1 | // SPDX-License-Identifier: GPL-2.0-only |
2 | /* |
3 | * Copyright(c) 2013-2015 Intel Corporation. All rights reserved. |
4 | */ |
5 | #include <linux/kstrtox.h> |
6 | #include <linux/module.h> |
7 | #include <linux/device.h> |
8 | #include <linux/sort.h> |
9 | #include <linux/slab.h> |
10 | #include <linux/list.h> |
11 | #include <linux/nd.h> |
12 | #include "nd-core.h" |
13 | #include "pmem.h" |
14 | #include "pfn.h" |
15 | #include "nd.h" |
16 | |
17 | static void namespace_io_release(struct device *dev) |
18 | { |
19 | struct nd_namespace_io *nsio = to_nd_namespace_io(dev); |
20 | |
21 | kfree(objp: nsio); |
22 | } |
23 | |
24 | static void namespace_pmem_release(struct device *dev) |
25 | { |
26 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); |
27 | struct nd_region *nd_region = to_nd_region(dev: dev->parent); |
28 | |
29 | if (nspm->id >= 0) |
30 | ida_free(&nd_region->ns_ida, id: nspm->id); |
31 | kfree(objp: nspm->alt_name); |
32 | kfree(objp: nspm->uuid); |
33 | kfree(objp: nspm); |
34 | } |
35 | |
36 | static bool is_namespace_pmem(const struct device *dev); |
37 | static bool is_namespace_io(const struct device *dev); |
38 | |
39 | static int is_uuid_busy(struct device *dev, void *data) |
40 | { |
41 | uuid_t *uuid1 = data, *uuid2 = NULL; |
42 | |
43 | if (is_namespace_pmem(dev)) { |
44 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); |
45 | |
46 | uuid2 = nspm->uuid; |
47 | } else if (is_nd_btt(dev)) { |
48 | struct nd_btt *nd_btt = to_nd_btt(dev); |
49 | |
50 | uuid2 = nd_btt->uuid; |
51 | } else if (is_nd_pfn(dev)) { |
52 | struct nd_pfn *nd_pfn = to_nd_pfn(dev); |
53 | |
54 | uuid2 = nd_pfn->uuid; |
55 | } |
56 | |
57 | if (uuid2 && uuid_equal(u1: uuid1, u2: uuid2)) |
58 | return -EBUSY; |
59 | |
60 | return 0; |
61 | } |
62 | |
63 | static int is_namespace_uuid_busy(struct device *dev, void *data) |
64 | { |
65 | if (is_nd_region(dev)) |
66 | return device_for_each_child(dev, data, fn: is_uuid_busy); |
67 | return 0; |
68 | } |
69 | |
70 | /** |
71 | * nd_is_uuid_unique - verify that no other namespace has @uuid |
72 | * @dev: any device on a nvdimm_bus |
73 | * @uuid: uuid to check |
74 | * |
75 | * Returns: %true if the uuid is unique, %false if not |
76 | */ |
77 | bool nd_is_uuid_unique(struct device *dev, uuid_t *uuid) |
78 | { |
79 | struct nvdimm_bus *nvdimm_bus = walk_to_nvdimm_bus(nd_dev: dev); |
80 | |
81 | if (!nvdimm_bus) |
82 | return false; |
83 | WARN_ON_ONCE(!is_nvdimm_bus_locked(&nvdimm_bus->dev)); |
84 | if (device_for_each_child(dev: &nvdimm_bus->dev, data: uuid, |
85 | fn: is_namespace_uuid_busy) != 0) |
86 | return false; |
87 | return true; |
88 | } |
89 | |
90 | bool pmem_should_map_pages(struct device *dev) |
91 | { |
92 | struct nd_region *nd_region = to_nd_region(dev: dev->parent); |
93 | struct nd_namespace_common *ndns = to_ndns(dev); |
94 | struct nd_namespace_io *nsio; |
95 | |
96 | if (!IS_ENABLED(CONFIG_ZONE_DEVICE)) |
97 | return false; |
98 | |
99 | if (!test_bit(ND_REGION_PAGEMAP, &nd_region->flags)) |
100 | return false; |
101 | |
102 | if (is_nd_pfn(dev) || is_nd_btt(dev)) |
103 | return false; |
104 | |
105 | if (ndns->force_raw) |
106 | return false; |
107 | |
108 | nsio = to_nd_namespace_io(dev); |
109 | if (region_intersects(offset: nsio->res.start, size: resource_size(res: &nsio->res), |
110 | IORESOURCE_SYSTEM_RAM, |
111 | desc: IORES_DESC_NONE) == REGION_MIXED) |
112 | return false; |
113 | |
114 | return ARCH_MEMREMAP_PMEM == MEMREMAP_WB; |
115 | } |
116 | EXPORT_SYMBOL(pmem_should_map_pages); |
117 | |
118 | unsigned int pmem_sector_size(struct nd_namespace_common *ndns) |
119 | { |
120 | if (is_namespace_pmem(dev: &ndns->dev)) { |
121 | struct nd_namespace_pmem *nspm; |
122 | |
123 | nspm = to_nd_namespace_pmem(dev: &ndns->dev); |
124 | if (nspm->lbasize == 0 || nspm->lbasize == 512) |
125 | /* default */; |
126 | else if (nspm->lbasize == 4096) |
127 | return 4096; |
128 | else |
129 | dev_WARN(&ndns->dev, "unsupported sector size: %ld\n" , |
130 | nspm->lbasize); |
131 | } |
132 | |
133 | /* |
134 | * There is no namespace label (is_namespace_io()), or the label |
135 | * indicates the default sector size. |
136 | */ |
137 | return 512; |
138 | } |
139 | EXPORT_SYMBOL(pmem_sector_size); |
140 | |
141 | const char *nvdimm_namespace_disk_name(struct nd_namespace_common *ndns, |
142 | char *name) |
143 | { |
144 | struct nd_region *nd_region = to_nd_region(dev: ndns->dev.parent); |
145 | const char *suffix = NULL; |
146 | |
147 | if (ndns->claim && is_nd_btt(dev: ndns->claim)) |
148 | suffix = "s" ; |
149 | |
150 | if (is_namespace_pmem(dev: &ndns->dev) || is_namespace_io(dev: &ndns->dev)) { |
151 | int nsidx = 0; |
152 | |
153 | if (is_namespace_pmem(dev: &ndns->dev)) { |
154 | struct nd_namespace_pmem *nspm; |
155 | |
156 | nspm = to_nd_namespace_pmem(dev: &ndns->dev); |
157 | nsidx = nspm->id; |
158 | } |
159 | |
160 | if (nsidx) |
161 | sprintf(buf: name, fmt: "pmem%d.%d%s" , nd_region->id, nsidx, |
162 | suffix ? suffix : "" ); |
163 | else |
164 | sprintf(buf: name, fmt: "pmem%d%s" , nd_region->id, |
165 | suffix ? suffix : "" ); |
166 | } else { |
167 | return NULL; |
168 | } |
169 | |
170 | return name; |
171 | } |
172 | EXPORT_SYMBOL(nvdimm_namespace_disk_name); |
173 | |
174 | const uuid_t *nd_dev_to_uuid(struct device *dev) |
175 | { |
176 | if (dev && is_namespace_pmem(dev)) { |
177 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); |
178 | |
179 | return nspm->uuid; |
180 | } |
181 | return &uuid_null; |
182 | } |
183 | EXPORT_SYMBOL(nd_dev_to_uuid); |
184 | |
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: dev->parent); |
189 | |
190 | return sprintf(buf, fmt: "%d\n" , nd_region_to_nstype(nd_region)); |
191 | } |
192 | static DEVICE_ATTR_RO(nstype); |
193 | |
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(s: 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(size: NSLABEL_NAME_LEN, GFP_KERNEL); |
221 | if (!alt_name) { |
222 | rc = -ENOMEM; |
223 | goto out; |
224 | } |
225 | kfree(objp: *ns_altname); |
226 | *ns_altname = alt_name; |
227 | sprintf(buf: *ns_altname, fmt: "%s" , pos); |
228 | rc = len; |
229 | |
230 | out: |
231 | kfree(objp: input); |
232 | return rc; |
233 | } |
234 | |
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(res: &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 | |
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: 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 | |
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, fmt: "%s\n" , ns_altname ? ns_altname : "" ); |
293 | } |
294 | static DEVICE_ATTR_RW(alt_name); |
295 | |
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, start: res->start, size: resource_size(res) - n); |
323 | if (rc == 0) |
324 | res->flags |= DPA_RESOURCE_ADJUSTED; |
325 | nd_dbg_dpa(nd_region, ndd, res, "shrink %d\n" , rc); |
326 | break; |
327 | } |
328 | |
329 | return rc; |
330 | } |
331 | |
332 | /** |
333 | * shrink_dpa_allocation - for each dimm in region free n bytes for label_id |
334 | * @nd_region: the set of dimms to reclaim @n bytes from |
335 | * @label_id: unique identifier for the namespace consuming this dpa range |
336 | * @n: number of bytes per-dimm to release |
337 | * |
338 | * Assumes resources are ordered. Starting from the end try to |
339 | * adjust_resource() the allocation to @n, but if @n is larger than the |
340 | * allocation delete it and find the 'new' last allocation in the label |
341 | * set. |
342 | * |
343 | * Returns: %0 on success on -errno on error |
344 | */ |
345 | static int shrink_dpa_allocation(struct nd_region *nd_region, |
346 | struct nd_label_id *label_id, resource_size_t n) |
347 | { |
348 | int i; |
349 | |
350 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
351 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; |
352 | int rc; |
353 | |
354 | rc = scan_free(nd_region, nd_mapping, label_id, n); |
355 | if (rc) |
356 | return rc; |
357 | } |
358 | |
359 | return 0; |
360 | } |
361 | |
362 | static resource_size_t init_dpa_allocation(struct nd_label_id *label_id, |
363 | struct nd_region *nd_region, struct nd_mapping *nd_mapping, |
364 | resource_size_t n) |
365 | { |
366 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
367 | struct resource *res; |
368 | int rc = 0; |
369 | |
370 | /* first resource allocation for this label-id or dimm */ |
371 | res = nvdimm_allocate_dpa(ndd, label_id, start: nd_mapping->start, n); |
372 | if (!res) |
373 | rc = -EBUSY; |
374 | |
375 | nd_dbg_dpa(nd_region, ndd, res, "init %d\n" , rc); |
376 | return rc ? n : 0; |
377 | } |
378 | |
379 | |
380 | /** |
381 | * space_valid() - validate free dpa space against constraints |
382 | * @nd_region: hosting region of the free space |
383 | * @ndd: dimm device data for debug |
