Loading...
Note: File does not exist in v3.1.
1/*
2 * Copyright(c) 2013-2015 Intel Corporation. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 */
13#include <linux/module.h>
14#include <linux/device.h>
15#include <linux/slab.h>
16#include <linux/pmem.h>
17#include <linux/nd.h>
18#include "nd-core.h"
19#include "nd.h"
20
21static void namespace_io_release(struct device *dev)
22{
23 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
24
25 kfree(nsio);
26}
27
28static void namespace_pmem_release(struct device *dev)
29{
30 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
31
32 kfree(nspm->alt_name);
33 kfree(nspm->uuid);
34 kfree(nspm);
35}
36
37static void namespace_blk_release(struct device *dev)
38{
39 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
40 struct nd_region *nd_region = to_nd_region(dev->parent);
41
42 if (nsblk->id >= 0)
43 ida_simple_remove(&nd_region->ns_ida, nsblk->id);
44 kfree(nsblk->alt_name);
45 kfree(nsblk->uuid);
46 kfree(nsblk->res);
47 kfree(nsblk);
48}
49
50static struct device_type namespace_io_device_type = {
51 .name = "nd_namespace_io",
52 .release = namespace_io_release,
53};
54
55static struct device_type namespace_pmem_device_type = {
56 .name = "nd_namespace_pmem",
57 .release = namespace_pmem_release,
58};
59
60static struct device_type namespace_blk_device_type = {
61 .name = "nd_namespace_blk",
62 .release = namespace_blk_release,
63};
64
65static bool is_namespace_pmem(struct device *dev)
66{
67 return dev ? dev->type == &namespace_pmem_device_type : false;
68}
69
70static bool is_namespace_blk(struct device *dev)
71{
72 return dev ? dev->type == &namespace_blk_device_type : false;
73}
74
75static bool is_namespace_io(struct device *dev)
76{
77 return dev ? dev->type == &namespace_io_device_type : false;
78}
79
80static int is_uuid_busy(struct device *dev, void *data)
81{
82 u8 *uuid1 = data, *uuid2 = NULL;
83
84 if (is_namespace_pmem(dev)) {
85 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
86
87 uuid2 = nspm->uuid;
88 } else if (is_namespace_blk(dev)) {
89 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
90
91 uuid2 = nsblk->uuid;
92 } else if (is_nd_btt(dev)) {
93 struct nd_btt *nd_btt = to_nd_btt(dev);
94
95 uuid2 = nd_btt->uuid;
96 } else if (is_nd_pfn(dev)) {
97 struct nd_pfn *nd_pfn = to_nd_pfn(dev);
98
99 uuid2 = nd_pfn->uuid;
100 }
101
102 if (uuid2 && memcmp(uuid1, uuid2, NSLABEL_UUID_LEN) == 0)
103 return -EBUSY;
104
105 return 0;
106}
107
108static int is_namespace_uuid_busy(struct device *dev, void *data)
109{
110 if (is_nd_pmem(dev) || is_nd_blk(dev))
111 return device_for_each_child(dev, data, is_uuid_busy);
112 return 0;
113}
114
115/**
116 * nd_is_uuid_unique - verify that no other namespace has @uuid
117 * @dev: any device on a nvdimm_bus
118 * @uuid: uuid to check
119 */
120bool nd_is_uuid_unique(struct device *dev, u8 *uuid)
121{
122 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(dev);
123
124 if (!nvdimm_bus)
125 return false;
126 WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm_bus->dev));
127 if (device_for_each_child(&nvdimm_bus->dev, uuid,
128 is_namespace_uuid_busy) != 0)
129 return false;
130 return true;
131}
132
133bool pmem_should_map_pages(struct device *dev)
134{
135 struct nd_region *nd_region = to_nd_region(dev->parent);
136 struct nd_namespace_io *nsio;
137
138 if (!IS_ENABLED(CONFIG_ZONE_DEVICE))
139 return false;
140
141 if (!test_bit(ND_REGION_PAGEMAP, &nd_region->flags))
142 return false;
143
144 if (is_nd_pfn(dev) || is_nd_btt(dev))
145 return false;
146
147 nsio = to_nd_namespace_io(dev);
148 if (region_intersects(nsio->res.start, resource_size(&nsio->res),
149 IORESOURCE_SYSTEM_RAM,
150 IORES_DESC_NONE) == REGION_MIXED)
151 return false;
152
153#ifdef ARCH_MEMREMAP_PMEM
154 return ARCH_MEMREMAP_PMEM == MEMREMAP_WB;
155#else
156 return false;
157#endif
158}
159EXPORT_SYMBOL(pmem_should_map_pages);
160
161const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns,
162 char *name)
163{
164 struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
165 const char *suffix = NULL;
166
167 if (ndns->claim && is_nd_btt(ndns->claim))
168 suffix = "s";
169
170 if (is_namespace_pmem(&ndns->dev) || is_namespace_io(&ndns->dev)) {
171 sprintf(name, "pmem%d%s", nd_region->id, suffix ? suffix : "");
172 } else if (is_namespace_blk(&ndns->dev)) {
173 struct nd_namespace_blk *nsblk;
174
175 nsblk = to_nd_namespace_blk(&ndns->dev);
176 sprintf(name, "ndblk%d.%d%s", nd_region->id, nsblk->id,
177 suffix ? suffix : "");
178 } else {
179 return NULL;
180 }
181
182 return name;
183}
184EXPORT_SYMBOL(nvdimm_namespace_disk_name);
185
186const u8 *nd_dev_to_uuid(struct device *dev)
187{
188 static const u8 null_uuid[16];
189
190 if (!dev)
191 return null_uuid;
192
193 if (is_namespace_pmem(dev)) {
194 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
195
196 return nspm->uuid;
197 } else if (is_namespace_blk(dev)) {
198 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
199
200 return nsblk->uuid;
201 } else
202 return null_uuid;
203}
204EXPORT_SYMBOL(nd_dev_to_uuid);
205
206static ssize_t nstype_show(struct device *dev,
207 struct device_attribute *attr, char *buf)
208{
209 struct nd_region *nd_region = to_nd_region(dev->parent);
210
211 return sprintf(buf, "%d\n", nd_region_to_nstype(nd_region));
212}
213static DEVICE_ATTR_RO(nstype);
214
215static ssize_t __alt_name_store(struct device *dev, const char *buf,
216 const size_t len)
217{
218 char *input, *pos, *alt_name, **ns_altname;
219 ssize_t rc;
220
221 if (is_namespace_pmem(dev)) {
222 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
223
224 ns_altname = &nspm->alt_name;
225 } else if (is_namespace_blk(dev)) {
226 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
227
228 ns_altname = &nsblk->alt_name;
229 } else
230 return -ENXIO;
231
232 if (dev->driver || to_ndns(dev)->claim)
233 return -EBUSY;
234
235 input = kmemdup(buf, len + 1, GFP_KERNEL);
236 if (!input)
237 return -ENOMEM;
238
239 input[len] = '\0';
240 pos = strim(input);
241 if (strlen(pos) + 1 > NSLABEL_NAME_LEN) {
242 rc = -EINVAL;
243 goto out;
244 }
245
246 alt_name = kzalloc(NSLABEL_NAME_LEN, GFP_KERNEL);
247 if (!alt_name) {
248 rc = -ENOMEM;
249 goto out;
250 }
251 kfree(*ns_altname);
252 *ns_altname = alt_name;
253 sprintf(*ns_altname, "%s", pos);
254 rc = len;
255
256out:
257 kfree(input);
258 return rc;
259}
260
261static resource_size_t nd_namespace_blk_size(struct nd_namespace_blk *nsblk)
262{
263 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
264 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
265 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
266 struct nd_label_id label_id;
267 resource_size_t size = 0;
268 struct resource *res;
269
270 if (!nsblk->uuid)
271 return 0;
272 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
273 for_each_dpa_resource(ndd, res)
274 if (strcmp(res->name, label_id.id) == 0)
275 size += resource_size(res);
276 return size;
277}
278
279static bool __nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
280{
281 struct nd_region *nd_region = to_nd_region(nsblk->common.dev.parent);
282 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
283 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
284 struct nd_label_id label_id;
285 struct resource *res;
286 int count, i;
287
288 if (!nsblk->uuid || !nsblk->lbasize || !ndd)
289 return false;
290
291 count = 0;
292 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
293 for_each_dpa_resource(ndd, res) {
294 if (strcmp(res->name, label_id.id) != 0)
295 continue;
296 /*
297 * Resources with unacknoweldged adjustments indicate a
298 * failure to update labels
299 */
300 if (res->flags & DPA_RESOURCE_ADJUSTED)
301 return false;
302 count++;
303 }
304
305 /* These values match after a successful label update */
306 if (count != nsblk->num_resources)
307 return false;
308
309 for (i = 0; i < nsblk->num_resources; i++) {
310 struct resource *found = NULL;
311
312 for_each_dpa_resource(ndd, res)
313 if (res == nsblk->res[i]) {
314 found = res;
315 break;
316 }
317 /* stale resource */
318 if (!found)
319 return false;
320 }
321
322 return true;
323}
324
325resource_size_t nd_namespace_blk_validate(struct nd_namespace_blk *nsblk)
326{
327 resource_size_t size;
328
329 nvdimm_bus_lock(&nsblk->common.dev);
330 size = __nd_namespace_blk_validate(nsblk);
331 nvdimm_bus_unlock(&nsblk->common.dev);
332
333 return size;
334}
335EXPORT_SYMBOL(nd_namespace_blk_validate);
336
337
338static int nd_namespace_label_update(struct nd_region *nd_region,
339 struct device *dev)
340{
341 dev_WARN_ONCE(dev, dev->driver || to_ndns(dev)->claim,
342 "namespace must be idle during label update\n");
343 if (dev->driver || to_ndns(dev)->claim)
344 return 0;
345
346 /*
347 * Only allow label writes that will result in a valid namespace
348 * or deletion of an existing namespace.
