Loading...
1/*
2 * Block Translation Table
3 * Copyright (c) 2014-2015, Intel Corporation.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 */
14#include <linux/highmem.h>
15#include <linux/debugfs.h>
16#include <linux/blkdev.h>
17#include <linux/module.h>
18#include <linux/device.h>
19#include <linux/mutex.h>
20#include <linux/hdreg.h>
21#include <linux/genhd.h>
22#include <linux/sizes.h>
23#include <linux/ndctl.h>
24#include <linux/fs.h>
25#include <linux/nd.h>
26#include "btt.h"
27#include "nd.h"
28
29enum log_ent_request {
30 LOG_NEW_ENT = 0,
31 LOG_OLD_ENT
32};
33
34static int arena_read_bytes(struct arena_info *arena, resource_size_t offset,
35 void *buf, size_t n)
36{
37 struct nd_btt *nd_btt = arena->nd_btt;
38 struct nd_namespace_common *ndns = nd_btt->ndns;
39
40 /* arena offsets are 4K from the base of the device */
41 offset += SZ_4K;
42 return nvdimm_read_bytes(ndns, offset, buf, n);
43}
44
45static int arena_write_bytes(struct arena_info *arena, resource_size_t offset,
46 void *buf, size_t n)
47{
48 struct nd_btt *nd_btt = arena->nd_btt;
49 struct nd_namespace_common *ndns = nd_btt->ndns;
50
51 /* arena offsets are 4K from the base of the device */
52 offset += SZ_4K;
53 return nvdimm_write_bytes(ndns, offset, buf, n);
54}
55
56static int btt_info_write(struct arena_info *arena, struct btt_sb *super)
57{
58 int ret;
59
60 ret = arena_write_bytes(arena, arena->info2off, super,
61 sizeof(struct btt_sb));
62 if (ret)
63 return ret;
64
65 return arena_write_bytes(arena, arena->infooff, super,
66 sizeof(struct btt_sb));
67}
68
69static int btt_info_read(struct arena_info *arena, struct btt_sb *super)
70{
71 WARN_ON(!super);
72 return arena_read_bytes(arena, arena->infooff, super,
73 sizeof(struct btt_sb));
74}
75
76/*
77 * 'raw' version of btt_map write
78 * Assumptions:
79 * mapping is in little-endian
80 * mapping contains 'E' and 'Z' flags as desired
81 */
82static int __btt_map_write(struct arena_info *arena, u32 lba, __le32 mapping)
83{
84 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
85
86 WARN_ON(lba >= arena->external_nlba);
87 return arena_write_bytes(arena, ns_off, &mapping, MAP_ENT_SIZE);
88}
89
90static int btt_map_write(struct arena_info *arena, u32 lba, u32 mapping,
91 u32 z_flag, u32 e_flag)
92{
93 u32 ze;
94 __le32 mapping_le;
95
96 /*
97 * This 'mapping' is supposed to be just the LBA mapping, without
98 * any flags set, so strip the flag bits.
99 */
100 mapping &= MAP_LBA_MASK;
101
102 ze = (z_flag << 1) + e_flag;
103 switch (ze) {
104 case 0:
105 /*
106 * We want to set neither of the Z or E flags, and
107 * in the actual layout, this means setting the bit
108 * positions of both to '1' to indicate a 'normal'
109 * map entry
110 */
111 mapping |= MAP_ENT_NORMAL;
112 break;
113 case 1:
114 mapping |= (1 << MAP_ERR_SHIFT);
115 break;
116 case 2:
117 mapping |= (1 << MAP_TRIM_SHIFT);
118 break;
119 default:
120 /*
121 * The case where Z and E are both sent in as '1' could be
122 * construed as a valid 'normal' case, but we decide not to,
123 * to avoid confusion
124 */
125 WARN_ONCE(1, "Invalid use of Z and E flags\n");
126 return -EIO;
127 }
128
129 mapping_le = cpu_to_le32(mapping);
130 return __btt_map_write(arena, lba, mapping_le);
131}
132
133static int btt_map_read(struct arena_info *arena, u32 lba, u32 *mapping,
134 int *trim, int *error)
135{
136 int ret;
137 __le32 in;
138 u32 raw_mapping, postmap, ze, z_flag, e_flag;
139 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
140
141 WARN_ON(lba >= arena->external_nlba);
142
143 ret = arena_read_bytes(arena, ns_off, &in, MAP_ENT_SIZE);
144 if (ret)
145 return ret;
146
147 raw_mapping = le32_to_cpu(in);
148
149 z_flag = (raw_mapping & MAP_TRIM_MASK) >> MAP_TRIM_SHIFT;
150 e_flag = (raw_mapping & MAP_ERR_MASK) >> MAP_ERR_SHIFT;
151 ze = (z_flag << 1) + e_flag;
152 postmap = raw_mapping & MAP_LBA_MASK;
153
154 /* Reuse the {z,e}_flag variables for *trim and *error */
155 z_flag = 0;
156 e_flag = 0;
157
158 switch (ze) {
159 case 0:
160 /* Initial state. Return postmap = premap */
161 *mapping = lba;
162 break;
163 case 1:
164 *mapping = postmap;
165 e_flag = 1;
166 break;
167 case 2:
168 *mapping = postmap;
169 z_flag = 1;
170 break;
171 case 3:
172 *mapping = postmap;
173 break;
174 default:
175 return -EIO;
176 }
177
178 if (trim)
179 *trim = z_flag;
180 if (error)
181 *error = e_flag;
182
183 return ret;
184}
185
186static int btt_log_read_pair(struct arena_info *arena, u32 lane,
187 struct log_entry *ent)
188{
189 WARN_ON(!ent);
190 return arena_read_bytes(arena,
191 arena->logoff + (2 * lane * LOG_ENT_SIZE), ent,
192 2 * LOG_ENT_SIZE);
193}
194
195static struct dentry *debugfs_root;
196
197static void arena_debugfs_init(struct arena_info *a, struct dentry *parent,
198 int idx)
199{
200 char dirname[32];
201 struct dentry *d;
202
203 /* If for some reason, parent bttN was not created, exit */
204 if (!parent)
205 return;
206
207 snprintf(dirname, 32, "arena%d", idx);
208 d = debugfs_create_dir(dirname, parent);
209 if (IS_ERR_OR_NULL(d))
210 return;
211 a->debugfs_dir = d;
212
213 debugfs_create_x64("size", S_IRUGO, d, &a->size);
214 debugfs_create_x64("external_lba_start", S_IRUGO, d,
215 &a->external_lba_start);
216 debugfs_create_x32("internal_nlba", S_IRUGO, d, &a->internal_nlba);
217 debugfs_create_u32("internal_lbasize", S_IRUGO, d,
218 &a->internal_lbasize);
219 debugfs_create_x32("external_nlba", S_IRUGO, d, &a->external_nlba);
220 debugfs_create_u32("external_lbasize", S_IRUGO, d,
221 &a->external_lbasize);
222 debugfs_create_u32("nfree", S_IRUGO, d, &a->nfree);
223 debugfs_create_u16("version_major", S_IRUGO, d, &a->version_major);
224 debugfs_create_u16("version_minor", S_IRUGO, d, &a->version_minor);
225 debugfs_create_x64("nextoff", S_IRUGO, d, &a->nextoff);
226 debugfs_create_x64("infooff", S_IRUGO, d, &a->infooff);
227 debugfs_create_x64("dataoff", S_IRUGO, d, &a->dataoff);
228 debugfs_create_x64("mapoff", S_IRUGO, d, &a->mapoff);
229 debugfs_create_x64("logoff", S_IRUGO, d, &a->logoff);
230 debugfs_create_x64("info2off", S_IRUGO, d, &a->info2off);
231 debugfs_create_x32("flags", S_IRUGO, d, &a->flags);
232}
233
234static void btt_debugfs_init(struct btt *btt)
235{
236 int i = 0;
237 struct arena_info *arena;
238
239 btt->debugfs_dir = debugfs_create_dir(dev_name(&btt->nd_btt->dev),
240 debugfs_root);
241 if (IS_ERR_OR_NULL(btt->debugfs_dir))
242 return;
243
244 list_for_each_entry(arena, &btt->arena_list, list) {
245 arena_debugfs_init(arena, btt->debugfs_dir, i);
246 i++;
247 }
248}
249
250/*
251 * This function accepts two log entries, and uses the
252 * sequence number to find the 'older' entry.
253 * It also updates the sequence number in this old entry to
254 * make it the 'new' one if the mark_flag is set.
255 * Finally, it returns which of the entries was the older one.
256 *
257 * TODO The logic feels a bit kludge-y. make it better..
258 */
259static int btt_log_get_old(struct log_entry *ent)
260{
261 int old;
262
263 /*
264 * the first ever time this is seen, the entry goes into [0]
265 * the next time, the following logic works out to put this
266 * (next) entry into [1]
267 */
268 if (ent[0].seq == 0) {
269 ent[0].seq = cpu_to_le32(1);
270 return 0;
271 }
272
273 if (ent[0].seq == ent[1].seq)
274 return -EINVAL;
275 if (le32_to_cpu(ent[0].seq) + le32_to_cpu(ent[1].seq) > 5)
276 return -EINVAL;
277
278 if (le32_to_cpu(ent[0].seq) < le32_to_cpu(ent[1].seq)) {
279 if (le32_to_cpu(ent[1].seq) - le32_to_cpu(ent[0].seq) == 1)
280 old = 0;
281 else
282 old = 1;
283 } else {
284 if (le32_to_cpu(ent[0].seq) - le32_to_cpu(ent[1].seq) == 1)
285 old = 1;
286 else
287 old = 0;
288 }
289
290 return old;
291}
292
293static struct device *to_dev(struct arena_info *arena)
294{
295 return &arena->nd_btt->dev;
296}
297
298/*
299 * This function copies the desired (old/new) log entry into ent if
300 * it is not NULL. It returns the sub-slot number (0 or 1)
301 * where the desired log entry was found. Negative return values
302 * indicate errors.
303 */
304static int btt_log_read(struct arena_info *arena, u32 lane,
305 struct log_entry *ent, int old_flag)
306{
307 int ret;
308 int old_ent, ret_ent;
309 struct log_entry log[2];
310
311 ret = btt_log_read_pair(arena, lane, log);
312 if (ret)
313 return -EIO;
314
315 old_ent = btt_log_get_old(log);
316 if (old_ent < 0 || old_ent > 1) {
317 dev_info(to_dev(arena),
318 "log corruption (%d): lane %d seq [%d, %d]\n",
319 old_ent, lane, log[0].seq, log[1].seq);
320 /* TODO set error state? */
321 return -EIO;
322 }
323
324 ret_ent = (old_flag ? old_ent : (1 - old_ent));
325
326 if (ent != NULL)
327 memcpy(ent, &log[ret_ent], LOG_ENT_SIZE);
328
329 return ret_ent;
330}
331
332/*
333 * This function commits a log entry to media
334 * It does _not_ prepare the freelist entry for the next write
335 * btt_flog_write is the wrapper for updating the freelist elements
336 */
337static int __btt_log_write(struct arena_info *arena, u32 lane,
338 u32 sub, struct log_entry *ent)
339{
340 int ret;
341 /*
342 * Ignore the padding in log_entry for calculating log_half.
343 * The entry is 'committed' when we write the sequence number,
344 * and we want to ensure that that is the last thing written.
