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 <linux/backing-dev.h>
27#include "btt.h"
28#include "nd.h"
29
30enum log_ent_request {
31 LOG_NEW_ENT = 0,
32 LOG_OLD_ENT
33};
34
35static struct device *to_dev(struct arena_info *arena)
36{
37 return &arena->nd_btt->dev;
38}
39
40static u64 adjust_initial_offset(struct nd_btt *nd_btt, u64 offset)
41{
42 return offset + nd_btt->initial_offset;
43}
44
45static int arena_read_bytes(struct arena_info *arena, resource_size_t offset,
46 void *buf, size_t n, unsigned long flags)
47{
48 struct nd_btt *nd_btt = arena->nd_btt;
49 struct nd_namespace_common *ndns = nd_btt->ndns;
50
51 /* arena offsets may be shifted from the base of the device */
52 offset = adjust_initial_offset(nd_btt, offset);
53 return nvdimm_read_bytes(ndns, offset, buf, n, flags);
54}
55
56static int arena_write_bytes(struct arena_info *arena, resource_size_t offset,
57 void *buf, size_t n, unsigned long flags)
58{
59 struct nd_btt *nd_btt = arena->nd_btt;
60 struct nd_namespace_common *ndns = nd_btt->ndns;
61
62 /* arena offsets may be shifted from the base of the device */
63 offset = adjust_initial_offset(nd_btt, offset);
64 return nvdimm_write_bytes(ndns, offset, buf, n, flags);
65}
66
67static int btt_info_write(struct arena_info *arena, struct btt_sb *super)
68{
69 int ret;
70
71 /*
72 * infooff and info2off should always be at least 512B aligned.
73 * We rely on that to make sure rw_bytes does error clearing
74 * correctly, so make sure that is the case.
75 */
76 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->infooff, 512),
77 "arena->infooff: %#llx is unaligned\n", arena->infooff);
78 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->info2off, 512),
79 "arena->info2off: %#llx is unaligned\n", arena->info2off);
80
81 ret = arena_write_bytes(arena, arena->info2off, super,
82 sizeof(struct btt_sb), 0);
83 if (ret)
84 return ret;
85
86 return arena_write_bytes(arena, arena->infooff, super,
87 sizeof(struct btt_sb), 0);
88}
89
90static int btt_info_read(struct arena_info *arena, struct btt_sb *super)
91{
92 return arena_read_bytes(arena, arena->infooff, super,
93 sizeof(struct btt_sb), 0);
94}
95
96/*
97 * 'raw' version of btt_map write
98 * Assumptions:
99 * mapping is in little-endian
100 * mapping contains 'E' and 'Z' flags as desired
101 */
102static int __btt_map_write(struct arena_info *arena, u32 lba, __le32 mapping,
103 unsigned long flags)
104{
105 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
106
107 if (unlikely(lba >= arena->external_nlba))
108 dev_err_ratelimited(to_dev(arena),
109 "%s: lba %#x out of range (max: %#x)\n",
110 __func__, lba, arena->external_nlba);
111 return arena_write_bytes(arena, ns_off, &mapping, MAP_ENT_SIZE, flags);
112}
113
114static int btt_map_write(struct arena_info *arena, u32 lba, u32 mapping,
115 u32 z_flag, u32 e_flag, unsigned long rwb_flags)
116{
117 u32 ze;
118 __le32 mapping_le;
119
120 /*
121 * This 'mapping' is supposed to be just the LBA mapping, without
122 * any flags set, so strip the flag bits.
123 */
124 mapping = ent_lba(mapping);
125
126 ze = (z_flag << 1) + e_flag;
127 switch (ze) {
128 case 0:
129 /*
130 * We want to set neither of the Z or E flags, and
131 * in the actual layout, this means setting the bit
132 * positions of both to '1' to indicate a 'normal'
133 * map entry
134 */
135 mapping |= MAP_ENT_NORMAL;
136 break;
137 case 1:
138 mapping |= (1 << MAP_ERR_SHIFT);
139 break;
140 case 2:
141 mapping |= (1 << MAP_TRIM_SHIFT);
142 break;
143 default:
144 /*
145 * The case where Z and E are both sent in as '1' could be
146 * construed as a valid 'normal' case, but we decide not to,
147 * to avoid confusion
148 */
149 dev_err_ratelimited(to_dev(arena),
150 "Invalid use of Z and E flags\n");
151 return -EIO;
152 }
153
154 mapping_le = cpu_to_le32(mapping);
155 return __btt_map_write(arena, lba, mapping_le, rwb_flags);
156}
157
158static int btt_map_read(struct arena_info *arena, u32 lba, u32 *mapping,
159 int *trim, int *error, unsigned long rwb_flags)
160{
161 int ret;
162 __le32 in;
163 u32 raw_mapping, postmap, ze, z_flag, e_flag;
164 u64 ns_off = arena->mapoff + (lba * MAP_ENT_SIZE);
165
166 if (unlikely(lba >= arena->external_nlba))
167 dev_err_ratelimited(to_dev(arena),
168 "%s: lba %#x out of range (max: %#x)\n",
169 __func__, lba, arena->external_nlba);
170
171 ret = arena_read_bytes(arena, ns_off, &in, MAP_ENT_SIZE, rwb_flags);
172 if (ret)
173 return ret;
174
175 raw_mapping = le32_to_cpu(in);
176
177 z_flag = ent_z_flag(raw_mapping);
178 e_flag = ent_e_flag(raw_mapping);
179 ze = (z_flag << 1) + e_flag;
180 postmap = ent_lba(raw_mapping);
181
182 /* Reuse the {z,e}_flag variables for *trim and *error */
183 z_flag = 0;
184 e_flag = 0;
185
186 switch (ze) {
187 case 0:
188 /* Initial state. Return postmap = premap */
189 *mapping = lba;
190 break;
191 case 1:
192 *mapping = postmap;
193 e_flag = 1;
194 break;
195 case 2:
196 *mapping = postmap;
197 z_flag = 1;
198 break;
199 case 3:
200 *mapping = postmap;
201 break;
202 default:
203 return -EIO;
204 }
205
206 if (trim)
207 *trim = z_flag;
208 if (error)
209 *error = e_flag;
210
211 return ret;
212}
213
214static int btt_log_group_read(struct arena_info *arena, u32 lane,
215 struct log_group *log)
216{
217 return arena_read_bytes(arena,
218 arena->logoff + (lane * LOG_GRP_SIZE), log,
219 LOG_GRP_SIZE, 0);
220}
221
222static struct dentry *debugfs_root;
223
224static void arena_debugfs_init(struct arena_info *a, struct dentry *parent,
225 int idx)
226{
227 char dirname[32];
228 struct dentry *d;
229
230 /* If for some reason, parent bttN was not created, exit */
231 if (!parent)
232 return;
233
234 snprintf(dirname, 32, "arena%d", idx);
235 d = debugfs_create_dir(dirname, parent);
236 if (IS_ERR_OR_NULL(d))
237 return;
238 a->debugfs_dir = d;
239
240 debugfs_create_x64("size", S_IRUGO, d, &a->size);
241 debugfs_create_x64("external_lba_start", S_IRUGO, d,
242 &a->external_lba_start);
243 debugfs_create_x32("internal_nlba", S_IRUGO, d, &a->internal_nlba);
244 debugfs_create_u32("internal_lbasize", S_IRUGO, d,
245 &a->internal_lbasize);
246 debugfs_create_x32("external_nlba", S_IRUGO, d, &a->external_nlba);
247 debugfs_create_u32("external_lbasize", S_IRUGO, d,
248 &a->external_lbasize);
249 debugfs_create_u32("nfree", S_IRUGO, d, &a->nfree);
250 debugfs_create_u16("version_major", S_IRUGO, d, &a->version_major);
251 debugfs_create_u16("version_minor", S_IRUGO, d, &a->version_minor);
252 debugfs_create_x64("nextoff", S_IRUGO, d, &a->nextoff);
253 debugfs_create_x64("infooff", S_IRUGO, d, &a->infooff);
254 debugfs_create_x64("dataoff", S_IRUGO, d, &a->dataoff);
255 debugfs_create_x64("mapoff", S_IRUGO, d, &a->mapoff);
256 debugfs_create_x64("logoff", S_IRUGO, d, &a->logoff);
257 debugfs_create_x64("info2off", S_IRUGO, d, &a->info2off);
258 debugfs_create_x32("flags", S_IRUGO, d, &a->flags);
259 debugfs_create_u32("log_index_0", S_IRUGO, d, &a->log_index[0]);
260 debugfs_create_u32("log_index_1", S_IRUGO, d, &a->log_index[1]);
261}
262
263static void btt_debugfs_init(struct btt *btt)
264{
265 int i = 0;
266 struct arena_info *arena;
267
268 btt->debugfs_dir = debugfs_create_dir(dev_name(&btt->nd_btt->dev),
269 debugfs_root);
270 if (IS_ERR_OR_NULL(btt->debugfs_dir))
271 return;
272
273 list_for_each_entry(arena, &btt->arena_list, list) {
274 arena_debugfs_init(arena, btt->debugfs_dir, i);
275 i++;
276 }
277}
278
279static u32 log_seq(struct log_group *log, int log_idx)
280{
281 return le32_to_cpu(log->ent[log_idx].seq);
282}
283
284/*
285 * This function accepts two log entries, and uses the
286 * sequence number to find the 'older' entry.
