, "%s",
1008 new_label_id.id);
1009
1010 mutex_lock(&nd_mapping->lock);
1011 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1012 struct nd_namespace_label *nd_label = label_ent->label;
1013 struct nd_label_id label_id;
1014 uuid_t uuid;
1015
1016 if (!nd_label)
1017 continue;
1018 nsl_get_uuid(ndd, nd_label, &uuid);
1019 nd_label_gen_id(&label_id, &uuid,
1020 nsl_get_flags(ndd, nd_label));
1021 if (strcmp(old_label_id.id, label_id.id) == 0)
1022 set_bit(ND_LABEL_REAP, &label_ent->flags);
1023 }
1024 mutex_unlock(&nd_mapping->lock);
1025 }
1026 kfree(*old_uuid);
1027 out:
1028 *old_uuid = new_uuid;
1029 return 0;
1030}
1031
1032static ssize_t uuid_store(struct device *dev,
1033 struct device_attribute *attr, const char *buf, size_t len)
1034{
1035 struct nd_region *nd_region = to_nd_region(dev->parent);
1036 uuid_t *uuid = NULL;
1037 uuid_t **ns_uuid;
1038 ssize_t rc = 0;
1039
1040 if (is_namespace_pmem(dev)) {
1041 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1042
1043 ns_uuid = &nspm->uuid;
1044 } else
1045 return -ENXIO;
1046
1047 device_lock(dev);
1048 nvdimm_bus_lock(dev);
1049 wait_nvdimm_bus_probe_idle(dev);
1050 if (to_ndns(dev)->claim)
1051 rc = -EBUSY;
1052 if (rc >= 0)
1053 rc = nd_uuid_store(dev, &uuid, buf, len);
1054 if (rc >= 0)
1055 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
1056 if (rc >= 0)
1057 rc = nd_namespace_label_update(nd_region, dev);
1058 else
1059 kfree(uuid);
1060 dev_dbg(dev, "result: %zd wrote: %s%s", rc, buf,
1061 buf[len - 1] == '\n' ? "" : "\n");
1062 nvdimm_bus_unlock(dev);
1063 device_unlock(dev);
1064
1065 return rc < 0 ? rc : len;
1066}
1067static DEVICE_ATTR_RW(uuid);
1068
1069static ssize_t resource_show(struct device *dev,
1070 struct device_attribute *attr, char *buf)
1071{
1072 struct resource *res;
1073
1074 if (is_namespace_pmem(dev)) {
1075 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1076
1077 res = &nspm->nsio.res;
1078 } else if (is_namespace_io(dev)) {
1079 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1080
1081 res = &nsio->res;
1082 } else
1083 return -ENXIO;
1084
1085 /* no address to convey if the namespace has no allocation */
1086 if (resource_size(res) == 0)
1087 return -ENXIO;
1088 return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
1089}
1090static DEVICE_ATTR_ADMIN_RO(resource);
1091
1092static const unsigned long pmem_lbasize_supported[] = { 512, 4096, 0 };
1093
1094static ssize_t sector_size_show(struct device *dev,
1095 struct device_attribute *attr, char *buf)
1096{
1097 if (is_namespace_pmem(dev)) {
1098 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1099
1100 return nd_size_select_show(nspm->lbasize,
1101 pmem_lbasize_supported, buf);
1102 }
1103 return -ENXIO;
1104}
1105
1106static ssize_t sector_size_store(struct device *dev,
1107 struct device_attribute *attr, const char *buf, size_t len)
1108{
1109 struct nd_region *nd_region = to_nd_region(dev->parent);
1110 const unsigned long *supported;
1111 unsigned long *lbasize;
1112 ssize_t rc = 0;
1113
1114 if (is_namespace_pmem(dev)) {
1115 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1116
1117 lbasize = &nspm->lbasize;
1118 supported = pmem_lbasize_supported;
1119 } else
1120 return -ENXIO;
1121
1122 device_lock(dev);
1123 nvdimm_bus_lock(dev);
1124 if (to_ndns(dev)->claim)
1125 rc = -EBUSY;
1126 if (rc >= 0)
1127 rc = nd_size_select_store(dev, buf, lbasize, supported);
1128 if (rc >= 0)
1129 rc = nd_namespace_label_update(nd_region, dev);
1130 dev_dbg(dev, "result: %zd %s: %s%s", rc, rc < 0 ? "tried" : "wrote",
1131 buf, buf[len - 1] == '\n' ? "" : "\n");
1132 nvdimm_bus_unlock(dev);
1133 device_unlock(dev);
1134
1135 return rc ? rc : len;
1136}
1137static DEVICE_ATTR_RW(sector_size);
1138
1139static ssize_t dpa_extents_show(struct device *dev,
1140 struct device_attribute *attr, char *buf)
1141{
1142 struct nd_region *nd_region = to_nd_region(dev->parent);
1143 struct nd_label_id label_id;
1144 uuid_t *uuid = NULL;
1145 int count = 0, i;
1146 u32 flags = 0;
1147
1148 nvdimm_bus_lock(dev);
1149 if (is_namespace_pmem(dev)) {
1150 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1151
1152 uuid = nspm->uuid;
1153 flags = 0;
1154 }
1155
1156 if (!uuid)
1157 goto out;
1158
1159 nd_label_gen_id(&label_id, uuid, flags);
1160 for (i = 0; i < nd_region->ndr_mappings; i++) {
1161 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1162 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1163 struct resource *res;
1164
1165 for_each_dpa_resource(ndd, res)
1166 if (strcmp(res->name, label_id.id) == 0)
1167 count++;
1168 }
1169 out:
1170 nvdimm_bus_unlock(dev);
1171
1172 return sprintf(buf, "%d\n", count);
1173}
1174static DEVICE_ATTR_RO(dpa_extents);
1175
1176static int btt_claim_class(struct device *dev)
1177{
1178 struct nd_region *nd_region = to_nd_region(dev->parent);
1179 int i, loop_bitmask = 0;
1180
1181 for (i = 0; i < nd_region->ndr_mappings; i++) {
1182 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1183 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1184 struct nd_namespace_index *nsindex;
1185
1186 /*
1187 * If any of the DIMMs do not support labels the only
1188 * possible BTT format is v1.
