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 | |
23 | enum log_ent_request { |
24 | LOG_NEW_ENT = 0, |
25 | LOG_OLD_ENT |
26 | }; |
27 | |
28 | static struct device *to_dev(struct arena_info *arena) |
29 | { |
30 | return &arena->nd_btt->dev; |
31 | } |
32 | |
33 | static u64 adjust_initial_offset(struct nd_btt *nd_btt, u64 offset) |
34 | { |
35 | return offset + nd_btt->initial_offset; |
36 | } |
37 | |
38 | static 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, size: n, flags); |
47 | } |
48 | |
49 | static 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, size: n, flags); |
58 | } |
59 | |
60 | static 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, offset: arena->info2off, buf: super, |
75 | n: sizeof(struct btt_sb), flags: 0); |
76 | if (ret) |
77 | return ret; |
78 | |
79 | return arena_write_bytes(arena, offset: arena->infooff, buf: super, |
80 | n: sizeof(struct btt_sb), flags: 0); |
81 | } |
82 | |
83 | static int btt_info_read(struct arena_info *arena, struct btt_sb *super) |
84 | { |
85 | return arena_read_bytes(arena, offset: arena->infooff, buf: super, |
86 | n: sizeof(struct btt_sb), flags: 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 | */ |
95 | static 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, offset: ns_off, buf: &mapping, MAP_ENT_SIZE, flags); |
105 | } |
106 | |
107 | static 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: mapping_le, flags: rwb_flags); |
149 | } |
150 | |
151 | static 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, offset: ns_off, buf: &in, MAP_ENT_SIZE, flags: 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 | |
207 | static int btt_log_group_read(struct arena_info *arena, u32 lane, |
208 | struct log_group *log) |
209 | { |
210 | return arena_read_bytes(arena, |
211 | offset: arena->logoff + (lane * LOG_GRP_SIZE), buf: log, |
212 | LOG_GRP_SIZE, flags: 0); |
213 | } |
214 | |
215 | static struct dentry *debugfs_root; |
216 | |
217 | static 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(buf: dirname, size: 32, fmt: "arena%d" , idx); |
228 | d = debugfs_create_dir(name: dirname, parent); |
229 | if (IS_ERR_OR_NULL(ptr: d)) |
230 | return; |
231 | a->debugfs_dir = d; |
232 | |
233 | debugfs_create_x64(name: "size" , S_IRUGO, parent: d, value: &a->size); |
234 | debugfs_create_x64(name: "external_lba_start" , S_IRUGO, parent: d, |
235 | value: &a->external_lba_start); |
236 | debugfs_create_x32(name: "internal_nlba" , S_IRUGO, parent: d, value: &a->internal_nlba); |
237 | debugfs_create_u32(name: "internal_lbasize" , S_IRUGO, parent: d, |
238 | value: &a->internal_lbasize); |
239 | debugfs_create_x32(name: "external_nlba" , S_IRUGO, parent: d, value: &a->external_nlba); |
240 | debugfs_create_u32(name: "external_lbasize" , S_IRUGO, parent: d, |
241 | value: &a->external_lbasize); |
242 | debugfs_create_u32(name: "nfree" , S_IRUGO, parent: d, value: &a->nfree); |
243 | debugfs_create_u16(name: "version_major" , S_IRUGO, parent: d, value: &a->version_major); |
244 | debugfs_create_u16(name: "version_minor" , S_IRUGO, parent: d, value: &a->version_minor); |
245 | debugfs_create_x64(name: "nextoff" , S_IRUGO, parent: d, value: &a->nextoff); |
246 | debugfs_create_x64(name: "infooff" , S_IRUGO, parent: d, value: &a->infooff); |
247 | debugfs_create_x64(name: "dataoff" , S_IRUGO, parent: d, value: &a->dataoff); |
248 | debugfs_create_x64(name: "mapoff" , S_IRUGO, parent: d, value: &a->mapoff); |
249 | debugfs_create_x64(name: "logoff" , S_IRUGO, parent: d, value: &a->logoff); |
250 | debugfs_create_x64(name: "info2off" , S_IRUGO, parent: d, value: &a->info2off); |
251 | debugfs_create_x32(name: "flags" , S_IRUGO, parent: d, value: &a->flags); |
252 | debugfs_create_u32(name: "log_index_0" , S_IRUGO, parent: d, value: &a->log_index[0]); |
253 | debugfs_create_u32(name: "log_index_1" , S_IRUGO, parent: d, value: &a->log_index[1]); |
254 | } |
255 | |