384 | * @label_id: namespace id to allocate space |
385 | * @prev: potential allocation that precedes free space |
386 | * @next: allocation that follows the given free space range |
387 | * @exist: first allocation with same id in the mapping |
388 | * @n: range that must satisfied for pmem allocations |
389 | * @valid: free space range to validate |
390 | * |
391 | * BLK-space is valid as long as it does not precede a PMEM |
392 | * allocation in a given region. PMEM-space must be contiguous |
393 | * and adjacent to an existing allocation (if one |
394 | * exists). If reserving PMEM any space is valid. |
395 | */ |
396 | static void space_valid(struct nd_region *nd_region, struct nvdimm_drvdata *ndd, |
397 | struct nd_label_id *label_id, struct resource *prev, |
398 | struct resource *next, struct resource *exist, |
399 | resource_size_t n, struct resource *valid) |
400 | { |
401 | bool is_reserve = strcmp(label_id->id, "pmem-reserve" ) == 0; |
402 | unsigned long align; |
403 | |
404 | align = nd_region->align / nd_region->ndr_mappings; |
405 | valid->start = ALIGN(valid->start, align); |
406 | valid->end = ALIGN_DOWN(valid->end + 1, align) - 1; |
407 | |
408 | if (valid->start >= valid->end) |
409 | goto invalid; |
410 | |
411 | if (is_reserve) |
412 | return; |
413 | |
414 | /* allocation needs to be contiguous, so this is all or nothing */ |
415 | if (resource_size(res: valid) < n) |
416 | goto invalid; |
417 | |
418 | /* we've got all the space we need and no existing allocation */ |
419 | if (!exist) |
420 | return; |
421 | |
422 | /* allocation needs to be contiguous with the existing namespace */ |
423 | if (valid->start == exist->end + 1 |
424 | || valid->end == exist->start - 1) |
425 | return; |
426 | |
427 | invalid: |
428 | /* truncate @valid size to 0 */ |
429 | valid->end = valid->start - 1; |
430 | } |
431 | |
432 | enum alloc_loc { |
433 | ALLOC_ERR = 0, ALLOC_BEFORE, ALLOC_MID, ALLOC_AFTER, |
434 | }; |
435 | |
436 | static resource_size_t scan_allocate(struct nd_region *nd_region, |
437 | struct nd_mapping *nd_mapping, struct nd_label_id *label_id, |
438 | resource_size_t n) |
439 | { |
440 | resource_size_t mapping_end = nd_mapping->start + nd_mapping->size - 1; |
441 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
442 | struct resource *res, *exist = NULL, valid; |
443 | const resource_size_t to_allocate = n; |
444 | int first; |
445 | |
446 | for_each_dpa_resource(ndd, res) |
447 | if (strcmp(label_id->id, res->name) == 0) |
448 | exist = res; |
449 | |
450 | valid.start = nd_mapping->start; |
451 | valid.end = mapping_end; |
452 | valid.name = "free space" ; |
453 | retry: |
454 | first = 0; |
455 | for_each_dpa_resource(ndd, res) { |
456 | struct resource *next = res->sibling, *new_res = NULL; |
457 | resource_size_t allocate, available = 0; |
458 | enum alloc_loc loc = ALLOC_ERR; |
459 | const char *action; |
460 | int rc = 0; |
461 | |
462 | /* ignore resources outside this nd_mapping */ |
463 | if (res->start > mapping_end) |
464 | continue; |
465 | if (res->end < nd_mapping->start) |
466 | continue; |
467 | |
468 | /* space at the beginning of the mapping */ |
469 | if (!first++ && res->start > nd_mapping->start) { |
470 | valid.start = nd_mapping->start; |
471 | valid.end = res->start - 1; |
472 | space_valid(nd_region, ndd, label_id, NULL, next, exist, |
473 | n: to_allocate, valid: &valid); |
474 | available = resource_size(res: &valid); |
475 | if (available) |
476 | loc = ALLOC_BEFORE; |
477 | } |
478 | |
479 | /* space between allocations */ |
480 | if (!loc && next) { |
481 | valid.start = res->start + resource_size(res); |
482 | valid.end = min(mapping_end, next->start - 1); |
483 | space_valid(nd_region, ndd, label_id, prev: res, next, exist, |
484 | n: to_allocate, valid: &valid); |
485 | available = resource_size(res: &valid); |
486 | if (available) |
487 | loc = ALLOC_MID; |
488 | } |
489 | |
490 | /* space at the end of the mapping */ |
491 | if (!loc && !next) { |
492 | valid.start = res->start + resource_size(res); |
493 | valid.end = mapping_end; |
494 | space_valid(nd_region, ndd, label_id, prev: res, next, exist, |
495 | n: to_allocate, valid: &valid); |
496 | available = resource_size(res: &valid); |
497 | if (available) |
498 | loc = ALLOC_AFTER; |
499 | } |
500 | |
501 | if (!loc || !available) |
502 | continue; |
503 | allocate = min(available, n); |
504 | switch (loc) { |
505 | case ALLOC_BEFORE: |
506 | if (strcmp(res->name, label_id->id) == 0) { |
507 | /* adjust current resource up */ |
508 | rc = adjust_resource(res, start: res->start - allocate, |
509 | size: resource_size(res) + allocate); |
510 | action = "cur grow up" ; |
511 | } else |
512 | action = "allocate" ; |
513 | break; |
514 | case ALLOC_MID: |
515 | if (strcmp(next->name, label_id->id) == 0) { |
516 | /* adjust next resource up */ |
517 | rc = adjust_resource(res: next, start: next->start |
518 | - allocate, size: resource_size(res: next) |
519 | + allocate); |
520 | new_res = next; |
521 | action = "next grow up" ; |
522 | } else if (strcmp(res->name, label_id->id) == 0) { |
523 | action = "grow down" ; |
524 | } else |
525 | action = "allocate" ; |
526 | break; |
527 | case ALLOC_AFTER: |
528 | if (strcmp(res->name, label_id->id) == 0) |
529 | action = "grow down" ; |
530 | else |
531 | action = "allocate" ; |
532 | break; |
533 | default: |
534 | return n; |
535 | } |
536 | |
537 | if (strcmp(action, "allocate" ) == 0) { |
538 | new_res = nvdimm_allocate_dpa(ndd, label_id, |
539 | start: valid.start, n: allocate); |
540 | if (!new_res) |
541 | rc = -EBUSY; |
542 | } else if (strcmp(action, "grow down" ) == 0) { |
543 | /* adjust current resource down */ |
544 | rc = adjust_resource(res, start: res->start, size: resource_size(res) |
545 | + allocate); |
546 | if (rc == 0) |
547 | res->flags |= DPA_RESOURCE_ADJUSTED; |
548 | } |
549 | |
550 | if (!new_res) |
551 | new_res = res; |
552 | |
553 | nd_dbg_dpa(nd_region, ndd, new_res, "%s(%d) %d\n" , |
554 | action, loc, rc); |
555 | |
556 | if (rc) |
557 | return n; |
558 | |
559 | n -= allocate; |
560 | if (n) { |
561 | /* |
562 | * Retry scan with newly inserted resources. |
563 | * For example, if we did an ALLOC_BEFORE |
564 | * insertion there may also have been space |
565 | * available for an ALLOC_AFTER insertion, so we |
566 | * need to check this same resource again |
567 | */ |
568 | goto retry; |
569 | } else |
570 | return 0; |
571 | } |
572 | |
573 | if (n == to_allocate) |
574 | return init_dpa_allocation(label_id, nd_region, nd_mapping, n); |
575 | return n; |
576 | } |
577 | |
578 | static int merge_dpa(struct nd_region *nd_region, |
579 | struct nd_mapping *nd_mapping, struct nd_label_id *label_id) |
580 | { |
581 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
582 | struct resource *res; |
583 | |
584 | if (strncmp("pmem" , label_id->id, 4) == 0) |
585 | return 0; |
586 | retry: |
587 | for_each_dpa_resource(ndd, res) { |
588 | int rc; |
589 | struct resource *next = res->sibling; |
590 | resource_size_t end = res->start + resource_size(res); |
591 | |
592 | if (!next || strcmp(res->name, label_id->id) != 0 |
593 | || strcmp(next->name, label_id->id) != 0 |
594 | || end != next->start) |
595 | continue; |
596 | end += resource_size(res: next); |
597 | nvdimm_free_dpa(ndd, res: next); |
598 | rc = adjust_resource(res, start: res->start, size: end - res->start); |
599 | nd_dbg_dpa(nd_region, ndd, res, "merge %d\n" , rc); |
600 | if (rc) |
601 | return rc; |
602 | res->flags |= DPA_RESOURCE_ADJUSTED; |
603 | goto retry; |
604 | } |
605 | |
606 | return 0; |
607 | } |
608 | |
609 | int __reserve_free_pmem(struct device *dev, void *data) |
610 | { |
611 | struct nvdimm *nvdimm = data; |
612 | struct nd_region *nd_region; |
613 | struct nd_label_id label_id; |
614 | int i; |
615 | |
616 | if (!is_memory(dev)) |
617 | return 0; |
618 | |
619 | nd_region = to_nd_region(dev); |
620 | if (nd_region->ndr_mappings == 0) |
621 | return 0; |
622 | |
623 | memset(&label_id, 0, sizeof(label_id)); |
624 | strcat(p: label_id.id, q: "pmem-reserve" ); |
625 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
626 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; |
627 | resource_size_t n, rem = 0; |
628 | |
629 | if (nd_mapping->nvdimm != nvdimm) |
630 | continue; |
631 | |
632 | n = nd_pmem_available_dpa(nd_region, nd_mapping); |
633 | if (n == 0) |
634 | return 0; |
635 | rem = scan_allocate(nd_region, nd_mapping, label_id: &label_id, n); |
636 | dev_WARN_ONCE(&nd_region->dev, rem, |
637 | "pmem reserve underrun: %#llx of %#llx bytes\n" , |
638 | (unsigned long long) n - rem, |
639 | (unsigned long long) n); |
640 | return rem ? -ENXIO : 0; |
641 | } |
642 | |
643 | return 0; |
644 | } |
645 | |