1189 */
1190 if (!ndd) {
1191 loop_bitmask = 0;
1192 break;
1193 }
1194
1195 nsindex = to_namespace_index(ndd, ndd->ns_current);
1196 if (nsindex == NULL)
1197 loop_bitmask |= 1;
1198 else {
1199 /* check whether existing labels are v1.1 or v1.2 */
1200 if (__le16_to_cpu(nsindex->major) == 1
1201 && __le16_to_cpu(nsindex->minor) == 1)
1202 loop_bitmask |= 2;
1203 else
1204 loop_bitmask |= 4;
1205 }
1206 }
1207 /*
1208 * If nsindex is null loop_bitmask's bit 0 will be set, and if an index
1209 * block is found, a v1.1 label for any mapping will set bit 1, and a
1210 * v1.2 label will set bit 2.
1211 *
1212 * At the end of the loop, at most one of the three bits must be set.
1213 * If multiple bits were set, it means the different mappings disagree
1214 * about their labels, and this must be cleaned up first.
1215 *
1216 * If all the label index blocks are found to agree, nsindex of NULL
1217 * implies labels haven't been initialized yet, and when they will,
1218 * they will be of the 1.2 format, so we can assume BTT2.0
1219 *
1220 * If 1.1 labels are found, we enforce BTT1.1, and if 1.2 labels are
1221 * found, we enforce BTT2.0
1222 *
1223 * If the loop was never entered, default to BTT1.1 (legacy namespaces)
1224 */
1225 switch (loop_bitmask) {
1226 case 0:
1227 case 2:
1228 return NVDIMM_CCLASS_BTT;
1229 case 1:
1230 case 4:
1231 return NVDIMM_CCLASS_BTT2;
1232 default:
1233 return -ENXIO;
1234 }
1235}
1236
1237static ssize_t holder_show(struct device *dev,
1238 struct device_attribute *attr, char *buf)
1239{
1240 struct nd_namespace_common *ndns = to_ndns(dev);
1241 ssize_t rc;
1242
1243 device_lock(dev);
1244 rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
1245 device_unlock(dev);
1246
1247 return rc;
1248}
1249static DEVICE_ATTR_RO(holder);
1250
1251static int __holder_class_store(struct device *dev, const char *buf)
1252{
1253 struct nd_namespace_common *ndns = to_ndns(dev);
1254
1255 if (dev->driver || ndns->claim)
1256 return -EBUSY;
1257
1258 if (sysfs_streq(buf, "btt")) {
1259 int rc = btt_claim_class(dev);
1260
1261 if (rc < NVDIMM_CCLASS_NONE)
1262 return rc;
1263 ndns->claim_class = rc;
1264 } else if (sysfs_streq(buf, "pfn"))
1265 ndns->claim_class = NVDIMM_CCLASS_PFN;
1266 else if (sysfs_streq(buf, "dax"))
1267 ndns->claim_class = NVDIMM_CCLASS_DAX;
1268 else if (sysfs_streq(buf, ""))
1269 ndns->claim_class = NVDIMM_CCLASS_NONE;
1270 else
1271 return -EINVAL;
1272
1273 return 0;
1274}
1275
1276static ssize_t holder_class_store(struct device *dev,
1277 struct device_attribute *attr, const char *buf, size_t len)
1278{
1279 struct nd_region *nd_region = to_nd_region(dev->parent);
1280 int rc;
1281
1282 device_lock(dev);
1283 nvdimm_bus_lock(dev);
1284 wait_nvdimm_bus_probe_idle(dev);
1285 rc = __holder_class_store(dev, buf);
1286 if (rc >= 0)
1287 rc = nd_namespace_label_update(nd_region, dev);
1288 dev_dbg(dev, "%s(%d)\n", rc < 0 ? "fail " : "", rc);
1289 nvdimm_bus_unlock(dev);
1290 device_unlock(dev);
1291
1292 return rc < 0 ? rc : len;
1293}
1294
1295static ssize_t holder_class_show(struct device *dev,
1296 struct device_attribute *attr, char *buf)
1297{
1298 struct nd_namespace_common *ndns = to_ndns(dev);
1299 ssize_t rc;
1300
1301 device_lock(dev);
1302 if (ndns->claim_class == NVDIMM_CCLASS_NONE)
1303 rc = sprintf(buf, "\n");
1304 else if ((ndns->claim_class == NVDIMM_CCLASS_BTT) ||
1305 (ndns->claim_class == NVDIMM_CCLASS_BTT2))
1306 rc = sprintf(buf, "btt\n");
1307 else if (ndns->claim_class == NVDIMM_CCLASS_PFN)
1308 rc = sprintf(buf, "pfn\n");
1309 else if (ndns->claim_class == NVDIMM_CCLASS_DAX)
1310 rc = sprintf(buf, "dax\n");
1311 else
1312 rc = sprintf(buf, "<unknown>\n");
1313 device_unlock(dev);
1314
1315 return rc;
1316}
1317static DEVICE_ATTR_RW(holder_class);
1318
1319static ssize_t mode_show(struct device *dev,
1320 struct device_attribute *attr, char *buf)