256 | static 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(name: dev_name(dev: &btt->nd_btt->dev), |
262 | parent: debugfs_root); |
263 | if (IS_ERR_OR_NULL(ptr: btt->debugfs_dir)) |
264 | return; |
265 | |
266 | list_for_each_entry(arena, &btt->arena_list, list) { |
267 | arena_debugfs_init(a: arena, parent: btt->debugfs_dir, idx: i); |
268 | i++; |
269 | } |
270 | } |
271 | |
272 | static 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 | */ |
286 | static 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, log_idx: idx0) == 0) { |
298 | log->ent[idx0].seq = cpu_to_le32(1); |
299 | return 0; |
300 | } |
301 | |
302 | if (log_seq(log, log_idx: idx0) == log_seq(log, log_idx: idx1)) |
303 | return -EINVAL; |
304 | if (log_seq(log, log_idx: idx0) + log_seq(log, log_idx: idx1) > 5) |
305 | return -EINVAL; |
306 | |
307 | if (log_seq(log, log_idx: idx0) < log_seq(log, log_idx: idx1)) { |
308 | if ((log_seq(log, log_idx: idx1) - log_seq(log, log_idx: idx0)) == 1) |
309 | old = 0; |
310 | else |
311 | old = 1; |
312 | } else { |
313 | if ((log_seq(log, log_idx: idx0) - log_seq(log, log_idx: 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 | */ |
328 | static 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: &log); |
336 | if (ret) |
337 | return -EIO; |
338 | |
339 | old_ent = btt_log_get_old(a: arena, log: &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 | */ |
362 | static 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, offset: ns_off, buf: src, n: 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, offset: ns_off, buf: src, n: log_half, flags); |
381 | } |
382 | |
383 | static 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, flags: 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 | */ |
407 | static 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(size: 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, offset: arena->mapoff + offset, buf: zerobuf, |
433 | n: size, flags: 0); |
434 | if (ret) |
435 | goto free; |
436 | |
437 | offset += size; |
438 | mapsize -= size; |
439 | cond_resched(); |
440 | } |
441 | |
442 | free: |
443 | kfree(objp: 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 | */ |
451 | static 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(size: 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, offset: arena->logoff + offset, buf: zerobuf, |
477 | n: size, flags: 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, lane: i, sub: 0, ent: &ent, flags: 0); |
492 | if (ret) |
493 | goto free; |
494 | } |
495 | |
496 | free: |
497 | kfree(objp: zerobuf); |
498 | return ret; |
499 | } |
500 | |
501 | static u64 to_namespace_offset(struct arena_info *arena, u64 lba) |
502 | { |
503 | return arena->dataoff + ((u64)lba * arena->internal_lbasize); |
504 | } |
505 | |
506 | static 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, offset: nsoff, buf: zero_page, |
522 | n: chunk, flags: 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(lock: &arena->err_lock); |
531 | } |
532 | return ret; |
533 | } |
534 | |
535 | static 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(n: arena->nfree, size: 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, lane: i, ent: &log_new, old_flag: 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, lane: 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), mapping: &map_entry, |
579 | NULL, NULL, rwb_flags: 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), z_flag: 0, e_flag: 0, rwb_flags: 0); |
597 | if (ret) |
598 | return ret; |
599 | } |
600 | } |
601 | |
602 | return 0; |
603 | } |
604 | |
605 | static 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 | */ |
624 | static 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, lane: i, log: &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(ent: &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(ent: &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(ent: &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 | |