345 * We don't bother writing the padding as that would be extra
346 * media wear and write amplification
347 */
348 unsigned int log_half = (LOG_ENT_SIZE - 2 * sizeof(u64)) / 2;
349 u64 ns_off = arena->logoff + (((2 * lane) + sub) * LOG_ENT_SIZE);
350 void *src = ent;
351
352 /* split the 16B write into atomic, durable halves */
353 ret = arena_write_bytes(arena, ns_off, src, log_half);
354 if (ret)
355 return ret;
356
357 ns_off += log_half;
358 src += log_half;
359 return arena_write_bytes(arena, ns_off, src, log_half);
360}
361
362static int btt_flog_write(struct arena_info *arena, u32 lane, u32 sub,
363 struct log_entry *ent)
364{
365 int ret;
366
367 ret = __btt_log_write(arena, lane, sub, ent);
368 if (ret)
369 return ret;
370
371 /* prepare the next free entry */
372 arena->freelist[lane].sub = 1 - arena->freelist[lane].sub;
373 if (++(arena->freelist[lane].seq) == 4)
374 arena->freelist[lane].seq = 1;
375 arena->freelist[lane].block = le32_to_cpu(ent->old_map);
376
377 return ret;
378}
379
380/*
381 * This function initializes the BTT map to the initial state, which is
382 * all-zeroes, and indicates an identity mapping
383 */
384static int btt_map_init(struct arena_info *arena)
385{
386 int ret = -EINVAL;
387 void *zerobuf;
388 size_t offset = 0;
389 size_t chunk_size = SZ_2M;
390 size_t mapsize = arena->logoff - arena->mapoff;
391
392 zerobuf = kzalloc(chunk_size, GFP_KERNEL);
393 if (!zerobuf)
394 return -ENOMEM;
395
396 while (mapsize) {
397 size_t size = min(mapsize, chunk_size);
398
399 ret = arena_write_bytes(arena, arena->mapoff + offset, zerobuf,
400 size);
401 if (ret)
402 goto free;
403
404 offset += size;
405 mapsize -= size;
406 cond_resched();
407 }
408
409 free:
410 kfree(zerobuf);
411 return ret;
412}
413
414/*
415 * This function initializes the BTT log with 'fake' entries pointing
416 * to the initial reserved set of blocks as being free
417 */
418static int btt_log_init(struct arena_info *arena)
419{
420 int ret;
421 u32 i;
422 struct log_entry log, zerolog;
423
424 memset(&zerolog, 0, sizeof(zerolog));
425
426 for (i = 0; i < arena->nfree; i++) {
427 log.lba = cpu_to_le32(i);
428 log.old_map = cpu_to_le32(arena->external_nlba + i);
429 log.new_map = cpu_to_le32(arena->external_nlba + i);
430 log.seq = cpu_to_le32(LOG_SEQ_INIT);
431 ret = __btt_log_write(arena, i, 0, &log);
432 if (ret)
433 return ret;
434 ret = __btt_log_write(arena, i, 1, &zerolog);
435 if (ret)
436 return ret;
437 }
438
439 return 0;
440}
441
442static int btt_freelist_init(struct arena_info *arena)
443{
444 int old, new, ret;
445 u32 i, map_entry;
446 struct log_entry log_new, log_old;
447
448 arena->freelist = kcalloc(arena->nfree, sizeof(struct free_entry),
449 GFP_KERNEL);
450 if (!arena->freelist)
451 return -ENOMEM;
452
453 for (i = 0; i < arena->nfree; i++) {
454 old = btt_log_read(arena, i, &log_old, LOG_OLD_ENT);
455 if (old < 0)
456 return old;
457
458 new = btt_log_read(arena, i, &log_new, LOG_NEW_ENT);
459 if (new < 0)
460 return new;
461
462 /* sub points to the next one to be overwritten */
463 arena->freelist[i].sub = 1 - new;
464 arena->freelist[i].seq = nd_inc_seq(le32_to_cpu(log_new.seq));
465 arena->freelist[i].block = le32_to_cpu(log_new.old_map);
466
467 /* This implies a newly created or untouched flog entry */
468 if (log_new.old_map == log_new.new_map)
469 continue;
470
471 /* Check if map recovery is needed */
472 ret = btt_map_read(arena, le32_to_cpu(log_new.lba), &map_entry,
473 NULL, NULL);
474 if (ret)
475 return ret;
476 if ((le32_to_cpu(log_new.new_map) != map_entry) &&
477 (le32_to_cpu(log_new.old_map) == map_entry)) {
478 /*
479 * Last transaction wrote the flog, but wasn't able
480 * to complete the map write. So fix up the map.
481 */
482 ret = btt_map_write(arena, le32_to_cpu(log_new.lba),
483 le32_to_cpu(log_new.new_map), 0, 0);
484 if (ret)
485 return ret;
486 }
487
488 }
489
490 return 0;
491}
492
493static int btt_rtt_init(struct arena_info *arena)
494{
495 arena->rtt = kcalloc(arena->nfree, sizeof(u32), GFP_KERNEL);
496 if (arena->rtt == NULL)
497 return -ENOMEM;
498
499 return 0;
500}
501
502static int btt_maplocks_init(struct arena_info *arena)
503{
504 u32 i;
505
506 arena->map_locks = kcalloc(arena->nfree, sizeof(struct aligned_lock),
507 GFP_KERNEL);
508 if (!arena->map_locks)
509 return -ENOMEM;
510
511 for (i = 0; i < arena->nfree; i++)
512 spin_lock_init(&arena->map_locks[i].lock);
513
514 return 0;
515}
516
517static struct arena_info *alloc_arena(struct btt *btt, size_t size,
518 size_t start, size_t arena_off)
519{
520 struct arena_info *arena;
521 u64 logsize, mapsize, datasize;
522 u64 available = size;
523
524 arena = kzalloc(sizeof(struct arena_info), GFP_KERNEL);
525 if (!arena)
526 return NULL;
527 arena->nd_btt = btt->nd_btt;
528
529 if (!size)
530 return arena;
531
532 arena->size = size;
533 arena->external_lba_start = start;
534 arena->external_lbasize = btt->lbasize;
535 arena->internal_lbasize = roundup(arena->external_lbasize,
536 INT_LBASIZE_ALIGNMENT);
537 arena->nfree = BTT_DEFAULT_NFREE;
538 arena->version_major = 1;
539 arena->version_minor = 1;
540
541 if (available % BTT_PG_SIZE)
542 available -= (available % BTT_PG_SIZE);
543
544 /* Two pages are reserved for the super block and its copy */
545 available -= 2 * BTT_PG_SIZE;
546
547 /* The log takes a fixed amount of space based on nfree */
548 logsize = roundup(2 * arena->nfree * sizeof(struct log_entry),
549 BTT_PG_SIZE);
550 available -= logsize;
551
552 /* Calculate optimal split between map and data area */
553 arena->internal_nlba = div_u64(available - BTT_PG_SIZE,
554 arena->internal_lbasize + MAP_ENT_SIZE);
555 arena->external_nlba = arena->internal_nlba - arena->nfree;
556
557 mapsize = roundup((arena->external_nlba * MAP_ENT_SIZE), BTT_PG_SIZE);
558 datasize = available - mapsize;
559
560 /* 'Absolute' values, relative to start of storage space */
561 arena->infooff = arena_off;
562 arena->dataoff = arena->infooff + BTT_PG_SIZE;
563 arena->mapoff = arena->dataoff + datasize;
564 arena->logoff = arena->mapoff + mapsize;
565 arena->info2off = arena->logoff + logsize;
566 return arena;
567}
568
569static void free_arenas(struct btt *btt)
570{
571 struct arena_info *arena, *next;
572
573 list_for_each_entry_safe(arena, next, &btt->arena_list, list) {
574 list_del(&arena->list);
575 kfree(arena->rtt);
576 kfree(arena->map_locks);
577 kfree(arena->freelist);
578 debugfs_remove_recursive(arena->debugfs_dir);
579 kfree(arena);
580 }
581}
582
583/*
584 * This function reads an existing valid btt superblock and
585 * populates the corresponding arena_info struct
586 */
587static void parse_arena_meta(struct arena_info *arena, struct btt_sb *super,
588 u64 arena_off)
589{
590 arena->internal_nlba = le32_to_cpu(super->internal_nlba);
591 arena->internal_lbasize = le32_to_cpu(super->internal_lbasize);
592 arena->external_nlba = le32_to_cpu(super->external_nlba);
593 arena->external_lbasize = le32_to_cpu(super->external_lbasize);
594 arena->nfree = le32_to_cpu(super->nfree);
595 arena->version_major = le16_to_cpu(super->version_major);
596 arena->version_minor = le16_to_cpu(super->version_minor);
597
598 arena->nextoff = (super->nextoff == 0) ? 0 : (arena_off +
599 le64_to_cpu(super->nextoff));
600 arena->infooff = arena_off;
601 arena->dataoff = arena_off + le64_to_cpu(super->dataoff);
602 arena->mapoff = arena_off + le64_to_cpu(super->mapoff);
603 arena->logoff = arena_off + le64_to_cpu(super->logoff);
604 arena->info2off = arena_off + le64_to_cpu(super->info2off);
605
606 arena->size = (le64_to_cpu(super->nextoff) > 0)
607 ? (le64_to_cpu(super->nextoff))
608 : (arena->info2off - arena->infooff + BTT_PG_SIZE);
609
610 arena->flags = le32_to_cpu(super->flags);
611}
612
613static int discover_arenas(struct btt *btt)
614{
615 int ret = 0;
616 struct arena_info *arena;
617 struct btt_sb *super;
618 size_t remaining = btt->rawsize;
619 u64 cur_nlba = 0;
620 size_t cur_off = 0;
621 int num_arenas = 0;
622
623 super = kzalloc(sizeof(*super), GFP_KERNEL);
624 if (!super)
625 return -ENOMEM;
626
627 while (remaining) {
628 /* Alloc memory for arena */
629 arena = alloc_arena(btt, 0, 0, 0);
630 if (!arena) {
631 ret = -ENOMEM;
632 goto out_super;
633 }
634
635 arena->infooff = cur_off;
636 ret = btt_info_read(arena, super);
637 if (ret)
638 goto out;
639
640 if (!nd_btt_arena_is_valid(btt->nd_btt, super)) {
641 if (remaining == btt->rawsize) {
642 btt->init_state = INIT_NOTFOUND;
643 dev_info(to_dev(arena), "No existing arenas\n");
644 goto out;
645 } else {
646 dev_info(to_dev(arena),
647 "Found corrupted metadata!\n");
648 ret = -ENODEV;
649 goto out;
650 }
651 }
652
653 arena->external_lba_start = cur_nlba;
654 parse_arena_meta(arena, super, cur_off);
655
656 ret = btt_freelist_init(arena);
657 if (ret)
658 goto out;
659
660 ret = btt_rtt_init(arena);
661 if (ret)
662 goto out;
663
664 ret = btt_maplocks_init(arena);
665 if (ret)
666 goto out;
667
668 list_add_tail(&arena->list, &btt->arena_list);
669
670 remaining -= arena->size;
671 cur_off += arena->size;
672 cur_nlba += arena->external_nlba;
673 num_arenas++;
674
675 if (arena->nextoff == 0)
676 break;
677 }
678 btt->num_arenas = num_arenas;
679 btt->nlba = cur_nlba;
680 btt->init_state = INIT_READY;
681
682 kfree(super);
683 return ret;
684
685 out:
686 kfree(arena);
687 free_arenas(btt);
688 out_super:
689 kfree(super);
690 return ret;
691}
692
693static int create_arenas(struct btt *btt)
694{
695 size_t remaining = btt->rawsize;
696 size_t cur_off = 0;
697
698 while (remaining) {
699 struct arena_info *arena;
700 size_t arena_size = min_t(u64, ARENA_MAX_SIZE, remaining);
701
702 remaining -= arena_size;
703 if (arena_size < ARENA_MIN_SIZE)
704 break;
705
706 arena = alloc_arena(btt, arena_size, btt->nlba, cur_off);
707 if (!arena) {
708 free_arenas(btt);
709 return -ENOMEM;
710 }
711 btt->nlba += arena->external_nlba;
712 if (remaining >= ARENA_MIN_SIZE)
713 arena->nextoff = arena->size;
714 else
715 arena->nextoff = 0;
716 cur_off += arena_size;
717 list_add_tail(&arena->list, &btt->arena_list);
718 }
719
720 return 0;
721}
722
723/*
724 * This function completes arena initialization by writing
725 * all the metadata.
726 * It is only called for an uninitialized arena when a write
727 * to that arena occurs for the first time.
728 */
729static int btt_arena_write_layout(struct arena_info *arena)
730{
731 int ret;
732 u64 sum;
733 struct btt_sb *super;
734 struct nd_btt *nd_btt = arena->nd_btt;
735 const u8 *parent_uuid = nd_dev_to_uuid(&nd_btt->ndns->dev);
736
737 ret = btt_map_init(arena);
738 if (ret)
739 return ret;
740
741 ret = btt_log_init(arena);
742 if (ret)
743 return ret;
744
745 super = kzalloc(sizeof(struct btt_sb), GFP_NOIO);
746 if (!super)
747 return -ENOMEM;
748
749 strncpy(super->signature, BTT_SIG, BTT_SIG_LEN);
750 memcpy(super->uuid, nd_btt->uuid, 16);
751 memcpy(super->parent_uuid, parent_uuid, 16);
752 super->flags = cpu_to_le32(arena->flags);
753 super->version_major = cpu_to_le16(arena->version_major);
754 super->version_minor = cpu_to_le16(arena->version_minor);
755 super->external_lbasize = cpu_to_le32(arena->external_lbasize);
756 super->external_nlba = cpu_to_le32(arena->external_nlba);
757 super->internal_lbasize = cpu_to_le32(arena->internal_lbasize);
758 super->internal_nlba = cpu_to_le32(arena->internal_nlba);
759 super->nfree = cpu_to_le32(arena->nfree);
760 super->infosize = cpu_to_le32(sizeof(struct btt_sb));
761 super->nextoff = cpu_to_le64(arena->nextoff);
762 /*
763 * Subtract arena->infooff (arena start) so numbers are relative
764 * to 'this' arena
765 */
766 super->dataoff = cpu_to_le64(arena->dataoff - arena->infooff);
767 super->mapoff = cpu_to_le64(arena->mapoff - arena->infooff);
768 super->logoff = cpu_to_le64(arena->logoff - arena->infooff);
769 super->info2off = cpu_to_le64(arena->info2off - arena->infooff);
770
771 super->flags = 0;
772 sum = nd_sb_checksum((struct nd_gen_sb *) super);
773 super->checksum = cpu_to_le64(sum);
774
775 ret = btt_info_write(arena, super);
776
777 kfree(super);
778 return ret;
779}
780
781/*
782 * This function completes the initialization for the BTT namespace
783 * such that it is ready to accept IOs
784 */
785static int btt_meta_init(struct btt *btt)
786{
787 int ret = 0;
788 struct arena_info *arena;
789
790 mutex_lock(&btt->init_lock);
791 list_for_each_entry(arena, &btt->arena_list, list) {
792 ret = btt_arena_write_layout(arena);
793 if (ret)
794 goto unlock;
795
796 ret = btt_freelist_init(arena);
797 if (ret)
798 goto unlock;
799
800 ret = btt_rtt_init(arena);
801 if (ret)
802 goto unlock;
803
804 ret = btt_maplocks_init(arena);
805 if (ret)
806 goto unlock;
807 }
808
809 btt->init_state = INIT_READY;
810
811 unlock:
812 mutex_unlock(&btt->init_lock);
813 return ret;
814}
815
816static u32 btt_meta_size(struct btt *btt)
817{
818 return btt->lbasize - btt->sector_size;
819}
820
821/*
822 * This function calculates the arena in which the given LBA lies
823 * by doing a linear walk. This is acceptable since we expect only
824 * a few arenas. If we have backing devices that get much larger,
825 * we can construct a balanced binary tree of arenas at init time
826 * so that this range search becomes faster.