646 | void release_free_pmem(struct nvdimm_bus *nvdimm_bus, |
647 | struct nd_mapping *nd_mapping) |
648 | { |
649 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
650 | struct resource *res, *_res; |
651 | |
652 | for_each_dpa_resource_safe(ndd, res, _res) |
653 | if (strcmp(res->name, "pmem-reserve" ) == 0) |
654 | nvdimm_free_dpa(ndd, res); |
655 | } |
656 | |
657 | /** |
658 | * grow_dpa_allocation - for each dimm allocate n bytes for @label_id |
659 | * @nd_region: the set of dimms to allocate @n more bytes from |
660 | * @label_id: unique identifier for the namespace consuming this dpa range |
661 | * @n: number of bytes per-dimm to add to the existing allocation |
662 | * |
663 | * Assumes resources are ordered. For BLK regions, first consume |
664 | * BLK-only available DPA free space, then consume PMEM-aliased DPA |
665 | * space starting at the highest DPA. For PMEM regions start |
666 | * allocations from the start of an interleave set and end at the first |
667 | * BLK allocation or the end of the interleave set, whichever comes |
668 | * first. |
669 | * |
670 | * Returns: %0 on success on -errno on error |
671 | */ |
672 | static int grow_dpa_allocation(struct nd_region *nd_region, |
673 | struct nd_label_id *label_id, resource_size_t n) |
674 | { |
675 | int i; |
676 | |
677 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
678 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; |
679 | resource_size_t rem = n; |
680 | int rc; |
681 | |
682 | rem = scan_allocate(nd_region, nd_mapping, label_id, n: rem); |
683 | dev_WARN_ONCE(&nd_region->dev, rem, |
684 | "allocation underrun: %#llx of %#llx bytes\n" , |
685 | (unsigned long long) n - rem, |
686 | (unsigned long long) n); |
687 | if (rem) |
688 | return -ENXIO; |
689 | |
690 | rc = merge_dpa(nd_region, nd_mapping, label_id); |
691 | if (rc) |
692 | return rc; |
693 | } |
694 | |
695 | return 0; |
696 | } |
697 | |
698 | static void nd_namespace_pmem_set_resource(struct nd_region *nd_region, |
699 | struct nd_namespace_pmem *nspm, resource_size_t size) |
700 | { |
701 | struct resource *res = &nspm->nsio.res; |
702 | resource_size_t offset = 0; |
703 | |
704 | if (size && !nspm->uuid) { |
705 | WARN_ON_ONCE(1); |
706 | size = 0; |
707 | } |
708 | |
709 | if (size && nspm->uuid) { |
710 | struct nd_mapping *nd_mapping = &nd_region->mapping[0]; |
711 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
712 | struct nd_label_id label_id; |
713 | struct resource *res; |
714 | |
715 | if (!ndd) { |
716 | size = 0; |
717 | goto out; |
718 | } |
719 | |
720 | nd_label_gen_id(label_id: &label_id, uuid: nspm->uuid, flags: 0); |
721 | |
722 | /* calculate a spa offset from the dpa allocation offset */ |
723 | for_each_dpa_resource(ndd, res) |
724 | if (strcmp(res->name, label_id.id) == 0) { |
725 | offset = (res->start - nd_mapping->start) |
726 | * nd_region->ndr_mappings; |
727 | goto out; |
728 | } |
729 | |
730 | WARN_ON_ONCE(1); |
731 | size = 0; |
732 | } |
733 | |
734 | out: |
735 | res->start = nd_region->ndr_start + offset; |
736 | res->end = res->start + size - 1; |
737 | } |
738 | |
739 | static bool uuid_not_set(const uuid_t *uuid, struct device *dev, |
740 | const char *where) |
741 | { |
742 | if (!uuid) { |
743 | dev_dbg(dev, "%s: uuid not set\n" , where); |
744 | return true; |
745 | } |
746 | return false; |
747 | } |
748 | |
749 | static ssize_t __size_store(struct device *dev, unsigned long long val) |
750 | { |
751 | resource_size_t allocated = 0, available = 0; |
752 | struct nd_region *nd_region = to_nd_region(dev: dev->parent); |
753 | struct nd_namespace_common *ndns = to_ndns(dev); |
754 | struct nd_mapping *nd_mapping; |
755 | struct nvdimm_drvdata *ndd; |
756 | struct nd_label_id label_id; |
757 | u32 flags = 0, remainder; |
758 | int rc, i, id = -1; |
759 | uuid_t *uuid = NULL; |
760 | |
761 | if (dev->driver || ndns->claim) |
762 | return -EBUSY; |
763 | |
764 | if (is_namespace_pmem(dev)) { |
765 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); |
766 | |
767 | uuid = nspm->uuid; |
768 | id = nspm->id; |
769 | } |
770 | |
771 | /* |
772 | * We need a uuid for the allocation-label and dimm(s) on which |
773 | * to store the label. |
774 | */ |
775 | if (uuid_not_set(uuid, dev, where: __func__)) |
776 | return -ENXIO; |
777 | if (nd_region->ndr_mappings == 0) { |
778 | dev_dbg(dev, "not associated with dimm(s)\n" ); |
779 | return -ENXIO; |
780 | } |
781 | |
782 | div_u64_rem(dividend: val, divisor: nd_region->align, remainder: &remainder); |
783 | if (remainder) { |
784 | dev_dbg(dev, "%llu is not %ldK aligned\n" , val, |
785 | nd_region->align / SZ_1K); |
786 | return -EINVAL; |
787 | } |
788 | |
789 | nd_label_gen_id(label_id: &label_id, uuid, flags); |
790 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
791 | nd_mapping = &nd_region->mapping[i]; |
792 | ndd = to_ndd(nd_mapping); |
793 | |
794 | /* |
795 | * All dimms in an interleave set, need to be enabled |
796 | * for the size to be changed. |
797 | */ |
798 | if (!ndd) |
799 | return -ENXIO; |
800 | |
801 | allocated += nvdimm_allocated_dpa(ndd, label_id: &label_id); |
802 | } |
803 | available = nd_region_allocatable_dpa(nd_region); |
804 | |
805 | if (val > available + allocated) |
806 | return -ENOSPC; |
807 | |
808 | if (val == allocated) |
809 | return 0; |
810 | |
811 | val = div_u64(dividend: val, divisor: nd_region->ndr_mappings); |
812 | allocated = div_u64(dividend: allocated, divisor: nd_region->ndr_mappings); |
813 | if (val < allocated) |
814 | rc = shrink_dpa_allocation(nd_region, label_id: &label_id, |
815 | n: allocated - val); |
816 | else |
817 | rc = grow_dpa_allocation(nd_region, label_id: &label_id, n: val - allocated); |
818 | |
819 | if (rc) |
820 | return rc; |
821 | |
822 | if (is_namespace_pmem(dev)) { |
823 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); |
824 | |
825 | nd_namespace_pmem_set_resource(nd_region, nspm, |
826 | size: val * nd_region->ndr_mappings); |
827 | } |
828 | |
829 | /* |
830 | * Try to delete the namespace if we deleted all of its |
831 | * allocation, this is not the seed or 0th device for the |
832 | * region, and it is not actively claimed by a btt, pfn, or dax |
833 | * instance. |
834 | */ |
835 | if (val == 0 && id != 0 && nd_region->ns_seed != dev && !ndns->claim) |
836 | nd_device_unregister(dev, mode: ND_ASYNC); |
837 | |
838 | return rc; |
839 | } |
840 | |
841 | static ssize_t size_store(struct device *dev, |
842 | struct device_attribute *attr, const char *buf, size_t len) |
843 | { |
844 | struct nd_region *nd_region = to_nd_region(dev: dev->parent); |
845 | unsigned long long val; |
846 | int rc; |
847 | |
848 | rc = kstrtoull(s: buf, base: 0, res: &val); |
849 | if (rc) |
850 | return rc; |
851 | |
852 | device_lock(dev); |
853 | nvdimm_bus_lock(dev); |
854 | wait_nvdimm_bus_probe_idle(dev); |
855 | rc = __size_store(dev, val); |
856 | if (rc >= 0) |
857 | rc = nd_namespace_label_update(nd_region, dev); |
858 | |
859 | /* setting size zero == 'delete namespace' */ |
860 | if (rc == 0 && val == 0 && is_namespace_pmem(dev)) { |
861 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); |
862 | |
863 | kfree(objp: nspm->uuid); |
864 | nspm->uuid = NULL; |
865 | } |
866 | |
867 | dev_dbg(dev, "%llx %s (%d)\n" , val, rc < 0 ? "fail" : "success" , rc); |
868 | |
869 | nvdimm_bus_unlock(dev); |
870 | device_unlock(dev); |
871 | |
872 | return rc < 0 ? rc : len; |
873 | } |
874 | |
875 | resource_size_t __nvdimm_namespace_capacity(struct nd_namespace_common *ndns) |
876 | { |
877 | struct device *dev = &ndns->dev; |
878 | |
879 | if (is_namespace_pmem(dev)) { |
880 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); |
881 | |
882 | return resource_size(res: &nspm->nsio.res); |
883 | } else if (is_namespace_io(dev)) { |
884 | struct nd_namespace_io *nsio = to_nd_namespace_io(dev); |
885 | |
886 | return resource_size(res: &nsio->res); |
887 | } else |
888 | WARN_ONCE(1, "unknown namespace type\n" ); |
889 | return 0; |
890 | } |
891 | |
892 | resource_size_t nvdimm_namespace_capacity(struct nd_namespace_common *ndns) |
893 | { |
894 | resource_size_t size; |
895 | |
896 | nvdimm_bus_lock(dev: &ndns->dev); |
897 | size = __nvdimm_namespace_capacity(ndns); |
898 | nvdimm_bus_unlock(dev: &ndns->dev); |
899 | |
900 | return size; |
901 | } |
902 | EXPORT_SYMBOL(nvdimm_namespace_capacity); |
903 | |
904 | bool nvdimm_namespace_locked(struct nd_namespace_common *ndns) |
905 | { |
906 | int i; |
907 | bool locked = false; |
908 | struct device *dev = &ndns->dev; |
909 | struct nd_region *nd_region = to_nd_region(dev: dev->parent); |
910 | |
911 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
912 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; |
913 | struct nvdimm *nvdimm = nd_mapping->nvdimm; |
