FILE
zoned.c
File preview with syntax highlighting.
linux-6.19.14/fs/btrfs/zoned.c
1// SPDX-License-Identifier: GPL-2.0 2 3#include <linux/bitops.h> 4#include <linux/slab.h> 5#include <linux/blkdev.h> 6#include <linux/sched/mm.h> 7#include <linux/atomic.h> 8#include <linux/vmalloc.h> 9#include "ctree.h" 10#include "volumes.h" 11#include "zoned.h" 12#include "disk-io.h" 13#include "block-group.h" 14#include "dev-replace.h" 15#include "space-info.h" 16#include "fs.h" 17#include "accessors.h" 18#include "bio.h" 19#include "transaction.h" 20#include "sysfs.h" 21 22/* Maximum number of zones to report per blkdev_report_zones() call */ 23#define BTRFS_REPORT_NR_ZONES 4096 24/* Invalid allocation pointer value for missing devices */ 25#define WP_MISSING_DEV ((u64)-1) 26/* Pseudo write pointer value for conventional zone */ 27#define WP_CONVENTIONAL ((u64)-2) 28 29/* 30 * Location of the first zone of superblock logging zone pairs. 31 * 32 * - primary superblock: 0B (zone 0) 33 * - first copy: 512G (zone starting at that offset) 34 * - second copy: 4T (zone starting at that offset) 35 */ 36#define BTRFS_SB_LOG_PRIMARY_OFFSET (0ULL) 37#define BTRFS_SB_LOG_FIRST_OFFSET (512ULL * SZ_1G) 38#define BTRFS_SB_LOG_SECOND_OFFSET (4096ULL * SZ_1G) 39 40#define BTRFS_SB_LOG_FIRST_SHIFT ilog2(BTRFS_SB_LOG_FIRST_OFFSET) 41#define BTRFS_SB_LOG_SECOND_SHIFT ilog2(BTRFS_SB_LOG_SECOND_OFFSET) 42 43/* Number of superblock log zones */ 44#define BTRFS_NR_SB_LOG_ZONES 2 45 46/* Default number of max active zones when the device has no limits. */ 47#define BTRFS_DEFAULT_MAX_ACTIVE_ZONES 128 48 49/* 50 * Minimum of active zones we need: 51 * 52 * - BTRFS_SUPER_MIRROR_MAX zones for superblock mirrors 53 * - 3 zones to ensure at least one zone per SYSTEM, META and DATA block group 54 * - 1 zone for tree-log dedicated block group 55 * - 1 zone for relocation 56 */ 57#define BTRFS_MIN_ACTIVE_ZONES (BTRFS_SUPER_MIRROR_MAX + 5) 58 59/* 60 * Minimum / maximum supported zone size. Currently, SMR disks have a zone 61 * size of 256MiB, and we are expecting ZNS drives to be in the 1-4GiB range. 62 * We do not expect the zone size to become larger than 8GiB or smaller than 63 * 4MiB in the near future. 64 */ 65#define BTRFS_MAX_ZONE_SIZE SZ_8G 66#define BTRFS_MIN_ZONE_SIZE SZ_4M 67 68#define SUPER_INFO_SECTORS ((u64)BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT) 69 70static void wait_eb_writebacks(struct btrfs_block_group *block_group); 71static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_written); 72 73static inline bool sb_zone_is_full(const struct blk_zone *zone) 74{ 75 return (zone->cond == BLK_ZONE_COND_FULL) || 76 (zone->wp + SUPER_INFO_SECTORS > zone->start + zone->capacity); 77} 78 79static int copy_zone_info_cb(struct blk_zone *zone, unsigned int idx, void *data) 80{ 81 struct blk_zone *zones = data; 82 83 memcpy(&zones[idx], zone, sizeof(*zone)); 84 85 return 0; 86} 87 88static int sb_write_pointer(struct block_device *bdev, struct blk_zone *zones, 89 u64 *wp_ret) 90{ 91 bool empty[BTRFS_NR_SB_LOG_ZONES]; 92 bool full[BTRFS_NR_SB_LOG_ZONES]; 93 sector_t sector; 94 95 for (int i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) { 96 ASSERT(zones[i].type != BLK_ZONE_TYPE_CONVENTIONAL, 97 "zones[%d].type=%d", i, zones[i].type); 98 empty[i] = (zones[i].cond == BLK_ZONE_COND_EMPTY); 99 full[i] = sb_zone_is_full(&zones[i]); 100 } 101 102 /* 103 * Possible states of log buffer zones 104 * 105 * Empty[0] In use[0] Full[0] 106 * Empty[1] * 0 1 107 * In use[1] x x 1 108 * Full[1] 0 0 C 109 * 110 * Log position: 111 * *: Special case, no superblock is written 112 * 0: Use write pointer of zones[0] 113 * 1: Use write pointer of zones[1] 114 * C: Compare super blocks from zones[0] and zones[1], use the latest 115 * one determined by generation 116 * x: Invalid state 117 */ 118 119 if (empty[0] && empty[1]) { 120 /* Special case to distinguish no superblock to read */ 121 *wp_ret = zones[0].start << SECTOR_SHIFT; 122 return -ENOENT; 123 } else if (full[0] && full[1]) { 124 /* Compare two super blocks */ 125 struct address_space *mapping = bdev->bd_mapping; 126 struct page *page[BTRFS_NR_SB_LOG_ZONES]; 127 struct btrfs_super_block *super[BTRFS_NR_SB_LOG_ZONES]; 128 129 for (int