827 */
828static int lba_to_arena(struct btt *btt, sector_t sector, __u32 *premap,
829 struct arena_info **arena)
830{
831 struct arena_info *arena_list;
832 __u64 lba = div_u64(sector << SECTOR_SHIFT, btt->sector_size);
833
834 list_for_each_entry(arena_list, &btt->arena_list, list) {
835 if (lba < arena_list->external_nlba) {
836 *arena = arena_list;
837 *premap = lba;
838 return 0;
839 }
840 lba -= arena_list->external_nlba;
841 }
842
843 return -EIO;
844}
845
846/*
847 * The following (lock_map, unlock_map) are mostly just to improve
848 * readability, since they index into an array of locks
849 */
850static void lock_map(struct arena_info *arena, u32 premap)
851 __acquires(&arena->map_locks[idx].lock)
852{
853 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
854
855 spin_lock(&arena->map_locks[idx].lock);
856}
857
858static void unlock_map(struct arena_info *arena, u32 premap)
859 __releases(&arena->map_locks[idx].lock)
860{
861 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
862
863 spin_unlock(&arena->map_locks[idx].lock);
864}
865
866static u64 to_namespace_offset(struct arena_info *arena, u64 lba)
867{
868 return arena->dataoff + ((u64)lba * arena->internal_lbasize);
869}
870
871static int btt_data_read(struct arena_info *arena, struct page *page,
872 unsigned int off, u32 lba, u32 len)
873{
874 int ret;
875 u64 nsoff = to_namespace_offset(arena, lba);
876 void *mem = kmap_atomic(page);
877
878 ret = arena_read_bytes(arena, nsoff, mem + off, len);
879 kunmap_atomic(mem);
880
881 return ret;
882}
883
884static int btt_data_write(struct arena_info *arena, u32 lba,
885 struct page *page, unsigned int off, u32 len)
886{
887 int ret;
888 u64 nsoff = to_namespace_offset(arena, lba);
889 void *mem = kmap_atomic(page);
890
891 ret = arena_write_bytes(arena, nsoff, mem + off, len);
892 kunmap_atomic(mem);
893
894 return ret;
895}
896
897static void zero_fill_data(struct page *page, unsigned int off, u32 len)
898{
899 void *mem = kmap_atomic(page);
900
901 memset(mem + off, 0, len);
902 kunmap_atomic(mem);
903}
904
905#ifdef CONFIG_BLK_DEV_INTEGRITY
906static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
907 struct arena_info *arena, u32 postmap, int rw)
908{
909 unsigned int len = btt_meta_size(btt);
910 u64 meta_nsoff;
911 int ret = 0;
912
913 if (bip == NULL)
914 return 0;
915
916 meta_nsoff = to_namespace_offset(arena, postmap) + btt->sector_size;
917
918 while (len) {
919 unsigned int cur_len;
920 struct bio_vec bv;
921 void *mem;
922
923 bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
924 /*
925 * The 'bv' obtained from bvec_iter_bvec has its .bv_len and
926 * .bv_offset already adjusted for iter->bi_bvec_done, and we
927 * can use those directly
928 */
929
930 cur_len = min(len, bv.bv_len);
931 mem = kmap_atomic(bv.bv_page);
932 if (rw)
933 ret = arena_write_bytes(arena, meta_nsoff,
934 mem + bv.bv_offset, cur_len);
935 else
936 ret = arena_read_bytes(arena, meta_nsoff,
937 mem + bv.bv_offset, cur_len);
938
939 kunmap_atomic(mem);
940 if (ret)
941 return ret;
942
943 len -= cur_len;
944 meta_nsoff += cur_len;
945 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len);
946 }
947
948 return ret;
949}
950
951#else /* CONFIG_BLK_DEV_INTEGRITY */
952static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
953 struct arena_info *arena, u32 postmap, int rw)
954{
955 return 0;
956}
957#endif
958
959static int btt_read_pg(struct btt *btt, struct bio_integrity_payload *bip,
960 struct page *page, unsigned int off, sector_t sector,
961 unsigned int len)
962{
963 int ret = 0;
964 int t_flag, e_flag;
965 struct arena_info *arena = NULL;
966 u32 lane = 0, premap, postmap;
967
968 while (len) {
969 u32 cur_len;
970
971 lane = nd_region_acquire_lane(btt->nd_region);
972
973 ret = lba_to_arena(btt, sector, &premap, &arena);
974 if (ret)
975 goto out_lane;
976
977 cur_len = min(btt->sector_size, len);
978
979 ret = btt_map_read(arena, premap, &postmap, &t_flag, &e_flag);
980 if (ret)
981 goto out_lane;
982
983 /*
984 * We loop to make sure that the post map LBA didn't change
985 * from under us between writing the RTT and doing the actual
986 * read.
987 */
988 while (1) {
989 u32 new_map;
990
991 if (t_flag) {
992 zero_fill_data(page, off, cur_len);
993 goto out_lane;
994 }
995
996 if (e_flag) {
997 ret = -EIO;
998 goto out_lane;
999 }
1000
1001 arena->rtt[lane] = RTT_VALID | postmap;
1002 /*
1003 * Barrier to make sure this write is not reordered
1004 * to do the verification map_read before the RTT store
1005 */
1006 barrier();
1007
1008 ret = btt_map_read(arena, premap, &new_map, &t_flag,
1009 &e_flag);
1010 if (ret)
1011 goto out_rtt;
1012
1013 if (postmap == new_map)
1014 break;
1015
1016 postmap = new_map;
1017 }
1018
1019 ret = btt_data_read(arena, page, off, postmap, cur_len);
1020 if (ret)
1021 goto out_rtt;
1022
1023 if (bip) {
1024 ret = btt_rw_integrity(btt, bip, arena, postmap, READ);
1025 if (ret)
1026 goto out_rtt;
1027 }
1028
1029 arena->rtt[lane] = RTT_INVALID;
1030 nd_region_release_lane(btt->nd_region, lane);
1031
1032 len -= cur_len;
1033 off += cur_len;
1034 sector += btt->sector_size >> SECTOR_SHIFT;
1035 }
1036
1037 return 0;
1038
1039 out_rtt:
1040 arena->rtt[lane] = RTT_INVALID;
1041 out_lane:
1042 nd_region_release_lane(btt->nd_region, lane);
1043 return ret;
1044}
1045
1046static int btt_write_pg(struct btt *btt, struct bio_integrity_payload *bip,
1047 sector_t sector, struct page *page, unsigned int off,
1048 unsigned int len)
1049{
1050 int ret = 0;
1051 struct arena_info *arena = NULL;
1052 u32 premap = 0, old_postmap, new_postmap, lane = 0, i;
1053 struct log_entry log;
1054 int sub;
1055
1056 while (len) {
1057 u32 cur_len;
1058
1059 lane = nd_region_acquire_lane(btt->nd_region);
1060
1061 ret = lba_to_arena(btt, sector, &premap, &arena);
1062 if (ret)
1063 goto out_lane;
1064 cur_len = min(btt->sector_size, len);
1065
1066 if ((arena->flags & IB_FLAG_ERROR_MASK) != 0) {
1067 ret = -EIO;
1068 goto out_lane;
1069 }
1070
1071 new_postmap = arena->freelist[lane].block;
1072
1073 /* Wait if the new block is being read from */
1074 for (i = 0; i < arena->nfree; i++)
1075 while (arena->rtt[i] == (RTT_VALID | new_postmap))
1076 cpu_relax();
1077
1078
1079 if (new_postmap >= arena->internal_nlba) {
1080 ret = -EIO;
1081 goto out_lane;
1082 }
1083
1084 ret = btt_data_write(arena, new_postmap, page, off, cur_len);
1085 if (ret)
1086 goto out_lane;
1087
1088 if (bip) {
1089 ret = btt_rw_integrity(btt, bip, arena, new_postmap,
1090 WRITE);
1091 if (ret)
1092 goto out_lane;
1093 }
1094
1095 lock_map(arena, premap);
1096 ret = btt_map_read(arena, premap, &old_postmap, NULL, NULL);
1097 if (ret)
1098 goto out_map;
1099 if (old_postmap >= arena->internal_nlba) {
1100 ret = -EIO;
1101 goto out_map;
1102 }
1103
1104 log.lba = cpu_to_le32(premap);
1105 log.old_map = cpu_to_le32(old_postmap);
1106 log.new_map = cpu_to_le32(new_postmap);
1107 log.seq = cpu_to_le32(arena->freelist[lane].seq);
1108 sub = arena->freelist[lane].sub;
1109 ret = btt_flog_write(arena, lane, sub, &log);
1110 if (ret)
1111 goto out_map;
1112
1113 ret = btt_map_write(arena, premap, new_postmap, 0, 0);
1114 if (ret)
1115 goto out_map;
1116
1117 unlock_map(arena, premap);
1118 nd_region_release_lane(btt->nd_region, lane);
1119
1120 len -= cur_len;
1121 off += cur_len;
1122 sector += btt->sector_size >> SECTOR_SHIFT;
1123 }
1124
1125 return 0;
1126
1127 out_map:
1128 unlock_map(arena, premap);
1129 out_lane:
1130 nd_region_release_lane(btt->nd_region, lane);
1131 return ret;
1132}
1133
1134static int btt_do_bvec(struct btt *btt, struct bio_integrity_payload *bip,
1135 struct page *page, unsigned int len, unsigned int off,
1136 bool is_write, sector_t sector)
1137{
1138 int ret;
1139
1140 if (!is_write) {
1141 ret = btt_read_pg(btt, bip, page, off, sector, len);
1142 flush_dcache_page(page);
1143 } else {
1144 flush_dcache_page(page);
1145 ret = btt_write_pg(btt, bip, sector, page, off, len);
1146 }
1147
1148 return ret;
1149}
1150
1151static blk_qc_t btt_make_request(struct request_queue *q, struct bio *bio)
1152{
1153 struct bio_integrity_payload *bip = bio_integrity(bio);
1154 struct btt *btt = q->queuedata;
1155 struct bvec_iter iter;
1156 unsigned long start;
1157 struct bio_vec bvec;
1158 int err = 0;
1159 bool do_acct;
1160
1161 /*
1162 * bio_integrity_enabled also checks if the bio already has an
1163 * integrity payload attached. If it does, we *don't* do a
1164 * bio_integrity_prep here - the payload has been generated by
1165 * another kernel subsystem, and we just pass it through.