914 | |
915 | if (test_bit(NDD_LOCKED, &nvdimm->flags)) { |
916 | dev_dbg(dev, "%s locked\n" , nvdimm_name(nvdimm)); |
917 | locked = true; |
918 | } |
919 | } |
920 | return locked; |
921 | } |
922 | EXPORT_SYMBOL(nvdimm_namespace_locked); |
923 | |
924 | static ssize_t size_show(struct device *dev, |
925 | struct device_attribute *attr, char *buf) |
926 | { |
927 | return sprintf(buf, fmt: "%llu\n" , (unsigned long long) |
928 | nvdimm_namespace_capacity(to_ndns(dev))); |
929 | } |
930 | static DEVICE_ATTR(size, 0444, size_show, size_store); |
931 | |
932 | static uuid_t *namespace_to_uuid(struct device *dev) |
933 | { |
934 | if (is_namespace_pmem(dev)) { |
935 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); |
936 | |
937 | return nspm->uuid; |
938 | } |
939 | return ERR_PTR(error: -ENXIO); |
940 | } |
941 | |
942 | static ssize_t uuid_show(struct device *dev, struct device_attribute *attr, |
943 | char *buf) |
944 | { |
945 | uuid_t *uuid = namespace_to_uuid(dev); |
946 | |
947 | if (IS_ERR(ptr: uuid)) |
948 | return PTR_ERR(ptr: uuid); |
949 | if (uuid) |
950 | return sprintf(buf, fmt: "%pUb\n" , uuid); |
951 | return sprintf(buf, fmt: "\n" ); |
952 | } |
953 | |
954 | /** |
955 | * namespace_update_uuid - check for a unique uuid and whether we're "renaming" |
956 | * @nd_region: parent region so we can updates all dimms in the set |
957 | * @dev: namespace type for generating label_id |
958 | * @new_uuid: incoming uuid |
959 | * @old_uuid: reference to the uuid storage location in the namespace object |
960 | * |
961 | * Returns: %0 on success on -errno on error |
962 | */ |
963 | static int namespace_update_uuid(struct nd_region *nd_region, |
964 | struct device *dev, uuid_t *new_uuid, |
965 | uuid_t **old_uuid) |
966 | { |
967 | struct nd_label_id old_label_id; |
968 | struct nd_label_id new_label_id; |
969 | int i; |
970 | |
971 | if (!nd_is_uuid_unique(dev, uuid: new_uuid)) |
972 | return -EINVAL; |
973 | |
974 | if (*old_uuid == NULL) |
975 | goto out; |
976 | |
977 | /* |
978 | * If we've already written a label with this uuid, then it's |
979 | * too late to rename because we can't reliably update the uuid |
980 | * without losing the old namespace. Userspace must delete this |
981 | * namespace to abandon the old uuid. |
982 | */ |
983 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
984 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; |
985 | |
986 | /* |
987 | * This check by itself is sufficient because old_uuid |
988 | * would be NULL above if this uuid did not exist in the |
989 | * currently written set. |
990 | * |
991 | * FIXME: can we delete uuid with zero dpa allocated? |
992 | */ |
993 | if (list_empty(head: &nd_mapping->labels)) |
994 | return -EBUSY; |
995 | } |
996 | |
997 | nd_label_gen_id(label_id: &old_label_id, uuid: *old_uuid, flags: 0); |
998 | nd_label_gen_id(label_id: &new_label_id, uuid: new_uuid, flags: 0); |
999 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
1000 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; |
1001 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
1002 | struct nd_label_ent *label_ent; |
1003 | struct resource *res; |
1004 | |
1005 | for_each_dpa_resource(ndd, res) |
1006 | if (strcmp(res->name, old_label_id.id) == 0) |
1007 | sprintf(buf: (void *) res->name, fmt: "%s" , |
1008 | new_label_id.id); |
1009 | |
1010 | mutex_lock(&nd_mapping->lock); |
1011 | list_for_each_entry(label_ent, &nd_mapping->labels, list) { |
1012 | struct nd_namespace_label *nd_label = label_ent->label; |
1013 | struct nd_label_id label_id; |
1014 | uuid_t uuid; |
1015 | |
1016 | if (!nd_label) |
1017 | continue; |
1018 | nsl_get_uuid(ndd, nd_label, uuid: &uuid); |
1019 | nd_label_gen_id(label_id: &label_id, uuid: &uuid, |
1020 | flags: nsl_get_flags(ndd, nd_label)); |
1021 | if (strcmp(old_label_id.id, label_id.id) == 0) |
1022 | set_bit(nr: ND_LABEL_REAP, addr: &label_ent->flags); |
1023 | } |
1024 | mutex_unlock(lock: &nd_mapping->lock); |
1025 | } |
1026 | kfree(objp: *old_uuid); |
1027 | out: |
1028 | *old_uuid = new_uuid; |
1029 | return 0; |
1030 | } |
1031 | |
1032 | static ssize_t uuid_store(struct device *dev, |
1033 | struct device_attribute *attr, const char *buf, size_t len) |
1034 | { |
1035 | struct nd_region *nd_region = to_nd_region(dev: dev->parent); |
1036 | uuid_t *uuid = NULL; |
1037 | uuid_t **ns_uuid; |
1038 | ssize_t rc = 0; |
1039 | |
1040 | if (is_namespace_pmem(dev)) { |
1041 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); |
1042 | |
1043 | ns_uuid = &nspm->uuid; |
1044 | } else |
1045 | return -ENXIO; |
1046 | |
1047 | device_lock(dev); |
1048 | nvdimm_bus_lock(dev); |
1049 | wait_nvdimm_bus_probe_idle(dev); |
1050 | if (to_ndns(dev)->claim) |
1051 | rc = -EBUSY; |
1052 | if (rc >= 0) |
1053 | rc = nd_uuid_store(dev, uuid_out: &uuid, buf, len); |
1054 | if (rc >= 0) |
1055 | rc = namespace_update_uuid(nd_region, dev, new_uuid: uuid, old_uuid: ns_uuid); |
1056 | if (rc >= 0) |
1057 | rc = nd_namespace_label_update(nd_region, dev); |
1058 | else |
1059 | kfree(objp: uuid); |
1060 | dev_dbg(dev, "result: %zd wrote: %s%s" , rc, buf, |
1061 | buf[len - 1] == '\n' ? "" : "\n" ); |
1062 | nvdimm_bus_unlock(dev); |
1063 | device_unlock(dev); |
1064 | |
1065 | return rc < 0 ? rc : len; |
1066 | } |
1067 | static DEVICE_ATTR_RW(uuid); |
1068 | |
1069 | static ssize_t resource_show(struct device *dev, |
1070 | struct device_attribute *attr, char *buf) |
1071 | { |
1072 | struct resource *res; |
1073 | |
1074 | if (is_namespace_pmem(dev)) { |
1075 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); |
1076 | |
1077 | res = &nspm->nsio.res; |
1078 | } else if (is_namespace_io(dev)) { |
1079 | struct nd_namespace_io *nsio = to_nd_namespace_io(dev); |
1080 | |
1081 | res = &nsio->res; |
1082 | } else |
1083 | return -ENXIO; |
1084 | |
1085 | /* no address to convey if the namespace has no allocation */ |
1086 | if (resource_size(res) == 0) |
1087 | return -ENXIO; |
1088 | return sprintf(buf, fmt: "%#llx\n" , (unsigned long long) res->start); |
1089 | } |
1090 | static DEVICE_ATTR_ADMIN_RO(resource); |
1091 | |
1092 | static const unsigned long pmem_lbasize_supported[] = { 512, 4096, 0 }; |
1093 | |
1094 | static ssize_t sector_size_show(struct device *dev, |
1095 | struct device_attribute *attr, char *buf) |
1096 | { |
1097 | if (is_namespace_pmem(dev)) { |
1098 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); |
1099 | |
1100 | return nd_size_select_show(current_size: nspm->lbasize, |
1101 | supported: pmem_lbasize_supported, buf); |
1102 | } |
1103 | return -ENXIO; |
1104 | } |
1105 | |
1106 | static ssize_t sector_size_store(struct device *dev, |
1107 | struct device_attribute *attr, const char *buf, size_t len) |
1108 | { |
1109 | struct nd_region *nd_region = to_nd_region(dev: dev->parent); |
1110 | const unsigned long *supported; |
1111 | unsigned long *lbasize; |
1112 | ssize_t rc = 0; |
1113 | |
1114 | if (is_namespace_pmem(dev)) { |
1115 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); |
1116 | |
1117 | lbasize = &nspm->lbasize; |
1118 | supported = pmem_lbasize_supported; |
1119 | } else |
1120 | return -ENXIO; |
1121 | |
1122 | device_lock(dev); |
1123 | nvdimm_bus_lock(dev); |
1124 | if (to_ndns(dev)->claim) |
1125 | rc = -EBUSY; |
1126 | if (rc >= 0) |
1127 | rc = nd_size_select_store(dev, buf, current_size: lbasize, supported); |
1128 | if (rc >= 0) |
1129 | rc = nd_namespace_label_update(nd_region, dev); |
1130 | dev_dbg(dev, "result: %zd %s: %s%s" , rc, rc < 0 ? "tried" : "wrote" , |
1131 | buf, buf[len - 1] == '\n' ? "" : "\n" ); |
1132 | nvdimm_bus_unlock(dev); |
1133 | device_unlock(dev); |
1134 | |
1135 | return rc ? rc : len; |
1136 | } |
1137 | static DEVICE_ATTR_RW(sector_size); |
1138 | |
1139 | static ssize_t dpa_extents_show(struct device *dev, |
1140 | struct device_attribute *attr, char *buf) |
1141 | { |
1142 | struct nd_region *nd_region = to_nd_region(dev: dev->parent); |
1143 | struct nd_label_id label_id; |
1144 | uuid_t *uuid = NULL; |
1145 | int count = 0, i; |
1146 | u32 flags = 0; |
1147 | |
1148 | nvdimm_bus_lock(dev); |
1149 | if (is_namespace_pmem(dev)) { |
1150 | struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(dev); |
1151 | |
1152 | uuid = nspm->uuid; |
1153 | flags = 0; |
1154 | } |
1155 | |
1156 | if (!uuid) |
1157 | goto out; |
1158 | |
1159 | nd_label_gen_id(label_id: &label_id, uuid, flags); |
1160 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
1161 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; |
1162 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
1163 | struct resource *res; |
1164 | |
1165 | for_each_dpa_resource(ndd, res) |
1166 | if (strcmp(res->name, label_id.id) == 0) |
1167 | count++; |
1168 | } |