1321{
1322 struct nd_namespace_common *ndns = to_ndns(dev);
1323 struct device *claim;
1324 char *mode;
1325 ssize_t rc;
1326
1327 device_lock(dev);
1328 claim = ndns->claim;
1329 if (claim && is_nd_btt(claim))
1330 mode = "safe";
1331 else if (claim && is_nd_pfn(claim))
1332 mode = "memory";
1333 else if (claim && is_nd_dax(claim))
1334 mode = "dax";
1335 else if (!claim && pmem_should_map_pages(dev))
1336 mode = "memory";
1337 else
1338 mode = "raw";
1339 rc = sprintf(buf, "%s\n", mode);
1340 device_unlock(dev);
1341
1342 return rc;
1343}
1344static DEVICE_ATTR_RO(mode);
1345
1346static ssize_t force_raw_store(struct device *dev,
1347 struct device_attribute *attr, const char *buf, size_t len)
1348{
1349 bool force_raw;
1350 int rc = kstrtobool(buf, &force_raw);
1351
1352 if (rc)
1353 return rc;
1354
1355 to_ndns(dev)->force_raw = force_raw;
1356 return len;
1357}
1358
1359static ssize_t force_raw_show(struct device *dev,
1360 struct device_attribute *attr, char *buf)
1361{
1362 return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
1363}
1364static DEVICE_ATTR_RW(force_raw);
1365
1366static struct attribute *nd_namespace_attributes[] = {
1367 &dev_attr_nstype.attr,
1368 &dev_attr_size.attr,
1369 &dev_attr_mode.attr,
1370 &dev_attr_uuid.attr,
1371 &dev_attr_holder.attr,
1372 &dev_attr_resource.attr,
1373 &dev_attr_alt_name.attr,
1374 &dev_attr_force_raw.attr,
1375 &dev_attr_sector_size.attr,
1376 &dev_attr_dpa_extents.attr,
1377 &dev_attr_holder_class.attr,
1378 NULL,
1379};
1380
1381static umode_t namespace_visible(struct kobject *kobj,
1382 struct attribute *a, int n)
1383{
1384 struct device *dev = container_of(kobj, struct device, kobj);
1385
1386 if (is_namespace_pmem(dev)) {
1387 if (a == &dev_attr_size.attr)
1388 return 0644;
1389
1390 return a->mode;
1391 }
1392
1393 /* base is_namespace_io() attributes */
1394 if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr ||
1395 a == &dev_attr_holder.attr || a == &dev_attr_holder_class.attr ||
1396 a == &dev_attr_force_raw.attr || a == &dev_attr_mode.attr ||
1397 a == &dev_attr_resource.attr)
1398 return a->mode;
1399
1400 return 0;
1401}
1402
1403static struct attribute_group nd_namespace_attribute_group = {
1404 .attrs = nd_namespace_attributes,
1405 .is_visible = namespace_visible,
1406};
1407
1408static const struct attribute_group *nd_namespace_attribute_groups[] = {
1409 &nd_device_attribute_group,
1410 &nd_namespace_attribute_group,
1411 &nd_numa_attribute_group,
1412 NULL,
1413};
1414
1415static const struct device_type namespace_io_device_type = {
1416 .name = "nd_namespace_io",
1417 .release = namespace_io_release,
1418 .groups = nd_namespace_attribute_groups,
1419};
1420
1421static const struct device_type namespace_pmem_device_type = {
1422 .name = "nd_namespace_pmem",
1423 .release = namespace_pmem_release,
1424 .groups = nd_namespace_attribute_groups,
1425};
1426
1427static bool is_namespace_pmem(const struct device *dev)
1428{
1429 return dev ? dev->type == &namespace_pmem_device_type : false;
1430}
1431
1432static bool is_namespace_io(const struct device *dev)
1433{
1434 return dev ? dev->type == &namespace_io_device_type : false;
1435}
1436
1437struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
1438{
1439 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
1440 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
1441 struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL;
1442 struct nd_namespace_common *ndns = NULL;
1443 resource_size_t size;
1444
1445 if (nd_btt || nd_pfn || nd_dax) {
1446 if (nd_btt)
1447 ndns = nd_btt->ndns;
1448 else if (nd_pfn)
1449 ndns = nd_pfn->ndns;
1450 else if (nd_dax)
1451 ndns = nd_dax->nd_pfn.ndns;
1452
1453 if (!ndns)
1454 return ERR_PTR(-ENODEV);
1455
1456 /*
1457 * Flush any in-progess probes / removals in the driver
1458 * for the raw personality of this namespace.