287 * It also updates the sequence number in this old entry to
288 * make it the 'new' one if the mark_flag is set.
289 * Finally, it returns which of the entries was the older one.
290 *
291 * TODO The logic feels a bit kludge-y. make it better..
292 */
293static int btt_log_get_old(struct arena_info *a, struct log_group *log)
294{
295 int idx0 = a->log_index[0];
296 int idx1 = a->log_index[1];
297 int old;
298
299 /*
300 * the first ever time this is seen, the entry goes into [0]
301 * the next time, the following logic works out to put this
302 * (next) entry into [1]
303 */
304 if (log_seq(log, idx0) == 0) {
305 log->ent[idx0].seq = cpu_to_le32(1);
306 return 0;
307 }
308
309 if (log_seq(log, idx0) == log_seq(log, idx1))
310 return -EINVAL;
311 if (log_seq(log, idx0) + log_seq(log, idx1) > 5)
312 return -EINVAL;
313
314 if (log_seq(log, idx0) < log_seq(log, idx1)) {
315 if ((log_seq(log, idx1) - log_seq(log, idx0)) == 1)
316 old = 0;
317 else
318 old = 1;
319 } else {
320 if ((log_seq(log, idx0) - log_seq(log, idx1)) == 1)
321 old = 1;
322 else
323 old = 0;
324 }
325
326 return old;
327}
328
329/*
330 * This function copies the desired (old/new) log entry into ent if
331 * it is not NULL. It returns the sub-slot number (0 or 1)
332 * where the desired log entry was found. Negative return values
333 * indicate errors.
334 */
335static int btt_log_read(struct arena_info *arena, u32 lane,
336 struct log_entry *ent, int old_flag)
337{
338 int ret;
339 int old_ent, ret_ent;
340 struct log_group log;
341
342 ret = btt_log_group_read(arena, lane, &log);
343 if (ret)
344 return -EIO;
345
346 old_ent = btt_log_get_old(arena, &log);
347 if (old_ent < 0 || old_ent > 1) {
348 dev_err(to_dev(arena),
349 "log corruption (%d): lane %d seq [%d, %d]\n",
350 old_ent, lane, log.ent[arena->log_index[0]].seq,
351 log.ent[arena->log_index[1]].seq);
352 /* TODO set error state? */
353 return -EIO;
354 }
355
356 ret_ent = (old_flag ? old_ent : (1 - old_ent));
357
358 if (ent != NULL)
359 memcpy(ent, &log.ent[arena->log_index[ret_ent]], LOG_ENT_SIZE);
360
361 return ret_ent;
362}
363
364/*
365 * This function commits a log entry to media
366 * It does _not_ prepare the freelist entry for the next write
367 * btt_flog_write is the wrapper for updating the freelist elements
368 */
369static int __btt_log_write(struct arena_info *arena, u32 lane,
370 u32 sub, struct log_entry *ent, unsigned long flags)
371{
372 int ret;
373 u32 group_slot = arena->log_index[sub];
374 unsigned int log_half = LOG_ENT_SIZE / 2;
375 void *src = ent;
376 u64 ns_off;
377
378 ns_off = arena->logoff + (lane * LOG_GRP_SIZE) +
379 (group_slot * LOG_ENT_SIZE);
380 /* split the 16B write into atomic, durable halves */
381 ret = arena_write_bytes(arena, ns_off, src, log_half, flags);
382 if (ret)
383 return ret;
384
385 ns_off += log_half;
386 src += log_half;
387 return arena_write_bytes(arena, ns_off, src, log_half, flags);
388}
389
390static int btt_flog_write(struct arena_info *arena, u32 lane, u32 sub,
391 struct log_entry *ent)
392{
393 int ret;
394
395 ret = __btt_log_write(arena, lane, sub, ent, NVDIMM_IO_ATOMIC);
396 if (ret)
397 return ret;
398
399 /* prepare the next free entry */
400 arena->freelist[lane].sub = 1 - arena->freelist[lane].sub;
401 if (++(arena->freelist[lane].seq) == 4)
402 arena->freelist[lane].seq = 1;
403 if (ent_e_flag(ent->old_map))
404 arena->freelist[lane].has_err = 1;
405 arena->freelist[lane].block = le32_to_cpu(ent_lba(ent->old_map));
406
407 return ret;
408}
409
410/*
411 * This function initializes the BTT map to the initial state, which is
412 * all-zeroes, and indicates an identity mapping
413 */
414static int btt_map_init(struct arena_info *arena)
415{
416 int ret = -EINVAL;
417 void *zerobuf;
418 size_t offset = 0;
419 size_t chunk_size = SZ_2M;
420 size_t mapsize = arena->logoff - arena->mapoff;
421
422 zerobuf = kzalloc(chunk_size, GFP_KERNEL);
423 if (!zerobuf)
424 return -ENOMEM;
425
426 /*
427 * mapoff should always be at least 512B aligned. We rely on that to
428 * make sure rw_bytes does error clearing correctly, so make sure that
429 * is the case.
430 */
431 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->mapoff, 512),
432 "arena->mapoff: %#llx is unaligned\n", arena->mapoff);
433
434 while (mapsize) {
435 size_t size = min(mapsize, chunk_size);
436
437 dev_WARN_ONCE(to_dev(arena), size < 512,
438 "chunk size: %#zx is unaligned\n", size);
439 ret = arena_write_bytes(arena, arena->mapoff + offset, zerobuf,
440 size, 0);
441 if (ret)
442 goto free;
443
444 offset += size;
445 mapsize -= size;
446 cond_resched();
447 }
448
449 free:
450 kfree(zerobuf);
451 return ret;
452}
453
454/*
455 * This function initializes the BTT log with 'fake' entries pointing
456 * to the initial reserved set of blocks as being free
457 */
458static int btt_log_init(struct arena_info *arena)
459{
460 size_t logsize = arena->info2off - arena->logoff;
461 size_t chunk_size = SZ_4K, offset = 0;
462 struct log_entry ent;
463 void *zerobuf;
464 int ret;
465 u32 i;
466
467 zerobuf = kzalloc(chunk_size, GFP_KERNEL);
468 if (!zerobuf)
469 return -ENOMEM;
470 /*
471 * logoff should always be at least 512B aligned. We rely on that to
472 * make sure rw_bytes does error clearing correctly, so make sure that
473 * is the case.