1166 */
1167 if (bio_integrity_enabled(bio) && bio_integrity_prep(bio)) {
1168 bio->bi_error = -EIO;
1169 goto out;
1170 }
1171
1172 do_acct = nd_iostat_start(bio, &start);
1173 bio_for_each_segment(bvec, bio, iter) {
1174 unsigned int len = bvec.bv_len;
1175
1176 BUG_ON(len > PAGE_SIZE);
1177 /* Make sure len is in multiples of sector size. */
1178 /* XXX is this right? */
1179 BUG_ON(len < btt->sector_size);
1180 BUG_ON(len % btt->sector_size);
1181
1182 err = btt_do_bvec(btt, bip, bvec.bv_page, len, bvec.bv_offset,
1183 op_is_write(bio_op(bio)), iter.bi_sector);
1184 if (err) {
1185 dev_info(&btt->nd_btt->dev,
1186 "io error in %s sector %lld, len %d,\n",
1187 (op_is_write(bio_op(bio))) ? "WRITE" :
1188 "READ",
1189 (unsigned long long) iter.bi_sector, len);
1190 bio->bi_error = err;
1191 break;
1192 }
1193 }
1194 if (do_acct)
1195 nd_iostat_end(bio, start);
1196
1197out:
1198 bio_endio(bio);
1199 return BLK_QC_T_NONE;
1200}
1201
1202static int btt_rw_page(struct block_device *bdev, sector_t sector,
1203 struct page *page, bool is_write)
1204{
1205 struct btt *btt = bdev->bd_disk->private_data;
1206
1207 btt_do_bvec(btt, NULL, page, PAGE_SIZE, 0, is_write, sector);
1208 page_endio(page, is_write, 0);
1209 return 0;
1210}
1211
1212
1213static int btt_getgeo(struct block_device *bd, struct hd_geometry *geo)
1214{
1215 /* some standard values */
1216 geo->heads = 1 << 6;
1217 geo->sectors = 1 << 5;
1218 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
1219 return 0;
1220}
1221
1222static const struct block_device_operations btt_fops = {
1223 .owner = THIS_MODULE,
1224 .rw_page = btt_rw_page,
1225 .getgeo = btt_getgeo,
1226 .revalidate_disk = nvdimm_revalidate_disk,
1227};
1228
1229static int btt_blk_init(struct btt *btt)
1230{
1231 struct nd_btt *nd_btt = btt->nd_btt;
1232 struct nd_namespace_common *ndns = nd_btt->ndns;
1233
1234 /* create a new disk and request queue for btt */
1235 btt->btt_queue = blk_alloc_queue(GFP_KERNEL);
1236 if (!btt->btt_queue)
1237 return -ENOMEM;
1238
1239 btt->btt_disk = alloc_disk(0);
1240 if (!btt->btt_disk) {
1241 blk_cleanup_queue(btt->btt_queue);
1242 return -ENOMEM;
1243 }
1244
1245 nvdimm_namespace_disk_name(ndns, btt->btt_disk->disk_name);
1246 btt->btt_disk->first_minor = 0;
1247 btt->btt_disk->fops = &btt_fops;
1248 btt->btt_disk->private_data = btt;
1249 btt->btt_disk->queue = btt->btt_queue;
1250 btt->btt_disk->flags = GENHD_FL_EXT_DEVT;
1251
1252 blk_queue_make_request(btt->btt_queue, btt_make_request);
1253 blk_queue_logical_block_size(btt->btt_queue, btt->sector_size);
1254 blk_queue_max_hw_sectors(btt->btt_queue, UINT_MAX);
1255 blk_queue_bounce_limit(btt->btt_queue, BLK_BOUNCE_ANY);
1256 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, btt->btt_queue);
1257 btt->btt_queue->queuedata = btt;
1258
1259 set_capacity(btt->btt_disk, 0);
1260 device_add_disk(&btt->nd_btt->dev, btt->btt_disk);
1261 if (btt_meta_size(btt)) {
1262 int rc = nd_integrity_init(btt->btt_disk, btt_meta_size(btt));
1263
1264 if (rc) {
1265 del_gendisk(btt->btt_disk);
1266 put_disk(btt->btt_disk);
1267 blk_cleanup_queue(btt->btt_queue);
1268 return rc;
1269 }
1270 }
1271 set_capacity(btt->btt_disk, btt->nlba * btt->sector_size >> 9);
1272 btt->nd_btt->size = btt->nlba * (u64)btt->sector_size;
1273 revalidate_disk(btt->btt_disk);
1274
1275 return 0;
1276}
1277
1278static void btt_blk_cleanup(struct btt *btt)
1279{
1280 del_gendisk(btt->btt_disk);
1281 put_disk(btt->btt_disk);
1282 blk_cleanup_queue(btt->btt_queue);
1283}
1284
1285/**
1286 * btt_init - initialize a block translation table for the given device
1287 * @nd_btt: device with BTT geometry and backing device info
1288 * @rawsize: raw size in bytes of the backing device
1289 * @lbasize: lba size of the backing device
1290 * @uuid: A uuid for the backing device - this is stored on media
1291 * @maxlane: maximum number of parallel requests the device can handle
1292 *
1293 * Initialize a Block Translation Table on a backing device to provide
1294 * single sector power fail atomicity.
1295 *
1296 * Context:
1297 * Might sleep.
1298 *
1299 * Returns:
1300 * Pointer to a new struct btt on success, NULL on failure.
1301 */
1302static struct btt *btt_init(struct nd_btt *nd_btt, unsigned long long rawsize,
1303 u32 lbasize, u8 *uuid, struct nd_region *nd_region)
1304{
1305 int ret;
1306 struct btt *btt;
1307 struct device *dev = &nd_btt->dev;
1308
1309 btt = devm_kzalloc(dev, sizeof(struct btt), GFP_KERNEL);
1310 if (!btt)
1311 return NULL;
1312
1313 btt->nd_btt = nd_btt;
1314 btt->rawsize = rawsize;
1315 btt->lbasize = lbasize;
1316 btt->sector_size = ((lbasize >= 4096) ? 4096 : 512);
1317 INIT_LIST_HEAD(&btt->arena_list);
1318 mutex_init(&btt->init_lock);
1319 btt->nd_region = nd_region;
1320
1321 ret = discover_arenas(btt);
1322 if (ret) {
1323 dev_err(dev, "init: error in arena_discover: %d\n", ret);
1324 return NULL;
1325 }
1326
1327 if (btt->init_state != INIT_READY && nd_region->ro) {
1328 dev_info(dev, "%s is read-only, unable to init btt metadata\n",
1329 dev_name(&nd_region->dev));
1330 return NULL;
1331 } else if (btt->init_state != INIT_READY) {
1332 btt->num_arenas = (rawsize / ARENA_MAX_SIZE) +
1333 ((rawsize % ARENA_MAX_SIZE) ? 1 : 0);
1334 dev_dbg(dev, "init: %d arenas for %llu rawsize\n",
1335 btt->num_arenas, rawsize);
1336
1337 ret = create_arenas(btt);
1338 if (ret) {
1339 dev_info(dev, "init: create_arenas: %d\n", ret);
1340 return NULL;
1341 }
1342
1343 ret = btt_meta_init(btt);
1344 if (ret) {
1345 dev_err(dev, "init: error in meta_init: %d\n", ret);
1346 return NULL;
1347 }
1348 }
1349
1350 ret = btt_blk_init(btt);
1351 if (ret) {
1352 dev_err(dev, "init: error in blk_init: %d\n", ret);
1353 return NULL;
1354 }
1355
1356 btt_debugfs_init(btt);
1357
1358 return btt;
1359}
1360
1361/**
1362 * btt_fini - de-initialize a BTT
1363 * @btt: the BTT handle that was generated by btt_init
1364 *
1365 * De-initialize a Block Translation Table on device removal
1366 *
1367 * Context:
1368 * Might sleep.
1369 */
1370static void btt_fini(struct btt *btt)
1371{
1372 if (btt) {
1373 btt_blk_cleanup(btt);
1374 free_arenas(btt);
1375 debugfs_remove_recursive(btt->debugfs_dir);
1376 }
1377}
1378
1379int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns)
1380{
1381 struct nd_btt *nd_btt = to_nd_btt(ndns->claim);
1382 struct nd_region *nd_region;
1383 struct btt *btt;
1384 size_t rawsize;
1385
1386 if (!nd_btt->uuid || !nd_btt->ndns || !nd_btt->lbasize) {
1387 dev_dbg(&nd_btt->dev, "incomplete btt configuration\n");
1388 return -ENODEV;
1389 }
1390
1391 rawsize = nvdimm_namespace_capacity(ndns) - SZ_4K;
1392 if (rawsize < ARENA_MIN_SIZE) {
1393 dev_dbg(&nd_btt->dev, "%s must be at least %ld bytes\n",
1394 dev_name(&ndns->dev), ARENA_MIN_SIZE + SZ_4K);
1395 return -ENXIO;
1396 }
1397 nd_region = to_nd_region(nd_btt->dev.parent);
1398 btt = btt_init(nd_btt, rawsize, nd_btt->lbasize, nd_btt->uuid,
1399 nd_region);
1400 if (!btt)
1401 return -ENOMEM;
1402 nd_btt->btt = btt;
1403
1404 return 0;
1405}
1406EXPORT_SYMBOL(nvdimm_namespace_attach_btt);
1407
1408int nvdimm_namespace_detach_btt(struct nd_btt *nd_btt)
1409{
1410 struct btt *btt = nd_btt->btt;
1411
1412 btt_fini(btt);
1413 nd_btt->btt = NULL;
1414
1415 return 0;
1416}
1417EXPORT_SYMBOL(nvdimm_namespace_detach_btt);
1418
1419static int __init nd_btt_init(void)
1420{
1421 int rc = 0;
1422
1423 debugfs_root = debugfs_create_dir("btt", NULL);
1424 if (IS_ERR_OR_NULL(debugfs_root))
1425 rc = -ENXIO;
1426
1427 return rc;
1428}
1429
1430static void __exit nd_btt_exit(void)
1431{
1432 debugfs_remove_recursive(debugfs_root);
1433}
1434
1435MODULE_ALIAS_ND_DEVICE(ND_DEVICE_BTT);
1436MODULE_AUTHOR("Vishal Verma <vishal.l.verma@linux.intel.com>");
1437MODULE_LICENSE("GPL v2");
1438module_init(nd_btt_init);
1439module_exit(nd_btt_exit);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Block Translation Table
4 * Copyright (c) 2014-2015, Intel Corporation.
5 */
6#include <linux/highmem.h>
7#include <linux/debugfs.h>
8#include <linux/blkdev.h>
9#include <linux/pagemap.h>
10#include <linux/module.h>
11#include <linux/device.h>
12#include <linux/mutex.h>
13#include <linux/hdreg.h>
14#include <linux/sizes.h>
15#include <linux/ndctl.h>
16#include <linux/fs.h>
17#include <linux/nd.h>
18#include <linux/backing-dev.h>
19#include <linux/cleanup.h>
20#include "btt.h"
21#include "nd.h"
22
23enum log_ent_request {
24 LOG_NEW_ENT = 0,
25 LOG_OLD_ENT
26};
27
28static struct device *to_dev(struct arena_info *arena)
29{
30 return &arena->nd_btt->dev;
31}
32
33static u64 adjust_initial_offset(struct nd_btt *nd_btt, u64 offset)
34{
35 return offset + nd_btt->initial_offset;
36}
37
38static int arena_read_bytes(struct arena_info *arena, resource_size_t offset,
39 void *buf, size_t n, unsigned long flags)
40{
41 struct nd_btt *nd_btt = arena->nd_btt;
42 struct nd_namespace_common *ndns = nd_btt->ndns;
43
44 /* arena offsets may be shifted from the base of the device */
45 offset = adjust_initial_offset(nd_btt, offset);
46 return nvdimm_read_bytes(ndns, offset, buf, n, flags);
47}
48
49static int arena_write_bytes(struct arena_info *arena, resource_size_t offset,
50 void *buf, size_t n, unsigned long flags)
51{
52 struct nd_btt *nd_btt = arena->nd_btt;
53 struct nd_namespace_common *ndns = nd_btt->ndns;
54
55 /* arena offsets may be shifted from the base of the device */
56 offset = adjust_initial_offset(nd_btt, offset);
57 return nvdimm_write_bytes(ndns, offset, buf, n, flags);
58}
59
60static int btt_info_write(struct arena_info *arena, struct btt_sb *super)
61{
62 int ret;
63
64 /*
65 * infooff and info2off should always be at least 512B aligned.
66 * We rely on that to make sure rw_bytes does error clearing
67 * correctly, so make sure that is the case.
68 */
69 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->infooff, 512),
70 "arena->infooff: %#llx is unaligned\n", arena->infooff);
71 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->info2off, 512),
72 "arena->info2off: %#llx is unaligned\n", arena->info2off);
73
74 ret = arena_write_bytes(arena, arena->info2off, super,
75 sizeof(struct btt_sb), 0);
76 if (ret)
77 return ret;
78
79 return arena_write_bytes(arena, arena->infooff, super,
80 sizeof(struct btt_sb), 0);
81}
82
83static int btt_info_read(struct arena_info *arena, struct btt_sb *super)
84{
85 return arena_read_bytes(arena, arena->infooff, super,
86 sizeof(struct btt_sb), 0);
87}
88
89/*
90 * 'raw' version of btt_map write
91 * Assumptions:
92 * mapping is in little-endian
93 * mapping contains 'E' and 'Z' flags as desired
94 */
95static int __btt_map_write(struct arena_info *arena, u32 lba, __le32 mapping,
96 unsigned long flags)
97{
98 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
99
100 if (unlikely(lba >= arena->external_nlba))
101 dev_err_ratelimited(to_dev(arena),
102 "%s: lba %#x out of range (max: %#x)\n",
103 __func__, lba, arena->external_nlba);
104 return arena_write_bytes(arena, ns_off, &mapping, MAP_ENT_SIZE, flags);
105}
106
107static int btt_map_write(struct arena_info *arena, u32 lba, u32 mapping,
108 u32 z_flag, u32 e_flag, unsigned long rwb_flags)
109{
110 u32 ze;
111 __le32 mapping_le;
112
113 /*
114 * This 'mapping' is supposed to be just the LBA mapping, without
115 * any flags set, so strip the flag bits.