722 | static int btt_rtt_init(struct arena_info *arena) |
723 | { |
724 | arena->rtt = kcalloc(n: arena->nfree, size: sizeof(u32), GFP_KERNEL); |
725 | if (arena->rtt == NULL) |
726 | return -ENOMEM; |
727 | |
728 | return 0; |
729 | } |
730 | |
731 | static int btt_maplocks_init(struct arena_info *arena) |
732 | { |
733 | u32 i; |
734 | |
735 | arena->map_locks = kcalloc(n: arena->nfree, size: 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 | |
746 | static 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(size: 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(dividend: available - BTT_PG_SIZE, |
784 | divisor: 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 | |
803 | static 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(entry: &arena->list); |
809 | kfree(objp: arena->rtt); |
810 | kfree(objp: arena->map_locks); |
811 | kfree(objp: arena->freelist); |
812 | debugfs_remove_recursive(dentry: arena->debugfs_dir); |
813 | kfree(objp: arena); |
814 | } |
815 | } |
816 | |
817 | /* |
818 | * This function reads an existing valid btt superblock and |
819 | * populates the corresponding arena_info struct |
820 | */ |
821 | static 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 | |
847 | static 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(size: sizeof(*super), GFP_KERNEL); |
857 | if (!super) |
858 | return -ENOMEM; |
859 | |
860 | while (remaining) { |
861 | /* Alloc memory for arena */ |
862 | arena = alloc_arena(btt, size: 0, start: 0, arena_off: 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(nd_btt: 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, arena_off: 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(new: &arena->list, head: &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(objp: arena); |
924 | free_arenas(btt); |
925 | return ret; |
926 | } |
927 | |
928 | static 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, size: arena_size, start: btt->nlba, arena_off: 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(new: &arena->list, head: &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 | */ |
964 | static 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(dev: &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(size: 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(dst: super->uuid, src: nd_btt->uuid); |
986 | export_uuid(dst: super->parent_uuid, src: 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(sb: (struct nd_gen_sb *) super); |
1008 | super->checksum = cpu_to_le64(sum); |
1009 | |
1010 | ret = btt_info_write(arena, super); |
1011 | |
1012 | kfree(objp: 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 | */ |
1020 | static 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(lock: &btt->init_lock); |
1048 | return ret; |
1049 | } |
1050 | |
1051 | static 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 | */ |
1063 | static 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(dividend: sector << SECTOR_SHIFT, divisor: 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 | */ |
1085 | static 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(lock: &arena->map_locks[idx].lock); |
1091 | } |
1092 | |
1093 | static 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(lock: &arena->map_locks[idx].lock); |
1099 | } |
1100 | |
1101 | static 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, offset: nsoff, buf: mem + off, n: len, flags: NVDIMM_IO_ATOMIC); |
1109 | kunmap_atomic(mem); |
1110 | |
1111 | return ret; |
1112 | } |
1113 | |
1114 | static 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, offset: nsoff, buf: mem + off, n: len, flags: NVDIMM_IO_ATOMIC); |
1122 | kunmap_atomic(mem); |
1123 | |
1124 | return ret; |
1125 | } |
1126 | |
1127 | static 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 |