474 */
475 dev_WARN_ONCE(to_dev(arena), !IS_ALIGNED(arena->logoff, 512),
476 "arena->logoff: %#llx is unaligned\n", arena->logoff);
477
478 while (logsize) {
479 size_t size = min(logsize, chunk_size);
480
481 dev_WARN_ONCE(to_dev(arena), size < 512,
482 "chunk size: %#zx is unaligned\n", size);
483 ret = arena_write_bytes(arena, arena->logoff + offset, zerobuf,
484 size, 0);
485 if (ret)
486 goto free;
487
488 offset += size;
489 logsize -= size;
490 cond_resched();
491 }
492
493 for (i = 0; i < arena->nfree; i++) {
494 ent.lba = cpu_to_le32(i);
495 ent.old_map = cpu_to_le32(arena->external_nlba + i);
496 ent.new_map = cpu_to_le32(arena->external_nlba + i);
497 ent.seq = cpu_to_le32(LOG_SEQ_INIT);
498 ret = __btt_log_write(arena, i, 0, &ent, 0);
499 if (ret)
500 goto free;
501 }
502
503 free:
504 kfree(zerobuf);
505 return ret;
506}
507
508static u64 to_namespace_offset(struct arena_info *arena, u64 lba)
509{
510 return arena->dataoff + ((u64)lba * arena->internal_lbasize);
511}
512
513static int arena_clear_freelist_error(struct arena_info *arena, u32 lane)
514{
515 int ret = 0;
516
517 if (arena->freelist[lane].has_err) {
518 void *zero_page = page_address(ZERO_PAGE(0));
519 u32 lba = arena->freelist[lane].block;
520 u64 nsoff = to_namespace_offset(arena, lba);
521 unsigned long len = arena->sector_size;
522
523 mutex_lock(&arena->err_lock);
524
525 while (len) {
526 unsigned long chunk = min(len, PAGE_SIZE);
527
528 ret = arena_write_bytes(arena, nsoff, zero_page,
529 chunk, 0);
530 if (ret)
531 break;
532 len -= chunk;
533 nsoff += chunk;
534 if (len == 0)
535 arena->freelist[lane].has_err = 0;
536 }
537 mutex_unlock(&arena->err_lock);
538 }
539 return ret;
540}
541
542static int btt_freelist_init(struct arena_info *arena)
543{
544 int new, ret;
545 struct log_entry log_new;
546 u32 i, map_entry, log_oldmap, log_newmap;
547
548 arena->freelist = kcalloc(arena->nfree, sizeof(struct free_entry),
549 GFP_KERNEL);
550 if (!arena->freelist)
551 return -ENOMEM;
552
553 for (i = 0; i < arena->nfree; i++) {
554 new = btt_log_read(arena, i, &log_new, LOG_NEW_ENT);
555 if (new < 0)
556 return new;
557
558 /* old and new map entries with any flags stripped out */
559 log_oldmap = ent_lba(le32_to_cpu(log_new.old_map));
560 log_newmap = ent_lba(le32_to_cpu(log_new.new_map));
561
562 /* sub points to the next one to be overwritten */
563 arena->freelist[i].sub = 1 - new;
564 arena->freelist[i].seq = nd_inc_seq(le32_to_cpu(log_new.seq));
565 arena->freelist[i].block = log_oldmap;
566
567 /*
568 * FIXME: if error clearing fails during init, we want to make
569 * the BTT read-only
570 */
571 if (ent_e_flag(log_new.old_map) &&
572 !ent_normal(log_new.old_map)) {
573 arena->freelist[i].has_err = 1;
574 ret = arena_clear_freelist_error(arena, i);
575 if (ret)
576 dev_err_ratelimited(to_dev(arena),
577 "Unable to clear known errors\n");
578 }
579
580 /* This implies a newly created or untouched flog entry */
581 if (log_oldmap == log_newmap)
582 continue;
583
584 /* Check if map recovery is needed */
585 ret = btt_map_read(arena, le32_to_cpu(log_new.lba), &map_entry,
586 NULL, NULL, 0);
587 if (ret)
588 return ret;
589
590 /*
591 * The map_entry from btt_read_map is stripped of any flag bits,
592 * so use the stripped out versions from the log as well for
593 * testing whether recovery is needed. For restoration, use the
594 * 'raw' version of the log entries as that captured what we
595 * were going to write originally.
596 */
597 if ((log_newmap != map_entry) && (log_oldmap == map_entry)) {
598 /*
599 * Last transaction wrote the flog, but wasn't able
600 * to complete the map write. So fix up the map.
601 */
602 ret = btt_map_write(arena, le32_to_cpu(log_new.lba),
603 le32_to_cpu(log_new.new_map), 0, 0, 0);
604 if (ret)
605 return ret;
606 }
607 }
608
609 return 0;
610}
611
612static bool ent_is_padding(struct log_entry *ent)
613{
614 return (ent->lba == 0) && (ent->old_map == 0) && (ent->new_map == 0)
615 && (ent->seq == 0);
616}
617
618/*
619 * Detecting valid log indices: We read a log group (see the comments in btt.h
620 * for a description of a 'log_group' and its 'slots'), and iterate over its
621 * four slots. We expect that a padding slot will be all-zeroes, and use this
622 * to detect a padding slot vs. an actual entry.
623 *
624 * If a log_group is in the initial state, i.e. hasn't been used since the
625 * creation of this BTT layout, it will have three of the four slots with
626 * zeroes. We skip over these log_groups for the detection of log_index. If
627 * all log_groups are in the initial state (i.e. the BTT has never been
628 * written to), it is safe to assume the 'new format' of log entries in slots
629 * (0, 1).
630 */
631static int log_set_indices(struct arena_info *arena)
632{
633 bool idx_set = false, initial_state = true;
634 int ret, log_index[2] = {-1, -1};
635 u32 i, j, next_idx = 0;
636 struct log_group log;
637 u32 pad_count = 0;
638
639 for (i = 0; i < arena->nfree; i++) {
640 ret = btt_log_group_read(arena, i, &log);
641 if (ret < 0)
642 return ret;
643
644 for (j = 0; j < 4; j++) {
645 if (!idx_set) {
646 if (ent_is_padding(&log.ent[j])) {
647 pad_count++;
648 continue;
649 } else {
650 /* Skip if index has been recorded */
651 if ((next_idx == 1) &&
652 (j == log_index[0]))
653 continue;
654 /* valid entry, record index */
655 log_index[next_idx] = j;
656 next_idx++;
657 }
658 if (next_idx == 2) {
659 /* two valid entries found */
660 idx_set = true;
661 } else if (next_idx > 2) {
662 /* too many valid indices */
663 return -ENXIO;
664 }
665 } else {
666 /*
667 * once the indices have been set, just verify
668 * that all subsequent log groups are either in
669 * their initial state or follow the same
670 * indices.