349 */
350 if (is_namespace_pmem(dev)) {
351 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
352 resource_size_t size = resource_size(&nspm->nsio.res);
353
354 if (size == 0 && nspm->uuid)
355 /* delete allocation */;
356 else if (!nspm->uuid)
357 return 0;
358
359 return nd_pmem_namespace_label_update(nd_region, nspm, size);
360 } else if (is_namespace_blk(dev)) {
361 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
362 resource_size_t size = nd_namespace_blk_size(nsblk);
363
364 if (size == 0 && nsblk->uuid)
365 /* delete allocation */;
366 else if (!nsblk->uuid || !nsblk->lbasize)
367 return 0;
368
369 return nd_blk_namespace_label_update(nd_region, nsblk, size);
370 } else
371 return -ENXIO;
372}
373
374static ssize_t alt_name_store(struct device *dev,
375 struct device_attribute *attr, const char *buf, size_t len)
376{
377 struct nd_region *nd_region = to_nd_region(dev->parent);
378 ssize_t rc;
379
380 device_lock(dev);
381 nvdimm_bus_lock(dev);
382 wait_nvdimm_bus_probe_idle(dev);
383 rc = __alt_name_store(dev, buf, len);
384 if (rc >= 0)
385 rc = nd_namespace_label_update(nd_region, dev);
386 dev_dbg(dev, "%s: %s(%zd)\n", __func__, rc < 0 ? "fail " : "", rc);
387 nvdimm_bus_unlock(dev);
388 device_unlock(dev);
389
390 return rc < 0 ? rc : len;
391}
392
393static ssize_t alt_name_show(struct device *dev,
394 struct device_attribute *attr, char *buf)
395{
396 char *ns_altname;
397
398 if (is_namespace_pmem(dev)) {
399 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
400
401 ns_altname = nspm->alt_name;
402 } else if (is_namespace_blk(dev)) {
403 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
404
405 ns_altname = nsblk->alt_name;
406 } else
407 return -ENXIO;
408
409 return sprintf(buf, "%s\n", ns_altname ? ns_altname : "");
410}
411static DEVICE_ATTR_RW(alt_name);
412
413static int scan_free(struct nd_region *nd_region,
414 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
415 resource_size_t n)
416{
417 bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
418 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
419 int rc = 0;
420
421 while (n) {
422 struct resource *res, *last;
423 resource_size_t new_start;
424
425 last = NULL;
426 for_each_dpa_resource(ndd, res)
427 if (strcmp(res->name, label_id->id) == 0)
428 last = res;
429 res = last;
430 if (!res)
431 return 0;
432
433 if (n >= resource_size(res)) {
434 n -= resource_size(res);
435 nd_dbg_dpa(nd_region, ndd, res, "delete %d\n", rc);
436 nvdimm_free_dpa(ndd, res);
437 /* retry with last resource deleted */
438 continue;
439 }
440
441 /*
442 * Keep BLK allocations relegated to high DPA as much as
443 * possible
444 */
445 if (is_blk)
446 new_start = res->start + n;
447 else
448 new_start = res->start;
449
450 rc = adjust_resource(res, new_start, resource_size(res) - n);
451 if (rc == 0)
452 res->flags |= DPA_RESOURCE_ADJUSTED;
453 nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n", rc);
454 break;
455 }
456
457 return rc;
458}
459
460/**
461 * shrink_dpa_allocation - for each dimm in region free n bytes for label_id
462 * @nd_region: the set of dimms to reclaim @n bytes from
463 * @label_id: unique identifier for the namespace consuming this dpa range
464 * @n: number of bytes per-dimm to release
465 *
466 * Assumes resources are ordered. Starting from the end try to
467 * adjust_resource() the allocation to @n, but if @n is larger than the
468 * allocation delete it and find the 'new' last allocation in the label
469 * set.
470 */
471static int shrink_dpa_allocation(struct nd_region *nd_region,
472 struct nd_label_id *label_id, resource_size_t n)
473{
474 int i;
475
476 for (i = 0; i < nd_region->ndr_mappings; i++) {
477 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
478 int rc;
479
480 rc = scan_free(nd_region, nd_mapping, label_id, n);
481 if (rc)
482 return rc;
483 }
484
485 return 0;
486}
487
488static resource_size_t init_dpa_allocation(struct nd_label_id *label_id,
489 struct nd_region *nd_region, struct nd_mapping *nd_mapping,
490 resource_size_t n)
491{
492 bool is_blk = strncmp(label_id->id, "blk", 3) == 0;
493 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
494 resource_size_t first_dpa;
495 struct resource *res;
496 int rc = 0;
497
498 /* allocate blk from highest dpa first */
499 if (is_blk)
500 first_dpa = nd_mapping->start + nd_mapping->size - n;
501 else
502 first_dpa = nd_mapping->start;
503
504 /* first resource allocation for this label-id or dimm */
505 res = nvdimm_allocate_dpa(ndd, label_id, first_dpa, n);
506 if (!res)
507 rc = -EBUSY;
508
509 nd_dbg_dpa(nd_region, ndd, res, "init %d\n", rc);
510 return rc ? n : 0;
511}
512
513static bool space_valid(bool is_pmem, bool is_reserve,
514 struct nd_label_id *label_id, struct resource *res)
515{
516 /*
517 * For BLK-space any space is valid, for PMEM-space, it must be
518 * contiguous with an existing allocation unless we are
519 * reserving pmem.