1459 */
1460 device_lock(&ndns->dev);
1461 device_unlock(&ndns->dev);
1462 if (ndns->dev.driver) {
1463 dev_dbg(&ndns->dev, "is active, can't bind %s\n",
1464 dev_name(dev));
1465 return ERR_PTR(-EBUSY);
1466 }
1467 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != dev,
1468 "host (%s) vs claim (%s) mismatch\n",
1469 dev_name(dev),
1470 dev_name(ndns->claim)))
1471 return ERR_PTR(-ENXIO);
1472 } else {
1473 ndns = to_ndns(dev);
1474 if (ndns->claim) {
1475 dev_dbg(dev, "claimed by %s, failing probe\n",
1476 dev_name(ndns->claim));
1477
1478 return ERR_PTR(-ENXIO);
1479 }
1480 }
1481
1482 if (nvdimm_namespace_locked(ndns))
1483 return ERR_PTR(-EACCES);
1484
1485 size = nvdimm_namespace_capacity(ndns);
1486 if (size < ND_MIN_NAMESPACE_SIZE) {
1487 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
1488 &size, ND_MIN_NAMESPACE_SIZE);
1489 return ERR_PTR(-ENODEV);
1490 }
1491
1492 /*
1493 * Note, alignment validation for fsdax and devdax mode
1494 * namespaces happens in nd_pfn_validate() where infoblock
1495 * padding parameters can be applied.
1496 */
1497 if (pmem_should_map_pages(dev)) {
1498 struct nd_namespace_io *nsio = to_nd_namespace_io(&ndns->dev);
1499 struct resource *res = &nsio->res;
1500
1501 if (!IS_ALIGNED(res->start | (res->end + 1),
1502 memremap_compat_align())) {
1503 dev_err(&ndns->dev, "%pr misaligned, unable to map\n", res);
1504 return ERR_PTR(-EOPNOTSUPP);
1505 }
1506 }
1507
1508 if (is_namespace_pmem(&ndns->dev)) {
1509 struct nd_namespace_pmem *nspm;
1510
1511 nspm = to_nd_namespace_pmem(&ndns->dev);
1512 if (uuid_not_set(nspm->uuid, &ndns->dev, __func__))
1513 return ERR_PTR(-ENODEV);
1514 }
1515
1516 return ndns;
1517}
1518EXPORT_SYMBOL(nvdimm_namespace_common_probe);
1519
1520int devm_namespace_enable(struct device *dev, struct nd_namespace_common *ndns,
1521 resource_size_t size)
1522{
1523 return devm_nsio_enable(dev, to_nd_namespace_io(&ndns->dev), size);
1524}
1525EXPORT_SYMBOL_GPL(devm_namespace_enable);
1526
1527void devm_namespace_disable(struct device *dev, struct nd_namespace_common *ndns)
1528{
1529 devm_nsio_disable(dev, to_nd_namespace_io(&ndns->dev));
1530}
1531EXPORT_SYMBOL_GPL(devm_namespace_disable);
1532
1533static struct device **create_namespace_io(struct nd_region *nd_region)
1534{
1535 struct nd_namespace_io *nsio;
1536 struct device *dev, **devs;
1537 struct resource *res;
1538
1539 nsio = kzalloc(sizeof(*nsio), GFP_KERNEL);
1540 if (!nsio)
1541 return NULL;
1542
1543 devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1544 if (!devs) {
1545 kfree(nsio);
1546 return NULL;
1547 }
1548
1549 dev = &nsio->common.dev;
1550 dev->type = &namespace_io_device_type;
1551 dev->parent = &nd_region->dev;
1552 res = &nsio->res;
1553 res->name = dev_name(&nd_region->dev);
1554 res->flags = IORESOURCE_MEM;
1555 res->start = nd_region->ndr_start;
1556 res->end = res->start + nd_region->ndr_size - 1;
1557
1558 devs[0] = dev;
1559 return devs;
1560}
1561
1562static bool has_uuid_at_pos(struct nd_region *nd_region, const uuid_t *uuid,
1563 u64 cookie, u16 pos)
1564{
1565 struct nd_namespace_label *found = NULL;
1566 int i;
1567
1568 for (i = 0; i < nd_region->ndr_mappings; i++) {
1569 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1570 struct nd_interleave_set *nd_set = nd_region->nd_set;
1571 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1572 struct nd_label_ent *label_ent;
1573 bool found_uuid = false;
1574
1575 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1576 struct nd_namespace_label *nd_label = label_ent->label;
1577 u16 position;
1578
1579 if (!nd_label)
1580 continue;
1581 position = nsl_get_position(ndd, nd_label);
1582
1583 if (!nsl_validate_isetcookie(ndd, nd_label, cookie))
1584 continue;
1585
1586 if (!nsl_uuid_equal(ndd, nd_label, uuid))
1587 continue;
1588
1589 if (!nsl_validate_type_guid(ndd, nd_label,
1590 &nd_set->type_guid))
1591 continue;
1592
1593 if (found_uuid) {
1594 dev_dbg(ndd->dev, "duplicate entry for uuid\n");
1595 return false;
1596 }
1597 found_uuid = true;
1598 if (!nsl_validate_nlabel(nd_region, ndd, nd_label))
1599 continue;
1600 if (position != pos)
1601 continue;
1602 found = nd_label;
1603 break;
1604 }
1605 if (found)
1606 break;
1607 }
1608 return found != NULL;
1609}
1610
1611static int select_pmem_id(struct nd_region *nd_region, const uuid_t *pmem_id)