116 */
117 mapping = ent_lba(mapping);
118
119 ze = (z_flag << 1) + e_flag;
120 switch (ze) {
121 case 0:
122 /*
123 * We want to set neither of the Z or E flags, and
124 * in the actual layout, this means setting the bit
125 * positions of both to '1' to indicate a 'normal'
126 * map entry
127 */
128 mapping |= MAP_ENT_NORMAL;
129 break;
130 case 1:
131 mapping |= (1 << MAP_ERR_SHIFT);
132 break;
133 case 2:
134 mapping |= (1 << MAP_TRIM_SHIFT);
135 break;
136 default:
137 /*
138 * The case where Z and E are both sent in as '1' could be
139 * construed as a valid 'normal' case, but we decide not to,
140 * to avoid confusion
141 */
142 dev_err_ratelimited(to_dev(arena),
143 "Invalid use of Z and E flags\n");
144 return -EIO;
145 }
146
147 mapping_le = cpu_to_le32(mapping);
148 return __btt_map_write(arena, lba, mapping_le, rwb_flags);
149}
150
151static int btt_map_read(struct arena_info *arena, u32 lba, u32 *mapping,
152 int *trim, int *error, unsigned long rwb_flags)
153{
154 int ret;
155 __le32 in;
156 u32 raw_mapping, postmap, ze, z_flag, e_flag;
157 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
158
159 if (unlikely(lba >= arena->external_nlba))
160 dev_err_ratelimited(to_dev(arena),
161 "%s: lba %#x out of range (max: %#x)\n",
162 __func__, lba, arena->external_nlba);
163
164 ret = arena_read_bytes(arena, ns_off, &in, MAP_ENT_SIZE, rwb_flags);
165 if (ret)
166 return ret;
167
168 raw_mapping = le32_to_cpu(in);
169
170 z_flag = ent_z_flag(raw_mapping);
171 e_flag = ent_e_flag(raw_mapping);
172 ze = (z_flag << 1) + e_flag;
173 postmap = ent_lba(raw_mapping);
174
175 /* Reuse the {z,e}_flag variables for *trim and *error */
176 z_flag = 0;
177 e_flag = 0;
178
179 switch (ze) {
180 case 0:
181 /* Initial state. Return postmap = premap */
182 *mapping = lba;
183 break;
184 case 1:
185 *mapping = postmap;
186 e_flag = 1;
187 break;
188 case 2:
189 *mapping = postmap;
190 z_flag = 1;
191 break;
192 case 3:
193 *mapping = postmap;
194 break;
195 default:
196 return -EIO;
197 }
198
199 if (trim)
200 *trim = z_flag;
201 if (error)
202 *error = e_flag;
203
204 return ret;
205}
206
207static int btt_log_group_read(struct arena_info *arena, u32 lane,
208 struct log_group *log)
209{
210 return arena_read_bytes(arena,
211 arena->logoff + (lane * LOG_GRP_SIZE), log,
212 LOG_GRP_SIZE, 0);
213}
214
215static struct dentry *debugfs_root;
216
217static void arena_debugfs_init(struct arena_info *a, struct dentry *parent,
218 int idx)
219{
220 char dirname[32];
221 struct dentry *d;
222
223 /* If for some reason, parent bttN was not created, exit */
224 if (!parent)
225 return;
226
227 snprintf(dirname, 32, "arena%d", idx);
228 d = debugfs_create_dir(dirname, parent);
229 if (IS_ERR_OR_NULL(d))
230 return;
231 a->debugfs_dir = d;
232
233 debugfs_create_x64("size", S_IRUGO, d, &a->size);
234 debugfs_create_x64("external_lba_start", S_IRUGO, d,
235 &a->external_lba_start);
236 debugfs_create_x32("internal_nlba", S_IRUGO, d, &a->internal_nlba);
237 debugfs_create_u32("internal_lbasize", S_IRUGO, d,
238 &a->internal_lbasize);
239 debugfs_create_x32("external_nlba", S_IRUGO, d, &a->external_nlba);
240 debugfs_create_u32("external_lbasize", S_IRUGO, d,
241 &a->external_lbasize);
242 debugfs_create_u32("nfree", S_IRUGO, d, &a->nfree);
243 debugfs_create_u16("version_major", S_IRUGO, d, &a->version_major);
244 debugfs_create_u16("version_minor", S_IRUGO, d, &a->version_minor);
245 debugfs_create_x64("nextoff", S_IRUGO, d, &a->nextoff);
246 debugfs_create_x64("infooff", S_IRUGO, d, &a->infooff);
247 debugfs_create_x64("dataoff", S_IRUGO, d, &a->dataoff);
248 debugfs_create_x64("mapoff", S_IRUGO, d, &a->mapoff);
249 debugfs_create_x64("logoff", S_IRUGO, d, &a->logoff);
250 debugfs_create_x64("info2off", S_IRUGO, d, &a->info2off);
251 debugfs_create_x32("flags", S_IRUGO, d, &a->flags);
252 debugfs_create_u32("log_index_0", S_IRUGO, d, &a->log_index[0]);
253 debugfs_create_u32("log_index_1", S_IRUGO, d, &a->log_index[1]);
254}
255
256static void btt_debugfs_init(struct btt *btt)
257{
258 int i = 0;
259 struct arena_info *arena;
260
261 btt->debugfs_dir = debugfs_create_dir(dev_name(&btt->nd_btt->dev),
262 debugfs_root);
263 if (IS_ERR_OR_NULL(btt->debugfs_dir))
264 return;
265
266 list_for_each_entry(arena, &btt->arena_list, list) {
267 arena_debugfs_init(arena, btt->debugfs_dir, i);
268 i++;
269 }
270}
271
272static u32 log_seq(struct log_group *log, int log_idx)
273{
274 return le32_to_cpu(log->ent[log_idx].seq);
275}
276
277/*
278 * This function accepts two log entries, and uses the
279 * sequence number to find the 'older' entry.
280 * It also updates the sequence number in this old entry to
281 * make it the 'new' one if the mark_flag is set.
282 * Finally, it returns which of the entries was the older one.
283 *
284 * TODO The logic feels a bit kludge-y. make it better..
285 */
286static int btt_log_get_old(struct arena_info *a, struct log_group *log)
287{
288 int idx0 = a->log_index[0];
289 int idx1 = a->log_index[1];
290 int old;
291
292 /*
293 * the first ever time this is seen, the entry goes into [0]
294 * the next time, the following logic works out to put this
295 * (next) entry into [1]
296 */
297 if (log_seq(log, idx0) == 0) {
298 log->ent[idx0].seq = cpu_to_le32(1);
299 return 0;
300 }
301
302 if (log_seq(log, idx0) == log_seq(log, idx1))
303 return -EINVAL;
304 if (log_seq(log, idx0) + log_seq(log, idx1) > 5)
305 return -EINVAL;
306
307 if (log_seq(log, idx0) < log_seq(log, idx1)) {
308 if ((log_seq(log, idx1) - log_seq(log, idx0)) == 1)
309 old = 0;
310 else
311 old = 1;
312 } else {
313 if ((log_seq(log, idx0) - log_seq(log, idx1)) == 1)
314 old = 1;
315 else
316 old = 0;
317 }
318
319 return old;
320}
321
322/*
323 * This function copies the desired (old/new) log entry into ent if
324 * it is not NULL. It returns the sub-slot number (0 or 1)
325 * where the desired log entry was found. Negative return values
326 * indicate errors.
327 */
328static int btt_log_read(struct arena_info *arena, u32 lane,
329 struct log_entry *ent, int old_flag)
330{
331 int ret;
332 int old_ent, ret_ent;
333 struct log_group log;
334
335 ret = btt_log_group_read(arena, lane, &log);
336 if (ret)
337 return -EIO;
338
339 old_ent = btt_log_get_old(arena, &log);
340 if (old_ent < 0 || old_ent > 1) {
341 dev_err(to_dev(arena),
342 "log corruption (%d): lane %d seq [%d, %d]\n",
343 old_ent, lane, log.ent[arena->log_index[0]].seq,
344 log.ent[arena->log_index[1]].seq);
345 /* TODO set error state? */
346 return -EIO;
347 }
348
349 ret_ent = (old_flag ? old_ent : (1 - old_ent));
350
351 if (ent != NULL)
352 memcpy(ent, &log.ent[arena->log_index[ret_ent]], LOG_ENT_SIZE);
353
354 return ret_ent;
355}
356
357/*
358 * This function commits a log entry to media
359 * It does _not_ prepare the freelist entry for the next write
360 * btt_flog_write is the wrapper for updating the freelist elements
361 */
362static int __btt_log_write(struct arena_info *arena, u32 lane,
363 u32 sub, struct log_entry *ent, unsigned long flags)
364{
365 int ret;
366 u32 group_slot = arena->log_index[sub];
367 unsigned int log_half = LOG_ENT_SIZE / 2;
368 void *src = ent;
369 u64 ns_off;
370
371 ns_off = arena->logoff + (lane * LOG_GRP_SIZE) +
372 (group_slot * LOG_ENT_SIZE);
373 /* split the 16B write into atomic, durable halves */
374 ret = arena_write_bytes(arena, ns_off, src, log_half, flags);
375 if (ret)
376 return ret;
377
378 ns_off += log_half;
379 src += log_half;
380 return arena_write_bytes(arena, ns_off, src, log_half, flags);
381}
382
383static int btt_flog_write(struct arena_info *arena, u32 lane, u32 sub,
384 struct log_entry *ent)
385{
386 int ret;
387
388 ret = __btt_log_write(arena, lane, sub, ent, NVDIMM_IO_ATOMIC);
389 if (ret)
390 return ret;
391
392 /* prepare the next free entry */
393 arena->freelist[lane].sub = 1 - arena->freelist[lane].sub;
394 if (++(arena->freelist[lane].seq) == 4)
395 arena->freelist[lane].seq = 1;
396 if (ent_e_flag(le32_to_cpu(ent->old_map)))
397 arena->freelist[lane].has_err = 1;
398 arena->freelist[lane].block = ent_lba(le32_to_cpu(ent->old_map));
399
400 return ret;
401}
402
403/*
404 * This function initializes the BTT map to the initial state, which is
405 * all-zeroes, and indicates an identity mapping
406 */
407static int btt_map_init(struct arena_info *arena)
408{
409 int ret = -EINVAL;
410 void *zerobuf;
411 size_t offset = 0;
412 size_t chunk_size = SZ_2M;
413 size_t mapsize = arena->logoff - arena->mapoff;
414
415 zerobuf = kzalloc(chunk_size, GFP_KERNEL);
416 if (!zerobuf)
417 return -ENOMEM;
418
419 /*
420 * mapoff should always be at least 512B aligned. We rely on that to
421 * make sure rw_bytes does error clearing correctly, so make sure that
422 * is the case.
423 */
424 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->mapoff, 512),
425 "arena->mapoff: %#llx is unaligned\n", arena->mapoff);
426
427 while (mapsize) {
428 size_t size = min(mapsize, chunk_size);
429
430 dev_WARN_ONCE(to_dev(arena), size < 512,
431 "chunk size: %#zx is unaligned\n", size);
432 ret = arena_write_bytes(arena, arena->mapoff + offset, zerobuf,
433 size, 0);
434 if (ret)
435 goto free;
436
437 offset += size;
438 mapsize -= size;
439 cond_resched();
440 }
441
442 free:
443 kfree(zerobuf);
444 return ret;
445}
446
447/*
448 * This function initializes the BTT log with 'fake' entries pointing
449 * to the initial reserved set of blocks as being free
450 */
451static int btt_log_init(struct arena_info *arena)
452{
453 size_t logsize = arena->info2off - arena->logoff;
454 size_t chunk_size = SZ_4K, offset = 0;
455 struct log_entry ent;
456 void *zerobuf;
457 int ret;
458 u32 i;
459
460 zerobuf = kzalloc(chunk_size, GFP_KERNEL);
461 if (!zerobuf)
462 return -ENOMEM;
463 /*
464 * logoff should always be at least 512B aligned. We rely on that to
465 * make sure rw_bytes does error clearing correctly, so make sure that
466 * is the case.
467 */
468 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->logoff, 512),
469 "arena->logoff: %#llx is unaligned\n", arena->logoff);
470
471 while (logsize) {
472 size_t size = min(logsize, chunk_size);
473
474 dev_WARN_ONCE(to_dev(arena), size < 512,
475 "chunk size: %#zx is unaligned\n", size);
476 ret = arena_write_bytes(arena, arena->logoff + offset, zerobuf,
477 size, 0);
478 if (ret)
479 goto free;
480
481 offset += size;
482 logsize -= size;
483 cond_resched();
484 }
485
486 for (i = 0; i < arena->nfree; i++) {
487 ent.lba = cpu_to_le32(i);
488 ent.old_map = cpu_to_le32(arena->external_nlba + i);
489 ent.new_map = cpu_to_le32(arena->external_nlba + i);
490 ent.seq = cpu_to_le32(LOG_SEQ_INIT);
491 ret = __btt_log_write(arena, i, 0, &ent, 0);
492 if (ret)
493 goto free;
494 }
495
496 free:
497 kfree(zerobuf);
498 return ret;
499}
500
501static u64 to_namespace_offset(struct arena_info *arena, u64 lba)
502{
503 return arena->dataoff + ((u64)lba * arena->internal_lbasize);
504}
505
506static int arena_clear_freelist_error(struct arena_info *arena, u32 lane)
507{
508 int ret = 0;
509
510 if (arena->freelist[lane].has_err) {
511 void *zero_page = page_address(ZERO_PAGE(0));
512 u32 lba = arena->freelist[lane].block;
513 u64 nsoff = to_namespace_offset(arena, lba);
514 unsigned long len = arena->sector_size;
515
516 mutex_lock(&arena->err_lock);
517
518 while (len) {
519 unsigned long chunk = min(len, PAGE_SIZE);
520
521 ret = arena_write_bytes(arena, nsoff, zero_page,
522 chunk, 0);
523 if (ret)
524 break;
525 len -= chunk;
526 nsoff += chunk;
527 if (len == 0)
528 arena->freelist[lane].has_err = 0;
529 }
530 mutex_unlock(&arena->err_lock);
531 }
532 return ret;
533}
534
535static int btt_freelist_init(struct arena_info *arena)
536{
537 int new, ret;
538 struct log_entry log_new;
539 u32 i, map_entry, log_oldmap, log_newmap;
540
541 arena->freelist = kcalloc(arena->nfree, sizeof(struct free_entry),
542 GFP_KERNEL);
543 if (!arena->freelist)
544 return -ENOMEM;
545
546 for (i = 0; i < arena->nfree; i++) {
547 new = btt_log_read(arena, i, &log_new, LOG_NEW_ENT);
548 if (new < 0)
549 return new;
550
551 /* old and new map entries with any flags stripped out */
552 log_oldmap = ent_lba(le32_to_cpu(log_new.old_map));
553 log_newmap = ent_lba(le32_to_cpu(log_new.new_map));
554
555 /* sub points to the next one to be overwritten */
556 arena->freelist[i].sub = 1 - new;
557 arena->freelist[i].seq = nd_inc_seq(le32_to_cpu(log_new.seq));
558 arena->freelist[i].block = log_oldmap;
559
560 /*
561 * FIXME: if error clearing fails during init, we want to make
562 * the BTT read-only
563 */
564 if (ent_e_flag(le32_to_cpu(log_new.old_map)) &&
565 !ent_normal(le32_to_cpu(log_new.old_map))) {
566 arena->freelist[i].has_err = 1;
567 ret = arena_clear_freelist_error(arena, i);
568 if (ret)
569 dev_err_ratelimited(to_dev(arena),
570 "Unable to clear known errors\n");
571 }
572
573 /* This implies a newly created or untouched flog entry */
574 if (log_oldmap == log_newmap)
575 continue;
576
577 /* Check if map recovery is needed */
578 ret = btt_map_read(arena, le32_to_cpu(log_new.lba), &map_entry,
579 NULL, NULL, 0);
580 if (ret)
581 return ret;
582
583 /*
584 * The map_entry from btt_read_map is stripped of any flag bits,
585 * so use the stripped out versions from the log as well for
586 * testing whether recovery is needed. For restoration, use the
587 * 'raw' version of the log entries as that captured what we
588 * were going to write originally.