520 */
521 if (is_reserve || !is_pmem)
522 return true;
523 if (!res || strcmp(res->name, label_id->id) == 0)
524 return true;
525 return false;
526}
527
528enum alloc_loc {
529 ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER,
530};
531
532static resource_size_t scan_allocate(struct nd_region *nd_region,
533 struct nd_mapping *nd_mapping, struct nd_label_id *label_id,
534 resource_size_t n)
535{
536 resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1;
537 bool is_reserve = strcmp(label_id->id, "pmem-reserve") == 0;
538 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
539 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
540 const resource_size_t to_allocate = n;
541 struct resource *res;
542 int first;
543
544 retry:
545 first = 0;
546 for_each_dpa_resource(ndd, res) {
547 resource_size_t allocate, available = 0, free_start, free_end;
548 struct resource *next = res->sibling, *new_res = NULL;
549 enum alloc_loc loc = ALLOC_ERR;
550 const char *action;
551 int rc = 0;
552
553 /* ignore resources outside this nd_mapping */
554 if (res->start > mapping_end)
555 continue;
556 if (res->end < nd_mapping->start)
557 continue;
558
559 /* space at the beginning of the mapping */
560 if (!first++ && res->start > nd_mapping->start) {
561 free_start = nd_mapping->start;
562 available = res->start - free_start;
563 if (space_valid(is_pmem, is_reserve, label_id, NULL))
564 loc = ALLOC_BEFORE;
565 }
566
567 /* space between allocations */
568 if (!loc && next) {
569 free_start = res->start + resource_size(res);
570 free_end = min(mapping_end, next->start - 1);
571 if (space_valid(is_pmem, is_reserve, label_id, res)
572 && free_start < free_end) {
573 available = free_end + 1 - free_start;
574 loc = ALLOC_MID;
575 }
576 }
577
578 /* space at the end of the mapping */
579 if (!loc && !next) {
580 free_start = res->start + resource_size(res);
581 free_end = mapping_end;
582 if (space_valid(is_pmem, is_reserve, label_id, res)
583 && free_start < free_end) {
584 available = free_end + 1 - free_start;
585 loc = ALLOC_AFTER;
586 }
587 }
588
589 if (!loc || !available)
590 continue;
591 allocate = min(available, n);
592 switch (loc) {
593 case ALLOC_BEFORE:
594 if (strcmp(res->name, label_id->id) == 0) {
595 /* adjust current resource up */
596 if (is_pmem && !is_reserve)
597 return n;
598 rc = adjust_resource(res, res->start - allocate,
599 resource_size(res) + allocate);
600 action = "cur grow up";
601 } else
602 action = "allocate";
603 break;
604 case ALLOC_MID:
605 if (strcmp(next->name, label_id->id) == 0) {
606 /* adjust next resource up */
607 if (is_pmem && !is_reserve)
608 return n;
609 rc = adjust_resource(next, next->start
610 - allocate, resource_size(next)
611 + allocate);
612 new_res = next;
613 action = "next grow up";
614 } else if (strcmp(res->name, label_id->id) == 0) {
615 action = "grow down";
616 } else
617 action = "allocate";
618 break;
619 case ALLOC_AFTER:
620 if (strcmp(res->name, label_id->id) == 0)
621 action = "grow down";
622 else
623 action = "allocate";
624 break;
625 default:
626 return n;
627 }
628
629 if (strcmp(action, "allocate") == 0) {
630 /* BLK allocate bottom up */
631 if (!is_pmem)
632 free_start += available - allocate;
633 else if (!is_reserve && free_start != nd_mapping->start)
634 return n;
635
636 new_res = nvdimm_allocate_dpa(ndd, label_id,
637 free_start, allocate);
638 if (!new_res)
639 rc = -EBUSY;
640 } else if (strcmp(action, "grow down") == 0) {
641 /* adjust current resource down */
642 rc = adjust_resource(res, res->start, resource_size(res)
643 + allocate);
644 if (rc == 0)
645 res->flags |= DPA_RESOURCE_ADJUSTED;
646 }
647
648 if (!new_res)
649 new_res = res;
650
651 nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n",
652 action, loc, rc);
653
654 if (rc)
655 return n;
656
657 n -= allocate;
658 if (n) {
659 /*
660 * Retry scan with newly inserted resources.
661 * For example, if we did an ALLOC_BEFORE
662 * insertion there may also have been space
663 * available for an ALLOC_AFTER insertion, so we
664 * need to check this same resource again
665 */
666 goto retry;
667 } else
668 return 0;
669 }
670
671 /*
672 * If we allocated nothing in the BLK case it may be because we are in
673 * an initial "pmem-reserve pass". Only do an initial BLK allocation
674 * when none of the DPA space is reserved.
675 */
676 if ((is_pmem || !ndd->dpa.child) && n == to_allocate)
677 return init_dpa_allocation(label_id, nd_region, nd_mapping, n);
678 return n;
679}
680
681static int merge_dpa(struct nd_region *nd_region,
682 struct nd_mapping *nd_mapping, struct nd_label_id *label_id)
683{
684 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
685 struct resource *res;
686
687 if (strncmp("pmem", label_id->id, 4) == 0)
688 return 0;
689 retry:
690 for_each_dpa_resource(ndd, res) {
691 int rc;
692 struct resource *next = res->sibling;
693 resource_size_t end = res->start + resource_size(res);
694
695 if (!next || strcmp(res->name, label_id->id) != 0
696 || strcmp(next->name, label_id->id) != 0
697 || end != next->start)
698 continue;
699 end += resource_size(next);
700 nvdimm_free_dpa(ndd, next);
701 rc = adjust_resource(res, res->start, end - res->start);
702 nd_dbg_dpa(nd_region, ndd, res, "merge %d\n", rc);
703 if (rc)
704 return rc;
705 res->flags |= DPA_RESOURCE_ADJUSTED;
706 goto retry;
707 }
708
709 return 0;
710}
711
712static int __reserve_free_pmem(struct device *dev, void *data)
713{
714 struct nvdimm *nvdimm = data;
715 struct nd_region *nd_region;
716 struct nd_label_id label_id;
717 int i;
718
719 if (!is_nd_pmem(dev))
720 return 0;
721
722 nd_region = to_nd_region(dev);
723 if (nd_region->ndr_mappings == 0)
724 return 0;
725
726 memset(&label_id, 0, sizeof(label_id));
727 strcat(label_id.id, "pmem-reserve");
728 for (i = 0; i < nd_region->ndr_mappings; i++) {
729 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
730 resource_size_t n, rem = 0;
731
732 if (nd_mapping->nvdimm != nvdimm)
733 continue;
734
735 n = nd_pmem_available_dpa(nd_region, nd_mapping, &rem);
736 if (n == 0)
737 return 0;
738 rem = scan_allocate(nd_region, nd_mapping, &label_id, n);
739 dev_WARN_ONCE(&nd_region->dev, rem,
740 "pmem reserve underrun: %#llx of %#llx bytes\n",
741 (unsigned long long) n - rem,
742 (unsigned long long) n);
743 return rem ? -ENXIO : 0;
744 }
745
746 return 0;
747}
748
749static void release_free_pmem(struct nvdimm_bus *nvdimm_bus,
750 struct nd_mapping *nd_mapping)
751{
752 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
753 struct resource *res, *_res;
754
755 for_each_dpa_resource_safe(ndd, res, _res)
756 if (strcmp(res->name, "pmem-reserve") == 0)
757 nvdimm_free_dpa(ndd, res);
758}
759
760static int reserve_free_pmem(struct nvdimm_bus *nvdimm_bus,
761 struct nd_mapping *nd_mapping)
762{
763 struct nvdimm *nvdimm = nd_mapping->nvdimm;
764 int rc;
765
766 rc = device_for_each_child(&nvdimm_bus->dev, nvdimm,
767 __reserve_free_pmem);
768 if (rc)
769 release_free_pmem(nvdimm_bus, nd_mapping);
770 return rc;
771}
772
773/**
774 * grow_dpa_allocation - for each dimm allocate n bytes for @label_id
775 * @nd_region: the set of dimms to allocate @n more bytes from
776 * @label_id: unique identifier for the namespace consuming this dpa range
777 * @n: number of bytes per-dimm to add to the existing allocation
778 *
779 * Assumes resources are ordered. For BLK regions, first consume
780 * BLK-only available DPA free space, then consume PMEM-aliased DPA
781 * space starting at the highest DPA. For PMEM regions start
782 * allocations from the start of an interleave set and end at the first
783 * BLK allocation or the end of the interleave set, whichever comes
784 * first.