1612{
1613 int i;
1614
1615 if (!pmem_id)
1616 return -ENODEV;
1617
1618 for (i = 0; i < nd_region->ndr_mappings; i++) {
1619 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1620 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1621 struct nd_namespace_label *nd_label = NULL;
1622 u64 hw_start, hw_end, pmem_start, pmem_end;
1623 struct nd_label_ent *label_ent;
1624
1625 lockdep_assert_held(&nd_mapping->lock);
1626 list_for_each_entry(label_ent, &nd_mapping->labels, list) {
1627 nd_label = label_ent->label;
1628 if (!nd_label)
1629 continue;
1630 if (nsl_uuid_equal(ndd, nd_label, pmem_id))
1631 break;
1632 nd_label = NULL;
1633 }
1634
1635 if (!nd_label) {
1636 WARN_ON(1);
1637 return -EINVAL;
1638 }
1639
1640 /*
1641 * Check that this label is compliant with the dpa
1642 * range published in NFIT
1643 */
1644 hw_start = nd_mapping->start;
1645 hw_end = hw_start + nd_mapping->size;
1646 pmem_start = nsl_get_dpa(ndd, nd_label);
1647 pmem_end = pmem_start + nsl_get_rawsize(ndd, nd_label);
1648 if (pmem_start >= hw_start && pmem_start < hw_end
1649 && pmem_end <= hw_end && pmem_end > hw_start)
1650 /* pass */;
1651 else {
1652 dev_dbg(&nd_region->dev, "%s invalid label for %pUb\n",
1653 dev_name(ndd->dev),
1654 nsl_uuid_raw(ndd, nd_label));
1655 return -EINVAL;
1656 }
1657
1658 /* move recently validated label to the front of the list */
1659 list_move(&label_ent->list, &nd_mapping->labels);
1660 }
1661 return 0;
1662}
1663
1664/**
1665 * create_namespace_pmem - validate interleave set labelling, retrieve label0
1666 * @nd_region: region with mappings to validate
1667 * @nd_mapping: container of dpa-resource-root + labels
1668 * @nd_label: target pmem namespace label to evaluate
1669 *
1670 * Returns: the created &struct device on success or ERR_PTR(-errno) on error
1671 */
1672static struct device *create_namespace_pmem(struct nd_region *nd_region,
1673 struct nd_mapping *nd_mapping,
1674 struct nd_namespace_label *nd_label)
1675{
1676 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1677 struct nd_namespace_index *nsindex =
1678 to_namespace_index(ndd, ndd->ns_current);
1679 u64 cookie = nd_region_interleave_set_cookie(nd_region, nsindex);
1680 u64 altcookie = nd_region_interleave_set_altcookie(nd_region);
1681 struct nd_label_ent *label_ent;
1682 struct nd_namespace_pmem *nspm;
1683 resource_size_t size = 0;
1684 struct resource *res;
1685 struct device *dev;
1686 uuid_t uuid;
1687 int rc = 0;
1688 u16 i;
1689
1690 if (cookie == 0) {
1691 dev_dbg(&nd_region->dev, "invalid interleave-set-cookie\n");
1692 return ERR_PTR(-ENXIO);
1693 }
1694
1695 if (!nsl_validate_isetcookie(ndd, nd_label, cookie)) {
1696 dev_dbg(&nd_region->dev, "invalid cookie in label: %pUb\n",
1697 nsl_uuid_raw(ndd, nd_label));
1698 if (!nsl_validate_isetcookie(ndd, nd_label, altcookie))
1699 return ERR_PTR(-EAGAIN);
1700
1701 dev_dbg(&nd_region->dev, "valid altcookie in label: %pUb\n",
1702 nsl_uuid_raw(ndd, nd_label));
1703 }
1704
1705 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1706 if (!nspm)
1707 return ERR_PTR(-ENOMEM);
1708
1709 nspm->id = -1;
1710 dev = &nspm->nsio.common.dev;
1711 dev->type = &namespace_pmem_device_type;
1712 dev->parent = &nd_region->dev;
1713 res = &nspm->nsio.res;
1714 res->name = dev_name(&nd_region->dev);
1715 res->flags = IORESOURCE_MEM;
1716
1717 for (i = 0; i < nd_region->ndr_mappings; i++) {
1718 nsl_get_uuid(ndd, nd_label, &uuid);
1719 if (has_uuid_at_pos(nd_region, &uuid, cookie, i))
1720 continue;
1721 if (has_uuid_at_pos(nd_region, &uuid, altcookie, i))
1722 continue;
1723 break;
1724 }
1725
1726 if (i < nd_region->ndr_mappings) {
1727 struct nvdimm *nvdimm = nd_region->mapping[i].nvdimm;
1728
1729 /*
1730 * Give up if we don't find an instance of a uuid at each
1731 * position (from 0 to nd_region->ndr_mappings - 1), or if we
1732 * find a dimm with two instances of the same uuid.
1733 */
1734 dev_err(&nd_region->dev, "%s missing label for %pUb\n",
1735 nvdimm_name(nvdimm), nsl_uuid_raw(ndd, nd_label));
1736 rc = -EINVAL;
1737 goto err;
1738 }
1739
1740 /*
1741 * Fix up each mapping's 'labels' to have the validated pmem label for
1742 * that position at labels[0], and NULL at labels[1]. In the process,
1743 * check that the namespace aligns with interleave-set.