589 */
590 if ((log_newmap != map_entry) && (log_oldmap == map_entry)) {
591 /*
592 * Last transaction wrote the flog, but wasn't able
593 * to complete the map write. So fix up the map.
594 */
595 ret = btt_map_write(arena, le32_to_cpu(log_new.lba),
596 le32_to_cpu(log_new.new_map), 0, 0, 0);
597 if (ret)
598 return ret;
599 }
600 }
601
602 return 0;
603}
604
605static bool ent_is_padding(struct log_entry *ent)
606{
607 return (ent->lba == 0) && (ent->old_map == 0) && (ent->new_map == 0)
608 && (ent->seq == 0);
609}
610
611/*
612 * Detecting valid log indices: We read a log group (see the comments in btt.h
613 * for a description of a 'log_group' and its 'slots'), and iterate over its
614 * four slots. We expect that a padding slot will be all-zeroes, and use this
615 * to detect a padding slot vs. an actual entry.
616 *
617 * If a log_group is in the initial state, i.e. hasn't been used since the
618 * creation of this BTT layout, it will have three of the four slots with
619 * zeroes. We skip over these log_groups for the detection of log_index. If
620 * all log_groups are in the initial state (i.e. the BTT has never been
621 * written to), it is safe to assume the 'new format' of log entries in slots
622 * (0, 1).
623 */
624static int log_set_indices(struct arena_info *arena)
625{
626 bool idx_set = false, initial_state = true;
627 int ret, log_index[2] = {-1, -1};
628 u32 i, j, next_idx = 0;
629 struct log_group log;
630 u32 pad_count = 0;
631
632 for (i = 0; i < arena->nfree; i++) {
633 ret = btt_log_group_read(arena, i, &log);
634 if (ret < 0)
635 return ret;
636
637 for (j = 0; j < 4; j++) {
638 if (!idx_set) {
639 if (ent_is_padding(&log.ent[j])) {
640 pad_count++;
641 continue;
642 } else {
643 /* Skip if index has been recorded */
644 if ((next_idx == 1) &&
645 (j == log_index[0]))
646 continue;
647 /* valid entry, record index */
648 log_index[next_idx] = j;
649 next_idx++;
650 }
651 if (next_idx == 2) {
652 /* two valid entries found */
653 idx_set = true;
654 } else if (next_idx > 2) {
655 /* too many valid indices */
656 return -ENXIO;
657 }
658 } else {
659 /*
660 * once the indices have been set, just verify
661 * that all subsequent log groups are either in
662 * their initial state or follow the same
663 * indices.
664 */
665 if (j == log_index[0]) {
666 /* entry must be 'valid' */
667 if (ent_is_padding(&log.ent[j]))
668 return -ENXIO;
669 } else if (j == log_index[1]) {
670 ;
671 /*
672 * log_index[1] can be padding if the
673 * lane never got used and it is still
674 * in the initial state (three 'padding'
675 * entries)
676 */
677 } else {
678 /* entry must be invalid (padding) */
679 if (!ent_is_padding(&log.ent[j]))
680 return -ENXIO;
681 }
682 }
683 }
684 /*
685 * If any of the log_groups have more than one valid,
686 * non-padding entry, then the we are no longer in the
687 * initial_state
688 */
689 if (pad_count < 3)
690 initial_state = false;
691 pad_count = 0;
692 }
693
694 if (!initial_state && !idx_set)
695 return -ENXIO;
696
697 /*
698 * If all the entries in the log were in the initial state,
699 * assume new padding scheme
700 */
701 if (initial_state)
702 log_index[1] = 1;
703
704 /*
705 * Only allow the known permutations of log/padding indices,
706 * i.e. (0, 1), and (0, 2)
707 */
708 if ((log_index[0] == 0) && ((log_index[1] == 1) || (log_index[1] == 2)))
709 ; /* known index possibilities */
710 else {
711 dev_err(to_dev(arena), "Found an unknown padding scheme\n");
712 return -ENXIO;
713 }
714
715 arena->log_index[0] = log_index[0];
716 arena->log_index[1] = log_index[1];
717 dev_dbg(to_dev(arena), "log_index_0 = %d\n", log_index[0]);
718 dev_dbg(to_dev(arena), "log_index_1 = %d\n", log_index[1]);
719 return 0;
720}
721
722static int btt_rtt_init(struct arena_info *arena)
723{
724 arena->rtt = kcalloc(arena->nfree, sizeof(u32), GFP_KERNEL);
725 if (arena->rtt == NULL)
726 return -ENOMEM;
727
728 return 0;
729}
730
731static int btt_maplocks_init(struct arena_info *arena)
732{
733 u32 i;
734
735 arena->map_locks = kcalloc(arena->nfree, sizeof(struct aligned_lock),
736 GFP_KERNEL);
737 if (!arena->map_locks)
738 return -ENOMEM;
739
740 for (i = 0; i < arena->nfree; i++)
741 spin_lock_init(&arena->map_locks[i].lock);
742
743 return 0;
744}
745
746static struct arena_info *alloc_arena(struct btt *btt, size_t size,
747 size_t start, size_t arena_off)
748{
749 struct arena_info *arena;
750 u64 logsize, mapsize, datasize;
751 u64 available = size;
752
753 arena = kzalloc(sizeof(struct arena_info), GFP_KERNEL);
754 if (!arena)
755 return NULL;
756 arena->nd_btt = btt->nd_btt;
757 arena->sector_size = btt->sector_size;
758 mutex_init(&arena->err_lock);
759
760 if (!size)
761 return arena;
762
763 arena->size = size;
764 arena->external_lba_start = start;
765 arena->external_lbasize = btt->lbasize;
766 arena->internal_lbasize = roundup(arena->external_lbasize,
767 INT_LBASIZE_ALIGNMENT);
768 arena->nfree = BTT_DEFAULT_NFREE;
769 arena->version_major = btt->nd_btt->version_major;
770 arena->version_minor = btt->nd_btt->version_minor;
771
772 if (available % BTT_PG_SIZE)
773 available -= (available % BTT_PG_SIZE);
774
775 /* Two pages are reserved for the super block and its copy */
776 available -= 2 * BTT_PG_SIZE;
777
778 /* The log takes a fixed amount of space based on nfree */
779 logsize = roundup(arena->nfree * LOG_GRP_SIZE, BTT_PG_SIZE);
780 available -= logsize;
781
782 /* Calculate optimal split between map and data area */
783 arena->internal_nlba = div_u64(available - BTT_PG_SIZE,
784 arena->internal_lbasize + MAP_ENT_SIZE);
785 arena->external_nlba = arena->internal_nlba - arena->nfree;
786
787 mapsize = roundup((arena->external_nlba * MAP_ENT_SIZE), BTT_PG_SIZE);
788 datasize = available - mapsize;
789
790 /* 'Absolute' values, relative to start of storage space */
791 arena->infooff = arena_off;
792 arena->dataoff = arena->infooff + BTT_PG_SIZE;
793 arena->mapoff = arena->dataoff + datasize;
794 arena->logoff = arena->mapoff + mapsize;
795 arena->info2off = arena->logoff + logsize;
796
797 /* Default log indices are (0,1) */
798 arena->log_index[0] = 0;
799 arena->log_index[1] = 1;
800 return arena;
801}
802
803static void free_arenas(struct btt *btt)
804{
805 struct arena_info *arena, *next;
806
807 list_for_each_entry_safe(arena, next, &btt->arena_list, list) {
808 list_del(&arena->list);
809 kfree(arena->rtt);
810 kfree(arena->map_locks);
811 kfree(arena->freelist);
812 debugfs_remove_recursive(arena->debugfs_dir);
813 kfree(arena);
814 }
815}
816
817/*
818 * This function reads an existing valid btt superblock and
819 * populates the corresponding arena_info struct
820 */
821static void parse_arena_meta(struct arena_info *arena, struct btt_sb *super,
822 u64 arena_off)
823{
824 arena->internal_nlba = le32_to_cpu(super->internal_nlba);
825 arena->internal_lbasize = le32_to_cpu(super->internal_lbasize);
826 arena->external_nlba = le32_to_cpu(super->external_nlba);
827 arena->external_lbasize = le32_to_cpu(super->external_lbasize);
828 arena->nfree = le32_to_cpu(super->nfree);
829 arena->version_major = le16_to_cpu(super->version_major);
830 arena->version_minor = le16_to_cpu(super->version_minor);
831
832 arena->nextoff = (super->nextoff == 0) ? 0 : (arena_off +
833 le64_to_cpu(super->nextoff));
834 arena->infooff = arena_off;
835 arena->dataoff = arena_off + le64_to_cpu(super->dataoff);
836 arena->mapoff = arena_off + le64_to_cpu(super->mapoff);
837 arena->logoff = arena_off + le64_to_cpu(super->logoff);
838 arena->info2off = arena_off + le64_to_cpu(super->info2off);
839
840 arena->size = (le64_to_cpu(super->nextoff) > 0)
841 ? (le64_to_cpu(super->nextoff))
842 : (arena->info2off - arena->infooff + BTT_PG_SIZE);
843
844 arena->flags = le32_to_cpu(super->flags);
845}
846
847static int discover_arenas(struct btt *btt)
848{
849 int ret = 0;
850 struct arena_info *arena;
851 size_t remaining = btt->rawsize;
852 u64 cur_nlba = 0;
853 size_t cur_off = 0;
854 int num_arenas = 0;
855
856 struct btt_sb *super __free(kfree) = kzalloc(sizeof(*super), GFP_KERNEL);
857 if (!super)
858 return -ENOMEM;
859
860 while (remaining) {
861 /* Alloc memory for arena */
862 arena = alloc_arena(btt, 0, 0, 0);
863 if (!arena)
864 return -ENOMEM;
865
866 arena->infooff = cur_off;
867 ret = btt_info_read(arena, super);
868 if (ret)
869 goto out;
870
871 if (!nd_btt_arena_is_valid(btt->nd_btt, super)) {
872 if (remaining == btt->rawsize) {
873 btt->init_state = INIT_NOTFOUND;
874 dev_info(to_dev(arena), "No existing arenas\n");
875 goto out;
876 } else {
877 dev_err(to_dev(arena),
878 "Found corrupted metadata!\n");
879 ret = -ENODEV;
880 goto out;
881 }
882 }
883
884 arena->external_lba_start = cur_nlba;
885 parse_arena_meta(arena, super, cur_off);
886
887 ret = log_set_indices(arena);
888 if (ret) {
889 dev_err(to_dev(arena),
890 "Unable to deduce log/padding indices\n");
891 goto out;
892 }
893
894 ret = btt_freelist_init(arena);
895 if (ret)
896 goto out;
897
898 ret = btt_rtt_init(arena);
899 if (ret)
900 goto out;
901
902 ret = btt_maplocks_init(arena);
903 if (ret)
904 goto out;
905
906 list_add_tail(&arena->list, &btt->arena_list);
907
908 remaining -= arena->size;
909 cur_off += arena->size;
910 cur_nlba += arena->external_nlba;
911 num_arenas++;
912
913 if (arena->nextoff == 0)
914 break;
915 }
916 btt->num_arenas = num_arenas;
917 btt->nlba = cur_nlba;
918 btt->init_state = INIT_READY;
919
920 return ret;
921
922 out:
923 kfree(arena);
924 free_arenas(btt);
925 return ret;
926}
927
928static int create_arenas(struct btt *btt)
929{
930 size_t remaining = btt->rawsize;
931 size_t cur_off = 0;
932
933 while (remaining) {
934 struct arena_info *arena;
935 size_t arena_size = min_t(u64, ARENA_MAX_SIZE, remaining);
936
937 remaining -= arena_size;
938 if (arena_size < ARENA_MIN_SIZE)
939 break;
940
941 arena = alloc_arena(btt, arena_size, btt->nlba, cur_off);
942 if (!arena) {
943 free_arenas(btt);
944 return -ENOMEM;
945 }
946 btt->nlba += arena->external_nlba;
947 if (remaining >= ARENA_MIN_SIZE)
948 arena->nextoff = arena->size;
949 else
950 arena->nextoff = 0;
951 cur_off += arena_size;
952 list_add_tail(&arena->list, &btt->arena_list);
953 }
954
955 return 0;
956}
957
958/*
959 * This function completes arena initialization by writing
960 * all the metadata.