1136 | static 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, lba: 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(bvec: &bv); |
1162 | if (rw) |
1163 | ret = arena_write_bytes(arena, offset: meta_nsoff, buf: mem, n: cur_len, |
1164 | flags: NVDIMM_IO_ATOMIC); |
1165 | else |
1166 | ret = arena_read_bytes(arena, offset: meta_nsoff, buf: mem, n: cur_len, |
1167 | flags: 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(bv: bip->bip_vec, iter: &bip->bip_iter, bytes: cur_len)) |
1176 | return -EIO; |
1177 | } |
1178 | |
1179 | return ret; |
1180 | } |
1181 | |
1182 | #else /* CONFIG_BLK_DEV_INTEGRITY */ |
1183 | static 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 | |
1190 | static 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(nd_region: btt->nd_region); |
1203 | |
1204 | ret = lba_to_arena(btt, sector, premap: &premap, arena: &arena); |
1205 | if (ret) |
1206 | goto out_lane; |
1207 | |
1208 | cur_len = min(btt->sector_size, len); |
1209 | |
1210 | ret = btt_map_read(arena, lba: premap, mapping: &postmap, trim: &t_flag, error: &e_flag, |
1211 | rwb_flags: 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, len: 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, lba: premap, mapping: &new_map, trim: &new_t, |
1242 | error: &new_e, rwb_flags: 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, lba: postmap, len: cur_len); |
1256 | if (ret) { |
1257 | /* Media error - set the e_flag */ |
1258 | if (btt_map_write(arena, lba: premap, mapping: postmap, z_flag: 0, e_flag: 1, rwb_flags: 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(nd_region: 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(nd_region: 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 | */ |
1293 | static bool btt_is_badblock(struct btt *btt, struct arena_info *arena, |
1294 | u32 postmap) |
1295 | { |
1296 | u64 nsoff = adjust_initial_offset(nd_btt: arena->nd_btt, |
1297 | offset: to_namespace_offset(arena, lba: postmap)); |
1298 | sector_t phys_sector = nsoff >> 9; |
1299 | |
1300 | return is_bad_pmem(bb: btt->phys_bb, sector: phys_sector, len: arena->internal_lbasize); |
1301 | } |
1302 | |
1303 | static 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(nd_region: btt->nd_region); |
1319 | |
1320 | ret = lba_to_arena(btt, sector, premap: &premap, arena: &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, postmap: arena->freelist[lane].block)) |
1331 | arena->freelist[lane].has_err = 1; |
1332 | |
1333 | if (mutex_is_locked(lock: &arena->err_lock) |
1334 | || arena->freelist[lane].has_err) { |
1335 | nd_region_release_lane(nd_region: 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, lba: new_postmap, page, off, len: cur_len); |
1359 | if (ret) |
1360 | goto out_lane; |
1361 | |
1362 | if (bip) { |
1363 | ret = btt_rw_integrity(btt, bip, arena, postmap: new_postmap, |
1364 | WRITE); |
1365 | if (ret) |
1366 | goto out_lane; |
1367 | } |
1368 | |
1369 | lock_map(arena, premap); |
1370 | ret = btt_map_read(arena, lba: premap, mapping: &old_postmap, NULL, error: &e_flag, |
1371 | rwb_flags: 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, ent: &log); |
1387 | if (ret) |
1388 | goto out_map; |
1389 | |
1390 | ret = btt_map_write(arena, lba: premap, mapping: new_postmap, z_flag: 0, e_flag: 0, |
1391 | rwb_flags: NVDIMM_IO_ATOMIC); |
1392 | if (ret) |
1393 | goto out_map; |
1394 | |
1395 | unlock_map(arena, premap); |
1396 | nd_region_release_lane(nd_region: 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(nd_region: btt->nd_region, lane); |
1415 | return ret; |
1416 | } |
1417 | |
1418 | static 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 | |
1435 | static 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, page: bvec.bv_page, len, off: bvec.bv_offset, |
1463 | op: bio_op(bio), sector: 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(errno: err); |
1471 | break; |
1472 | } |
1473 | } |
1474 | if (do_acct) |
1475 | bio_end_io_acct(bio, start_time: start); |
1476 | |