i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) { 130 u64 zone_end = (zones[i].start + zones[i].capacity) << SECTOR_SHIFT; 131 u64 bytenr = ALIGN_DOWN(zone_end, BTRFS_SUPER_INFO_SIZE) - 132 BTRFS_SUPER_INFO_SIZE; 133 134 page[i] = read_cache_page_gfp(mapping, 135 bytenr >> PAGE_SHIFT, GFP_NOFS); 136 if (IS_ERR(page[i])) { 137 if (i == 1) 138 btrfs_release_disk_super(super[0]); 139 return PTR_ERR(page[i]); 140 } 141 super[i] = page_address(page[i]); 142 } 143 144 if (btrfs_super_generation(super[0]) > 145 btrfs_super_generation(super[1])) 146 sector = zones[1].start; 147 else 148 sector = zones[0].start; 149 150 for (int i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) 151 btrfs_release_disk_super(super[i]); 152 } else if (!full[0] && (empty[1] || full[1])) { 153 sector = zones[0].wp; 154 } else if (full[0]) { 155 sector = zones[1].wp; 156 } else { 157 return -EUCLEAN; 158 } 159 *wp_ret = sector << SECTOR_SHIFT; 160 return 0; 161} 162 163/* 164 * Get the first zone number of the superblock mirror 165 */ 166static inline u32 sb_zone_number(int shift, int mirror) 167{ 168 u64 zone = U64_MAX; 169 170 ASSERT(mirror < BTRFS_SUPER_MIRROR_MAX, "mirror=%d", mirror); 171 switch (mirror) { 172 case 0: zone = 0; break; 173 case 1: zone = 1ULL << (BTRFS_SB_LOG_FIRST_SHIFT - shift); break; 174 case 2: zone = 1ULL << (BTRFS_SB_LOG_SECOND_SHIFT - shift); break; 175 } 176 177 ASSERT(zone <= U32_MAX, "zone=%llu", zone); 178 179 return (u32)zone; 180} 181 182static inline sector_t zone_start_sector(u32 zone_number, 183 struct block_device *bdev) 184{ 185 return (sector_t)zone_number << ilog2(bdev_zone_sectors(bdev)); 186} 187 188static inline u64 zone_start_physical(u32 zone_number, 189 struct btrfs_zoned_device_info *zone_info) 190{ 191 return (u64)zone_number << zone_info->zone_size_shift; 192} 193 194/* 195 * Emulate blkdev_report_zones() for a non-zoned device. It slices up the block 196 * device into static sized chunks and fake a conventional zone on each of 197 * them. 198 */ 199static int emulate_report_zones(struct btrfs_device *device, u64 pos, 200 struct blk_zone *zones, unsigned int nr_zones) 201{ 202 const sector_t zone_sectors = device->fs_info->zone_size >> SECTOR_SHIFT; 203 sector_t bdev_size = bdev_nr_sectors(device->bdev); 204 unsigned int i; 205 206 pos >>= SECTOR_SHIFT; 207 for (i = 0; i < nr_zones; i++) { 208 zones[i].start = i * zone_sectors + pos; 209 zones[i].len = zone_sectors; 210 zones[i].capacity = zone_sectors; 211 zones[i].wp = zones[i].start + zone_sectors; 212 zones[i].type = BLK_ZONE_TYPE_CONVENTIONAL; 213 zones[i].cond = BLK_ZONE_COND_NOT_WP; 214 215 if (zones[i].wp >= bdev_size) { 216 i++; 217 break; 218 } 219 } 220 221 return i; 222} 223 224static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos, 225 struct blk_zone *zones, unsigned int *nr_zones) 226{ 227 struct btrfs_zoned_device_info *zinfo = device->zone_info; 228 int ret; 229 230 if (!*nr_zones) 231 return 0; 232 233 if (!bdev_is_zoned(device->bdev)) { 234 ret = emulate_report_zones(device, pos, zones, *nr_zones); 235 *nr_zones = ret; 236 return 0; 237 } 238 239 /* Check cache */ 240 if (zinfo->zone_cache) { 241 unsigned int i; 242 u32 zno; 243 244 ASSERT(IS_ALIGNED(pos, zinfo->zone_size), 245 "pos=%llu zinfo->zone_size=%llu", pos, zinfo->zone_size); 246 zno = pos >> zinfo->zone_size_shift; 247 /* 248 * We cannot report zones beyond the zone end. So, it is OK to 249 * cap *nr_zones to at the end. 250 */ 251 *nr_zones = min_t(u32, *nr_zones, zinfo->nr_zones - zno); 252 253 for (i = 0; i < *nr_zones; i++) { 254 struct blk_zone *zone_info; 255 256 zone_info = &zinfo->zone_cache[zno + i]; 257 if (!zone_info->len) 258 break; 259 } 260 261 if (i == *nr_zones) { 262 /* Cache hit on all the zones */ 263 memcpy(zones, zinfo->zone_cache + zno, 264 sizeof(*zinfo->zone_cache) * *nr_zones); 265 return 0; 266 } 267 } 268 269 ret = blkdev_report_zones_cached(device->bdev, pos >> SECTOR_SHIFT, 270 *nr_zones, copy_zone_info_cb, zones); 271 if (ret < 0) { 272 btrfs_err(device->fs_info, 273 "zoned: failed to read zone %llu on %s (devid %llu)", 274 pos, rcu_dereference(device->name), 275 device->devid); 276 return ret; 277 } 278 *nr_zones = ret; 279 if (unlikely(!ret)) 280 return -EIO; 281 282 /* Populate cache */ 283 if (zinfo->zone_cache) { 284 u32 zno = pos >> zinfo->zone_size_shift; 285 286 memcpy(zinfo->zone_cache + zno, zones, 287 sizeof(*zinfo->zone_cache) * *nr_zones); 288 } 289 290 return 0; 291} 292 293/* The emulated zone size is determined from the size of device extent */ 294static int calculate_emulated_zone_size(struct btrfs_fs_info *fs_info) 295{ 296 BTRFS_PATH_AUTO_FREE(path); 297 struct btrfs_root *root = fs_info->dev_root; 298 struct btrfs_key key; 299 struct extent_buffer *leaf; 300 struct btrfs_dev_extent *dext; 301 int ret = 0; 302 303 key.objectid = 1; 304 key.type = BTRFS_DEV_EXTENT_KEY; 305 key.offset = 0; 306 307 path = btrfs_alloc_path(); 308 if (!path) 309 return -ENOMEM; 310 311 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 312 if (ret < 0) 313 return ret; 314 315 if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) { 316 ret = btrfs_next_leaf(root, path); 317 if (ret < 0) 318 return ret; 319 /* No dev extents at all? Not good */ 320 if (unlikely(ret > 0)) 321 return -EUCLEAN; 322 } 323 324 leaf = path->nodes[0]; 325 dext = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent); 326 fs_info->zone_size = btrfs_dev_extent_length(leaf, dext); 327 return 0; 328} 329 330int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info) 331{ 332 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 333 struct btrfs_device *device; 334 int ret = 0; 335 336 /* fs_info->zone_size might not set yet. Use the incomapt flag here. */ 337 if (!btrfs_fs_incompat(fs_info, ZONED)) 338 return 0; 339 340 /* 341 * No need to take the device_list mutex here, we're still in the mount 342 * path and devices cannot be added to or removed from the list yet. 343 */ 344 list_for_each_entry(device, &fs_devices->devices, dev_list) { 345 /* We can skip reading of zone info for missing devices */ 346 if (!device->bdev) 347 continue; 348 349 ret = btrfs_get_dev_zone_info(device, true); 350 if (ret) 351 break; 352 } 353 354 return ret; 355} 356 357int btrfs_get_dev_zone_info(struct btrfs_device *device, bool populate_cache) 358{ 359 struct btrfs_fs_info *fs_info = device->fs_info; 360 struct btrfs_zoned_device_info *zone_info = NULL; 361 struct block_device *bdev = device->bdev; 362 unsigned int max_active_zones; 363 unsigned int nactive; 364 sector_t nr_sectors; 365 sector_t sector = 0; 366 struct blk_zone *zones = NULL; 367 unsigned int i, nreported = 0, nr_zones; 368 sector_t zone_sectors; 369 char *model, *emulated; 370 int ret; 371 372 /* 373 * Cannot use btrfs_is_zoned here, since fs_info::zone_size might not 374 * yet be set. 375 */ 376 if (!btrfs_fs_incompat(fs_info, ZONED)) 377 return 0; 378 379 if (device->zone_info) 380 return 0; 381 382 zone_info = kzalloc(sizeof(*zone_info), GFP_KERNEL); 383 if (!zone_info) 384 return -ENOMEM; 385 386 device->zone_info = zone_info; 387 388 if (!bdev_is_zoned(bdev)) { 389 if (!fs_info->zone_size) { 390 ret = calculate_emulated_zone_size(fs_info); 391 if (ret) 392 goto out; 393 } 394 395 ASSERT(fs_info->zone_size); 396 zone_sectors = fs_info->zone_size >> SECTOR_SHIFT; 397 } else { 398 zone_sectors = bdev_zone_sectors(bdev); 399 } 400 401 ASSERT(is_power_of_two_u64(zone_sectors)); 402 zone_info->zone_size = zone_sectors << SECTOR_SHIFT; 403 404 /* We reject devices with a zone size larger than 8GB */ 405 if (zone_info->zone_size > BTRFS_MAX_ZONE_SIZE) { 406 btrfs_err(fs_info, 407 "zoned: %s: zone size %llu larger than supported maximum %llu", 408 rcu_dereference(device->name), 409 zone_info->zone_size, BTRFS_MAX_ZONE_SIZE); 410 ret = -EINVAL; 411 goto out; 412 } else if (zone_info->zone_size < BTRFS_MIN_ZONE_SIZE) { 413 btrfs_err(fs_info, 414 "zoned: %s: zone size %llu smaller than supported minimum %u", 415 rcu_dereference(device->name), 416 zone_info->zone_size, BTRFS_MIN_ZONE_SIZE); 417 ret = -EINVAL; 418 goto out; 419 } 420 421 nr_sectors = bdev_nr_sectors(bdev); 422 zone_info->zone_size_shift = ilog2(zone_info->zone_size); 423 zone_info->nr_zones = nr_sectors >> ilog2(zone_sectors); 424 if (!IS_ALIGNED(nr_sectors, zone_sectors)) 425 zone_info->nr_zones++; 426 427 max_active_zones = min_not_zero(bdev_max_active_zones(bdev), 428 bdev_max_open_zones(bdev)); 429 if (!max_active_zones && zone_info->nr_zones > BTRFS_DEFAULT_MAX_ACTIVE_ZONES) 430 max_active_zones = BTRFS_DEFAULT_MAX_ACTIVE_ZONES; 431 if (max_active_zones && max_active_zones < BTRFS_MIN_ACTIVE_ZONES) { 432 btrfs_err(fs_info, 433"zoned: %s: max active zones %u is too small, need at least %u active zones", 434 rcu_dereference(device->name), max_active_zones, 435 BTRFS_MIN_ACTIVE_ZONES); 436 ret = -EINVAL; 437 goto out; 438 } 439 zone_info->max_active_zones = max_active_zones; 440 441 zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); 442 if (!zone_info->seq_zones) { 443 ret = -ENOMEM; 444 goto out; 445 } 446 447 zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); 448 if (!zone_info->empty_zones) { 449 ret = -ENOMEM; 450 goto out; 451 } 452 453 zone_info->active_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); 454 if (!zone_info->active_zones) { 455 ret = -ENOMEM; 456 goto out; 457 } 458 459 zones = kvcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL); 460 if (!zones) { 461 ret = -ENOMEM; 462 goto out; 463 } 464 465 /* 466 * Enable zone cache only for a zoned device. On a non-zoned device, we 467 * fill the zone info with emulated CONVENTIONAL zones, so no need to 468 * use the cache. 469 */ 470 if (populate_cache && bdev_is_zoned(device->bdev)) { 471 zone_info->zone_cache = vcalloc(zone_info->nr_zones, 472 sizeof(struct blk_zone)); 473 if (!zone_info->zone_cache) { 474 btrfs_err(device->fs_info, 475 "zoned: failed to allocate zone cache for %s", 476 rcu_dereference(device->name)); 477 ret = -ENOMEM; 478 goto out; 479 } 480 } 481 482 /* Get zones type */ 483 nactive = 0; 484 while (sector < nr_sectors) { 485 nr_zones = BTRFS_REPORT_NR_ZONES; 486 ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones, 487 &nr_zones); 488 if (ret) 489 goto out; 490 491 for (i = 0; i < nr_zones; i++) { 492 if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ) 493 __set_bit(nreported, zone_info->seq_zones); 494 switch (zones[i].cond) { 495 case BLK_ZONE_COND_EMPTY: 496 __set_bit(nreported, zone_info->empty_zones); 497 break; 498 case BLK_ZONE_COND_IMP_OPEN: 499 case BLK_ZONE_COND_EXP_OPEN: 500 case BLK_ZONE_COND_CLOSED: 501 case BLK_ZONE_COND_ACTIVE: 502 __set_bit(nreported, zone_info->active_zones); 503 nactive++; 504 break; 505 } 506 nreported++; 507 } 508 sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len; 509 } 510 511 if (unlikely(nreported != zone_info->nr_zones)) { 512 btrfs_err(device->fs_info, 513 "inconsistent number of zones on %s (%u/%u)", 514 rcu_dereference(device->name), nreported, 515 zone_info->nr_zones); 516 ret = -EIO; 517 goto out; 518 } 519 520 if (max_active_zones) { 521 if (unlikely(nactive > max_active_zones)) { 522 if (bdev_max_active_zones(bdev) == 0) { 523 max_active_zones = 0; 524 zone_info->max_active_zones = 0; 525 goto validate; 526 } 527 btrfs_err(device->fs_info, 528 "zoned: %u active zones on %s exceeds max_active_zones %u", 529 nactive, rcu_dereference(device->name), 530 max_active_zones); 531 ret = -EIO; 532 goto out; 533 } 534 atomic_set(&zone_info->active_zones_left, 535 max_active_zones - nactive); 536 set_bit(BTRFS_FS_ACTIVE_ZONE_TRACKING, &fs_info->flags); 537 } 538 539validate: 540 /* Validate superblock log */ 541 nr_zones = BTRFS_NR_SB_LOG_ZONES; 542 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { 543 u32 sb_zone; 544 u64 sb_wp; 545 int sb_pos = BTRFS_NR_SB_LOG_ZONES * i; 546 547 sb_zone = sb_zone_number(zone_info->zone_size_shift, i); 548 if (sb_zone + 1 >= zone_info->nr_zones) 549 continue; 550 551 ret = btrfs_get_dev_zones(device, 552 zone_start_physical(sb_zone, zone_info), 553 &zone_info->sb_zones[sb_pos], 554 &nr_zones); 555 if (ret) 556 goto out; 557 558 if (unlikely(nr_zones != BTRFS_NR_SB_LOG_ZONES)) { 559 btrfs_err(device->fs_info, 560 "zoned: failed to read super block log zone info at devid %llu zone %u", 561 device->devid, sb_zone); 562 ret = -EUCLEAN; 563 goto out; 564 } 565 566 /* 567 * If zones[0] is conventional, always use the beginning of the 568 * zone to record superblock. No need to validate in that case. 569 */ 570 if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type == 571 BLK_ZONE_TYPE_CONVENTIONAL) 572 continue; 573 574 ret = sb_write_pointer(device->bdev, 575 &zone_info->sb_zones[sb_pos], &sb_wp); 