1169 | out: |
1170 | nvdimm_bus_unlock(dev); |
1171 | |
1172 | return sprintf(buf, fmt: "%d\n" , count); |
1173 | } |
1174 | static DEVICE_ATTR_RO(dpa_extents); |
1175 | |
1176 | static int btt_claim_class(struct device *dev) |
1177 | { |
1178 | struct nd_region *nd_region = to_nd_region(dev: dev->parent); |
1179 | int i, loop_bitmask = 0; |
1180 | |
1181 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
1182 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; |
1183 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
1184 | struct nd_namespace_index *nsindex; |
1185 | |
1186 | /* |
1187 | * If any of the DIMMs do not support labels the only |
1188 | * possible BTT format is v1. |
1189 | */ |
1190 | if (!ndd) { |
1191 | loop_bitmask = 0; |
1192 | break; |
1193 | } |
1194 | |
1195 | nsindex = to_namespace_index(ndd, i: ndd->ns_current); |
1196 | if (nsindex == NULL) |
1197 | loop_bitmask |= 1; |
1198 | else { |
1199 | /* check whether existing labels are v1.1 or v1.2 */ |
1200 | if (__le16_to_cpu(nsindex->major) == 1 |
1201 | && __le16_to_cpu(nsindex->minor) == 1) |
1202 | loop_bitmask |= 2; |
1203 | else |
1204 | loop_bitmask |= 4; |
1205 | } |
1206 | } |
1207 | /* |
1208 | * If nsindex is null loop_bitmask's bit 0 will be set, and if an index |
1209 | * block is found, a v1.1 label for any mapping will set bit 1, and a |
1210 | * v1.2 label will set bit 2. |
1211 | * |
1212 | * At the end of the loop, at most one of the three bits must be set. |
1213 | * If multiple bits were set, it means the different mappings disagree |
1214 | * about their labels, and this must be cleaned up first. |
1215 | * |
1216 | * If all the label index blocks are found to agree, nsindex of NULL |
1217 | * implies labels haven't been initialized yet, and when they will, |
1218 | * they will be of the 1.2 format, so we can assume BTT2.0 |
1219 | * |
1220 | * If 1.1 labels are found, we enforce BTT1.1, and if 1.2 labels are |
1221 | * found, we enforce BTT2.0 |
1222 | * |
1223 | * If the loop was never entered, default to BTT1.1 (legacy namespaces) |
1224 | */ |
1225 | switch (loop_bitmask) { |
1226 | case 0: |
1227 | case 2: |
1228 | return NVDIMM_CCLASS_BTT; |
1229 | case 1: |
1230 | case 4: |
1231 | return NVDIMM_CCLASS_BTT2; |
1232 | default: |
1233 | return -ENXIO; |
1234 | } |
1235 | } |
1236 | |
1237 | static ssize_t holder_show(struct device *dev, |
1238 | struct device_attribute *attr, char *buf) |
1239 | { |
1240 | struct nd_namespace_common *ndns = to_ndns(dev); |
1241 | ssize_t rc; |
1242 | |
1243 | device_lock(dev); |
1244 | rc = sprintf(buf, fmt: "%s\n" , ndns->claim ? dev_name(dev: ndns->claim) : "" ); |
1245 | device_unlock(dev); |
1246 | |
1247 | return rc; |
1248 | } |
1249 | static DEVICE_ATTR_RO(holder); |
1250 | |
1251 | static int __holder_class_store(struct device *dev, const char *buf) |
1252 | { |
1253 | struct nd_namespace_common *ndns = to_ndns(dev); |
1254 | |
1255 | if (dev->driver || ndns->claim) |
1256 | return -EBUSY; |
1257 | |
1258 | if (sysfs_streq(s1: buf, s2: "btt" )) { |
1259 | int rc = btt_claim_class(dev); |
1260 | |
1261 | if (rc < NVDIMM_CCLASS_NONE) |
1262 | return rc; |
1263 | ndns->claim_class = rc; |
1264 | } else if (sysfs_streq(s1: buf, s2: "pfn" )) |
1265 | ndns->claim_class = NVDIMM_CCLASS_PFN; |
1266 | else if (sysfs_streq(s1: buf, s2: "dax" )) |
1267 | ndns->claim_class = NVDIMM_CCLASS_DAX; |
1268 | else if (sysfs_streq(s1: buf, s2: "" )) |
1269 | ndns->claim_class = NVDIMM_CCLASS_NONE; |
1270 | else |
1271 | return -EINVAL; |
1272 | |
1273 | return 0; |
1274 | } |
1275 | |
1276 | static ssize_t holder_class_store(struct device *dev, |
1277 | struct device_attribute *attr, const char *buf, size_t len) |
1278 | { |
1279 | struct nd_region *nd_region = to_nd_region(dev: dev->parent); |
1280 | int rc; |
1281 | |
1282 | device_lock(dev); |
1283 | nvdimm_bus_lock(dev); |
1284 | wait_nvdimm_bus_probe_idle(dev); |
1285 | rc = __holder_class_store(dev, buf); |
1286 | if (rc >= 0) |
1287 | rc = nd_namespace_label_update(nd_region, dev); |
1288 | dev_dbg(dev, "%s(%d)\n" , rc < 0 ? "fail " : "" , rc); |
1289 | nvdimm_bus_unlock(dev); |
1290 | device_unlock(dev); |
1291 | |
1292 | return rc < 0 ? rc : len; |
1293 | } |
1294 | |
1295 | static ssize_t holder_class_show(struct device *dev, |
1296 | struct device_attribute *attr, char *buf) |
1297 | { |
1298 | struct nd_namespace_common *ndns = to_ndns(dev); |
1299 | ssize_t rc; |
1300 | |
1301 | device_lock(dev); |
1302 | if (ndns->claim_class == NVDIMM_CCLASS_NONE) |
1303 | rc = sprintf(buf, fmt: "\n" ); |
1304 | else if ((ndns->claim_class == NVDIMM_CCLASS_BTT) || |
1305 | (ndns->claim_class == NVDIMM_CCLASS_BTT2)) |
1306 | rc = sprintf(buf, fmt: "btt\n" ); |
1307 | else if (ndns->claim_class == NVDIMM_CCLASS_PFN) |
1308 | rc = sprintf(buf, fmt: "pfn\n" ); |
1309 | else if (ndns->claim_class == NVDIMM_CCLASS_DAX) |
1310 | rc = sprintf(buf, fmt: "dax\n" ); |
1311 | else |
1312 | rc = sprintf(buf, fmt: "<unknown>\n" ); |
1313 | device_unlock(dev); |
1314 | |
1315 | return rc; |
1316 | } |
1317 | static DEVICE_ATTR_RW(holder_class); |
1318 | |
1319 | static ssize_t mode_show(struct device *dev, |
1320 | struct device_attribute *attr, char *buf) |
1321 | { |
1322 | struct nd_namespace_common *ndns = to_ndns(dev); |
1323 | struct device *claim; |
1324 | char *mode; |
1325 | ssize_t rc; |
1326 | |
1327 | device_lock(dev); |
1328 | claim = ndns->claim; |
1329 | if (claim && is_nd_btt(dev: claim)) |
1330 | mode = "safe" ; |
1331 | else if (claim && is_nd_pfn(dev: claim)) |
1332 | mode = "memory" ; |
1333 | else if (claim && is_nd_dax(dev: claim)) |
1334 | mode = "dax" ; |
1335 | else if (!claim && pmem_should_map_pages(dev)) |
1336 | mode = "memory" ; |
1337 | else |
1338 | mode = "raw" ; |
1339 | rc = sprintf(buf, fmt: "%s\n" , mode); |
1340 | device_unlock(dev); |
1341 | |
1342 | return rc; |
1343 | } |
1344 | static DEVICE_ATTR_RO(mode); |
1345 | |
1346 | static ssize_t force_raw_store(struct device *dev, |
1347 | struct device_attribute *attr, const char *buf, size_t len) |
1348 | { |
1349 | bool force_raw; |
1350 | int rc = kstrtobool(s: buf, res: &force_raw); |
1351 | |
1352 | if (rc) |
1353 | return rc; |
1354 | |
1355 | to_ndns(dev)->force_raw = force_raw; |
1356 | return len; |
1357 | } |
1358 | |
1359 | static ssize_t force_raw_show(struct device *dev, |
1360 | struct device_attribute *attr, char *buf) |
1361 | { |
1362 | return sprintf(buf, fmt: "%d\n" , to_ndns(dev)->force_raw); |
1363 | } |
1364 | static DEVICE_ATTR_RW(force_raw); |
1365 | |
1366 | static struct attribute *nd_namespace_attributes[] = { |
1367 | &dev_attr_nstype.attr, |
1368 | &dev_attr_size.attr, |
1369 | &dev_attr_mode.attr, |
1370 | &dev_attr_uuid.attr, |
1371 | &dev_attr_holder.attr, |
1372 | &dev_attr_resource.attr, |
1373 | &dev_attr_alt_name.attr, |
1374 | &dev_attr_force_raw.attr, |
1375 | &dev_attr_sector_size.attr, |
1376 | &dev_attr_dpa_extents.attr, |
1377 | &dev_attr_holder_class.attr, |
1378 | NULL, |
1379 | }; |
1380 | |
1381 | static umode_t namespace_visible(struct kobject *kobj, |
1382 | struct attribute *a, int n) |
1383 | { |
1384 | struct device *dev = container_of(kobj, struct device, kobj); |
1385 | |
1386 | if (is_namespace_pmem(dev)) { |
1387 | if (a == &dev_attr_size.attr) |
1388 | return 0644; |
1389 | |
1390 | return a->mode; |
1391 | } |
1392 | |
1393 | /* base is_namespace_io() attributes */ |
1394 | if (a == &dev_attr_nstype.attr || a == &dev_attr_size.attr || |
1395 | a == &dev_attr_holder.attr || a == &dev_attr_holder_class.attr || |
1396 | a == &dev_attr_force_raw.attr || a == &dev_attr_mode.attr || |
1397 | a == &dev_attr_resource.attr) |
1398 | return a->mode; |
1399 | |
1400 | return 0; |
1401 | } |
1402 | |
1403 | static struct attribute_group nd_namespace_attribute_group = { |
1404 | .attrs = nd_namespace_attributes, |
1405 | .is_visible = namespace_visible, |
1406 | }; |
1407 | |
1408 | static const struct attribute_group *nd_namespace_attribute_groups[] = { |
1409 | &nd_device_attribute_group, |
1410 | &nd_namespace_attribute_group, |
1411 | &nd_numa_attribute_group, |
1412 | NULL, |
1413 | }; |
1414 | |
1415 | static const struct device_type namespace_io_device_type = { |
1416 | .name = "nd_namespace_io" , |
1417 | .release = namespace_io_release, |
1418 | .groups = nd_namespace_attribute_groups, |
1419 | }; |
1420 | |
1421 | static const struct device_type namespace_pmem_device_type = { |
1422 | .name = "nd_namespace_pmem" , |
1423 | .release = namespace_pmem_release, |
1424 | .groups = nd_namespace_attribute_groups, |
1425 | }; |
1426 | |
1427 | static bool is_namespace_pmem(const struct device *dev) |
1428 | { |
1429 | return dev ? dev->type == &namespace_pmem_device_type : false; |
1430 | } |
1431 | |