671 */
672 if (j == log_index[0]) {
673 /* entry must be 'valid' */
674 if (ent_is_padding(&log.ent[j]))
675 return -ENXIO;
676 } else if (j == log_index[1]) {
677 ;
678 /*
679 * log_index[1] can be padding if the
680 * lane never got used and it is still
681 * in the initial state (three 'padding'
682 * entries)
683 */
684 } else {
685 /* entry must be invalid (padding) */
686 if (!ent_is_padding(&log.ent[j]))
687 return -ENXIO;
688 }
689 }
690 }
691 /*
692 * If any of the log_groups have more than one valid,
693 * non-padding entry, then the we are no longer in the
694 * initial_state
695 */
696 if (pad_count < 3)
697 initial_state = false;
698 pad_count = 0;
699 }
700
701 if (!initial_state && !idx_set)
702 return -ENXIO;
703
704 /*
705 * If all the entries in the log were in the initial state,
706 * assume new padding scheme
707 */
708 if (initial_state)
709 log_index[1] = 1;
710
711 /*
712 * Only allow the known permutations of log/padding indices,
713 * i.e. (0, 1), and (0, 2)
714 */
715 if ((log_index[0] == 0) && ((log_index[1] == 1) || (log_index[1] == 2)))
716 ; /* known index possibilities */
717 else {
718 dev_err(to_dev(arena), "Found an unknown padding scheme\n");
719 return -ENXIO;
720 }
721
722 arena->log_index[0] = log_index[0];
723 arena->log_index[1] = log_index[1];
724 dev_dbg(to_dev(arena), "log_index_0 = %d\n", log_index[0]);
725 dev_dbg(to_dev(arena), "log_index_1 = %d\n", log_index[1]);
726 return 0;
727}
728
729static int btt_rtt_init(struct arena_info *arena)
730{
731 arena->rtt = kcalloc(arena->nfree, sizeof(u32), GFP_KERNEL);
732 if (arena->rtt == NULL)
733 return -ENOMEM;
734
735 return 0;
736}
737
738static int btt_maplocks_init(struct arena_info *arena)
739{
740 u32 i;
741
742 arena->map_locks = kcalloc(arena->nfree, sizeof(struct aligned_lock),
743 GFP_KERNEL);
744 if (!arena->map_locks)
745 return -ENOMEM;
746
747 for (i = 0; i < arena->nfree; i++)
748 spin_lock_init(&arena->map_locks[i].lock);
749
750 return 0;
751}
752
753static struct arena_info *alloc_arena(struct btt *btt, size_t size,
754 size_t start, size_t arena_off)
755{
756 struct arena_info *arena;
757 u64 logsize, mapsize, datasize;
758 u64 available = size;
759
760 arena = kzalloc(sizeof(struct arena_info), GFP_KERNEL);
761 if (!arena)
762 return NULL;
763 arena->nd_btt = btt->nd_btt;
764 arena->sector_size = btt->sector_size;
765 mutex_init(&arena->err_lock);
766
767 if (!size)
768 return arena;
769
770 arena->size = size;
771 arena->external_lba_start = start;
772 arena->external_lbasize = btt->lbasize;
773 arena->internal_lbasize = roundup(arena->external_lbasize,
774 INT_LBASIZE_ALIGNMENT);
775 arena->nfree = BTT_DEFAULT_NFREE;
776 arena->version_major = btt->nd_btt->version_major;
777 arena->version_minor = btt->nd_btt->version_minor;
778
779 if (available % BTT_PG_SIZE)
780 available -= (available % BTT_PG_SIZE);
781
782 /* Two pages are reserved for the super block and its copy */
783 available -= 2 * BTT_PG_SIZE;
784
785 /* The log takes a fixed amount of space based on nfree */
786 logsize = roundup(arena->nfree * LOG_GRP_SIZE, BTT_PG_SIZE);
787 available -= logsize;
788
789 /* Calculate optimal split between map and data area */
790 arena->internal_nlba = div_u64(available - BTT_PG_SIZE,
791 arena->internal_lbasize + MAP_ENT_SIZE);
792 arena->external_nlba = arena->internal_nlba - arena->nfree;
793
794 mapsize = roundup((arena->external_nlba * MAP_ENT_SIZE), BTT_PG_SIZE);
795 datasize = available - mapsize;
796
797 /* 'Absolute' values, relative to start of storage space */
798 arena->infooff = arena_off;
799 arena->dataoff = arena->infooff + BTT_PG_SIZE;
800 arena->mapoff = arena->dataoff + datasize;
801 arena->logoff = arena->mapoff + mapsize;
802 arena->info2off = arena->logoff + logsize;
803
804 /* Default log indices are (0,1) */
805 arena->log_index[0] = 0;
806 arena->log_index[1] = 1;
807 return arena;
808}
809
810static void free_arenas(struct btt *btt)
811{
812 struct arena_info *arena, *next;
813
814 list_for_each_entry_safe(arena, next, &btt->arena_list, list) {
815 list_del(&arena->list);
816 kfree(arena->rtt);
817 kfree(arena->map_locks);
818 kfree(arena->freelist);
819 debugfs_remove_recursive(arena->debugfs_dir);
820 kfree(arena);
821 }
822}
823
824/*
825 * This function reads an existing valid btt superblock and
826 * populates the corresponding arena_info struct
827 */
828static void parse_arena_meta(struct arena_info *arena, struct btt_sb *super,
829 u64 arena_off)
830{
831 arena->internal_nlba = le32_to_cpu(super->internal_nlba);
832 arena->internal_lbasize = le32_to_cpu(super->internal_lbasize);
833 arena->external_nlba = le32_to_cpu(super->external_nlba);
834 arena->external_lbasize = le32_to_cpu(super->external_lbasize);
835 arena->nfree = le32_to_cpu(super->nfree);
836 arena->version_major = le16_to_cpu(super->version_major);
837 arena->version_minor = le16_to_cpu(super->version_minor);
838
839 arena->nextoff = (super->nextoff == 0) ? 0 : (arena_off +
840 le64_to_cpu(super->nextoff));
841 arena->infooff = arena_off;
842 arena->dataoff = arena_off + le64_to_cpu(super->dataoff);
843 arena->mapoff = arena_off + le64_to_cpu(super->mapoff);
844 arena->logoff = arena_off + le64_to_cpu(super->logoff);
845 arena->info2off = arena_off + le64_to_cpu(super->info2off);
846
847 arena->size = (le64_to_cpu(super->nextoff) > 0)
848 ? (le64_to_cpu(super->nextoff))
849 : (arena->info2off - arena->infooff + BTT_PG_SIZE);
850
851 arena->flags = le32_to_cpu(super->flags);
852}
853
854static int discover_arenas(struct btt *btt)
855{
856 int ret = 0;
857 struct arena_info *arena;
858 struct btt_sb *super;
859 size_t remaining = btt->rawsize;
860 u64 cur_nlba = 0;
861 size_t cur_off = 0;
862 int num_arenas = 0;
863
864 super = kzalloc(sizeof(*super), GFP_KERNEL);
865 if (!super)
866 return -ENOMEM;
867
868 while (remaining) {
869 /* Alloc memory for arena */
870 arena = alloc_arena(btt, 0, 0, 0);
871 if (!arena) {
872 ret = -ENOMEM;
873 goto out_super;
874 }
875
876 arena->infooff = cur_off;
877 ret = btt_info_read(arena, super);
878 if (ret)
879 goto out;
880
881 if (!nd_btt_arena_is_valid(btt->nd_btt, super)) {
882 if (remaining == btt->rawsize) {
883 btt->init_state = INIT_NOTFOUND;
884 dev_info(to_dev(arena), "No existing arenas\n");
885 goto out;
886 } else {
887 dev_err(to_dev(arena),
888 "Found corrupted metadata!\n");
889 ret = -ENODEV;
890 goto out;
891 }
892 }
893
894 arena->external_lba_start = cur_nlba;
895 parse_arena_meta(arena, super, cur_off);
896
897 ret = log_set_indices(arena);
898 if (ret) {
899 dev_err(to_dev(arena),
900 "Unable to deduce log/padding indices\n");
901 goto out;
902 }
903
904 ret = btt_freelist_init(arena);
905 if (ret)
906 goto out;
907
908 ret = btt_rtt_init(arena);
909 if (ret)
910 goto out;
911
912 ret = btt_maplocks_init(arena);
913 if (ret)
914 goto out;
915
916 list_add_tail(&arena->list, &btt->arena_list);
917
918 remaining -= arena->size;
919 cur_off += arena->size;
920 cur_nlba += arena->external_nlba;
921 num_arenas++;
922
923 if (arena->nextoff == 0)
924 break;
925 }
926 btt->num_arenas = num_arenas;
927 btt->nlba = cur_nlba;
928 btt->init_state = INIT_READY;
929
930 kfree(super);
931 return ret;
932
933 out:
934 kfree(arena);
935 free_arenas(btt);
936 out_super:
937 kfree(super);
938 return ret;
939}
940
941static int create_arenas(struct btt *btt)
942{
943 size_t remaining = btt->rawsize;
944 size_t cur_off = 0;
945
946 while (remaining) {
947 struct arena_info *arena;
948 size_t arena_size = min_t(u64, ARENA_MAX_SIZE, remaining);
949
950 remaining -= arena_size;
951 if (arena_size < ARENA_MIN_SIZE)
952 break;
953
954 arena = alloc_arena(btt, arena_size, btt->nlba, cur_off);
955 if (!arena) {
956 free_arenas(btt);
957 return -ENOMEM;
958 }
959 btt->nlba += arena->external_nlba;
960 if (remaining >= ARENA_MIN_SIZE)
961 arena->nextoff = arena->size;
962 else
963 arena->nextoff = 0;
964 cur_off += arena_size;
965 list_add_tail(&arena->list, &btt->arena_list);
966 }
967
968 return 0;
969}
970
971/*
972 * This function completes arena initialization by writing
973 * all the metadata.