961 * It is only called for an uninitialized arena when a write
962 * to that arena occurs for the first time.
963 */
964static int btt_arena_write_layout(struct arena_info *arena)
965{
966 int ret;
967 u64 sum;
968 struct btt_sb *super;
969 struct nd_btt *nd_btt = arena->nd_btt;
970 const uuid_t *parent_uuid = nd_dev_to_uuid(&nd_btt->ndns->dev);
971
972 ret = btt_map_init(arena);
973 if (ret)
974 return ret;
975
976 ret = btt_log_init(arena);
977 if (ret)
978 return ret;
979
980 super = kzalloc(sizeof(struct btt_sb), GFP_NOIO);
981 if (!super)
982 return -ENOMEM;
983
984 strscpy(super->signature, BTT_SIG, sizeof(super->signature));
985 export_uuid(super->uuid, nd_btt->uuid);
986 export_uuid(super->parent_uuid, parent_uuid);
987 super->flags = cpu_to_le32(arena->flags);
988 super->version_major = cpu_to_le16(arena->version_major);
989 super->version_minor = cpu_to_le16(arena->version_minor);
990 super->external_lbasize = cpu_to_le32(arena->external_lbasize);
991 super->external_nlba = cpu_to_le32(arena->external_nlba);
992 super->internal_lbasize = cpu_to_le32(arena->internal_lbasize);
993 super->internal_nlba = cpu_to_le32(arena->internal_nlba);
994 super->nfree = cpu_to_le32(arena->nfree);
995 super->infosize = cpu_to_le32(sizeof(struct btt_sb));
996 super->nextoff = cpu_to_le64(arena->nextoff);
997 /*
998 * Subtract arena->infooff (arena start) so numbers are relative
999 * to 'this' arena
1000 */
1001 super->dataoff = cpu_to_le64(arena->dataoff - arena->infooff);
1002 super->mapoff = cpu_to_le64(arena->mapoff - arena->infooff);
1003 super->logoff = cpu_to_le64(arena->logoff - arena->infooff);
1004 super->info2off = cpu_to_le64(arena->info2off - arena->infooff);
1005
1006 super->flags = 0;
1007 sum = nd_sb_checksum((struct nd_gen_sb *) super);
1008 super->checksum = cpu_to_le64(sum);
1009
1010 ret = btt_info_write(arena, super);
1011
1012 kfree(super);
1013 return ret;
1014}
1015
1016/*
1017 * This function completes the initialization for the BTT namespace
1018 * such that it is ready to accept IOs
1019 */
1020static int btt_meta_init(struct btt *btt)
1021{
1022 int ret = 0;
1023 struct arena_info *arena;
1024
1025 mutex_lock(&btt->init_lock);
1026 list_for_each_entry(arena, &btt->arena_list, list) {
1027 ret = btt_arena_write_layout(arena);
1028 if (ret)
1029 goto unlock;
1030
1031 ret = btt_freelist_init(arena);
1032 if (ret)
1033 goto unlock;
1034
1035 ret = btt_rtt_init(arena);
1036 if (ret)
1037 goto unlock;
1038
1039 ret = btt_maplocks_init(arena);
1040 if (ret)
1041 goto unlock;
1042 }
1043
1044 btt->init_state = INIT_READY;
1045
1046 unlock:
1047 mutex_unlock(&btt->init_lock);
1048 return ret;
1049}
1050
1051static u32 btt_meta_size(struct btt *btt)
1052{
1053 return btt->lbasize - btt->sector_size;
1054}
1055
1056/*
1057 * This function calculates the arena in which the given LBA lies
1058 * by doing a linear walk. This is acceptable since we expect only
1059 * a few arenas. If we have backing devices that get much larger,
1060 * we can construct a balanced binary tree of arenas at init time
1061 * so that this range search becomes faster.
1062 */
1063static int lba_to_arena(struct btt *btt, sector_t sector, __u32 *premap,
1064 struct arena_info **arena)
1065{
1066 struct arena_info *arena_list;
1067 __u64 lba = div_u64(sector << SECTOR_SHIFT, btt->sector_size);
1068
1069 list_for_each_entry(arena_list, &btt->arena_list, list) {
1070 if (lba < arena_list->external_nlba) {
1071 *arena = arena_list;
1072 *premap = lba;
1073 return 0;
1074 }
1075 lba -= arena_list->external_nlba;
1076 }
1077
1078 return -EIO;
1079}
1080
1081/*
1082 * The following (lock_map, unlock_map) are mostly just to improve
1083 * readability, since they index into an array of locks
1084 */
1085static void lock_map(struct arena_info *arena, u32 premap)
1086 __acquires(&arena->map_locks[idx].lock)
1087{
1088 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
1089
1090 spin_lock(&arena->map_locks[idx].lock);
1091}
1092
1093static void unlock_map(struct arena_info *arena, u32 premap)
1094 __releases(&arena->map_locks[idx].lock)
1095{
1096 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
1097
1098 spin_unlock(&arena->map_locks[idx].lock);
1099}
1100
1101static int btt_data_read(struct arena_info *arena, struct page *page,
1102 unsigned int off, u32 lba, u32 len)
1103{
1104 int ret;
1105 u64 nsoff = to_namespace_offset(arena, lba);
1106 void *mem = kmap_atomic(page);
1107
1108 ret = arena_read_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
1109 kunmap_atomic(mem);
1110
1111 return ret;
1112}
1113
1114static int btt_data_write(struct arena_info *arena, u32 lba,
1115 struct page *page, unsigned int off, u32 len)
1116{
1117 int ret;
1118 u64 nsoff = to_namespace_offset(arena, lba);
1119 void *mem = kmap_atomic(page);
1120
1121 ret = arena_write_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
1122 kunmap_atomic(mem);
1123
1124 return ret;
1125}
1126
1127static void zero_fill_data(struct page *page, unsigned int off, u32 len)
1128{
1129 void *mem = kmap_atomic(page);
1130
1131 memset(mem + off, 0, len);
1132 kunmap_atomic(mem);
1133}
1134
1135#ifdef CONFIG_BLK_DEV_INTEGRITY
1136static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1137 struct arena_info *arena, u32 postmap, int rw)
1138{
1139 unsigned int len = btt_meta_size(btt);
1140 u64 meta_nsoff;
1141 int ret = 0;
1142
1143 if (bip == NULL)
1144 return 0;
1145
1146 meta_nsoff = to_namespace_offset(arena, postmap) + btt->sector_size;
1147
1148 while (len) {
1149 unsigned int cur_len;
1150 struct bio_vec bv;
1151 void *mem;
1152
1153 bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1154 /*
1155 * The 'bv' obtained from bvec_iter_bvec has its .bv_len and
1156 * .bv_offset already adjusted for iter->bi_bvec_done, and we
1157 * can use those directly
1158 */
1159
1160 cur_len = min(len, bv.bv_len);
1161 mem = bvec_kmap_local(&bv);
1162 if (rw)
1163 ret = arena_write_bytes(arena, meta_nsoff, mem, cur_len,
1164 NVDIMM_IO_ATOMIC);
1165 else
1166 ret = arena_read_bytes(arena, meta_nsoff, mem, cur_len,
1167 NVDIMM_IO_ATOMIC);
1168
1169 kunmap_local(mem);
1170 if (ret)
1171 return ret;
1172
1173 len -= cur_len;
1174 meta_nsoff += cur_len;
1175 if (!bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len))
1176 return -EIO;
1177 }
1178
1179 return ret;
1180}
1181
1182#else /* CONFIG_BLK_DEV_INTEGRITY */
1183static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1184 struct arena_info *arena, u32 postmap, int rw)
1185{
1186 return 0;
1187}
1188#endif
1189
1190static int btt_read_pg(struct btt *btt, struct bio_integrity_payload *bip,
1191 struct page *page, unsigned int off, sector_t sector,
1192 unsigned int len)
1193{
1194 int ret = 0;
1195 int t_flag, e_flag;
1196 struct arena_info *arena = NULL;
1197 u32 lane = 0, premap, postmap;
1198
1199 while (len) {
1200 u32 cur_len;
1201
1202 lane = nd_region_acquire_lane(btt->nd_region);
1203
1204 ret = lba_to_arena(btt, sector, &premap, &arena);
1205 if (ret)
1206 goto out_lane;
1207
1208 cur_len = min(btt->sector_size, len);
1209
1210 ret = btt_map_read(arena, premap, &postmap, &t_flag, &e_flag,
1211 NVDIMM_IO_ATOMIC);
1212 if (ret)
1213 goto out_lane;
1214
1215 /*
1216 * We loop to make sure that the post map LBA didn't change
1217 * from under us between writing the RTT and doing the actual
1218 * read.
1219 */
1220 while (1) {
1221 u32 new_map;
1222 int new_t, new_e;
1223
1224 if (t_flag) {
1225 zero_fill_data(page, off, cur_len);
1226 goto out_lane;
1227 }
1228
1229 if (e_flag) {
1230 ret = -EIO;
1231 goto out_lane;
1232 }
1233
1234 arena->rtt[lane] = RTT_VALID | postmap;
1235 /*
1236 * Barrier to make sure this write is not reordered
1237 * to do the verification map_read before the RTT store
1238 */
1239 barrier();
1240
1241 ret = btt_map_read(arena, premap, &new_map, &new_t,
1242 &new_e, NVDIMM_IO_ATOMIC);
1243 if (ret)
1244 goto out_rtt;
1245
1246 if ((postmap == new_map) && (t_flag == new_t) &&
1247 (e_flag == new_e))
1248 break;
1249
1250 postmap = new_map;
1251 t_flag = new_t;
1252 e_flag = new_e;
1253 }
1254
1255 ret = btt_data_read(arena, page, off, postmap, cur_len);
1256 if (ret) {
1257 /* Media error - set the e_flag */
1258 if (btt_map_write(arena, premap, postmap, 0, 1, NVDIMM_IO_ATOMIC))
1259 dev_warn_ratelimited(to_dev(arena),
1260 "Error persistently tracking bad blocks at %#x\n",
1261 premap);
1262 goto out_rtt;
1263 }
1264
1265 if (bip) {
1266 ret = btt_rw_integrity(btt, bip, arena, postmap, READ);
1267 if (ret)
1268 goto out_rtt;
1269 }
1270
1271 arena->rtt[lane] = RTT_INVALID;
1272 nd_region_release_lane(btt->nd_region, lane);
1273
1274 len -= cur_len;
1275 off += cur_len;
1276 sector += btt->sector_size >> SECTOR_SHIFT;
1277 }
1278
1279 return 0;
1280
1281 out_rtt:
1282 arena->rtt[lane] = RTT_INVALID;
1283 out_lane:
1284 nd_region_release_lane(btt->nd_region, lane);
1285 return ret;
1286}
1287
1288/*
1289 * Normally, arena_{read,write}_bytes will take care of the initial offset
1290 * adjustment, but in the case of btt_is_badblock, where we query is_bad_pmem,
1291 * we need the final, raw namespace offset here
1292 */
1293static bool btt_is_badblock(struct btt *btt, struct arena_info *arena,
1294 u32 postmap)
1295{
1296 u64 nsoff = adjust_initial_offset(arena->nd_btt,
1297 to_namespace_offset(arena, postmap));
1298 sector_t phys_sector = nsoff >> 9;
1299
1300 return is_bad_pmem(btt->phys_bb, phys_sector, arena->internal_lbasize);
1301}
1302
1303static int btt_write_pg(struct btt *btt, struct bio_integrity_payload *bip,
1304 sector_t sector, struct page *page, unsigned int off,
1305 unsigned int len)
1306{
1307 int ret = 0;
1308 struct arena_info *arena = NULL;
1309 u32 premap = 0, old_postmap, new_postmap, lane = 0, i;
1310 struct log_entry log;
1311 int sub;
1312
1313 while (len) {
1314 u32 cur_len;
1315 int e_flag;
1316
1317 retry:
1318 lane = nd_region_acquire_lane(btt->nd_region);
1319
1320 ret = lba_to_arena(btt, sector, &premap, &arena);
1321 if (ret)
1322 goto out_lane;
1323 cur_len = min(btt->sector_size, len);
1324
1325 if ((arena->flags & IB_FLAG_ERROR_MASK) != 0) {
1326 ret = -EIO;
1327 goto out_lane;
1328 }
1329
1330 if (btt_is_badblock(btt, arena, arena->freelist[lane].block))
1331 arena->freelist[lane].has_err = 1;
1332
1333 if (mutex_is_locked(&arena->err_lock)
1334 || arena->freelist[lane].has_err) {
1335 nd_region_release_lane(btt->nd_region, lane);
1336
1337 ret = arena_clear_freelist_error(arena, lane);
1338 if (ret)