1744 */
1745 nsl_get_uuid(ndd, nd_label, &uuid);
1746 rc = select_pmem_id(nd_region, &uuid);
1747 if (rc)
1748 goto err;
1749
1750 /* Calculate total size and populate namespace properties from label0 */
1751 for (i = 0; i < nd_region->ndr_mappings; i++) {
1752 struct nd_namespace_label *label0;
1753 struct nvdimm_drvdata *ndd;
1754
1755 nd_mapping = &nd_region->mapping[i];
1756 label_ent = list_first_entry_or_null(&nd_mapping->labels,
1757 typeof(*label_ent), list);
1758 label0 = label_ent ? label_ent->label : NULL;
1759
1760 if (!label0) {
1761 WARN_ON(1);
1762 continue;
1763 }
1764
1765 ndd = to_ndd(nd_mapping);
1766 size += nsl_get_rawsize(ndd, label0);
1767 if (nsl_get_position(ndd, label0) != 0)
1768 continue;
1769 WARN_ON(nspm->alt_name || nspm->uuid);
1770 nspm->alt_name = kmemdup(nsl_ref_name(ndd, label0),
1771 NSLABEL_NAME_LEN, GFP_KERNEL);
1772 nsl_get_uuid(ndd, label0, &uuid);
1773 nspm->uuid = kmemdup(&uuid, sizeof(uuid_t), GFP_KERNEL);
1774 nspm->lbasize = nsl_get_lbasize(ndd, label0);
1775 nspm->nsio.common.claim_class =
1776 nsl_get_claim_class(ndd, label0);
1777 }
1778
1779 if (!nspm->alt_name || !nspm->uuid) {
1780 rc = -ENOMEM;
1781 goto err;
1782 }
1783
1784 nd_namespace_pmem_set_resource(nd_region, nspm, size);
1785
1786 return dev;
1787 err:
1788 namespace_pmem_release(dev);
1789 switch (rc) {
1790 case -EINVAL:
1791 dev_dbg(&nd_region->dev, "invalid label(s)\n");
1792 break;
1793 case -ENODEV:
1794 dev_dbg(&nd_region->dev, "label not found\n");
1795 break;
1796 default:
1797 dev_dbg(&nd_region->dev, "unexpected err: %d\n", rc);
1798 break;
1799 }
1800 return ERR_PTR(rc);
1801}
1802
1803static struct device *nd_namespace_pmem_create(struct nd_region *nd_region)
1804{
1805 struct nd_namespace_pmem *nspm;
1806 struct resource *res;
1807 struct device *dev;
1808
1809 if (!is_memory(&nd_region->dev))
1810 return NULL;
1811
1812 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1813 if (!nspm)
1814 return NULL;
1815
1816 dev = &nspm->nsio.common.dev;
1817 dev->type = &namespace_pmem_device_type;
1818 dev->parent = &nd_region->dev;
1819 res = &nspm->nsio.res;
1820 res->name = dev_name(&nd_region->dev);
1821 res->flags = IORESOURCE_MEM;
1822
1823 nspm->id = ida_alloc(&nd_region->ns_ida, GFP_KERNEL);
1824 if (nspm->id < 0) {
1825 kfree(nspm);
1826 return NULL;
1827 }
1828 dev_set_name(dev, "namespace%d.%d", nd_region->id, nspm->id);
1829 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
1830
1831 return dev;
1832}
1833
1834static struct lock_class_key nvdimm_namespace_key;
1835
1836void nd_region_create_ns_seed(struct nd_region *nd_region)
1837{
1838 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1839
1840 if (nd_region_to_nstype(nd_region) == ND_DEVICE_NAMESPACE_IO)
1841 return;
1842
1843 nd_region->ns_seed = nd_namespace_pmem_create(nd_region);
1844
1845 /*
1846 * Seed creation failures are not fatal, provisioning is simply
1847 * disabled until memory becomes available
1848 */
1849 if (!nd_region->ns_seed)
1850 dev_err(&nd_region->dev, "failed to create namespace\n");
1851 else {
1852 device_initialize(nd_region->ns_seed);
1853 lockdep_set_class(&nd_region->ns_seed->mutex,
1854 &nvdimm_namespace_key);
1855 nd_device_register(nd_region->ns_seed);
1856 }
1857}
1858
1859void nd_region_create_dax_seed(struct nd_region *nd_region)
1860{
1861 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1862 nd_region->dax_seed = nd_dax_create(nd_region);
1863 /*
1864 * Seed creation failures are not fatal, provisioning is simply
1865 * disabled until memory becomes available
1866 */
1867 if (!nd_region->dax_seed)
1868 dev_err(&nd_region->dev, "failed to create dax namespace\n");
1869}
1870
1871void nd_region_create_pfn_seed(struct nd_region *nd_region)
1872{
1873 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1874 nd_region->pfn_seed = nd_pfn_create(nd_region);
1875 /*
1876 * Seed creation failures are not fatal, provisioning is simply
1877 * disabled until memory becomes available
1878 */
1879 if (!nd_region->pfn_seed)
1880 dev_err(&nd_region->dev, "failed to create pfn namespace\n");
1881}
1882
1883void nd_region_create_btt_seed(struct nd_region *nd_region)
1884{
1885 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1886 nd_region->btt_seed = nd_btt_create(nd_region);
1887 /*
1888 * Seed creation failures are not fatal, provisioning is simply
1889 * disabled until memory becomes available
1890 */
1891 if (!nd_region->btt_seed)
1892 dev_err(&nd_region->dev, "failed to create btt namespace\n");
1893}
1894
1895static int add_namespace_resource(struct nd_region *nd_region,
1896 struct nd_namespace_label *nd_label, struct device **devs,
1897 int count)
1898{
1899 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
1900 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1901 int i;
1902
1903 for (i = 0; i < count; i++) {
1904 uuid_t *uuid = namespace_to_uuid(devs[i]);
1905
1906 if (IS_ERR(uuid)) {
1907 WARN_ON(1);
1908 continue;
1909 }
1910
1911 if (!nsl_uuid_equal(ndd, nd_label, uuid))
1912 continue;
1913 dev_err(&nd_region->dev,
1914 "error: conflicting extents for uuid: %pUb\n", uuid);
1915 return -ENXIO;
1916 }
1917
1918 return i;
1919}
1920
1921static int cmp_dpa(const void *a, const void *b)