974 * It is only called for an uninitialized arena when a write
975 * to that arena occurs for the first time.
976 */
977static int btt_arena_write_layout(struct arena_info *arena)
978{
979 int ret;
980 u64 sum;
981 struct btt_sb *super;
982 struct nd_btt *nd_btt = arena->nd_btt;
983 const u8 *parent_uuid = nd_dev_to_uuid(&nd_btt->ndns->dev);
984
985 ret = btt_map_init(arena);
986 if (ret)
987 return ret;
988
989 ret = btt_log_init(arena);
990 if (ret)
991 return ret;
992
993 super = kzalloc(sizeof(struct btt_sb), GFP_NOIO);
994 if (!super)
995 return -ENOMEM;
996
997 strncpy(super->signature, BTT_SIG, BTT_SIG_LEN);
998 memcpy(super->uuid, nd_btt->uuid, 16);
999 memcpy(super->parent_uuid, parent_uuid, 16);
1000 super->flags = cpu_to_le32(arena->flags);
1001 super->version_major = cpu_to_le16(arena->version_major);
1002 super->version_minor = cpu_to_le16(arena->version_minor);
1003 super->external_lbasize = cpu_to_le32(arena->external_lbasize);
1004 super->external_nlba = cpu_to_le32(arena->external_nlba);
1005 super->internal_lbasize = cpu_to_le32(arena->internal_lbasize);
1006 super->internal_nlba = cpu_to_le32(arena->internal_nlba);
1007 super->nfree = cpu_to_le32(arena->nfree);
1008 super->infosize = cpu_to_le32(sizeof(struct btt_sb));
1009 super->nextoff = cpu_to_le64(arena->nextoff);
1010 /*
1011 * Subtract arena->infooff (arena start) so numbers are relative
1012 * to 'this' arena
1013 */
1014 super->dataoff = cpu_to_le64(arena->dataoff - arena->infooff);
1015 super->mapoff = cpu_to_le64(arena->mapoff - arena->infooff);
1016 super->logoff = cpu_to_le64(arena->logoff - arena->infooff);
1017 super->info2off = cpu_to_le64(arena->info2off - arena->infooff);
1018
1019 super->flags = 0;
1020 sum = nd_sb_checksum((struct nd_gen_sb *) super);
1021 super->checksum = cpu_to_le64(sum);
1022
1023 ret = btt_info_write(arena, super);
1024
1025 kfree(super);
1026 return ret;
1027}
1028
1029/*
1030 * This function completes the initialization for the BTT namespace
1031 * such that it is ready to accept IOs
1032 */
1033static int btt_meta_init(struct btt *btt)
1034{
1035 int ret = 0;
1036 struct arena_info *arena;
1037
1038 mutex_lock(&btt->init_lock);
1039 list_for_each_entry(arena, &btt->arena_list, list) {
1040 ret = btt_arena_write_layout(arena);
1041 if (ret)
1042 goto unlock;
1043
1044 ret = btt_freelist_init(arena);
1045 if (ret)
1046 goto unlock;
1047
1048 ret = btt_rtt_init(arena);
1049 if (ret)
1050 goto unlock;
1051
1052 ret = btt_maplocks_init(arena);
1053 if (ret)
1054 goto unlock;
1055 }
1056
1057 btt->init_state = INIT_READY;
1058
1059 unlock:
1060 mutex_unlock(&btt->init_lock);
1061 return ret;
1062}
1063
1064static u32 btt_meta_size(struct btt *btt)
1065{
1066 return btt->lbasize - btt->sector_size;
1067}
1068
1069/*
1070 * This function calculates the arena in which the given LBA lies
1071 * by doing a linear walk. This is acceptable since we expect only
1072 * a few arenas. If we have backing devices that get much larger,
1073 * we can construct a balanced binary tree of arenas at init time
1074 * so that this range search becomes faster.