1432 | static bool is_namespace_io(const struct device *dev) |
1433 | { |
1434 | return dev ? dev->type == &namespace_io_device_type : false; |
1435 | } |
1436 | |
1437 | struct nd_namespace_common *nvdimm_namespace_common_probe(struct device *dev) |
1438 | { |
1439 | struct nd_btt *nd_btt = is_nd_btt(dev) ? to_nd_btt(dev) : NULL; |
1440 | struct nd_pfn *nd_pfn = is_nd_pfn(dev) ? to_nd_pfn(dev) : NULL; |
1441 | struct nd_dax *nd_dax = is_nd_dax(dev) ? to_nd_dax(dev) : NULL; |
1442 | struct nd_namespace_common *ndns = NULL; |
1443 | resource_size_t size; |
1444 | |
1445 | if (nd_btt || nd_pfn || nd_dax) { |
1446 | if (nd_btt) |
1447 | ndns = nd_btt->ndns; |
1448 | else if (nd_pfn) |
1449 | ndns = nd_pfn->ndns; |
1450 | else if (nd_dax) |
1451 | ndns = nd_dax->nd_pfn.ndns; |
1452 | |
1453 | if (!ndns) |
1454 | return ERR_PTR(error: -ENODEV); |
1455 | |
1456 | /* |
1457 | * Flush any in-progess probes / removals in the driver |
1458 | * for the raw personality of this namespace. |
1459 | */ |
1460 | device_lock(dev: &ndns->dev); |
1461 | device_unlock(dev: &ndns->dev); |
1462 | if (ndns->dev.driver) { |
1463 | dev_dbg(&ndns->dev, "is active, can't bind %s\n" , |
1464 | dev_name(dev)); |
1465 | return ERR_PTR(error: -EBUSY); |
1466 | } |
1467 | if (dev_WARN_ONCE(&ndns->dev, ndns->claim != dev, |
1468 | "host (%s) vs claim (%s) mismatch\n" , |
1469 | dev_name(dev), |
1470 | dev_name(ndns->claim))) |
1471 | return ERR_PTR(error: -ENXIO); |
1472 | } else { |
1473 | ndns = to_ndns(dev); |
1474 | if (ndns->claim) { |
1475 | dev_dbg(dev, "claimed by %s, failing probe\n" , |
1476 | dev_name(ndns->claim)); |
1477 | |
1478 | return ERR_PTR(error: -ENXIO); |
1479 | } |
1480 | } |
1481 | |
1482 | if (nvdimm_namespace_locked(ndns)) |
1483 | return ERR_PTR(error: -EACCES); |
1484 | |
1485 | size = nvdimm_namespace_capacity(ndns); |
1486 | if (size < ND_MIN_NAMESPACE_SIZE) { |
1487 | dev_dbg(&ndns->dev, "%pa, too small must be at least %#x\n" , |
1488 | &size, ND_MIN_NAMESPACE_SIZE); |
1489 | return ERR_PTR(error: -ENODEV); |
1490 | } |
1491 | |
1492 | /* |
1493 | * Note, alignment validation for fsdax and devdax mode |
1494 | * namespaces happens in nd_pfn_validate() where infoblock |
1495 | * padding parameters can be applied. |
1496 | */ |
1497 | if (pmem_should_map_pages(dev)) { |
1498 | struct nd_namespace_io *nsio = to_nd_namespace_io(dev: &ndns->dev); |
1499 | struct resource *res = &nsio->res; |
1500 | |
1501 | if (!IS_ALIGNED(res->start | (res->end + 1), |
1502 | memremap_compat_align())) { |
1503 | dev_err(&ndns->dev, "%pr misaligned, unable to map\n" , res); |
1504 | return ERR_PTR(error: -EOPNOTSUPP); |
1505 | } |
1506 | } |
1507 | |
1508 | if (is_namespace_pmem(dev: &ndns->dev)) { |
1509 | struct nd_namespace_pmem *nspm; |
1510 | |
1511 | nspm = to_nd_namespace_pmem(dev: &ndns->dev); |
1512 | if (uuid_not_set(uuid: nspm->uuid, dev: &ndns->dev, where: __func__)) |
1513 | return ERR_PTR(error: -ENODEV); |
1514 | } |
1515 | |
1516 | return ndns; |
1517 | } |
1518 | EXPORT_SYMBOL(nvdimm_namespace_common_probe); |
1519 | |
1520 | int devm_namespace_enable(struct device *dev, struct nd_namespace_common *ndns, |
1521 | resource_size_t size) |
1522 | { |
1523 | return devm_nsio_enable(dev, nsio: to_nd_namespace_io(dev: &ndns->dev), size); |
1524 | } |
1525 | EXPORT_SYMBOL_GPL(devm_namespace_enable); |
1526 | |
1527 | void devm_namespace_disable(struct device *dev, struct nd_namespace_common *ndns) |
1528 | { |
1529 | devm_nsio_disable(dev, nsio: to_nd_namespace_io(dev: &ndns->dev)); |
1530 | } |
1531 | EXPORT_SYMBOL_GPL(devm_namespace_disable); |
1532 | |
1533 | static struct device **create_namespace_io(struct nd_region *nd_region) |
1534 | { |
1535 | struct nd_namespace_io *nsio; |
1536 | struct device *dev, **devs; |
1537 | struct resource *res; |
1538 | |
1539 | nsio = kzalloc(size: sizeof(*nsio), GFP_KERNEL); |
1540 | if (!nsio) |
1541 | return NULL; |
1542 | |
1543 | devs = kcalloc(n: 2, size: sizeof(struct device *), GFP_KERNEL); |
1544 | if (!devs) { |
1545 | kfree(objp: nsio); |
1546 | return NULL; |
1547 | } |
1548 | |
1549 | dev = &nsio->common.dev; |
1550 | dev->type = &namespace_io_device_type; |
1551 | dev->parent = &nd_region->dev; |
1552 | res = &nsio->res; |
1553 | res->name = dev_name(dev: &nd_region->dev); |
1554 | res->flags = IORESOURCE_MEM; |
1555 | res->start = nd_region->ndr_start; |
1556 | res->end = res->start + nd_region->ndr_size - 1; |
1557 | |
1558 | devs[0] = dev; |
1559 | return devs; |
1560 | } |
1561 | |
1562 | static bool has_uuid_at_pos(struct nd_region *nd_region, const uuid_t *uuid, |
1563 | u64 cookie, u16 pos) |
1564 | { |
1565 | struct nd_namespace_label *found = NULL; |
1566 | int i; |
1567 | |
1568 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
1569 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; |
1570 | struct nd_interleave_set *nd_set = nd_region->nd_set; |
1571 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
1572 | struct nd_label_ent *label_ent; |
1573 | bool found_uuid = false; |
1574 | |
1575 | list_for_each_entry(label_ent, &nd_mapping->labels, list) { |
1576 | struct nd_namespace_label *nd_label = label_ent->label; |
1577 | u16 position; |
1578 | |
1579 | if (!nd_label) |
1580 | continue; |
1581 | position = nsl_get_position(ndd, nd_label); |
1582 | |
1583 | if (!nsl_validate_isetcookie(ndd, nd_label, cookie)) |
1584 | continue; |
1585 | |
1586 | if (!nsl_uuid_equal(ndd, nd_label, uuid)) |
1587 | continue; |
1588 | |
1589 | if (!nsl_validate_type_guid(ndd, nd_label, |
1590 | guid: &nd_set->type_guid)) |
1591 | continue; |
1592 | |
1593 | if (found_uuid) { |
1594 | dev_dbg(ndd->dev, "duplicate entry for uuid\n" ); |
1595 | return false; |
1596 | } |
1597 | found_uuid = true; |
1598 | if (!nsl_validate_nlabel(nd_region, ndd, nd_label)) |
1599 | continue; |
1600 | if (position != pos) |
1601 | continue; |
1602 | found = nd_label; |
1603 | break; |
1604 | } |
1605 | if (found) |
1606 | break; |
1607 | } |
1608 | return found != NULL; |
1609 | } |
1610 | |
1611 | static int select_pmem_id(struct nd_region *nd_region, const uuid_t *pmem_id) |
1612 | { |
1613 | int i; |
1614 | |
1615 | if (!pmem_id) |
1616 | return -ENODEV; |
1617 | |
1618 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
1619 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; |
1620 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
1621 | struct nd_namespace_label *nd_label = NULL; |
1622 | u64 hw_start, hw_end, pmem_start, pmem_end; |
1623 | struct nd_label_ent *label_ent; |
1624 | |
1625 | lockdep_assert_held(&nd_mapping->lock); |
1626 | list_for_each_entry(label_ent, &nd_mapping->labels, list) { |
1627 | nd_label = label_ent->label; |
1628 | if (!nd_label) |
1629 | continue; |
1630 | if (nsl_uuid_equal(ndd, nd_label, uuid: pmem_id)) |
1631 | break; |
1632 | nd_label = NULL; |
1633 | } |
1634 | |
1635 | if (!nd_label) { |
1636 | WARN_ON(1); |
1637 | return -EINVAL; |
1638 | } |
1639 | |
1640 | /* |
1641 | * Check that this label is compliant with the dpa |
1642 | * range published in NFIT |
1643 | */ |
1644 | hw_start = nd_mapping->start; |
1645 | hw_end = hw_start + nd_mapping->size; |
1646 | pmem_start = nsl_get_dpa(ndd, nd_label); |
1647 | pmem_end = pmem_start + nsl_get_rawsize(ndd, nd_label); |
1648 | if (pmem_start >= hw_start && pmem_start < hw_end |
1649 | && pmem_end <= hw_end && pmem_end > hw_start) |
1650 | /* pass */; |
1651 | else { |
1652 | dev_dbg(&nd_region->dev, "%s invalid label for %pUb\n" , |
1653 | dev_name(ndd->dev), |
1654 | nsl_uuid_raw(ndd, nd_label)); |
1655 | return -EINVAL; |
1656 | } |
1657 | |
1658 | /* move recently validated label to the front of the list */ |
1659 | list_move(list: &label_ent->list, head: &nd_mapping->labels); |
1660 | } |
1661 | return 0; |
1662 | } |
1663 | |
1664 | /** |
1665 | * create_namespace_pmem - validate interleave set labelling, retrieve label0 |
1666 | * @nd_region: region with mappings to validate |
1667 | * @nd_mapping: container of dpa-resource-root + labels |
1668 | * @nd_label: target pmem namespace label to evaluate |
1669 | * |
1670 | * Returns: the created &struct device on success or ERR_PTR(-errno) on error |
1671 | */ |
1672 | static struct device *create_namespace_pmem(struct nd_region *nd_region, |
1673 | struct nd_mapping *nd_mapping, |
1674 | struct nd_namespace_label *nd_label) |
1675 | { |
1676 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
1677 | struct nd_namespace_index *nsindex = |
1678 | to_namespace_index(ndd, i: ndd->ns_current); |
1679 | u64 cookie = nd_region_interleave_set_cookie(nd_region, nsindex); |
1680 | u64 altcookie = nd_region_interleave_set_altcookie(nd_region); |
1681 | struct nd_label_ent *label_ent; |
1682 | struct nd_namespace_pmem *nspm; |
1683 | resource_size_t size = 0; |
