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