1922{
1923 const struct device *dev_a = *(const struct device **) a;
1924 const struct device *dev_b = *(const struct device **) b;
1925 struct nd_namespace_pmem *nspm_a, *nspm_b;
1926
1927 if (is_namespace_io(dev_a))
1928 return 0;
1929
1930 nspm_a = to_nd_namespace_pmem(dev_a);
1931 nspm_b = to_nd_namespace_pmem(dev_b);
1932
1933 return memcmp(&nspm_a->nsio.res.start, &nspm_b->nsio.res.start,
1934 sizeof(resource_size_t));
1935}
1936
1937static struct device **scan_labels(struct nd_region *nd_region)
1938{
1939 int i, count = 0;
1940 struct device *dev, **devs = NULL;
1941 struct nd_label_ent *label_ent, *e;
1942 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
1943 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1944 resource_size_t map_end = nd_mapping->start + nd_mapping->size - 1;
1945
1946 /* "safe" because create_namespace_pmem() might list_move() label_ent */
1947 list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) {
1948 struct nd_namespace_label *nd_label = label_ent->label;
1949 struct device **__devs;
1950
1951 if (!nd_label)
1952 continue;
1953
1954 /* skip labels that describe extents outside of the region */
1955 if (nsl_get_dpa(ndd, nd_label) < nd_mapping->start ||
1956 nsl_get_dpa(ndd, nd_label) > map_end)
1957 continue;
1958
1959 i = add_namespace_resource(nd_region, nd_label, devs, count);
1960 if (i < 0)
1961 goto err;
1962 if (i < count)
1963 continue;
1964 __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
1965 if (!__devs)
1966 goto err;
1967 memcpy(__devs, devs, sizeof(dev) * count);
1968 kfree(devs);
1969 devs = __devs;
1970
1971 dev = create_namespace_pmem(nd_region, nd_mapping, nd_label);
1972 if (IS_ERR(dev)) {
1973 switch (PTR_ERR(dev)) {
1974 case -EAGAIN:
1975 /* skip invalid labels */
1976 continue;
1977 case -ENODEV:
1978 /* fallthrough to seed creation */
1979 break;
1980 default:
1981 goto err;
1982 }
1983 } else
1984 devs[count++] = dev;
1985
1986 }
1987
1988 dev_dbg(&nd_region->dev, "discovered %d namespace%s\n", count,
1989 count == 1 ? "" : "s");
1990
1991 if (count == 0) {
1992 struct nd_namespace_pmem *nspm;
1993
1994 /* Publish a zero-sized namespace for userspace to configure. */
1995 nd_mapping_free_labels(nd_mapping);
1996
1997 devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
1998 if (!devs)
1999 goto err;
2000
2001 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
2002 if (!nspm)
2003 goto err;
2004 dev = &nspm->nsio.common.dev;
2005 dev->type = &namespace_pmem_device_type;
2006 nd_namespace_pmem_set_resource(nd_region, nspm, 0);
2007 dev->parent = &nd_region->dev;
2008 devs[count++] = dev;
2009 } else if (is_memory(&nd_region->dev)) {
2010 /* clean unselected labels */
2011 for (i = 0; i < nd_region->ndr_mappings; i++) {
2012 struct list_head *l, *e;
2013 LIST_HEAD(list);
2014 int j;
2015
2016 nd_mapping = &nd_region->mapping[i];
2017 if (list_empty(&nd_mapping->labels)) {
2018 WARN_ON(1);
2019 continue;
2020 }
2021
2022 j = count;
2023 list_for_each_safe(l, e, &nd_mapping->labels) {
2024 if (!j--)
2025 break;
2026 list_move_tail(l, &list);
2027 }
2028 nd_mapping_free_labels(nd_mapping);
2029 list_splice_init(&list, &nd_mapping->labels);
2030 }
2031 }
2032
2033 if (count > 1)
2034 sort(devs, count, sizeof(struct device *), cmp_dpa, NULL);
2035
2036 return devs;
2037
2038 err:
2039 if (devs) {
2040 for (i = 0; devs[i]; i++)
2041 namespace_pmem_release(devs[i]);
2042 kfree(devs);
2043 }
2044 return NULL;
2045}
2046
2047static struct device **create_namespaces(struct nd_region *nd_region)
2048{
2049 struct nd_mapping *nd_mapping;
2050 struct device **devs;
2051 int i;
2052
2053 if (nd_region->ndr_mappings == 0)
2054 return NULL;
2055
2056 /* lock down all mappings while we scan labels */
2057 for (i = 0; i < nd_region->ndr_mappings; i++) {
2058 nd_mapping = &nd_region->mapping[i];
2059 mutex_lock_nested(&nd_mapping->lock, i);
2060 }
2061
2062 devs = scan_labels(nd_region);
2063
2064 for (i = 0; i < nd_region->ndr_mappings; i++) {
2065 int reverse = nd_region->ndr_mappings - 1 - i;
2066
2067 nd_mapping = &nd_region->mapping[reverse];
2068 mutex_unlock(&nd_mapping->lock);
2069 }
2070
2071 return devs;
2072}
2073
2074static void deactivate_labels(void *region)
2075{
2076 struct nd_region *nd_region = region;
2077 int i;
2078
2079 for (i = 0; i < nd_region->ndr_mappings; i++) {
2080 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2081 struct nvdimm_drvdata *ndd = nd_mapping->ndd;
2082 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2083
2084 mutex_lock(&nd_mapping->lock);
2085 nd_mapping_free_labels(nd_mapping);
2086 mutex_unlock(&nd_mapping->lock);
2087
2088 put_ndd(ndd);
2089 nd_mapping->ndd = NULL;
2090 if (ndd)
2091 atomic_dec(&nvdimm->busy);
2092 }
2093}
2094
2095static int init_active_labels(struct nd_region *nd_region)
2096{
2097 int i, rc = 0;
2098
2099 for (i = 0; i < nd_region->ndr_mappings; i++) {
2100 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
2101 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
2102 struct nvdimm *nvdimm = nd_mapping->nvdimm;
2103 struct nd_label_ent *label_ent;
2104 int count, j;
2105
2106 /*
2107 * If the dimm is disabled then we may need to prevent
2108 * the region from being activated.