1477 | bio_endio(bio); |
1478 | } |
1479 | |
1480 | static 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(disk: bd->bd_disk) >> 11; |
1486 | return 0; |
1487 | } |
1488 | |
1489 | static const struct block_device_operations btt_fops = { |
1490 | .owner = THIS_MODULE, |
1491 | .submit_bio = btt_submit_bio, |
1492 | .getgeo = btt_getgeo, |
1493 | }; |
1494 | |
1495 | static 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 | struct queue_limits lim = { |
1500 | .logical_block_size = btt->sector_size, |
1501 | .max_hw_sectors = UINT_MAX, |
1502 | }; |
1503 | int rc; |
1504 | |
1505 | btt->btt_disk = blk_alloc_disk(&lim, NUMA_NO_NODE); |
1506 | if (IS_ERR(ptr: btt->btt_disk)) |
1507 | return PTR_ERR(ptr: btt->btt_disk); |
1508 | |
1509 | nvdimm_namespace_disk_name(ndns, name: btt->btt_disk->disk_name); |
1510 | btt->btt_disk->first_minor = 0; |
1511 | btt->btt_disk->fops = &btt_fops; |
1512 | btt->btt_disk->private_data = btt; |
1513 | |
1514 | blk_queue_flag_set(QUEUE_FLAG_NONROT, q: btt->btt_disk->queue); |
1515 | blk_queue_flag_set(QUEUE_FLAG_SYNCHRONOUS, q: btt->btt_disk->queue); |
1516 | |
1517 | if (btt_meta_size(btt)) { |
1518 | rc = nd_integrity_init(disk: btt->btt_disk, meta_size: btt_meta_size(btt)); |
1519 | if (rc) |
1520 | goto out_cleanup_disk; |
1521 | } |
1522 | |
1523 | set_capacity(disk: btt->btt_disk, size: btt->nlba * btt->sector_size >> 9); |
1524 | rc = device_add_disk(parent: &btt->nd_btt->dev, disk: btt->btt_disk, NULL); |
1525 | if (rc) |
1526 | goto out_cleanup_disk; |
1527 | |
1528 | btt->nd_btt->size = btt->nlba * (u64)btt->sector_size; |
1529 | nvdimm_check_and_set_ro(disk: btt->btt_disk); |
1530 | |
1531 | return 0; |
1532 | |
1533 | out_cleanup_disk: |
1534 | put_disk(disk: btt->btt_disk); |
1535 | return rc; |
1536 | } |
1537 | |
1538 | static void btt_blk_cleanup(struct btt *btt) |
1539 | { |
1540 | del_gendisk(gp: btt->btt_disk); |
1541 | put_disk(disk: btt->btt_disk); |
1542 | } |
1543 | |
1544 | /** |
1545 | * btt_init - initialize a block translation table for the given device |
1546 | * @nd_btt: device with BTT geometry and backing device info |
1547 | * @rawsize: raw size in bytes of the backing device |
1548 | * @lbasize: lba size of the backing device |
1549 | * @uuid: A uuid for the backing device - this is stored on media |
1550 | * @nd_region: &struct nd_region for the REGION device |
1551 | * |
1552 | * Initialize a Block Translation Table on a backing device to provide |
1553 | * single sector power fail atomicity. |
1554 | * |
1555 | * Context: |
1556 | * Might sleep. |
1557 | * |
1558 | * Returns: |
1559 | * Pointer to a new struct btt on success, NULL on failure. |
1560 | */ |
1561 | static struct btt *btt_init(struct nd_btt *nd_btt, unsigned long long rawsize, |
1562 | u32 lbasize, uuid_t *uuid, |
1563 | struct nd_region *nd_region) |
1564 | { |
1565 | int ret; |
1566 | struct btt *btt; |
1567 | struct nd_namespace_io *nsio; |
1568 | struct device *dev = &nd_btt->dev; |
1569 | |
1570 | btt = devm_kzalloc(dev, size: sizeof(struct btt), GFP_KERNEL); |
1571 | if (!btt) |
1572 | return NULL; |
1573 | |
1574 | btt->nd_btt = nd_btt; |
1575 | btt->rawsize = rawsize; |
1576 | btt->lbasize = lbasize; |
1577 | btt->sector_size = ((lbasize >= 4096) ? 4096 : 512); |
1578 | INIT_LIST_HEAD(list: &btt->arena_list); |
1579 | mutex_init(&btt->init_lock); |
1580 | btt->nd_region = nd_region; |
1581 | nsio = to_nd_namespace_io(dev: &nd_btt->ndns->dev); |
1582 | btt->phys_bb = &nsio->bb; |
1583 | |
1584 | ret = discover_arenas(btt); |
1585 | if (ret) { |
1586 | dev_err(dev, "init: error in arena_discover: %d\n" , ret); |
1587 | return NULL; |
1588 | } |
1589 | |
1590 | if (btt->init_state != INIT_READY && nd_region->ro) { |
1591 | dev_warn(dev, "%s is read-only, unable to init btt metadata\n" , |
1592 | dev_name(&nd_region->dev)); |
1593 | return NULL; |
1594 | } else if (btt->init_state != INIT_READY) { |