1075 */
1076static int lba_to_arena(struct btt *btt, sector_t sector, __u32 *premap,
1077 struct arena_info **arena)
1078{
1079 struct arena_info *arena_list;
1080 __u64 lba = div_u64(sector << SECTOR_SHIFT, btt->sector_size);
1081
1082 list_for_each_entry(arena_list, &btt->arena_list, list) {
1083 if (lba < arena_list->external_nlba) {
1084 *arena = arena_list;
1085 *premap = lba;
1086 return 0;
1087 }
1088 lba -= arena_list->external_nlba;
1089 }
1090
1091 return -EIO;
1092}
1093
1094/*
1095 * The following (lock_map, unlock_map) are mostly just to improve
1096 * readability, since they index into an array of locks
1097 */
1098static void lock_map(struct arena_info *arena, u32 premap)
1099 __acquires(&arena->map_locks[idx].lock)
1100{
1101 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
1102
1103 spin_lock(&arena->map_locks[idx].lock);
1104}
1105
1106static void unlock_map(struct arena_info *arena, u32 premap)
1107 __releases(&arena->map_locks[idx].lock)
1108{
1109 u32 idx = (premap * MAP_ENT_SIZE / L1_CACHE_BYTES) % arena->nfree;
1110
1111 spin_unlock(&arena->map_locks[idx].lock);
1112}
1113
1114static int btt_data_read(struct arena_info *arena, struct page *page,
1115 unsigned int off, u32 lba, u32 len)
1116{
1117 int ret;
1118 u64 nsoff = to_namespace_offset(arena, lba);
1119 void *mem = kmap_atomic(page);
1120
1121 ret = arena_read_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
1122 kunmap_atomic(mem);
1123
1124 return ret;
1125}
1126
1127static int btt_data_write(struct arena_info *arena, u32 lba,
1128 struct page *page, unsigned int off, u32 len)
1129{
1130 int ret;
1131 u64 nsoff = to_namespace_offset(arena, lba);
1132 void *mem = kmap_atomic(page);
1133
1134 ret = arena_write_bytes(arena, nsoff, mem + off, len, NVDIMM_IO_ATOMIC);
1135 kunmap_atomic(mem);
1136
1137 return ret;
1138}
1139
1140static void zero_fill_data(struct page *page, unsigned int off, u32 len)
1141{
1142 void *mem = kmap_atomic(page);
1143
1144 memset(mem + off, 0, len);
1145 kunmap_atomic(mem);
1146}
1147
1148#ifdef CONFIG_BLK_DEV_INTEGRITY
1149static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1150 struct arena_info *arena, u32 postmap, int rw)
1151{
1152 unsigned int len = btt_meta_size(btt);
1153 u64 meta_nsoff;
1154 int ret = 0;
1155
1156 if (bip == NULL)
1157 return 0;
1158
1159 meta_nsoff = to_namespace_offset(arena, postmap) + btt->sector_size;
1160
1161 while (len) {
1162 unsigned int cur_len;
1163 struct bio_vec bv;
1164 void *mem;
1165
1166 bv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1167 /*
1168 * The 'bv' obtained from bvec_iter_bvec has its .bv_len and
1169 * .bv_offset already adjusted for iter->bi_bvec_done, and we
1170 * can use those directly
1171 */
1172
1173 cur_len = min(len, bv.bv_len);
1174 mem = kmap_atomic(bv.bv_page);
1175 if (rw)
1176 ret = arena_write_bytes(arena, meta_nsoff,
1177 mem + bv.bv_offset, cur_len,
1178 NVDIMM_IO_ATOMIC);
1179 else
1180 ret = arena_read_bytes(arena, meta_nsoff,
1181 mem + bv.bv_offset, cur_len,
1182 NVDIMM_IO_ATOMIC);
1183
1184 kunmap_atomic(mem);
1185 if (ret)
1186 return ret;
1187
1188 len -= cur_len;
1189 meta_nsoff += cur_len;
1190 if (!bvec_iter_advance(bip->bip_vec, &bip->bip_iter, cur_len))
1191 return -EIO;
1192 }
1193
1194 return ret;
1195}
1196
1197#else /* CONFIG_BLK_DEV_INTEGRITY */
1198static int btt_rw_integrity(struct btt *btt, struct bio_integrity_payload *bip,
1199 struct arena_info *arena, u32 postmap, int rw)
1200{
1201 return 0;
1202}
1203#endif
1204
1205static int btt_read_pg(struct btt *btt, struct bio_integrity_payload *bip,
1206 struct page *page, unsigned int off, sector_t sector,
1207 unsigned int len)
1208{
1209 int ret = 0;
1210 int t_flag, e_flag;
1211 struct arena_info *arena = NULL;
1212 u32 lane = 0, premap, postmap;
1213
1214 while (len) {
1215 u32 cur_len;
1216
1217 lane = nd_region_acquire_lane(btt->nd_region);
1218
1219 ret = lba_to_arena(btt, sector, &premap, &arena);
1220 if (ret)
1221 goto out_lane;
1222
1223 cur_len = min(btt->sector_size, len);
1224
1225 ret = btt_map_read(arena, premap, &postmap, &t_flag, &e_flag,
1226 NVDIMM_IO_ATOMIC);
1227 if (ret)
1228 goto out_lane;
1229
1230 /*
1231 * We loop to make sure that the post map LBA didn't change
1232 * from under us between writing the RTT and doing the actual
1233 * read.
1234 */
1235 while (1) {
1236 u32 new_map;
1237 int new_t, new_e;
1238
1239 if (t_flag) {
1240 zero_fill_data(page, off, cur_len);
1241 goto out_lane;
1242 }
1243
1244 if (e_flag) {
1245 ret = -EIO;
1246 goto out_lane;
1247 }
1248
1249 arena->rtt[lane] = RTT_VALID | postmap;
1250 /*
1251 * Barrier to make sure this write is not reordered
1252 * to do the verification map_read before the RTT store
1253 */
1254 barrier();
1255
1256 ret = btt_map_read(arena, premap, &new_map, &new_t,
1257 &new_e, NVDIMM_IO_ATOMIC);
1258 if (ret)
1259 goto out_rtt;
1260
1261 if ((postmap == new_map) && (t_flag == new_t) &&
1262 (e_flag == new_e))
1263 break;
1264
1265 postmap = new_map;
1266 t_flag = new_t;
1267 e_flag = new_e;
1268 }
1269
1270 ret = btt_data_read(arena, page, off, postmap, cur_len);
1271 if (ret) {
1272 int rc;
1273
1274 /* Media error - set the e_flag */
1275 rc = btt_map_write(arena, premap, postmap, 0, 1,
1276 NVDIMM_IO_ATOMIC);
1277 goto out_rtt;
1278 }
1279
1280 if (bip) {
1281 ret = btt_rw_integrity(btt, bip, arena, postmap, READ);
1282 if (ret)
1283 goto out_rtt;
1284 }
1285
1286 arena->rtt[lane] = RTT_INVALID;
1287 nd_region_release_lane(btt->nd_region, lane);
1288
1289 len -= cur_len;
1290 off += cur_len;
1291 sector += btt->sector_size >> SECTOR_SHIFT;
1292 }
1293
1294 return 0;
1295
1296 out_rtt:
1297 arena->rtt[lane] = RTT_INVALID;
1298 out_lane:
1299 nd_region_release_lane(btt->nd_region, lane);
1300 return ret;
1301}
1302
1303/*
1304 * Normally, arena_{read,write}_bytes will take care of the initial offset
1305 * adjustment, but in the case of btt_is_badblock, where we query is_bad_pmem,
1306 * we need the final, raw namespace offset here
1307 */
1308static bool btt_is_badblock(struct btt *btt, struct arena_info *arena,
1309 u32 postmap)
1310{
1311 u64 nsoff = adjust_initial_offset(arena->nd_btt,
1312 to_namespace_offset(arena, postmap));
1313 sector_t phys_sector = nsoff >> 9;
1314
1315 return is_bad_pmem(btt->phys_bb, phys_sector, arena->internal_lbasize);
1316}
1317
1318static int btt_write_pg(struct btt *btt, struct bio_integrity_payload *bip,
1319 sector_t sector, struct page *page, unsigned int off,
1320 unsigned int len)
1321{
1322 int ret = 0;
1323 struct arena_info *arena = NULL;
1324 u32 premap = 0, old_postmap, new_postmap, lane = 0, i;
1325 struct log_entry log;
1326 int sub;
1327
1328 while (len) {
1329 u32 cur_len;
1330 int e_flag;
1331
1332 retry:
1333 lane = nd_region_acquire_lane(btt->nd_region);
1334
1335 ret = lba_to_arena(btt, sector, &premap, &arena);
1336 if (ret)
1337 goto out_lane;
1338 cur_len = min(btt->sector_size, len);
1339
1340 if ((arena->flags & IB_FLAG_ERROR_MASK) != 0) {
1341 ret = -EIO;
1342 goto out_lane;
1343 }
1344