785 */
786static int grow_dpa_allocation(struct nd_region *nd_region,
787 struct nd_label_id *label_id, resource_size_t n)
788{
789 struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(&nd_region->dev);
790 bool is_pmem = strncmp(label_id->id, "pmem", 4) == 0;
791 int i;
792
793 for (i = 0; i < nd_region->ndr_mappings; i++) {
794 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
795 resource_size_t rem = n;
796 int rc, j;
797
798 /*
799 * In the BLK case try once with all unallocated PMEM
800 * reserved, and once without
801 */
802 for (j = is_pmem; j < 2; j++) {
803 bool blk_only = j == 0;
804
805 if (blk_only) {
806 rc = reserve_free_pmem(nvdimm_bus, nd_mapping);
807 if (rc)
808 return rc;
809 }
810 rem = scan_allocate(nd_region, nd_mapping,
811 label_id, rem);
812 if (blk_only)
813 release_free_pmem(nvdimm_bus, nd_mapping);
814
815 /* try again and allow encroachments into PMEM */
816 if (rem == 0)
817 break;
818 }
819
820 dev_WARN_ONCE(&nd_region->dev, rem,
821 "allocation underrun: %#llx of %#llx bytes\n",
822 (unsigned long long) n - rem,
823 (unsigned long long) n);
824 if (rem)
825 return -ENXIO;
826
827 rc = merge_dpa(nd_region, nd_mapping, label_id);
828 if (rc)
829 return rc;
830 }
831
832 return 0;
833}
834
835static void nd_namespace_pmem_set_size(struct nd_region *nd_region,
836 struct nd_namespace_pmem *nspm, resource_size_t size)
837{
838 struct resource *res = &nspm->nsio.res;
839
840 res->start = nd_region->ndr_start;
841 res->end = nd_region->ndr_start + size - 1;
842}
843
844static bool uuid_not_set(const u8 *uuid, struct device *dev, const char *where)
845{
846 if (!uuid) {
847 dev_dbg(dev, "%s: uuid not set\n", where);
848 return true;
849 }
850 return false;
851}
852
853static ssize_t __size_store(struct device *dev, unsigned long long val)
854{
855 resource_size_t allocated = 0, available = 0;
856 struct nd_region *nd_region = to_nd_region(dev->parent);
857 struct nd_mapping *nd_mapping;
858 struct nvdimm_drvdata *ndd;
859 struct nd_label_id label_id;
860 u32 flags = 0, remainder;
861 u8 *uuid = NULL;
862 int rc, i;
863
864 if (dev->driver || to_ndns(dev)->claim)
865 return -EBUSY;
866
867 if (is_namespace_pmem(dev)) {
868 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
869
870 uuid = nspm->uuid;
871 } else if (is_namespace_blk(dev)) {
872 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
873
874 uuid = nsblk->uuid;
875 flags = NSLABEL_FLAG_LOCAL;
876 }
877
878 /*
879 * We need a uuid for the allocation-label and dimm(s) on which
880 * to store the label.
881 */
882 if (uuid_not_set(uuid, dev, __func__))
883 return -ENXIO;
884 if (nd_region->ndr_mappings == 0) {
885 dev_dbg(dev, "%s: not associated with dimm(s)\n", __func__);
886 return -ENXIO;
887 }
888
889 div_u64_rem(val, SZ_4K * nd_region->ndr_mappings, &remainder);
890 if (remainder) {
891 dev_dbg(dev, "%llu is not %dK aligned\n", val,
892 (SZ_4K * nd_region->ndr_mappings) / SZ_1K);
893 return -EINVAL;
894 }
895
896 nd_label_gen_id(&label_id, uuid, flags);
897 for (i = 0; i < nd_region->ndr_mappings; i++) {
898 nd_mapping = &nd_region->mapping[i];
899 ndd = to_ndd(nd_mapping);
900
901 /*
902 * All dimms in an interleave set, or the base dimm for a blk
903 * region, need to be enabled for the size to be changed.
904 */
905 if (!ndd)
906 return -ENXIO;
907
908 allocated += nvdimm_allocated_dpa(ndd, &label_id);
909 }
910 available = nd_region_available_dpa(nd_region);
911
912 if (val > available + allocated)
913 return -ENOSPC;
914
915 if (val == allocated)
916 return 0;
917
918 val = div_u64(val, nd_region->ndr_mappings);
919 allocated = div_u64(allocated, nd_region->ndr_mappings);
920 if (val < allocated)
921 rc = shrink_dpa_allocation(nd_region, &label_id,
922 allocated - val);
923 else
924 rc = grow_dpa_allocation(nd_region, &label_id, val - allocated);
925
926 if (rc)
927 return rc;
928
929 if (is_namespace_pmem(dev)) {
930 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
931
932 nd_namespace_pmem_set_size(nd_region, nspm,
933 val * nd_region->ndr_mappings);
934 } else if (is_namespace_blk(dev)) {
935 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
936
937 /*
938 * Try to delete the namespace if we deleted all of its
939 * allocation, this is not the seed device for the
940 * region, and it is not actively claimed by a btt
941 * instance.
942 */
943 if (val == 0 && nd_region->ns_seed != dev
944 && !nsblk->common.claim)
945 nd_device_unregister(dev, ND_ASYNC);
946 }
947
948 return rc;
949}
950
951static ssize_t size_store(struct device *dev,
952 struct device_attribute *attr, const char *buf, size_t len)
953{
954 struct nd_region *nd_region = to_nd_region(dev->parent);
955 unsigned long long val;
956 u8 **uuid = NULL;
957 int rc;
958
959 rc = kstrtoull(buf, 0, &val);
960 if (rc)
961 return rc;
962
963 device_lock(dev);
964 nvdimm_bus_lock(dev);
965 wait_nvdimm_bus_probe_idle(dev);
966 rc = __size_store(dev, val);
967 if (rc >= 0)
968 rc = nd_namespace_label_update(nd_region, dev);
969
970 if (is_namespace_pmem(dev)) {
971 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
972
973 uuid = &nspm->uuid;
974 } else if (is_namespace_blk(dev)) {
975 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
976
977 uuid = &nsblk->uuid;
978 }
979
980 if (rc == 0 && val == 0 && uuid) {
981 /* setting size zero == 'delete namespace' */
982 kfree(*uuid);
983 *uuid = NULL;
984 }
985
986 dev_dbg(dev, "%s: %llx %s (%d)\n", __func__, val, rc < 0
987 ? "fail" : "success", rc);
988
989 nvdimm_bus_unlock(dev);
990 device_unlock(dev);
991
992 return rc < 0 ? rc : len;
993}
994
995resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
996{
997 struct device *dev = &ndns->dev;
998
999 if (is_namespace_pmem(dev)) {
1000 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1001
1002 return resource_size(&nspm->nsio.res);
1003 } else if (is_namespace_blk(dev)) {
1004 return nd_namespace_blk_size(to_nd_namespace_blk(dev));
1005 } else if (is_namespace_io(dev)) {
1006 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1007
1008 return resource_size(&nsio->res);
1009 } else
1010 WARN_ONCE(1, "unknown namespace type\n");
1011 return 0;
1012}
1013
1014resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns)
1015{
1016 resource_size_t size;
1017
1018 nvdimm_bus_lock(&ndns->dev);
1019 size = __nvdimm_namespace_capacity(ndns);
1020 nvdimm_bus_unlock(&ndns->dev);
1021
1022 return size;
1023}
1024EXPORT_SYMBOL(nvdimm_namespace_capacity);
1025
1026static ssize_t size_show(struct device *dev,
1027 struct device_attribute *attr, char *buf)
1028{
1029 return sprintf(buf, "%llu\n", (unsigned long long)
1030 nvdimm_namespace_capacity(to_ndns(dev)));
1031}
1032static DEVICE_ATTR(size, S_IRUGO, size_show, size_store);
1033
1034static ssize_t uuid_show(struct device *dev,
1035 struct device_attribute *attr, char *buf)
1036{
1037 u8 *uuid;
1038
1039 if (is_namespace_pmem(dev)) {
1040 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1041
1042 uuid = nspm->uuid;
1043 } else if (is_namespace_blk(dev)) {
1044 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1045
1046 uuid = nsblk->uuid;
1047 } else
1048 return -ENXIO;
1049
1050 if (uuid)
1051 return sprintf(buf, "%pUb\n", uuid);
1052 return sprintf(buf, "\n");
1053}
1054
1055/**
1056 * namespace_update_uuid - check for a unique uuid and whether we're "renaming"
1057 * @nd_region: parent region so we can updates all dimms in the set
1058 * @dev: namespace type for generating label_id
1059 * @new_uuid: incoming uuid
1060 * @old_uuid: reference to the uuid storage location in the namespace object
1061 */
1062static int namespace_update_uuid(struct nd_region *nd_region,
1063 struct device *dev, u8 *new_uuid, u8 **old_uuid)
1064{
1065 u32 flags = is_namespace_blk(dev) ? NSLABEL_FLAG_LOCAL : 0;
1066 struct nd_label_id old_label_id;
1067 struct nd_label_id new_label_id;
1068 int i;
1069
1070 if (!nd_is_uuid_unique(dev, new_uuid))
1071 return -EINVAL;
1072
1073 if (*old_uuid == NULL)
1074 goto out;
1075
1076 /*
1077 * If we've already written a label with this uuid, then it's
1078 * too late to rename because we can't reliably update the uuid
1079 * without losing the old namespace. Userspace must delete this
1080 * namespace to abandon the old uuid.
1081 */
1082 for (i = 0; i < nd_region->ndr_mappings; i++) {
1083 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1084
1085 /*
1086 * This check by itself is sufficient because old_uuid
1087 * would be NULL above if this uuid did not exist in the
1088 * currently written set.
1089 *
1090 * FIXME: can we delete uuid with zero dpa allocated?