576 if (unlikely(ret != -ENOENT && ret)) { 577 btrfs_err(device->fs_info, 578 "zoned: super block log zone corrupted devid %llu zone %u", 579 device->devid, sb_zone); 580 ret = -EUCLEAN; 581 goto out; 582 } 583 } 584 585 586 kvfree(zones); 587 588 if (bdev_is_zoned(bdev)) { 589 model = "host-managed zoned"; 590 emulated = ""; 591 } else { 592 model = "regular"; 593 emulated = "emulated "; 594 } 595 596 btrfs_info(fs_info, 597 "%s block device %s, %u %szones of %llu bytes", 598 model, rcu_dereference(device->name), zone_info->nr_zones, 599 emulated, zone_info->zone_size); 600 601 return 0; 602 603out: 604 kvfree(zones); 605 btrfs_destroy_dev_zone_info(device); 606 return ret; 607} 608 609void btrfs_destroy_dev_zone_info(struct btrfs_device *device) 610{ 611 struct btrfs_zoned_device_info *zone_info = device->zone_info; 612 613 if (!zone_info) 614 return; 615 616 bitmap_free(zone_info->active_zones); 617 bitmap_free(zone_info->seq_zones); 618 bitmap_free(zone_info->empty_zones); 619 vfree(zone_info->zone_cache); 620 kfree(zone_info); 621 device->zone_info = NULL; 622} 623 624struct btrfs_zoned_device_info *btrfs_clone_dev_zone_info(struct btrfs_device *orig_dev) 625{ 626 struct btrfs_zoned_device_info *zone_info; 627 628 zone_info = kmemdup(orig_dev->zone_info, sizeof(*zone_info), GFP_KERNEL); 629 if (!zone_info) 630 return NULL; 631 632 zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); 633 if (!zone_info->seq_zones) 634 goto out; 635 636 bitmap_copy(zone_info->seq_zones, orig_dev->zone_info->seq_zones, 637 zone_info->nr_zones); 638 639 zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); 640 if (!zone_info->empty_zones) 641 goto out; 642 643 bitmap_copy(zone_info->empty_zones, orig_dev->zone_info->empty_zones, 644 zone_info->nr_zones); 645 646 zone_info->active_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL); 647 if (!zone_info->active_zones) 648 goto out; 649 650 bitmap_copy(zone_info->active_zones, orig_dev->zone_info->active_zones, 651 zone_info->nr_zones); 652 zone_info->zone_cache = NULL; 653 654 return zone_info; 655 656out: 657 bitmap_free(zone_info->seq_zones); 658 bitmap_free(zone_info->empty_zones); 659 bitmap_free(zone_info->active_zones); 660 kfree(zone_info); 661 return NULL; 662} 663 664static int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos, struct blk_zone *zone) 665{ 666 unsigned int nr_zones = 1; 667 int ret; 668 669 ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones); 670 if (ret != 0 || !nr_zones) 671 return ret ? ret : -EIO; 672 673 return 0; 674} 675 676static int btrfs_check_for_zoned_device(struct btrfs_fs_info *fs_info) 677{ 678 struct btrfs_device *device; 679 680 list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) { 681 if (device->bdev && bdev_is_zoned(device->bdev)) { 682 btrfs_err(fs_info, 683 "zoned: mode not enabled but zoned device found: %pg", 684 device->bdev); 685 return -EINVAL; 686 } 687 } 688 689 return 0; 690} 691 692int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info) 693{ 694 struct queue_limits *lim = &fs_info->limits; 695 struct btrfs_device *device; 696 u64 zone_size = 0; 697 int ret; 698 699 /* 700 * Host-Managed devices can't be used without the ZONED flag. With the 701 * ZONED all devices can be used, using zone emulation if required. 702 */ 703 if (!btrfs_fs_incompat(fs_info, ZONED)) 704 return btrfs_check_for_zoned_device(fs_info); 705 706 blk_set_stacking_limits(lim); 707 708 list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) { 709 struct btrfs_zoned_device_info *zone_info = device->zone_info; 710 711 if (!device->bdev) 712 continue; 713 714 if (!zone_size) { 715 zone_size = zone_info->zone_size; 716 } else if (zone_info->zone_size != zone_size) { 717 btrfs_err(fs_info, 718 "zoned: unequal block device zone sizes: have %llu found %llu", 719 zone_info->zone_size, zone_size); 720 return -EINVAL; 721 } 722 723 /* 724 * With the zoned emulation, we can have non-zoned device on the 725 * zoned mode. In this case, we don't have a valid max zone 726 * append size. 727 */ 728 if (bdev_is_zoned(device->bdev)) 729 blk_stack_limits(lim, bdev_limits(device->bdev), 0); 730 } 731 732 ret = blk_validate_limits(lim); 733 if (ret) { 734 btrfs_err(fs_info, "zoned: failed