2109 */
2110 if (!ndd) {
2111 if (test_bit(NDD_LOCKED, &nvdimm->flags))
2112 /* fail, label data may be unreadable */;
2113 else if (test_bit(NDD_LABELING, &nvdimm->flags))
2114 /* fail, labels needed to disambiguate dpa */;
2115 else
2116 continue;
2117
2118 dev_err(&nd_region->dev, "%s: is %s, failing probe\n",
2119 dev_name(&nd_mapping->nvdimm->dev),
2120 test_bit(NDD_LOCKED, &nvdimm->flags)
2121 ? "locked" : "disabled");
2122 rc = -ENXIO;
2123 goto out;
2124 }
2125 nd_mapping->ndd = ndd;
2126 atomic_inc(&nvdimm->busy);
2127 get_ndd(ndd);
2128
2129 count = nd_label_active_count(ndd);
2130 dev_dbg(ndd->dev, "count: %d\n", count);
2131 if (!count)
2132 continue;
2133 for (j = 0; j < count; j++) {
2134 struct nd_namespace_label *label;
2135
2136 label_ent = kzalloc(sizeof(*label_ent), GFP_KERNEL);
2137 if (!label_ent)
2138 break;
2139 label = nd_label_active(ndd, j);
2140 label_ent->label = label;
2141
2142 mutex_lock(&nd_mapping->lock);
2143 list_add_tail(&label_ent->list, &nd_mapping->labels);
2144 mutex_unlock(&nd_mapping->lock);
2145 }
2146
2147 if (j < count)
2148 break;
2149 }
2150
2151 if (i < nd_region->ndr_mappings)
2152 rc = -ENOMEM;
2153
2154out:
2155 if (rc) {
2156 deactivate_labels(nd_region);
2157 return rc;
2158 }
2159
2160 return devm_add_action_or_reset(&nd_region->dev, deactivate_labels,
2161 nd_region);
2162}
2163
2164int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
2165{
2166 struct device **devs = NULL;
2167 int i, rc = 0, type;
2168
2169 *err = 0;
2170 nvdimm_bus_lock(&nd_region->dev);
2171 rc = init_active_labels(nd_region);
2172 if (rc) {
2173 nvdimm_bus_unlock(&nd_region->dev);
2174 return rc;
2175 }
2176
2177 type = nd_region_to_nstype(nd_region);
2178 switch (type) {
2179 case ND_DEVICE_NAMESPACE_IO:
2180 devs = create_namespace_io(nd_region);
2181 break;
2182 case ND_DEVICE_NAMESPACE_PMEM:
2183 devs = create_namespaces(nd_region);
2184 break;
2185 default:
2186 break;
2187 }
2188 nvdimm_bus_unlock(&nd_region->dev);
2189
2190 if (!devs)
2191 return -ENODEV;
2192
2193 for (i = 0; devs[i]; i++) {
2194 struct device *dev = devs[i];
2195 int id;
2196
2197 if (type == ND_DEVICE_NAMESPACE_PMEM) {
2198 struct nd_namespace_pmem *nspm;
2199
2200 nspm = to_nd_namespace_pmem(dev);
2201 id = ida_alloc(&nd_region->ns_ida, GFP_KERNEL);
2202 nspm->id = id;
2203 } else
2204 id = i;
2205
2206 if (id < 0)
2207 break;
2208 dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
2209 device_initialize(dev);
2210 lockdep_set_class(&dev->mutex, &nvdimm_namespace_key);
2211 nd_device_register(dev);
2212 }
2213 if (i)
2214 nd_region->ns_seed = devs[0];
2215
2216 if (devs[i]) {
2217 int j;
2218
2219 for (j = i; devs[j]; j++) {
2220 struct device *dev = devs[j];
2221
2222 device_initialize(dev);
2223 put_device(dev);
2224 }
2225 *err = j - i;
2226 /*
2227 * All of the namespaces we tried to register failed, so
2228 * fail region activation.
2229 */
2230 if (*err == 0)
2231 rc = -ENODEV;
2232 }
2233 kfree(devs);
2234
2235 if (rc == -ENODEV)
2236 return rc;
2237
2238 return i;
2239}