1091 */
1092 if (nd_mapping->labels)
1093 return -EBUSY;
1094 }
1095
1096 nd_label_gen_id(&old_label_id, *old_uuid, flags);
1097 nd_label_gen_id(&new_label_id, new_uuid, flags);
1098 for (i = 0; i < nd_region->ndr_mappings; i++) {
1099 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1100 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1101 struct resource *res;
1102
1103 for_each_dpa_resource(ndd, res)
1104 if (strcmp(res->name, old_label_id.id) == 0)
1105 sprintf((void *) res->name, "%s",
1106 new_label_id.id);
1107 }
1108 kfree(*old_uuid);
1109 out:
1110 *old_uuid = new_uuid;
1111 return 0;
1112}
1113
1114static ssize_t uuid_store(struct device *dev,
1115 struct device_attribute *attr, const char *buf, size_t len)
1116{
1117 struct nd_region *nd_region = to_nd_region(dev->parent);
1118 u8 *uuid = NULL;
1119 ssize_t rc = 0;
1120 u8 **ns_uuid;
1121
1122 if (is_namespace_pmem(dev)) {
1123 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1124
1125 ns_uuid = &nspm->uuid;
1126 } else if (is_namespace_blk(dev)) {
1127 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1128
1129 ns_uuid = &nsblk->uuid;
1130 } else
1131 return -ENXIO;
1132
1133 device_lock(dev);
1134 nvdimm_bus_lock(dev);
1135 wait_nvdimm_bus_probe_idle(dev);
1136 if (to_ndns(dev)->claim)
1137 rc = -EBUSY;
1138 if (rc >= 0)
1139 rc = nd_uuid_store(dev, &uuid, buf, len);
1140 if (rc >= 0)
1141 rc = namespace_update_uuid(nd_region, dev, uuid, ns_uuid);
1142 if (rc >= 0)
1143 rc = nd_namespace_label_update(nd_region, dev);
1144 else
1145 kfree(uuid);
1146 dev_dbg(dev, "%s: result: %zd wrote: %s%s", __func__,
1147 rc, buf, buf[len - 1] == '\n' ? "" : "\n");
1148 nvdimm_bus_unlock(dev);
1149 device_unlock(dev);
1150
1151 return rc < 0 ? rc : len;
1152}
1153static DEVICE_ATTR_RW(uuid);
1154
1155static ssize_t resource_show(struct device *dev,
1156 struct device_attribute *attr, char *buf)
1157{
1158 struct resource *res;
1159
1160 if (is_namespace_pmem(dev)) {
1161 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1162
1163 res = &nspm->nsio.res;
1164 } else if (is_namespace_io(dev)) {
1165 struct nd_namespace_io *nsio = to_nd_namespace_io(dev);
1166
1167 res = &nsio->res;
1168 } else
1169 return -ENXIO;
1170
1171 /* no address to convey if the namespace has no allocation */
1172 if (resource_size(res) == 0)
1173 return -ENXIO;
1174 return sprintf(buf, "%#llx\n", (unsigned long long) res->start);
1175}
1176static DEVICE_ATTR_RO(resource);
1177
1178static const unsigned long ns_lbasize_supported[] = { 512, 520, 528,
1179 4096, 4104, 4160, 4224, 0 };
1180
1181static ssize_t sector_size_show(struct device *dev,
1182 struct device_attribute *attr, char *buf)
1183{
1184 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1185
1186 if (!is_namespace_blk(dev))
1187 return -ENXIO;
1188
1189 return nd_sector_size_show(nsblk->lbasize, ns_lbasize_supported, buf);
1190}
1191
1192static ssize_t sector_size_store(struct device *dev,
1193 struct device_attribute *attr, const char *buf, size_t len)
1194{
1195 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1196 struct nd_region *nd_region = to_nd_region(dev->parent);
1197 ssize_t rc = 0;
1198
1199 if (!is_namespace_blk(dev))
1200 return -ENXIO;
1201
1202 device_lock(dev);
1203 nvdimm_bus_lock(dev);
1204 if (to_ndns(dev)->claim)
1205 rc = -EBUSY;
1206 if (rc >= 0)
1207 rc = nd_sector_size_store(dev, buf, &nsblk->lbasize,
1208 ns_lbasize_supported);
1209 if (rc >= 0)
1210 rc = nd_namespace_label_update(nd_region, dev);
1211 dev_dbg(dev, "%s: result: %zd %s: %s%s", __func__,
1212 rc, rc < 0 ? "tried" : "wrote", buf,
1213 buf[len - 1] == '\n' ? "" : "\n");
1214 nvdimm_bus_unlock(dev);
1215 device_unlock(dev);
1216
1217 return rc ? rc : len;
1218}
1219static DEVICE_ATTR_RW(sector_size);
1220
1221static ssize_t dpa_extents_show(struct device *dev,
1222 struct device_attribute *attr, char *buf)
1223{
1224 struct nd_region *nd_region = to_nd_region(dev->parent);
1225 struct nd_label_id label_id;
1226 int count = 0, i;
1227 u8 *uuid = NULL;
1228 u32 flags = 0;
1229
1230 nvdimm_bus_lock(dev);
1231 if (is_namespace_pmem(dev)) {
1232 struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev);
1233
1234 uuid = nspm->uuid;
1235 flags = 0;
1236 } else if (is_namespace_blk(dev)) {
1237 struct nd_namespace_blk *nsblk = to_nd_namespace_blk(dev);
1238
1239 uuid = nsblk->uuid;
1240 flags = NSLABEL_FLAG_LOCAL;
1241 }
1242
1243 if (!uuid)
1244 goto out;
1245
1246 nd_label_gen_id(&label_id, uuid, flags);
1247 for (i = 0; i < nd_region->ndr_mappings; i++) {
1248 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1249 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1250 struct resource *res;
1251
1252 for_each_dpa_resource(ndd, res)
1253 if (strcmp(res->name, label_id.id) == 0)
1254 count++;
1255 }
1256 out:
1257 nvdimm_bus_unlock(dev);
1258
1259 return sprintf(buf, "%d\n", count);
1260}
1261static DEVICE_ATTR_RO(dpa_extents);
1262
1263static ssize_t holder_show(struct device *dev,
1264 struct device_attribute *attr, char *buf)
1265{
1266 struct nd_namespace_common *ndns = to_ndns(dev);
1267 ssize_t rc;
1268
1269 device_lock(dev);
1270 rc = sprintf(buf, "%s\n", ndns->claim ? dev_name(ndns->claim) : "");
1271 device_unlock(dev);
1272
1273 return rc;
1274}
1275static DEVICE_ATTR_RO(holder);
1276
1277static ssize_t mode_show(struct device *dev,
1278 struct device_attribute *attr, char *buf)
1279{
1280 struct nd_namespace_common *ndns = to_ndns(dev);
1281 struct device *claim;
1282 char *mode;
1283 ssize_t rc;
1284
1285 device_lock(dev);
1286 claim = ndns->claim;
1287 if (claim && is_nd_btt(claim))
1288 mode = "safe";
1289 else if (claim && is_nd_pfn(claim))
1290 mode = "memory";
1291 else if (!claim && pmem_should_map_pages(dev))
1292 mode = "memory";
1293 else
1294 mode = "raw";
1295 rc = sprintf(buf, "%s\n", mode);
1296 device_unlock(dev);
1297
1298 return rc;
1299}
1300static DEVICE_ATTR_RO(mode);
1301
1302static ssize_t force_raw_store(struct device *dev,
1303 struct device_attribute *attr, const char *buf, size_t len)
1304{
1305 bool force_raw;
1306 int rc = strtobool(buf, &force_raw);
1307
1308 if (rc)
1309 return rc;
1310
1311 to_ndns(dev)->force_raw = force_raw;
1312 return len;
1313}
1314
1315static ssize_t force_raw_show(struct device *dev,
1316 struct device_attribute *attr, char *buf)
1317{
1318 return sprintf(buf, "%d\n", to_ndns(dev)->force_raw);
1319}
1320static DEVICE_ATTR_RW(force_raw);
1321
1322static struct attribute *nd_namespace_attributes[] = {
1323 &dev_attr_nstype.attr,
1324 &dev_attr_size.attr,
1325 &dev_attr_mode.attr,
1326 &dev_attr_uuid.attr,
1327 &dev_attr_holder.attr,
1328 &dev_attr_resource.attr,
1329 &dev_attr_alt_name.attr,
1330 &dev_attr_force_raw.attr,
1331 &dev_attr_sector_size.attr,
1332 &dev_attr_dpa_extents.attr,
1333 NULL,
1334};
1335
1336static umode_t namespace_visible(struct kobject *kobj,
1337 struct attribute *a, int n)
1338{
1339 struct device *dev = container_of(kobj, struct device, kobj);
1340
1341 if (a == &dev_attr_resource.attr) {
1342 if (is_namespace_blk(dev))
1343 return 0;
1344 return a->mode;
1345 }
1346
1347 if (is_namespace_pmem(dev) || is_namespace_blk(dev)) {
1348 if (a == &dev_attr_size.attr)
1349 return S_IWUSR | S_IRUGO;
1350
1351 if (is_namespace_pmem(dev) && a == &dev_attr_sector_size.attr)
1352 return 0;
1353
1354 return a->mode;
1355 }
1356
1357 if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr
1358 || a == &dev_attr_holder.attr
1359 || a == &dev_attr_force_raw.attr
1360 || a == &dev_attr_mode.attr)
1361 return a->mode;
1362
1363 return 0;
1364}
1365
1366static struct attribute_group nd_namespace_attribute_group = {
1367 .attrs = nd_namespace_attributes,
1368 .is_visible = namespace_visible,
1369};
1370
1371static const struct attribute_group *nd_namespace_attribute_groups[] = {
1372 &nd_device_attribute_group,
1373 &nd_namespace_attribute_group,
1374 &nd_numa_attribute_group,
1375 NULL,
1376};
1377
1378struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev)
1379{
1380 struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL;
1381 struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL;
1382 struct nd_namespace_common *ndns;
1383 resource_size_t size;
1384
1385 if (nd_btt || nd_pfn) {
1386 struct device *host = NULL;
1387
1388 if (nd_btt) {
1389 host = &nd_btt->dev;
1390 ndns = nd_btt->ndns;
1391 } else if (nd_pfn) {
1392 host = &nd_pfn->dev;
1393 ndns = nd_pfn->ndns;
1394 }
1395
1396 if (!ndns || !host)
1397 return ERR_PTR(-ENODEV);
1398
1399 /*
1400 * Flush any in-progess probes / removals in the driver
1401 * for the raw personality of this namespace.