1595 | btt->num_arenas = (rawsize / ARENA_MAX_SIZE) + |
1596 | ((rawsize % ARENA_MAX_SIZE) ? 1 : 0); |
1597 | dev_dbg(dev, "init: %d arenas for %llu rawsize\n" , |
1598 | btt->num_arenas, rawsize); |
1599 | |
1600 | ret = create_arenas(btt); |
1601 | if (ret) { |
1602 | dev_info(dev, "init: create_arenas: %d\n" , ret); |
1603 | return NULL; |
1604 | } |
1605 | |
1606 | ret = btt_meta_init(btt); |
1607 | if (ret) { |
1608 | dev_err(dev, "init: error in meta_init: %d\n" , ret); |
1609 | return NULL; |
1610 | } |
1611 | } |
1612 | |
1613 | ret = btt_blk_init(btt); |
1614 | if (ret) { |
1615 | dev_err(dev, "init: error in blk_init: %d\n" , ret); |
1616 | return NULL; |
1617 | } |
1618 | |
1619 | btt_debugfs_init(btt); |
1620 | |
1621 | return btt; |
1622 | } |
1623 | |
1624 | /** |
1625 | * btt_fini - de-initialize a BTT |
1626 | * @btt: the BTT handle that was generated by btt_init |
1627 | * |
1628 | * De-initialize a Block Translation Table on device removal |
1629 | * |
1630 | * Context: |
1631 | * Might sleep. |
1632 | */ |
1633 | static void btt_fini(struct btt *btt) |
1634 | { |
1635 | if (btt) { |
1636 | btt_blk_cleanup(btt); |
1637 | free_arenas(btt); |
1638 | debugfs_remove_recursive(dentry: btt->debugfs_dir); |
1639 | } |
1640 | } |
1641 | |
1642 | int nvdimm_namespace_attach_btt(struct nd_namespace_common *ndns) |
1643 | { |
1644 | struct nd_btt *nd_btt = to_nd_btt(dev: ndns->claim); |
1645 | struct nd_region *nd_region; |
1646 | struct btt_sb *btt_sb; |
1647 | struct btt *btt; |
1648 | size_t size, rawsize; |
1649 | int rc; |
1650 | |
1651 | if (!nd_btt->uuid || !nd_btt->ndns || !nd_btt->lbasize) { |
1652 | dev_dbg(&nd_btt->dev, "incomplete btt configuration\n" ); |
1653 | return -ENODEV; |
1654 | } |
1655 | |
1656 | btt_sb = devm_kzalloc(dev: &nd_btt->dev, size: sizeof(*btt_sb), GFP_KERNEL); |
1657 | if (!btt_sb) |
1658 | return -ENOMEM; |
1659 | |
1660 | size = nvdimm_namespace_capacity(ndns); |
1661 | rc = devm_namespace_enable(dev: &nd_btt->dev, ndns, size); |
1662 | if (rc) |
1663 | return rc; |
1664 | |
1665 | /* |
1666 | * If this returns < 0, that is ok as it just means there wasn't |
1667 | * an existing BTT, and we're creating a new one. We still need to |
1668 | * call this as we need the version dependent fields in nd_btt to be |
1669 | * set correctly based on the holder class |
1670 | */ |
1671 | nd_btt_version(nd_btt, ndns, btt_sb); |
1672 | |
1673 | rawsize = size - nd_btt->initial_offset; |
1674 | if (rawsize < ARENA_MIN_SIZE) { |
1675 | dev_dbg(&nd_btt->dev, "%s must be at least %ld bytes\n" , |
1676 | dev_name(&ndns->dev), |
1677 | ARENA_MIN_SIZE + nd_btt->initial_offset); |
1678 | return -ENXIO; |
1679 | } |
1680 | nd_region = to_nd_region(dev: nd_btt->dev.parent); |
1681 | btt = btt_init(nd_btt, rawsize, lbasize: nd_btt->lbasize, uuid: nd_btt->uuid, |
1682 | nd_region); |
1683 | if (!btt) |
1684 | return -ENOMEM; |
1685 | nd_btt->btt = btt; |
1686 | |
1687 | return 0; |
1688 | } |
1689 | EXPORT_SYMBOL(nvdimm_namespace_attach_btt); |
1690 | |
1691 | int nvdimm_namespace_detach_btt(struct nd_btt *nd_btt) |
1692 | { |
1693 | struct btt *btt = nd_btt->btt; |
1694 | |
1695 | btt_fini(btt); |
1696 | nd_btt->btt = NULL; |
1697 | |
1698 | return 0; |
1699 | } |
1700 | EXPORT_SYMBOL(nvdimm_namespace_detach_btt); |
1701 | |
1702 | static int __init nd_btt_init(void) |
1703 | { |
1704 | int rc = 0; |
1705 | |
1706 | debugfs_root = debugfs_create_dir(name: "btt" , NULL); |
1707 | if (IS_ERR_OR_NULL(ptr: debugfs_root)) |
1708 | rc = -ENXIO; |
1709 | |
1710 | return rc; |
1711 | } |
1712 | |
1713 | static void __exit nd_btt_exit(void) |
1714 | { |
1715 | debugfs_remove_recursive(dentry: debugfs_root); |
1716 | } |
1717 | |
1718 | MODULE_ALIAS_ND_DEVICE(ND_DEVICE_BTT); |
1719 | MODULE_AUTHOR("Vishal Verma <vishal.l.verma@linux.intel.com>" ); |
1720 | MODULE_LICENSE("GPL v2" ); |
1721 | module_init(nd_btt_init); |
1722 | module_exit(nd_btt_exit); |
1723 | |