1345 if (btt_is_badblock(btt, arena, arena->freelist[lane].block))
1346 arena->freelist[lane].has_err = 1;
1347
1348 if (mutex_is_locked(&arena->err_lock)
1349 || arena->freelist[lane].has_err) {
1350 nd_region_release_lane(btt->nd_region, lane);
1351
1352 ret = arena_clear_freelist_error(arena, lane);
1353 if (ret)
1354 return ret;
1355
1356 /* OK to acquire a different lane/free block */
1357 goto retry;
1358 }
1359
1360 new_postmap = arena->freelist[lane].block;
1361
1362 /* Wait if the new block is being read from */
1363 for (i = 0; i < arena->nfree; i++)
1364 while (arena->rtt[i] == (RTT_VALID | new_postmap))
1365 cpu_relax();
1366
1367
1368 if (new_postmap >= arena->internal_nlba) {
1369 ret = -EIO;
1370 goto out_lane;
1371 }
1372
1373 ret = btt_data_write(arena, new_postmap, page, off, cur_len);
1374 if (ret)
1375 goto out_lane;
1376
1377 if (bip) {
1378 ret = btt_rw_integrity(btt, bip, arena, new_postmap,
1379 WRITE);
1380 if (ret)
1381 goto out_lane;
1382 }
1383
1384 lock_map(arena, premap);
1385 ret = btt_map_read(arena, premap, &old_postmap, NULL, &e_flag,
1386 NVDIMM_IO_ATOMIC);
1387 if (ret)
1388 goto out_map;
1389 if (old_postmap >= arena->internal_nlba) {
1390 ret = -EIO;
1391 goto out_map;
1392 }
1393 if (e_flag)
1394 set_e_flag(old_postmap);
1395
1396 log.lba = cpu_to_le32(premap);
1397 log.old_map = cpu_to_le32(old_postmap);
1398 log.new_map = cpu_to_le32(new_postmap);
1399 log.seq = cpu_to_le32(arena->freelist[lane].seq);
1400 sub = arena->freelist[lane].sub;
1401 ret = btt_flog_write(arena, lane, sub, &log);
1402 if (ret)
1403 goto out_map;
1404
1405 ret = btt_map_write(arena, premap, new_postmap, 0, 0,
1406 NVDIMM_IO_ATOMIC);
1407 if (ret)
1408 goto out_map;
1409
1410 unlock_map(arena, premap);
1411 nd_region_release_lane(btt->nd_region, lane);
1412
1413 if (e_flag) {
1414 ret = arena_clear_freelist_error(arena, lane);
1415 if (ret)
1416 return ret;
1417 }
1418
1419 len -= cur_len;
1420 off += cur_len;
1421 sector += btt->sector_size >> SECTOR_SHIFT;
1422 }
1423
1424 return 0;
1425
1426 out_map:
1427 unlock_map(arena, premap);
1428 out_lane:
1429 nd_region_release_lane(btt->nd_region, lane);
1430 return ret;
1431}
1432
1433static int btt_do_bvec(struct btt *btt, struct bio_integrity_payload *bip,
1434 struct page *page, unsigned int len, unsigned int off,
1435 unsigned int op, sector_t sector)
1436{
1437 int ret;
1438
1439 if (!op_is_write(op)) {
1440 ret = btt_read_pg(btt, bip, page, off, sector, len);
1441 flush_dcache_page(page);
1442 } else {
1443 flush_dcache_page(page);
1444 ret = btt_write_pg(btt, bip, sector, page, off, len);
1445 }
1446
1447 return ret;
1448}
1449
1450static blk_qc_t btt_make_request(struct request_queue *q, struct bio *bio)
1451{
1452 struct bio_integrity_payload *bip = bio_integrity(bio);
1453 struct btt *btt = q->queuedata;
1454 struct bvec_iter iter;
1455 unsigned long start;
1456 struct bio_vec bvec;
1457 int err = 0;
1458 bool do_acct;
1459
1460 if (!bio_integrity_prep(bio))
1461 return BLK_QC_T_NONE;
1462
1463 do_acct = nd_iostat_start(bio, &start);
1464 bio_for_each_segment(bvec, bio, iter) {
1465 unsigned int len = bvec.bv_len;
1466
1467 if (len > PAGE_SIZE || len < btt->sector_size ||
1468 len % btt->sector_size) {
1469 dev_err_ratelimited(&btt->nd_btt->dev,
1470 "unaligned bio segment (len: %d)\n", len);
1471 bio->bi_status = BLK_STS_IOERR;
1472 break;
1473 }
1474
1475 err = btt_do_bvec(btt, bip, bvec.bv_page, len, bvec.bv_offset,
1476 bio_op(bio), iter.bi_sector);
1477 if (err) {
1478 dev_err(&btt->nd_btt->dev,
1479 "io error in %s sector %lld, len %d,\n",
1480 (op_is_write(bio_op(bio))) ? "WRITE" :
1481 "READ",
1482 (unsigned long long) iter.bi_sector, len);
1483 bio->bi_status = errno_to_blk_status(err);
1484 break;
1485 }
1486 }
1487 if (do_acct)
1488 nd_iostat_end(bio, start);
1489
1490 bio_endio(bio);
1491 return BLK_QC_T_NONE;
1492}
1493
1494static int btt_rw_page(struct block_device *bdev, sector_t sector,
1495 struct page *page, unsigned int op)
1496{
1497 struct btt *btt = bdev->bd_disk->private_data;
1498 int rc;
1499 unsigned int len;
1500
1501 len = hpage_nr_pages(page) * PAGE_SIZE;
1502 rc = btt_do_bvec(btt, NULL, page, len, 0, op, sector);
1503 if (rc == 0)
1504 page_endio(page, op_is_write(op), 0);
1505
1506 return rc;
1507}
1508
1509
1510static int btt_getgeo(struct block_device *bd, struct hd_geometry *geo)
1511{
1512 /* some standard values */
1513 geo->heads = 1 << 6;
1514 geo->sectors = 1 << 5;
1515 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
1516 return 0;
1517}
1518
1519static const struct block_device_operations btt_fops = {
1520 .owner = THIS_MODULE,
1521 .rw_page = btt_rw_page,
1522 .getgeo = btt_getgeo,
1523 .revalidate_disk = nvdimm_revalidate_disk,
1524};
1525
1526static int btt_blk_init(struct btt *btt)
1527{
1528 struct nd_btt *nd_btt = btt->nd_btt;
1529 struct nd_namespace_common *ndns = nd_btt->ndns;
1530
1531 /* create a new disk and request queue for btt */
1532 btt->btt_queue = blk_alloc_queue(GFP_KERNEL);
1533 if (!btt->btt_queue)
1534 return -ENOMEM;
1535
1536 btt->btt_disk = alloc_disk(0);
1537 if (!btt->btt_disk) {
1538 blk_cleanup_queue(btt->btt_queue);
1539 return -ENOMEM;
1540 }
1541
1542 nvdimm_namespace_disk_name(ndns, btt->btt_disk->disk_name);
1543 btt->btt_disk->first_minor = 0;
1544 btt->btt_disk->fops = &btt_fops;
1545 btt->btt_disk->private_data = btt;
1546 btt->btt_disk->queue = btt->btt_queue;
1547 btt->btt_disk->flags = GENHD_FL_EXT_DEVT;
1548 btt->btt_disk->queue->backing_dev_info->capabilities |=
1549 BDI_CAP_SYNCHRONOUS_IO;
1550
1551 blk_queue_make_request(btt->btt_queue, btt_make_request);
1552 blk_queue_logical_block_size(btt->btt_queue, btt->sector_size);
1553 blk_queue_max_hw_sectors(btt->btt_queue, UINT_MAX);
1554 blk_queue_flag_set(QUEUE_FLAG_NONROT, btt->btt_queue);
1555 btt->btt_queue->queuedata = btt;
1556
1557 if (btt_meta_size(btt)) {
1558 int rc = nd_integrity_init(btt->btt_disk, btt_meta_size(btt));
1559
1560 if (rc) {
1561 del_gendisk(btt->btt_disk);
1562 put_disk(btt->btt_disk);
1563 blk_cleanup_queue(btt->btt_queue);
1564 return rc;
1565 }
1566 }
1567 set_capacity(btt->btt_disk, btt->nlba * btt->sector_size >> 9);
1568 device_add_disk(&btt->nd_btt->dev, btt->btt_disk, NULL);
1569 btt->nd_btt->size = btt->nlba * (u64)btt->sector_size;
1570 revalidate_disk(btt->btt_disk);
1571
1572 return 0;
1573}
1574
1575static void btt_blk_cleanup(struct btt *btt)
1576{
1577 del_gendisk(btt->btt_disk);
1578 put_disk(btt->btt_disk);
1579 blk_cleanup_queue(btt->btt_queue);
1580}
1581
1582/**
1583 * btt_init - initialize a block translation table for the given device
1584 * @nd_btt: device with BTT geometry and backing device info
1585 * @rawsize: raw size in bytes of the backing device
1586 * @lbasize: lba size of the backing device
1587 * @uuid: A uuid for the backing device - this is stored on media
1588 * @maxlane: maximum number of parallel requests the device can handle
1589 *
1590 * Initialize a Block Translation Table on a backing device to provide
1591 * single sector power fail atomicity.