to validate queue limits"); 735 return ret; 736 } 737 738 /* 739 * stripe_size is always aligned to BTRFS_STRIPE_LEN in 740 * btrfs_create_chunk(). Since we want stripe_len == zone_size, 741 * check the alignment here. 742 */ 743 if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) { 744 btrfs_err(fs_info, 745 "zoned: zone size %llu not aligned to stripe %u", 746 zone_size, BTRFS_STRIPE_LEN); 747 return -EINVAL; 748 } 749 750 if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) { 751 btrfs_err(fs_info, "zoned: mixed block groups not supported"); 752 return -EINVAL; 753 } 754 755 fs_info->zone_size = zone_size; 756 /* 757 * Also limit max_zone_append_size by max_segments * PAGE_SIZE. 758 * Technically, we can have multiple pages per segment. But, since 759 * we add the pages one by one to a bio, and cannot increase the 760 * metadata reservation even if it increases the number of extents, it 761 * is safe to stick with the limit. 762 */ 763 fs_info->max_zone_append_size = ALIGN_DOWN( 764 min3((u64)lim->max_zone_append_sectors << SECTOR_SHIFT, 765 (u64)lim->max_sectors << SECTOR_SHIFT, 766 (u64)lim->max_segments << PAGE_SHIFT), 767 fs_info->sectorsize); 768 fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED; 769 770 fs_info->max_extent_size = min_not_zero(fs_info->max_extent_size, 771 fs_info->max_zone_append_size); 772 773 /* 774 * Check mount options here, because we might change fs_info->zoned 775 * from fs_info->zone_size. 776 */ 777 ret = btrfs_check_mountopts_zoned(fs_info, &fs_info->mount_opt); 778 if (ret) 779 return ret; 780 781 btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size); 782 return 0; 783} 784 785int btrfs_check_mountopts_zoned(const struct btrfs_fs_info *info, 786 unsigned long long *mount_opt) 787{ 788 if (!btrfs_is_zoned(info)) 789 return 0; 790 791 /* 792 * Space cache writing is not COWed. Disable that to avoid write errors 793 * in sequential zones. 794 */ 795 if (btrfs_raw_test_opt(*mount_opt, SPACE_CACHE)) { 796 btrfs_err(info, "zoned: space cache v1 is not supported"); 797 return -EINVAL; 798 } 799 800 if (btrfs_raw_test_opt(*mount_opt, NODATACOW)) { 801 btrfs_err(info, "zoned: NODATACOW not supported"); 802 return -EINVAL; 803 } 804 805 if (btrfs_raw_test_opt(*mount_opt, DISCARD_ASYNC)) { 806 btrfs_info(info, 807 "zoned: async discard ignored and disabled for zoned mode"); 808 btrfs_clear_opt(*mount_opt, DISCARD_ASYNC); 809 } 810 811 return 0; 812} 813 814static int sb_log_location(struct block_device *bdev, struct blk_zone *zones, 815 int rw, u64 *bytenr_ret) 816{ 817 u64 wp; 818 int ret; 819 820 if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) { 821 *bytenr_ret = zones[0].start << SECTOR_SHIFT; 822 return 0; 823 } 824 825 ret = sb_write_pointer(bdev, zones, &wp); 826 if (ret != -ENOENT && ret < 0) 827 return ret; 828 829 if (rw == WRITE) { 830 struct blk_zone *reset = NULL; 831 832 if (wp == zones[0].start << SECTOR_SHIFT) 833 reset = &zones[0]; 834 else if (wp == zones[1].start << SECTOR_SHIFT) 835 reset = &zones[1]; 836 837 if (reset && reset->cond != BLK_ZONE_COND_EMPTY) { 838 unsigned int nofs_flags; 839 840 ASSERT(sb_zone_is_full(reset)); 841 842 nofs_flags = memalloc_nofs_save(); 843 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 844 reset->start, reset->len); 845 memalloc_nofs_restore(nofs_flags); 846 if (ret) 847 return ret; 848 849 reset->cond = BLK_ZONE_COND_EMPTY; 850 reset->wp = reset->start; 851 } 852 } else if (ret != -ENOENT) { 853 /* 854 * For READ, we want the previous one. Move write pointer to 855 * the end of a zone, if it is at the head of a zone. 856 */ 857 u64 zone_end = 0; 858 859 if (wp == zones[0].start << SECTOR_SHIFT) 860 zone_end = zones[1].start + zones[1].capacity; 861 else if (wp == zones[1].start << SECTOR_SHIFT) 862 zone_end = zones[0].start + zones[0].capacity; 863 if (zone_end) 864 wp = ALIGN_DOWN(zone_end << SECTOR_SHIFT, 865 BTRFS_SUPER_INFO_SIZE); 866 867 wp -= BTRFS_SUPER_INFO_SIZE; 868 } 869 870 *bytenr_ret = wp; 871 return 0; 872 873} 874 875int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw, 876 u64 *bytenr_ret) 877{ 878 struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES]; 879 sector_t zone_sectors; 880 u32 sb_zone; 881 int ret; 882 u8 