1339 return ret;
1340
1341 /* OK to acquire a different lane/free block */
1342 goto retry;
1343 }
1344
1345 new_postmap = arena->freelist[lane].block;
1346
1347 /* Wait if the new block is being read from */
1348 for (i = 0; i < arena->nfree; i++)
1349 while (arena->rtt[i] == (RTT_VALID | new_postmap))
1350 cpu_relax();
1351
1352
1353 if (new_postmap >= arena->internal_nlba) {
1354 ret = -EIO;
1355 goto out_lane;
1356 }
1357
1358 ret = btt_data_write(arena, new_postmap, page, off, cur_len);
1359 if (ret)
1360 goto out_lane;
1361
1362 if (bip) {
1363 ret = btt_rw_integrity(btt, bip, arena, new_postmap,
1364 WRITE);
1365 if (ret)
1366 goto out_lane;
1367 }
1368
1369 lock_map(arena, premap);
1370 ret = btt_map_read(arena, premap, &old_postmap, NULL, &e_flag,
1371 NVDIMM_IO_ATOMIC);
1372 if (ret)
1373 goto out_map;
1374 if (old_postmap >= arena->internal_nlba) {
1375 ret = -EIO;
1376 goto out_map;
1377 }
1378 if (e_flag)
1379 set_e_flag(old_postmap);
1380
1381 log.lba = cpu_to_le32(premap);
1382 log.old_map = cpu_to_le32(old_postmap);
1383 log.new_map = cpu_to_le32(new_postmap);
1384 log.seq = cpu_to_le32(arena->freelist[lane].seq);
1385 sub = arena->freelist[lane].sub;
1386 ret = btt_flog_write(arena, lane, sub, &log);
1387 if (ret)
1388 goto out_map;
1389
1390 ret = btt_map_write(arena, premap, new_postmap, 0, 0,
1391 NVDIMM_IO_ATOMIC);
1392 if (ret)
1393 goto out_map;
1394
1395 unlock_map(arena, premap);
1396 nd_region_release_lane(btt->nd_region, lane);
1397
1398 if (e_flag) {
1399 ret = arena_clear_freelist_error(arena, lane);
1400 if (ret)
1401 return ret;
1402 }
1403
1404 len -= cur_len;
1405 off += cur_len;
1406 sector += btt->sector_size >> SECTOR_SHIFT;
1407 }
1408
1409 return 0;
1410
1411 out_map:
1412 unlock_map(arena, premap);
1413 out_lane:
1414 nd_region_release_lane(btt->nd_region, lane);
1415 return ret;
1416}
1417
1418static int btt_do_bvec(struct btt *btt, struct bio_integrity_payload *bip,
1419 struct page *page, unsigned int len, unsigned int off,
1420 enum req_op op, sector_t sector)
1421{
1422 int ret;
1423
1424 if (!op_is_write(op)) {
1425 ret = btt_read_pg(btt, bip, page, off, sector, len);
1426 flush_dcache_page(page);
1427 } else {
1428 flush_dcache_page(page);
1429 ret = btt_write_pg(btt, bip, sector, page, off, len);
1430 }
1431
1432 return ret;
1433}
1434
1435static void btt_submit_bio(struct bio *bio)
1436{
1437 struct bio_integrity_payload *bip = bio_integrity(bio);
1438 struct btt *btt = bio->bi_bdev->bd_disk->private_data;
1439 struct bvec_iter iter;
1440 unsigned long start;
1441 struct bio_vec bvec;
1442 int err = 0;
1443 bool do_acct;
1444
1445 if (!bio_integrity_prep(bio))
1446 return;
1447
1448 do_acct = blk_queue_io_stat(bio->bi_bdev->bd_disk->queue);
1449 if (do_acct)
1450 start = bio_start_io_acct(bio);
1451 bio_for_each_segment(bvec, bio, iter) {
1452 unsigned int len = bvec.bv_len;
1453
1454 if (len > PAGE_SIZE || len < btt->sector_size ||
1455 len % btt->sector_size) {
1456 dev_err_ratelimited(&btt->nd_btt->dev,
1457 "unaligned bio segment (len: %d)\n", len);
1458 bio->bi_status = BLK_STS_IOERR;
1459 break;
1460 }
1461
1462 err = btt_do_bvec(btt, bip, bvec.bv_page, len, bvec.bv_offset,
1463 bio_op(bio), iter.bi_sector);
1464 if (err) {
1465 dev_err(&btt->nd_btt->dev,
1466 "io error in %s sector %lld, len %d,\n",
1467 (op_is_write(bio_op(bio))) ? "WRITE" :
1468 "READ",
1469 (unsigned long long) iter.bi_sector, len);
1470 bio->bi_status = errno_to_blk_status(err);
1471 break;
1472 }
1473 }
1474 if (do_acct)
1475 bio_end_io_acct(bio, start);
1476
1477 bio_endio(bio);
1478}
1479
1480static int btt_getgeo(struct block_device *bd, struct hd_geometry *geo)
1481{
1482 /* some standard values */
1483 geo->heads = 1 << 6;
1484 geo->sectors = 1 << 5;
1485 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
1486 return 0;
1487}
1488
1489static const struct block_device_operations btt_fops = {
1490 .owner = THIS_MODULE,
1491 .submit_bio = btt_submit_bio,
1492 .getgeo = btt_getgeo,
1493};
1494
1495static int btt_blk_init(struct btt *btt)
1496{
1497 struct nd_btt *nd_btt = btt->nd_btt;
1498 struct nd_namespace_common *ndns = nd_btt->ndns;
1499 int rc = -ENOMEM;
1500
1501 btt->btt_disk = blk_alloc_disk(NUMA_NO_NODE);
1502 if (!btt->btt_disk)
1503 return -ENOMEM;
1504
1505 nvdimm_namespace_disk_name(ndns, btt->btt_disk->disk_name);
1506 btt->btt_disk->first_minor = 0;
1507 btt->btt_disk->fops = &btt_fops;
1508 btt->btt_disk->private_data = btt;
1509
1510 blk_queue_logical_block_size(btt->btt_disk->queue, btt->sector_size);
1511 blk_queue_max_hw_sectors(btt->btt_disk->queue, UINT_MAX);
1512 blk_queue_flag_set(QUEUE_FLAG_NONROT, btt->btt_disk->queue);
1513 blk_queue_flag_set(QUEUE_FLAG_SYNCHRONOUS, btt->btt_disk->queue);
1514
1515 if (btt_meta_size(btt)) {
1516 rc = nd_integrity_init(btt->btt_disk, btt_meta_size(btt));
1517 if (rc)
1518 goto out_cleanup_disk;
1519 }
1520
1521 set_capacity(btt->btt_disk, btt->nlba * btt->sector_size >> 9);
1522 rc = device_add_disk(&btt->nd_btt->dev, btt->btt_disk, NULL);
1523 if (rc)
1524 goto out_cleanup_disk;
1525
1526 btt->nd_btt->size = btt->nlba * (u64)btt->sector_size;
1527 nvdimm_check_and_set_ro(btt->btt_disk);
1528
1529 return 0;
1530
1531out_cleanup_disk:
1532 put_disk(btt->btt_disk);
1533 return rc;
1534}
1535
1536static void btt_blk_cleanup(struct btt *btt)
1537{
1538 del_gendisk(btt->btt_disk);
1539 put_disk(btt->btt_disk);
1540}
1541
1542/**
1543 * btt_init - initialize a block translation table for the given device
1544 * @nd_btt: device with BTT geometry and backing device info
1545 * @rawsize: raw size in bytes of the backing device
1546 * @lbasize: lba size of the backing device
1547 * @uuid: A uuid for the backing device - this is stored on media
1548 * @nd_region: &struct nd_region for the REGION device
1549 *
1550 * Initialize a Block Translation Table on a backing device to provide
1551 * single sector power fail atomicity.
1552 *
1553 * Context:
1554 * Might sleep.
1555 *
1556 * Returns:
1557 * Pointer to a new struct btt on success, NULL on failure.
1558 */
1559static struct btt *btt_init(struct nd_btt *nd_btt, unsigned long long rawsize,
1560 u32 lbasize, uuid_t *uuid,
1561 struct nd_region *nd_region)
1562{
1563 int ret;
1564 struct btt *btt;
1565 struct nd_namespace_io *nsio;
1566 struct device *dev = &nd_btt->dev;
1567
1568 btt = devm_kzalloc(dev, sizeof(struct btt), GFP_KERNEL);
1569 if (!btt)
1570 return NULL;
1571
1572 btt->nd_btt = nd_btt;
1573 btt->rawsize = rawsize;
1574 btt->lbasize = lbasize;
1575 btt->sector_size = ((lbasize >= 4096) ? 4096 : 512);
1576 INIT_LIST_HEAD(&btt->arena_list);
1577 mutex_init(&btt->init_lock);
1578 btt->nd_region = nd_region;
1579 nsio = to_nd_namespace_io(&nd_btt->ndns->dev);
1580 btt->phys_bb = &nsio->bb;
1581
1582 ret = discover_arenas(btt);
1583 if (ret) {
1584 dev_err(dev, "init: error in arena_discover: %d\n", ret);
1585 return NULL;
1586 }
1587
1588 if (btt->init_state != INIT_READY && nd_region->ro) {
1589 dev_warn(dev, "%s is read-only, unable to init btt metadata\n",
1590 dev_name(&nd_region->dev));
1591 return NULL;
1592 } else if (btt->init_state != INIT_READY) {
1593 btt->num_arenas = (rawsize / ARENA_MAX_SIZE) +
1594 ((rawsize % ARENA_MAX_SIZE) ? 1 : 0);
1595 dev_dbg(dev, "init: %d arenas for %llu rawsize\n",
1596 btt->num_arenas, rawsize);
1597
1598 ret = create_arenas(btt);
1599 if (ret) {
1600 dev_info(dev, "init: create_arenas: %d\n", ret);
1601 return NULL;
1602 }
1603
1604 ret = btt_meta_init(btt);
1605 if (ret) {
1606 dev_err(dev, "init: error in meta_init: %d\n", ret);
1607 return NULL;
1608 }
1609 }
1610
1611 ret = btt_blk_init(btt);
1612 if (ret) {
1613 dev_err(dev, "init: error in blk_init: %d\n", ret);
1614 return NULL;
1615 }
1616
1617 btt_debugfs_init(btt);
1618
1619 return btt;
1620}
1621
1622/**
1623 * btt_fini - de-initialize a BTT
1624 * @btt: the BTT handle that was generated by btt_init
1625 *
1626 * De-initialize a Block Translation Table on device removal
1627 *
1628 * Context:
1629 * Might sleep.
1630 */
1631static void btt_fini(struct btt *btt)
1632{
1633 if (btt) {
1634 btt_blk_cleanup(btt);
1635 free_arenas(btt);
1636 debugfs_remove_recursive(btt->debugfs_dir);
1637 }
1638}
1639
1640int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns)
1641{
1642 struct nd_btt *nd_btt = to_nd_btt(ndns->claim);
1643 struct nd_region *nd_region;
1644 struct btt_sb *btt_sb;
1645 struct btt *btt;
1646 size_t size, rawsize;
1647 int rc;
1648
1649 if (!nd_btt->uuid || !nd_btt->ndns || !nd_btt->lbasize) {
1650 dev_dbg(&nd_btt->dev, "incomplete btt configuration\n");
1651 return -ENODEV;
1652 }
1653
1654 btt_sb = devm_kzalloc(&nd_btt->dev, sizeof(*btt_sb), GFP_KERNEL);
1655 if (!btt_sb)
1656 return -ENOMEM;
1657
1658 size = nvdimm_namespace_capacity(ndns);
1659 rc = devm_namespace_enable(&nd_btt->dev, ndns, size);
1660 if (rc)
1661 return rc;
1662
1663 /*
1664 * If this returns < 0, that is ok as it just means there wasn't
1665 * an existing BTT, and we're creating a new one. We still need to
1666 * call this as we need the version dependent fields in nd_btt to be
1667 * set correctly based on the holder class
1668 */
1669 nd_btt_version(nd_btt, ndns, btt_sb);
1670
1671 rawsize = size - nd_btt->initial_offset;
1672 if (rawsize < ARENA_MIN_SIZE) {
1673 dev_dbg(&nd_btt->dev, "%s must be at least %ld bytes\n",
1674 dev_name(&ndns->dev),
1675 ARENA_MIN_SIZE + nd_btt->initial_offset);
1676 return -ENXIO;
1677 }
1678 nd_region = to_nd_region(nd_btt->dev.parent);
1679 btt = btt_init(nd_btt, rawsize, nd_btt->lbasize, nd_btt->uuid,
1680 nd_region);
1681 if (!btt)
1682 return -ENOMEM;
1683 nd_btt->btt = btt;
1684
1685 return 0;
1686}
1687EXPORT_SYMBOL(nvdimm_namespace_attach_btt);
1688
1689int nvdimm_namespace_detach_btt(struct nd_btt *nd_btt)
1690{
1691 struct btt *btt = nd_btt->btt;
1692
1693 btt_fini(btt);
1694 nd_btt->btt = NULL;
1695
1696 return 0;
1697}
1698EXPORT_SYMBOL(nvdimm_namespace_detach_btt);
1699
1700static int __init nd_btt_init(void)
1701{
1702 int rc = 0;
1703
1704 debugfs_root = debugfs_create_dir("btt", NULL);
1705 if (IS_ERR_OR_NULL(debugfs_root))
1706 rc = -ENXIO;
1707
1708 return rc;
1709}
1710
1711static void __exit nd_btt_exit(void)
1712{
1713 debugfs_remove_recursive(debugfs_root);
1714}
1715
1716MODULE_ALIAS_ND_DEVICE(ND_DEVICE_BTT);
1717MODULE_AUTHOR("Vishal Verma <vishal.l.verma@linux.intel.com>");
1718MODULE_LICENSE("GPL v2");
1719module_init(nd_btt_init);
1720module_exit(nd_btt_exit);