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