1592 *
1593 * Context:
1594 * Might sleep.
1595 *
1596 * Returns:
1597 * Pointer to a new struct btt on success, NULL on failure.
1598 */
1599static struct btt *btt_init(struct nd_btt *nd_btt, unsigned long long rawsize,
1600 u32 lbasize, u8 *uuid, struct nd_region *nd_region)
1601{
1602 int ret;
1603 struct btt *btt;
1604 struct nd_namespace_io *nsio;
1605 struct device *dev = &nd_btt->dev;
1606
1607 btt = devm_kzalloc(dev, sizeof(struct btt), GFP_KERNEL);
1608 if (!btt)
1609 return NULL;
1610
1611 btt->nd_btt = nd_btt;
1612 btt->rawsize = rawsize;
1613 btt->lbasize = lbasize;
1614 btt->sector_size = ((lbasize >= 4096) ? 4096 : 512);
1615 INIT_LIST_HEAD(&btt->arena_list);
1616 mutex_init(&btt->init_lock);
1617 btt->nd_region = nd_region;
1618 nsio = to_nd_namespace_io(&nd_btt->ndns->dev);
1619 btt->phys_bb = &nsio->bb;
1620
1621 ret = discover_arenas(btt);
1622 if (ret) {
1623 dev_err(dev, "init: error in arena_discover: %d\n", ret);
1624 return NULL;
1625 }
1626
1627 if (btt->init_state != INIT_READY && nd_region->ro) {
1628 dev_warn(dev, "%s is read-only, unable to init btt metadata\n",
1629 dev_name(&nd_region->dev));
1630 return NULL;
1631 } else if (btt->init_state != INIT_READY) {
1632 btt->num_arenas = (rawsize / ARENA_MAX_SIZE) +
1633 ((rawsize % ARENA_MAX_SIZE) ? 1 : 0);
1634 dev_dbg(dev, "init: %d arenas for %llu rawsize\n",
1635 btt->num_arenas, rawsize);
1636
1637 ret = create_arenas(btt);
1638 if (ret) {
1639 dev_info(dev, "init: create_arenas: %d\n", ret);
1640 return NULL;
1641 }
1642
1643 ret = btt_meta_init(btt);
1644 if (ret) {
1645 dev_err(dev, "init: error in meta_init: %d\n", ret);
1646 return NULL;
1647 }
1648 }
1649
1650 ret = btt_blk_init(btt);
1651 if (ret) {
1652 dev_err(dev, "init: error in blk_init: %d\n", ret);
1653 return NULL;
1654 }
1655
1656 btt_debugfs_init(btt);
1657
1658 return btt;
1659}
1660
1661/**
1662 * btt_fini - de-initialize a BTT
1663 * @btt: the BTT handle that was generated by btt_init
1664 *
1665 * De-initialize a Block Translation Table on device removal
1666 *
1667 * Context:
1668 * Might sleep.
1669 */
1670static void btt_fini(struct btt *btt)
1671{
1672 if (btt) {
1673 btt_blk_cleanup(btt);
1674 free_arenas(btt);
1675 debugfs_remove_recursive(btt->debugfs_dir);
1676 }
1677}
1678
1679int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns)
1680{
1681 struct nd_btt *nd_btt = to_nd_btt(ndns->claim);
1682 struct nd_region *nd_region;
1683 struct btt_sb *btt_sb;
1684 struct btt *btt;
1685 size_t rawsize;
1686
1687 if (!nd_btt->uuid || !nd_btt->ndns || !nd_btt->lbasize) {
1688 dev_dbg(&nd_btt->dev, "incomplete btt configuration\n");
1689 return -ENODEV;
1690 }
1691
1692 btt_sb = devm_kzalloc(&nd_btt->dev, sizeof(*btt_sb), GFP_KERNEL);
1693 if (!btt_sb)
1694 return -ENOMEM;
1695
1696 /*
1697 * If this returns < 0, that is ok as it just means there wasn't
1698 * an existing BTT, and we're creating a new one. We still need to
1699 * call this as we need the version dependent fields in nd_btt to be
1700 * set correctly based on the holder class
1701 */
1702 nd_btt_version(nd_btt, ndns, btt_sb);
1703
1704 rawsize = nvdimm_namespace_capacity(ndns) - nd_btt->initial_offset;
1705 if (rawsize < ARENA_MIN_SIZE) {
1706 dev_dbg(&nd_btt->dev, "%s must be at least %ld bytes\n",
1707 dev_name(&ndns->dev),
1708 ARENA_MIN_SIZE + nd_btt->initial_offset);
1709 return -ENXIO;
1710 }
1711 nd_region = to_nd_region(nd_btt->dev.parent);
1712 btt = btt_init(nd_btt, rawsize, nd_btt->lbasize, nd_btt->uuid,
1713 nd_region);
1714 if (!btt)
1715 return -ENOMEM;
1716 nd_btt->btt = btt;
1717
1718 return 0;
1719}
1720EXPORT_SYMBOL(nvdimm_namespace_attach_btt);
1721
1722int nvdimm_namespace_detach_btt(struct nd_btt *nd_btt)
1723{
1724 struct btt *btt = nd_btt->btt;
1725
1726 btt_fini(btt);
1727 nd_btt->btt = NULL;
1728
1729 return 0;
1730}
1731EXPORT_SYMBOL(nvdimm_namespace_detach_btt);
1732
1733static int __init nd_btt_init(void)
1734{
1735 int rc = 0;
1736
1737 debugfs_root = debugfs_create_dir("btt", NULL);
1738 if (IS_ERR_OR_NULL(debugfs_root))
1739 rc = -ENXIO;
1740
1741 return rc;
1742}
1743
1744static void __exit nd_btt_exit(void)
1745{
1746 debugfs_remove_recursive(debugfs_root);
1747}
1748
1749MODULE_ALIAS_ND_DEVICE(ND_DEVICE_BTT);
1750MODULE_AUTHOR("Vishal Verma <vishal.l.verma@linux.intel.com>");
1751MODULE_LICENSE("GPL v2");
1752module_init(nd_btt_init);
1753module_exit(nd_btt_exit);
1754