1402 */
1403 device_lock(&ndns->dev);
1404 device_unlock(&ndns->dev);
1405 if (ndns->dev.driver) {
1406 dev_dbg(&ndns->dev, "is active, can't bind %s\n",
1407 dev_name(host));
1408 return ERR_PTR(-EBUSY);
1409 }
1410 if (dev_WARN_ONCE(&ndns->dev, ndns->claim != host,
1411 "host (%s) vs claim (%s) mismatch\n",
1412 dev_name(host),
1413 dev_name(ndns->claim)))
1414 return ERR_PTR(-ENXIO);
1415 } else {
1416 ndns = to_ndns(dev);
1417 if (ndns->claim) {
1418 dev_dbg(dev, "claimed by %s, failing probe\n",
1419 dev_name(ndns->claim));
1420
1421 return ERR_PTR(-ENXIO);
1422 }
1423 }
1424
1425 size = nvdimm_namespace_capacity(ndns);
1426 if (size < ND_MIN_NAMESPACE_SIZE) {
1427 dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n",
1428 &size, ND_MIN_NAMESPACE_SIZE);
1429 return ERR_PTR(-ENODEV);
1430 }
1431
1432 if (is_namespace_pmem(&ndns->dev)) {
1433 struct nd_namespace_pmem *nspm;
1434
1435 nspm = to_nd_namespace_pmem(&ndns->dev);
1436 if (uuid_not_set(nspm->uuid, &ndns->dev, __func__))
1437 return ERR_PTR(-ENODEV);
1438 } else if (is_namespace_blk(&ndns->dev)) {
1439 struct nd_namespace_blk *nsblk;
1440
1441 nsblk = to_nd_namespace_blk(&ndns->dev);
1442 if (uuid_not_set(nsblk->uuid, &ndns->dev, __func__))
1443 return ERR_PTR(-ENODEV);
1444 if (!nsblk->lbasize) {
1445 dev_dbg(&ndns->dev, "%s: sector size not set\n",
1446 __func__);
1447 return ERR_PTR(-ENODEV);
1448 }
1449 if (!nd_namespace_blk_validate(nsblk))
1450 return ERR_PTR(-ENODEV);
1451 }
1452
1453 return ndns;
1454}
1455EXPORT_SYMBOL(nvdimm_namespace_common_probe);
1456
1457static struct device **create_namespace_io(struct nd_region *nd_region)
1458{
1459 struct nd_namespace_io *nsio;
1460 struct device *dev, **devs;
1461 struct resource *res;
1462
1463 nsio = kzalloc(sizeof(*nsio), GFP_KERNEL);
1464 if (!nsio)
1465 return NULL;
1466
1467 devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1468 if (!devs) {
1469 kfree(nsio);
1470 return NULL;
1471 }
1472
1473 dev = &nsio->common.dev;
1474 dev->type = &namespace_io_device_type;
1475 dev->parent = &nd_region->dev;
1476 res = &nsio->res;
1477 res->name = dev_name(&nd_region->dev);
1478 res->flags = IORESOURCE_MEM;
1479 res->start = nd_region->ndr_start;
1480 res->end = res->start + nd_region->ndr_size - 1;
1481
1482 devs[0] = dev;
1483 return devs;
1484}
1485
1486static bool has_uuid_at_pos(struct nd_region *nd_region, u8 *uuid,
1487 u64 cookie, u16 pos)
1488{
1489 struct nd_namespace_label *found = NULL;
1490 int i;
1491
1492 for (i = 0; i < nd_region->ndr_mappings; i++) {
1493 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1494 struct nd_namespace_label *nd_label;
1495 bool found_uuid = false;
1496 int l;
1497
1498 for_each_label(l, nd_label, nd_mapping->labels) {
1499 u64 isetcookie = __le64_to_cpu(nd_label->isetcookie);
1500 u16 position = __le16_to_cpu(nd_label->position);
1501 u16 nlabel = __le16_to_cpu(nd_label->nlabel);
1502
1503 if (isetcookie != cookie)
1504 continue;
1505
1506 if (memcmp(nd_label->uuid, uuid, NSLABEL_UUID_LEN) != 0)
1507 continue;
1508
1509 if (found_uuid) {
1510 dev_dbg(to_ndd(nd_mapping)->dev,
1511 "%s duplicate entry for uuid\n",
1512 __func__);
1513 return false;
1514 }
1515 found_uuid = true;
1516 if (nlabel != nd_region->ndr_mappings)
1517 continue;
1518 if (position != pos)
1519 continue;
1520 found = nd_label;
1521 break;
1522 }
1523 if (found)
1524 break;
1525 }
1526 return found != NULL;
1527}
1528
1529static int select_pmem_id(struct nd_region *nd_region, u8 *pmem_id)
1530{
1531 struct nd_namespace_label *select = NULL;
1532 int i;
1533
1534 if (!pmem_id)
1535 return -ENODEV;
1536
1537 for (i = 0; i < nd_region->ndr_mappings; i++) {
1538 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1539 struct nd_namespace_label *nd_label;
1540 u64 hw_start, hw_end, pmem_start, pmem_end;
1541 int l;
1542
1543 for_each_label(l, nd_label, nd_mapping->labels)
1544 if (memcmp(nd_label->uuid, pmem_id, NSLABEL_UUID_LEN) == 0)
1545 break;
1546
1547 if (!nd_label) {
1548 WARN_ON(1);
1549 return -EINVAL;
1550 }
1551
1552 select = nd_label;
1553 /*
1554 * Check that this label is compliant with the dpa
1555 * range published in NFIT
1556 */
1557 hw_start = nd_mapping->start;
1558 hw_end = hw_start + nd_mapping->size;
1559 pmem_start = __le64_to_cpu(select->dpa);
1560 pmem_end = pmem_start + __le64_to_cpu(select->rawsize);
1561 if (pmem_start == hw_start && pmem_end <= hw_end)
1562 /* pass */;
1563 else
1564 return -EINVAL;
1565
1566 nd_mapping->labels[0] = select;
1567 nd_mapping->labels[1] = NULL;
1568 }
1569 return 0;
1570}
1571
1572/**
1573 * find_pmem_label_set - validate interleave set labelling, retrieve label0
1574 * @nd_region: region with mappings to validate
1575 */
1576static int find_pmem_label_set(struct nd_region *nd_region,
1577 struct nd_namespace_pmem *nspm)
1578{
1579 u64 cookie = nd_region_interleave_set_cookie(nd_region);
1580 struct nd_namespace_label *nd_label;
1581 u8 select_id[NSLABEL_UUID_LEN];
1582 resource_size_t size = 0;
1583 u8 *pmem_id = NULL;
1584 int rc = -ENODEV, l;
1585 u16 i;
1586
1587 if (cookie == 0)
1588 return -ENXIO;
1589
1590 /*
1591 * Find a complete set of labels by uuid. By definition we can start
1592 * with any mapping as the reference label
1593 */
1594 for_each_label(l, nd_label, nd_region->mapping[0].labels) {
1595 u64 isetcookie = __le64_to_cpu(nd_label->isetcookie);
1596
1597 if (isetcookie != cookie)
1598 continue;
1599
1600 for (i = 0; nd_region->ndr_mappings; i++)
1601 if (!has_uuid_at_pos(nd_region, nd_label->uuid,
1602 cookie, i))
1603 break;
1604 if (i < nd_region->ndr_mappings) {
1605 /*
1606 * Give up if we don't find an instance of a
1607 * uuid at each position (from 0 to
1608 * nd_region->ndr_mappings - 1), or if we find a
1609 * dimm with two instances of the same uuid.