zone_sectors_shift; 883 sector_t nr_sectors; 884 u32 nr_zones; 885 886 if (!bdev_is_zoned(bdev)) { 887 *bytenr_ret = btrfs_sb_offset(mirror); 888 return 0; 889 } 890 891 ASSERT(rw == READ || rw == WRITE); 892 893 zone_sectors = bdev_zone_sectors(bdev); 894 if (!is_power_of_2(zone_sectors)) 895 return -EINVAL; 896 zone_sectors_shift = ilog2(zone_sectors); 897 nr_sectors = bdev_nr_sectors(bdev); 898 nr_zones = nr_sectors >> zone_sectors_shift; 899 900 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror); 901 if (sb_zone + 1 >= nr_zones) 902 return -ENOENT; 903 904 ret = blkdev_report_zones_cached(bdev, zone_start_sector(sb_zone, bdev), 905 BTRFS_NR_SB_LOG_ZONES, 906 copy_zone_info_cb, zones); 907 if (ret < 0) 908 return ret; 909 if (unlikely(ret != BTRFS_NR_SB_LOG_ZONES)) 910 return -EIO; 911 912 return sb_log_location(bdev, zones, rw, bytenr_ret); 913} 914 915int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw, 916 u64 *bytenr_ret) 917{ 918 struct btrfs_zoned_device_info *zinfo = device->zone_info; 919 u32 zone_num; 920 921 /* 922 * For a zoned filesystem on a non-zoned block device, use the same 923 * super block locations as regular filesystem. Doing so, the super 924 * block can always be retrieved and the zoned flag of the volume 925 * detected from the super block information. 926 */ 927 if (!bdev_is_zoned(device->bdev)) { 928 *bytenr_ret = btrfs_sb_offset(mirror); 929 return 0; 930 } 931 932 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror); 933 if (zone_num + 1 >= zinfo->nr_zones) 934 return -ENOENT; 935 936 return sb_log_location(device->bdev, 937 &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror], 938 rw, bytenr_ret); 939} 940 941static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo, 942 int mirror) 943{ 944 u32 zone_num; 945 946 if (!zinfo) 947 return false; 948 949 zone_num = sb_zone_number(zinfo->zone_size_shift, mirror); 950 if (zone_num + 1 >= zinfo->nr_zones) 951 return false; 952 953 if (!test_bit(zone_num, zinfo->seq_zones)) 954 return false; 955 956 return true; 957} 958 959int btrfs_advance_sb_log(struct btrfs_device *device, int mirror) 960{ 961 struct btrfs_zoned_device_info *zinfo = device->zone_info; 962 struct blk_zone *zone; 963 int i; 964 965 if (!is_sb_log_zone(zinfo, mirror)) 966 return 0; 967 968 zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror]; 969 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) { 970 /* Advance the next zone */ 971 if (zone->cond == BLK_ZONE_COND_FULL) { 972 zone++; 973 continue; 974 } 975 976 if (zone->cond == BLK_ZONE_COND_EMPTY) 977 zone->cond = BLK_ZONE_COND_IMP_OPEN; 978 979 zone->wp += SUPER_INFO_SECTORS; 980 981 if (sb_zone_is_full(zone)) { 982 /* 983 * No room left to write new superblock. Since 984 * superblock is written with REQ_SYNC, it is safe to 985 * finish the zone now. 986 * 987 * If the write pointer is exactly at the capacity, 988 * explicit ZONE_FINISH is not necessary. 989 */ 990 if (zone->wp != zone->start + zone->capacity) { 991 unsigned int nofs_flags; 992 int ret; 993 994 nofs_flags = memalloc_nofs_save(); 995 ret = blkdev_zone_mgmt(device->bdev, 996 REQ_OP_ZONE_FINISH, zone->start, 997 zone->len); 998 memalloc_nofs_restore(nofs_flags); 999 if (ret) 1000 return ret; 1001 } 1002 1003 zone->wp = zone->start + zone->len; 1004 zone->cond = BLK_ZONE_COND_FULL; 1005 } 1006 return 0; 1007 } 1008 1009 /* All the zones are FULL. Should not reach here. */ 1010 DEBUG_WARN("unexpected state, all zones full"); 1011 return -EIO; 1012} 1013 1014int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror) 1015{ 1016 unsigned int nofs_flags; 1017 sector_t zone_sectors; 1018 sector_t nr_sectors; 1019 u8 zone_sectors_shift; 1020 u32 sb_zone; 1021 u32 nr_zones; 1022 int ret; 1023 1024 zone_sectors = bdev_zone_sectors(bdev); 1025 zone_sectors_shift = ilog2(zone_sectors); 1026 nr_sectors = bdev_nr_sectors(bdev); 1027 nr_zones = nr_sectors >> zone_sectors_shift; 1028 1029 sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror); 1030 if (sb_zone + 1 >= nr_zones) 1031 return -ENOENT; 1032 1033 nofs_flags = memalloc_nofs_save(); 1034 ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET, 1035 zone_start_sector(sb_zone, bdev), 1036 zone_sectors * BTRFS_NR_SB_LOG_ZONES); 1037 memalloc_nofs_restore(nofs_flags); 1038 return ret; 1039} 1040 1041/* 1042 * Find allocatable zones within a given region. 