1684 | struct resource *res; |
1685 | struct device *dev; |
1686 | uuid_t uuid; |
1687 | int rc = 0; |
1688 | u16 i; |
1689 | |
1690 | if (cookie == 0) { |
1691 | dev_dbg(&nd_region->dev, "invalid interleave-set-cookie\n" ); |
1692 | return ERR_PTR(error: -ENXIO); |
1693 | } |
1694 | |
1695 | if (!nsl_validate_isetcookie(ndd, nd_label, cookie)) { |
1696 | dev_dbg(&nd_region->dev, "invalid cookie in label: %pUb\n" , |
1697 | nsl_uuid_raw(ndd, nd_label)); |
1698 | if (!nsl_validate_isetcookie(ndd, nd_label, cookie: altcookie)) |
1699 | return ERR_PTR(error: -EAGAIN); |
1700 | |
1701 | dev_dbg(&nd_region->dev, "valid altcookie in label: %pUb\n" , |
1702 | nsl_uuid_raw(ndd, nd_label)); |
1703 | } |
1704 | |
1705 | nspm = kzalloc(size: sizeof(*nspm), GFP_KERNEL); |
1706 | if (!nspm) |
1707 | return ERR_PTR(error: -ENOMEM); |
1708 | |
1709 | nspm->id = -1; |
1710 | dev = &nspm->nsio.common.dev; |
1711 | dev->type = &namespace_pmem_device_type; |
1712 | dev->parent = &nd_region->dev; |
1713 | res = &nspm->nsio.res; |
1714 | res->name = dev_name(dev: &nd_region->dev); |
1715 | res->flags = IORESOURCE_MEM; |
1716 | |
1717 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
1718 | nsl_get_uuid(ndd, nd_label, uuid: &uuid); |
1719 | if (has_uuid_at_pos(nd_region, uuid: &uuid, cookie, pos: i)) |
1720 | continue; |
1721 | if (has_uuid_at_pos(nd_region, uuid: &uuid, cookie: altcookie, pos: i)) |
1722 | continue; |
1723 | break; |
1724 | } |
1725 | |
1726 | if (i < nd_region->ndr_mappings) { |
1727 | struct nvdimm *nvdimm = nd_region->mapping[i].nvdimm; |
1728 | |
1729 | /* |
1730 | * Give up if we don't find an instance of a uuid at each |
1731 | * position (from 0 to nd_region->ndr_mappings - 1), or if we |
1732 | * find a dimm with two instances of the same uuid. |
1733 | */ |
1734 | dev_err(&nd_region->dev, "%s missing label for %pUb\n" , |
1735 | nvdimm_name(nvdimm), nsl_uuid_raw(ndd, nd_label)); |
1736 | rc = -EINVAL; |
1737 | goto err; |
1738 | } |
1739 | |
1740 | /* |
1741 | * Fix up each mapping's 'labels' to have the validated pmem label for |
1742 | * that position at labels[0], and NULL at labels[1]. In the process, |
1743 | * check that the namespace aligns with interleave-set. |
1744 | */ |
1745 | nsl_get_uuid(ndd, nd_label, uuid: &uuid); |
1746 | rc = select_pmem_id(nd_region, pmem_id: &uuid); |
1747 | if (rc) |
1748 | goto err; |
1749 | |
1750 | /* Calculate total size and populate namespace properties from label0 */ |
1751 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
1752 | struct nd_namespace_label *label0; |
1753 | struct nvdimm_drvdata *ndd; |
1754 | |
1755 | nd_mapping = &nd_region->mapping[i]; |
1756 | label_ent = list_first_entry_or_null(&nd_mapping->labels, |
1757 | typeof(*label_ent), list); |
1758 | label0 = label_ent ? label_ent->label : NULL; |
1759 | |
1760 | if (!label0) { |
1761 | WARN_ON(1); |
1762 | continue; |
1763 | } |
1764 | |
1765 | ndd = to_ndd(nd_mapping); |
1766 | size += nsl_get_rawsize(ndd, nd_label: label0); |
1767 | if (nsl_get_position(ndd, nd_label: label0) != 0) |
1768 | continue; |
1769 | WARN_ON(nspm->alt_name || nspm->uuid); |
1770 | nspm->alt_name = kmemdup(p: nsl_ref_name(ndd, nd_label: label0), |
1771 | size: NSLABEL_NAME_LEN, GFP_KERNEL); |
1772 | nsl_get_uuid(ndd, nd_label: label0, uuid: &uuid); |
1773 | nspm->uuid = kmemdup(p: &uuid, size: sizeof(uuid_t), GFP_KERNEL); |
1774 | nspm->lbasize = nsl_get_lbasize(ndd, nd_label: label0); |
1775 | nspm->nsio.common.claim_class = |
1776 | nsl_get_claim_class(ndd, nd_label: label0); |
1777 | } |
1778 | |
1779 | if (!nspm->alt_name || !nspm->uuid) { |
1780 | rc = -ENOMEM; |
1781 | goto err; |
1782 | } |
1783 | |
1784 | nd_namespace_pmem_set_resource(nd_region, nspm, size); |
1785 | |
1786 | return dev; |
1787 | err: |
1788 | namespace_pmem_release(dev); |
1789 | switch (rc) { |
1790 | case -EINVAL: |
1791 | dev_dbg(&nd_region->dev, "invalid label(s)\n" ); |
1792 | break; |
1793 | case -ENODEV: |
1794 | dev_dbg(&nd_region->dev, "label not found\n" ); |
1795 | break; |
1796 | default: |
1797 | dev_dbg(&nd_region->dev, "unexpected err: %d\n" , rc); |
1798 | break; |
1799 | } |
1800 | return ERR_PTR(error: rc); |
1801 | } |
1802 | |
1803 | static struct device *nd_namespace_pmem_create(struct nd_region *nd_region) |
1804 | { |
1805 | struct nd_namespace_pmem *nspm; |
1806 | struct resource *res; |
1807 | struct device *dev; |
1808 | |
1809 | if (!is_memory(dev: &nd_region->dev)) |
1810 | return NULL; |
1811 | |
1812 | nspm = kzalloc(size: sizeof(*nspm), GFP_KERNEL); |
1813 | if (!nspm) |
1814 | return NULL; |
1815 | |
1816 | dev = &nspm->nsio.common.dev; |
1817 | dev->type = &namespace_pmem_device_type; |
1818 | dev->parent = &nd_region->dev; |
1819 | res = &nspm->nsio.res; |
1820 | res->name = dev_name(dev: &nd_region->dev); |
1821 | res->flags = IORESOURCE_MEM; |
1822 | |
1823 | nspm->id = ida_alloc(ida: &nd_region->ns_ida, GFP_KERNEL); |
1824 | if (nspm->id < 0) { |
1825 | kfree(objp: nspm); |
1826 | return NULL; |
1827 | } |
1828 | dev_set_name(dev, name: "namespace%d.%d" , nd_region->id, nspm->id); |
1829 | nd_namespace_pmem_set_resource(nd_region, nspm, size: 0); |
1830 | |
1831 | return dev; |
1832 | } |
1833 | |
1834 | static struct lock_class_key nvdimm_namespace_key; |
1835 | |
1836 | void nd_region_create_ns_seed(struct nd_region *nd_region) |
1837 | { |
1838 | WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev)); |
1839 | |
1840 | if (nd_region_to_nstype(nd_region) == ND_DEVICE_NAMESPACE_IO) |
1841 | return; |
1842 | |
1843 | nd_region->ns_seed = nd_namespace_pmem_create(nd_region); |
1844 | |
1845 | /* |
1846 | * Seed creation failures are not fatal, provisioning is simply |
1847 | * disabled until memory becomes available |
1848 | */ |
1849 | if (!nd_region->ns_seed) |
1850 | dev_err(&nd_region->dev, "failed to create namespace\n" ); |
1851 | else { |
1852 | device_initialize(dev: nd_region->ns_seed); |
1853 | lockdep_set_class(&nd_region->ns_seed->mutex, |
1854 | &nvdimm_namespace_key); |
1855 | nd_device_register(dev: nd_region->ns_seed); |
1856 | } |
1857 | } |
1858 | |
1859 | void nd_region_create_dax_seed(struct nd_region *nd_region) |
1860 | { |
1861 | WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev)); |
1862 | nd_region->dax_seed = nd_dax_create(nd_region); |
1863 | /* |
1864 | * Seed creation failures are not fatal, provisioning is simply |
1865 | * disabled until memory becomes available |
1866 | */ |
1867 | if (!nd_region->dax_seed) |
1868 | dev_err(&nd_region->dev, "failed to create dax namespace\n" ); |
1869 | } |
1870 | |
1871 | void nd_region_create_pfn_seed(struct nd_region *nd_region) |
1872 | { |
1873 | WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev)); |
1874 | nd_region->pfn_seed = nd_pfn_create(nd_region); |
1875 | /* |
1876 | * Seed creation failures are not fatal, provisioning is simply |
1877 | * disabled until memory becomes available |
1878 | */ |
1879 | if (!nd_region->pfn_seed) |
1880 | dev_err(&nd_region->dev, "failed to create pfn namespace\n" ); |
1881 | } |
1882 | |
1883 | void nd_region_create_btt_seed(struct nd_region *nd_region) |
1884 | { |
1885 | WARN_ON(!is_nvdimm_bus_locked(&nd_region->dev)); |
1886 | nd_region->btt_seed = nd_btt_create(nd_region); |
1887 | /* |
1888 | * Seed creation failures are not fatal, provisioning is simply |
1889 | * disabled until memory becomes available |
1890 | */ |
1891 | if (!nd_region->btt_seed) |
1892 | dev_err(&nd_region->dev, "failed to create btt namespace\n" ); |
1893 | } |
1894 | |
1895 | static int add_namespace_resource(struct nd_region *nd_region, |
1896 | struct nd_namespace_label *nd_label, struct device **devs, |
1897 | int count) |
1898 | { |
1899 | struct nd_mapping *nd_mapping = &nd_region->mapping[0]; |
1900 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
1901 | int i; |
1902 | |
1903 | for (i = 0; i < count; i++) { |
1904 | uuid_t *uuid = namespace_to_uuid(dev: devs[i]); |
1905 | |
1906 | if (IS_ERR(ptr: uuid)) { |
1907 | WARN_ON(1); |
1908 | continue; |
1909 | } |
1910 | |
1911 | if (!nsl_uuid_equal(ndd, nd_label, uuid)) |
1912 | continue; |
1913 | dev_err(&nd_region->dev, |
1914 | "error: conflicting extents for uuid: %pUb\n" , uuid); |
1915 | return -ENXIO; |
1916 | } |
1917 | |
1918 | return i; |
1919 | } |
1920 | |
1921 | static int cmp_dpa(const void *a, const void *b) |
1922 | { |
1923 | const struct device *dev_a = *(const struct device **) a; |
1924 | const struct device *dev_b = *(const struct device **) b; |
1925 | struct nd_namespace_pmem *nspm_a, *nspm_b; |
1926 | |
1927 | if (is_namespace_io(dev: dev_a)) |
1928 | return 0; |
1929 | |
1930 | nspm_a = to_nd_namespace_pmem(dev: dev_a); |
1931 | nspm_b = to_nd_namespace_pmem(dev: dev_b); |
1932 | |