1610 */
1611 rc = -EINVAL;
1612 goto err;
1613 } else if (pmem_id) {
1614 /*
1615 * If there is more than one valid uuid set, we
1616 * need userspace to clean this up.
1617 */
1618 rc = -EBUSY;
1619 goto err;
1620 }
1621 memcpy(select_id, nd_label->uuid, NSLABEL_UUID_LEN);
1622 pmem_id = select_id;
1623 }
1624
1625 /*
1626 * Fix up each mapping's 'labels' to have the validated pmem label for
1627 * that position at labels[0], and NULL at labels[1]. In the process,
1628 * check that the namespace aligns with interleave-set. We know
1629 * that it does not overlap with any blk namespaces by virtue of
1630 * the dimm being enabled (i.e. nd_label_reserve_dpa()
1631 * succeeded).
1632 */
1633 rc = select_pmem_id(nd_region, pmem_id);
1634 if (rc)
1635 goto err;
1636
1637 /* Calculate total size and populate namespace properties from label0 */
1638 for (i = 0; i < nd_region->ndr_mappings; i++) {
1639 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1640 struct nd_namespace_label *label0 = nd_mapping->labels[0];
1641
1642 size += __le64_to_cpu(label0->rawsize);
1643 if (__le16_to_cpu(label0->position) != 0)
1644 continue;
1645 WARN_ON(nspm->alt_name || nspm->uuid);
1646 nspm->alt_name = kmemdup((void __force *) label0->name,
1647 NSLABEL_NAME_LEN, GFP_KERNEL);
1648 nspm->uuid = kmemdup((void __force *) label0->uuid,
1649 NSLABEL_UUID_LEN, GFP_KERNEL);
1650 }
1651
1652 if (!nspm->alt_name || !nspm->uuid) {
1653 rc = -ENOMEM;
1654 goto err;
1655 }
1656
1657 nd_namespace_pmem_set_size(nd_region, nspm, size);
1658
1659 return 0;
1660 err:
1661 switch (rc) {
1662 case -EINVAL:
1663 dev_dbg(&nd_region->dev, "%s: invalid label(s)\n", __func__);
1664 break;
1665 case -ENODEV:
1666 dev_dbg(&nd_region->dev, "%s: label not found\n", __func__);
1667 break;
1668 default:
1669 dev_dbg(&nd_region->dev, "%s: unexpected err: %d\n",
1670 __func__, rc);
1671 break;
1672 }
1673 return rc;
1674}
1675
1676static struct device **create_namespace_pmem(struct nd_region *nd_region)
1677{
1678 struct nd_namespace_pmem *nspm;
1679 struct device *dev, **devs;
1680 struct resource *res;
1681 int rc;
1682
1683 nspm = kzalloc(sizeof(*nspm), GFP_KERNEL);
1684 if (!nspm)
1685 return NULL;
1686
1687 dev = &nspm->nsio.common.dev;
1688 dev->type = &namespace_pmem_device_type;
1689 dev->parent = &nd_region->dev;
1690 res = &nspm->nsio.res;
1691 res->name = dev_name(&nd_region->dev);
1692 res->flags = IORESOURCE_MEM;
1693 rc = find_pmem_label_set(nd_region, nspm);
1694 if (rc == -ENODEV) {
1695 int i;
1696
1697 /* Pass, try to permit namespace creation... */
1698 for (i = 0; i < nd_region->ndr_mappings; i++) {
1699 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1700
1701 kfree(nd_mapping->labels);
1702 nd_mapping->labels = NULL;
1703 }
1704
1705 /* Publish a zero-sized namespace for userspace to configure. */
1706 nd_namespace_pmem_set_size(nd_region, nspm, 0);
1707
1708 rc = 0;
1709 } else if (rc)
1710 goto err;
1711
1712 devs = kcalloc(2, sizeof(struct device *), GFP_KERNEL);
1713 if (!devs)
1714 goto err;
1715
1716 devs[0] = dev;
1717 return devs;
1718
1719 err:
1720 namespace_pmem_release(&nspm->nsio.common.dev);
1721 return NULL;
1722}
1723
1724struct resource *nsblk_add_resource(struct nd_region *nd_region,
1725 struct nvdimm_drvdata *ndd, struct nd_namespace_blk *nsblk,
1726 resource_size_t start)
1727{
1728 struct nd_label_id label_id;
1729 struct resource *res;
1730
1731 nd_label_gen_id(&label_id, nsblk->uuid, NSLABEL_FLAG_LOCAL);
1732 res = krealloc(nsblk->res,
1733 sizeof(void *) * (nsblk->num_resources + 1),
1734 GFP_KERNEL);
1735 if (!res)
1736 return NULL;
1737 nsblk->res = (struct resource **) res;
1738 for_each_dpa_resource(ndd, res)
1739 if (strcmp(res->name, label_id.id) == 0
1740 && res->start == start) {
1741 nsblk->res[nsblk->num_resources++] = res;
1742 return res;
1743 }
1744 return NULL;
1745}
1746
1747static struct device *nd_namespace_blk_create(struct nd_region *nd_region)
1748{
1749 struct nd_namespace_blk *nsblk;
1750 struct device *dev;
1751
1752 if (!is_nd_blk(&nd_region->dev))
1753 return NULL;
1754
1755 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
1756 if (!nsblk)
1757 return NULL;
1758
1759 dev = &nsblk->common.dev;
1760 dev->type = &namespace_blk_device_type;
1761 nsblk->id = ida_simple_get(&nd_region->ns_ida, 0, 0, GFP_KERNEL);
1762 if (nsblk->id < 0) {
1763 kfree(nsblk);
1764 return NULL;
1765 }
1766 dev_set_name(dev, "namespace%d.%d", nd_region->id, nsblk->id);
1767 dev->parent = &nd_region->dev;
1768 dev->groups = nd_namespace_attribute_groups;
1769
1770 return &nsblk->common.dev;
1771}
1772
1773void nd_region_create_blk_seed(struct nd_region *nd_region)
1774{
1775 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1776 nd_region->ns_seed = nd_namespace_blk_create(nd_region);
1777 /*
1778 * Seed creation failures are not fatal, provisioning is simply
1779 * disabled until memory becomes available
1780 */
1781 if (!nd_region->ns_seed)
1782 dev_err(&nd_region->dev, "failed to create blk namespace\n");
1783 else
1784 nd_device_register(nd_region->ns_seed);
1785}
1786
1787void nd_region_create_pfn_seed(struct nd_region *nd_region)
1788{
1789 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1790 nd_region->pfn_seed = nd_pfn_create(nd_region);
1791 /*
1792 * Seed creation failures are not fatal, provisioning is simply
1793 * disabled until memory becomes available
1794 */
1795 if (!nd_region->pfn_seed)
1796 dev_err(&nd_region->dev, "failed to create pfn namespace\n");
1797}
1798
1799void nd_region_create_btt_seed(struct nd_region *nd_region)
1800{
1801 WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev));
1802 nd_region->btt_seed = nd_btt_create(nd_region);
1803 /*
1804 * Seed creation failures are not fatal, provisioning is simply
1805 * disabled until memory becomes available
1806 */
1807 if (!nd_region->btt_seed)
1808 dev_err(&nd_region->dev, "failed to create btt namespace\n");
1809}
1810
1811static struct device **create_namespace_blk(struct nd_region *nd_region)
1812{
1813 struct nd_mapping *nd_mapping = &nd_region->mapping[0];
1814 struct nd_namespace_label *nd_label;
1815 struct device *dev, **devs = NULL;
1816 struct nd_namespace_blk *nsblk;
1817 struct nvdimm_drvdata *ndd;
1818 int i, l, count = 0;
1819 struct resource *res;
1820
1821 if (nd_region->ndr_mappings == 0)
1822 return NULL;
1823
1824 ndd = to_ndd(nd_mapping);
1825 for_each_label(l, nd_label, nd_mapping->labels) {
1826 u32 flags = __le32_to_cpu(nd_label->flags);
1827 char *name[NSLABEL_NAME_LEN];
1828 struct device **__devs;
1829
1830 if (flags & NSLABEL_FLAG_LOCAL)
1831 /* pass */;
1832 else
1833 continue;
1834
1835 for (i = 0; i < count; i++) {
1836 nsblk = to_nd_namespace_blk(devs[i]);
1837 if (memcmp(nsblk->uuid, nd_label->uuid,
1838 NSLABEL_UUID_LEN) == 0) {
1839 res = nsblk_add_resource(nd_region, ndd, nsblk,
1840 __le64_to_cpu(nd_label->dpa));
1841 if (!res)
1842 goto err;
1843 nd_dbg_dpa(nd_region, ndd, res, "%s assign\n",
1844 dev_name(&nsblk->common.dev));
1845 break;
1846 }
1847 }
1848 if (i < count)
1849 continue;
1850 __devs = kcalloc(count + 2, sizeof(dev), GFP_KERNEL);
1851 if (!__devs)
1852 goto err;
1853 memcpy(__devs, devs, sizeof(dev) * count);
1854 kfree(devs);
1855 devs = __devs;
1856
1857 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
1858 if (!nsblk)
1859 goto err;
1860 dev = &nsblk->common.dev;
1861 dev->type = &namespace_blk_device_type;
1862 dev->parent = &nd_region->dev;
1863 dev_set_name(dev, "namespace%d.%d", nd_region->id, count);
1864 devs[count++] = dev;
1865 nsblk->id = -1;
1866 nsblk->lbasize = __le64_to_cpu(nd_label->lbasize);
1867 nsblk->uuid = kmemdup(nd_label->uuid, NSLABEL_UUID_LEN,
1868 GFP_KERNEL);
1869 if (!nsblk->uuid)
1870 goto err;
1871 memcpy(name, nd_label->name, NSLABEL_NAME_LEN);
1872 if (name[0])
1873 nsblk->alt_name = kmemdup(name, NSLABEL_NAME_LEN,
1874 GFP_KERNEL);
1875 res = nsblk_add_resource(nd_region, ndd, nsblk,
1876 __le64_to_cpu(nd_label->dpa));
1877 if (!res)
1878 goto err;
1879 nd_dbg_dpa(nd_region, ndd, res, "%s assign\n",
1880 dev_name(&nsblk->common.dev));
1881 }
1882
1883 dev_dbg(&nd_region->dev, "%s: discovered %d blk namespace%s\n",
1884 __func__, count, count == 1 ? "" : "s");
1885
1886 if (count == 0) {
1887 /* Publish a zero-sized namespace for userspace to configure. */
1888 for (i = 0; i < nd_region->ndr_mappings; i++) {
1889 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1890
1891 kfree(nd_mapping->labels);
1892 nd_mapping->labels = NULL;
1893 }
1894
1895 devs = kcalloc(2, sizeof(dev), GFP_KERNEL);
1896 if (!devs)
1897 goto err;
1898 nsblk = kzalloc(sizeof(*nsblk), GFP_KERNEL);
1899 if (!nsblk)
1900 goto err;
1901 dev = &nsblk->common.dev;
1902 dev->type = &namespace_blk_device_type;
1903 dev->parent = &nd_region->dev;
1904 devs[count++] = dev;
1905 }
1906
1907 return devs;
1908
1909err:
1910 for (i = 0; i < count; i++) {
1911 nsblk = to_nd_namespace_blk(devs[i]);
1912 namespace_blk_release(&nsblk->common.dev);
1913 }
1914 kfree(devs);
1915 return NULL;
1916}
1917
1918static int init_active_labels(struct nd_region *nd_region)
1919{
1920 int i;
1921
1922 for (i = 0; i < nd_region->ndr_mappings; i++) {
1923 struct nd_mapping *nd_mapping = &nd_region->mapping[i];
1924 struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
1925 struct nvdimm *nvdimm = nd_mapping->nvdimm;
1926 int count, j;
1927
1928 /*
1929 * If the dimm is disabled then prevent the region from
1930 * being activated if it aliases DPA.
1931 */
1932 if (!ndd) {
1933 if ((nvdimm->flags & NDD_ALIASING) == 0)
1934 return 0;
1935 dev_dbg(&nd_region->dev, "%s: is disabled, failing probe\n",
1936 dev_name(&nd_mapping->nvdimm->dev));
1937 return -ENXIO;
1938 }
1939 nd_mapping->ndd = ndd;
1940 atomic_inc(&nvdimm->busy);
1941 get_ndd(ndd);
1942
1943 count = nd_label_active_count(ndd);
1944 dev_dbg(ndd->dev, "%s: %d\n", __func__, count);
1945 if (!count)
1946 continue;
1947 nd_mapping->labels = kcalloc(count + 1, sizeof(void *),
1948 GFP_KERNEL);
1949 if (!nd_mapping->labels)
1950 return -ENOMEM;
1951 for (j = 0; j < count; j++) {
1952 struct nd_namespace_label *label;
1953
1954 label = nd_label_active(ndd, j);
1955 nd_mapping->labels[j] = label;
1956 }
1957 }
1958
1959 return 0;
1960}
1961
1962int nd_region_register_namespaces(struct nd_region *nd_region, int *err)
1963{
1964 struct device **devs = NULL;
1965 int i, rc = 0, type;
1966
1967 *err = 0;
1968 nvdimm_bus_lock(&nd_region->dev);
1969 rc = init_active_labels(nd_region);
1970 if (rc) {
1971 nvdimm_bus_unlock(&nd_region->dev);
1972 return rc;
1973 }
1974
1975 type = nd_region_to_nstype(nd_region);
1976 switch (type) {
1977 case ND_DEVICE_NAMESPACE_IO:
1978 devs = create_namespace_io(nd_region);
1979 break;
1980 case ND_DEVICE_NAMESPACE_PMEM:
1981 devs = create_namespace_pmem(nd_region);
1982 break;
1983 case ND_DEVICE_NAMESPACE_BLK:
1984 devs = create_namespace_blk(nd_region);
1985 break;
1986 default:
1987 break;
1988 }
1989 nvdimm_bus_unlock(&nd_region->dev);
1990
1991 if (!devs)
1992 return -ENODEV;
1993
1994 for (i = 0; devs[i]; i++) {
1995 struct device *dev = devs[i];
1996 int id;
1997
1998 if (type == ND_DEVICE_NAMESPACE_BLK) {
1999 struct nd_namespace_blk *nsblk;
2000
2001 nsblk = to_nd_namespace_blk(dev);
2002 id = ida_simple_get(&nd_region->ns_ida, 0, 0,
2003 GFP_KERNEL);
2004 nsblk->id = id;
2005 } else
2006 id = i;
2007
2008 if (id < 0)
2009 break;
2010 dev_set_name(dev, "namespace%d.%d", nd_region->id, id);
2011 dev->groups = nd_namespace_attribute_groups;
2012 nd_device_register(dev);
2013 }
2014 if (i)
2015 nd_region->ns_seed = devs[0];
2016
2017 if (devs[i]) {
2018 int j;
2019
2020 for (j = i; devs[j]; j++) {
2021 struct device *dev = devs[j];
2022
2023 device_initialize(dev);
2024 put_device(dev);
2025 }
2026 *err = j - i;
2027 /*
2028 * All of the namespaces we tried to register failed, so
2029 * fail region activation.
2030 */
2031 if (*err == 0)
2032 rc = -ENODEV;
2033 }
2034 kfree(devs);
2035
2036 if (rc == -ENODEV)
2037 return rc;
2038
2039 return i;
2040}