1933 | return memcmp(p: &nspm_a->nsio.res.start, q: &nspm_b->nsio.res.start, |
1934 | size: sizeof(resource_size_t)); |
1935 | } |
1936 | |
1937 | static struct device **scan_labels(struct nd_region *nd_region) |
1938 | { |
1939 | int i, count = 0; |
1940 | struct device *dev, **devs = NULL; |
1941 | struct nd_label_ent *label_ent, *e; |
1942 | struct nd_mapping *nd_mapping = &nd_region->mapping[0]; |
1943 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
1944 | resource_size_t map_end = nd_mapping->start + nd_mapping->size - 1; |
1945 | |
1946 | /* "safe" because create_namespace_pmem() might list_move() label_ent */ |
1947 | list_for_each_entry_safe(label_ent, e, &nd_mapping->labels, list) { |
1948 | struct nd_namespace_label *nd_label = label_ent->label; |
1949 | struct device **__devs; |
1950 | |
1951 | if (!nd_label) |
1952 | continue; |
1953 | |
1954 | /* skip labels that describe extents outside of the region */ |
1955 | if (nsl_get_dpa(ndd, nd_label) < nd_mapping->start || |
1956 | nsl_get_dpa(ndd, nd_label) > map_end) |
1957 | continue; |
1958 | |
1959 | i = add_namespace_resource(nd_region, nd_label, devs, count); |
1960 | if (i < 0) |
1961 | goto err; |
1962 | if (i < count) |
1963 | continue; |
1964 | __devs = kcalloc(n: count + 2, size: sizeof(dev), GFP_KERNEL); |
1965 | if (!__devs) |
1966 | goto err; |
1967 | memcpy(__devs, devs, sizeof(dev) * count); |
1968 | kfree(objp: devs); |
1969 | devs = __devs; |
1970 | |
1971 | dev = create_namespace_pmem(nd_region, nd_mapping, nd_label); |
1972 | if (IS_ERR(ptr: dev)) { |
1973 | switch (PTR_ERR(ptr: dev)) { |
1974 | case -EAGAIN: |
1975 | /* skip invalid labels */ |
1976 | continue; |
1977 | case -ENODEV: |
1978 | /* fallthrough to seed creation */ |
1979 | break; |
1980 | default: |
1981 | goto err; |
1982 | } |
1983 | } else |
1984 | devs[count++] = dev; |
1985 | |
1986 | } |
1987 | |
1988 | dev_dbg(&nd_region->dev, "discovered %d namespace%s\n" , count, |
1989 | count == 1 ? "" : "s" ); |
1990 | |
1991 | if (count == 0) { |
1992 | struct nd_namespace_pmem *nspm; |
1993 | |
1994 | /* Publish a zero-sized namespace for userspace to configure. */ |
1995 | nd_mapping_free_labels(nd_mapping); |
1996 | |
1997 | devs = kcalloc(n: 2, size: sizeof(dev), GFP_KERNEL); |
1998 | if (!devs) |
1999 | goto err; |
2000 | |
2001 | nspm = kzalloc(size: sizeof(*nspm), GFP_KERNEL); |
2002 | if (!nspm) |
2003 | goto err; |
2004 | dev = &nspm->nsio.common.dev; |
2005 | dev->type = &namespace_pmem_device_type; |
2006 | nd_namespace_pmem_set_resource(nd_region, nspm, size: 0); |
2007 | dev->parent = &nd_region->dev; |
2008 | devs[count++] = dev; |
2009 | } else if (is_memory(dev: &nd_region->dev)) { |
2010 | /* clean unselected labels */ |
2011 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
2012 | struct list_head *l, *e; |
2013 | LIST_HEAD(list); |
2014 | int j; |
2015 | |
2016 | nd_mapping = &nd_region->mapping[i]; |
2017 | if (list_empty(head: &nd_mapping->labels)) { |
2018 | WARN_ON(1); |
2019 | continue; |
2020 | } |
2021 | |
2022 | j = count; |
2023 | list_for_each_safe(l, e, &nd_mapping->labels) { |
2024 | if (!j--) |
2025 | break; |
2026 | list_move_tail(list: l, head: &list); |
2027 | } |
2028 | nd_mapping_free_labels(nd_mapping); |
2029 | list_splice_init(list: &list, head: &nd_mapping->labels); |
2030 | } |
2031 | } |
2032 | |
2033 | if (count > 1) |
2034 | sort(base: devs, num: count, size: sizeof(struct device *), cmp_func: cmp_dpa, NULL); |
2035 | |
2036 | return devs; |
2037 | |
2038 | err: |
2039 | if (devs) { |
2040 | for (i = 0; devs[i]; i++) |
2041 | namespace_pmem_release(dev: devs[i]); |
2042 | kfree(objp: devs); |
2043 | } |
2044 | return NULL; |
2045 | } |
2046 | |
2047 | static struct device **create_namespaces(struct nd_region *nd_region) |
2048 | { |
2049 | struct nd_mapping *nd_mapping; |
2050 | struct device **devs; |
2051 | int i; |
2052 | |
2053 | if (nd_region->ndr_mappings == 0) |
2054 | return NULL; |
2055 | |
2056 | /* lock down all mappings while we scan labels */ |
2057 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
2058 | nd_mapping = &nd_region->mapping[i]; |
2059 | mutex_lock_nested(lock: &nd_mapping->lock, subclass: i); |
2060 | } |
2061 | |
2062 | devs = scan_labels(nd_region); |
2063 | |
2064 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
2065 | int reverse = nd_region->ndr_mappings - 1 - i; |
2066 | |
2067 | nd_mapping = &nd_region->mapping[reverse]; |
2068 | mutex_unlock(lock: &nd_mapping->lock); |
2069 | } |
2070 | |
2071 | return devs; |
2072 | } |
2073 | |
2074 | static void deactivate_labels(void *region) |
2075 | { |
2076 | struct nd_region *nd_region = region; |
2077 | int i; |
2078 | |
2079 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
2080 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; |
2081 | struct nvdimm_drvdata *ndd = nd_mapping->ndd; |
2082 | struct nvdimm *nvdimm = nd_mapping->nvdimm; |
2083 | |
2084 | mutex_lock(&nd_mapping->lock); |
2085 | nd_mapping_free_labels(nd_mapping); |
2086 | mutex_unlock(lock: &nd_mapping->lock); |
2087 | |
2088 | put_ndd(ndd); |
2089 | nd_mapping->ndd = NULL; |
2090 | if (ndd) |
2091 | atomic_dec(v: &nvdimm->busy); |
2092 | } |
2093 | } |
2094 | |
2095 | static int init_active_labels(struct nd_region *nd_region) |
2096 | { |
2097 | int i, rc = 0; |
2098 | |
2099 | for (i = 0; i < nd_region->ndr_mappings; i++) { |
2100 | struct nd_mapping *nd_mapping = &nd_region->mapping[i]; |
2101 | struct nvdimm_drvdata *ndd = to_ndd(nd_mapping); |
2102 | struct nvdimm *nvdimm = nd_mapping->nvdimm; |
2103 | struct nd_label_ent *label_ent; |
2104 | int count, j; |
2105 | |
2106 | /* |
2107 | * If the dimm is disabled then we may need to prevent |
2108 | * the region from being activated. |
2109 | */ |
2110 | if (!ndd) { |
2111 | if (test_bit(NDD_LOCKED, &nvdimm->flags)) |
2112 | /* fail, label data may be unreadable */; |
2113 | else if (test_bit(NDD_LABELING, &nvdimm->flags)) |
2114 | /* fail, labels needed to disambiguate dpa */; |
2115 | else |
2116 | continue; |
2117 | |
2118 | dev_err(&nd_region->dev, "%s: is %s, failing probe\n" , |
2119 | dev_name(&nd_mapping->nvdimm->dev), |
2120 | test_bit(NDD_LOCKED, &nvdimm->flags) |
2121 | ? "locked" : "disabled" ); |
2122 | rc = -ENXIO; |
2123 | goto out; |
2124 | } |
2125 | nd_mapping->ndd = ndd; |
2126 | atomic_inc(v: &nvdimm->busy); |
2127 | get_ndd(ndd); |
2128 | |
2129 | count = nd_label_active_count(ndd); |
2130 | dev_dbg(ndd->dev, "count: %d\n" , count); |
2131 | if (!count) |
2132 | continue; |
2133 | for (j = 0; j < count; j++) { |
2134 | struct nd_namespace_label *label; |
2135 | |
2136 | label_ent = kzalloc(size: sizeof(*label_ent), GFP_KERNEL); |
2137 | if (!label_ent) |
2138 | break; |
2139 | label = nd_label_active(ndd, n: j); |
2140 | label_ent->label = label; |
2141 | |
2142 | mutex_lock(&nd_mapping->lock); |
2143 | list_add_tail(new: &label_ent->list, head: &nd_mapping->labels); |
2144 | mutex_unlock(lock: &nd_mapping->lock); |
2145 | } |
2146 | |
2147 | if (j < count) |
2148 | break; |
2149 | } |
2150 | |
2151 | if (i < nd_region->ndr_mappings) |
2152 | rc = -ENOMEM; |
2153 | |
2154 | out: |
2155 | if (rc) { |
2156 | deactivate_labels(region: nd_region); |
2157 | return rc; |
2158 | } |
2159 | |
2160 | return devm_add_action_or_reset(&nd_region->dev, deactivate_labels, |
2161 | nd_region); |
2162 | } |
2163 | |
2164 | int nd_region_register_namespaces(struct nd_region *nd_region, int *err) |
2165 | { |
2166 | struct device **devs = NULL; |
2167 | int i, rc = 0, type; |
2168 | |
2169 | *err = 0; |
2170 | nvdimm_bus_lock(dev: &nd_region->dev); |
2171 | rc = init_active_labels(nd_region); |
2172 | if (rc) { |
2173 | nvdimm_bus_unlock(dev: &nd_region->dev); |
2174 | return rc; |
2175 | } |
2176 | |
2177 | type = nd_region_to_nstype(nd_region); |
2178 | switch (type) { |
2179 | case ND_DEVICE_NAMESPACE_IO: |
2180 | devs = create_namespace_io(nd_region); |
2181 | break; |
2182 | case ND_DEVICE_NAMESPACE_PMEM: |
2183 | devs = create_namespaces(nd_region); |
2184 | break; |
2185 | default: |
2186 | break; |
2187 | } |
2188 | nvdimm_bus_unlock(dev: &nd_region->dev); |
2189 | |
2190 | if (!devs) |
2191 | return -ENODEV; |
2192 | |
2193 | for (i = 0; devs[i]; i++) { |
2194 | struct device *dev = devs[i]; |
2195 | int id; |
2196 | |
2197 | if (type == ND_DEVICE_NAMESPACE_PMEM) { |
2198 | struct nd_namespace_pmem *nspm; |
2199 | |
2200 | nspm = to_nd_namespace_pmem(dev); |
2201 | id = ida_alloc(ida: &nd_region->ns_ida, GFP_KERNEL); |
2202 | nspm->id = id; |
2203 | } else |
2204 | id = i; |
2205 | |
2206 | if (id < 0) |
2207 | break; |
2208 | dev_set_name(dev, name: "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(objp: devs); |
2234 | |
2235 | if (rc == -ENODEV) |
2236 | return rc; |
2237 | |
2238 | return i; |
2239 | } |
2240 | |