1043 * 1044 * @device: the device to allocate a region on 1045 * @hole_start: the position of the hole to allocate the region 1046 * @num_bytes: size of wanted region 1047 * @hole_end: the end of the hole 1048 * @return: position of allocatable zones 1049 * 1050 * Allocatable region should not contain any superblock locations. 1051 */ 1052u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start, 1053 u64 hole_end, u64 num_bytes) 1054{ 1055 struct btrfs_zoned_device_info *zinfo = device->zone_info; 1056 const u8 shift = zinfo->zone_size_shift; 1057 u64 nzones = num_bytes >> shift; 1058 u64 pos = hole_start; 1059 u64 begin, end; 1060 bool have_sb; 1061 int i; 1062 1063 ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size), 1064 "hole_start=%llu zinfo->zone_size=%llu", hole_start, zinfo->zone_size); 1065 ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size), 1066 "num_bytes=%llu zinfo->zone_size=%llu", num_bytes, zinfo->zone_size); 1067 1068 while (pos < hole_end) { 1069 begin = pos >> shift; 1070 end = begin + nzones; 1071 1072 if (end > zinfo->nr_zones) 1073 return hole_end; 1074 1075 /* Check if zones in the region are all empty */ 1076 if (btrfs_dev_is_sequential(device, pos) && 1077 !bitmap_test_range_all_set(zinfo->empty_zones, begin, nzones)) { 1078 pos += zinfo->zone_size; 1079 continue; 1080 } 1081 1082 have_sb = false; 1083 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) { 1084 u32 sb_zone; 1085 u64 sb_pos; 1086 1087 sb_zone = sb_zone_number(shift, i); 1088 if (!(end <= sb_zone || 1089 sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) { 1090 have_sb = true; 1091 pos = zone_start_physical( 1092 sb_zone + BTRFS_NR_SB_LOG_ZONES, zinfo); 1093 break; 1094 } 1095 1096 /* We also need to exclude regular superblock positions */ 1097 sb_pos = btrfs_sb_offset(i); 1098 if (!(pos + num_bytes <= sb_pos || 1099 sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) { 1100 have_sb = true; 1101 pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE, 1102 zinfo->zone_size); 1103 break; 1104 } 1105 } 1106 if (!have_sb) 1107 break; 1108 } 1109 1110 return pos; 1111} 1112 1113static bool btrfs_dev_set_active_zone(struct btrfs_device *device, u64 pos) 1114{ 1115 struct btrfs_zoned_device_info *zone_info = device->zone_info; 1116 unsigned int zno = (pos >> zone_info->zone_size_shift); 1117 1118 /* We can use any number of zones */ 1119 if (zone_info->max_active_zones == 0) 1120 return true; 1121 1122 if (!test_bit(zno, zone_info->active_zones)) { 1123 /* Active zone left? */ 1124 if (atomic_dec_if_positive(&zone_info->active_zones_left) < 0) 1125 return false; 1126 if (test_and_set_bit(zno, zone_info->active_zones)) { 1127 /* Someone already set the bit */ 1128 atomic_inc(&zone_info->active_zones_left); 1129 } 1130 } 1131 1132 return true; 1133} 1134 1135static void btrfs_dev_clear_active_zone(struct btrfs_device *device, u64 pos) 1136{ 1137 struct btrfs_zoned_device_info *zone_info = device->zone_info; 1138 unsigned int zno = (pos >> zone_info->zone_size_shift); 1139 1140 /* We can use any number of zones */ 1141 if (zone_info->max_active_zones == 0) 1142 return; 1143 1144 if (test_and_clear_bit(zno, zone_info->active_zones)) 1145 atomic_inc(&zone_info->active_zones_left); 1146} 1147 1148int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical, 1149 u64 length, u64 *bytes) 1150{ 1151 unsigned int nofs_flags; 1152 int ret; 1153 1154 *bytes = 0; 1155 nofs_flags = memalloc_nofs_save(); 1156 ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET, 1157 physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT); 1158 memalloc_nofs_restore(nofs_flags); 1159 if (ret) 1160 return ret; 1161 1162 *bytes = length; 1163 while (length) { 1164 btrfs_dev_set_zone_empty(device, physical); 1165 btrfs_dev_clear_active_zone(device, physical); 1166 physical += device->zone_info->zone_size; 1167 length -= device->zone_info->zone_size; 1168 } 1169 1170 return 0; 1171} 1172 1173int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size) 1174{ 1175 struct btrfs_zoned_device_info *zinfo = device->zone_info; 1176 const u8 shift = zinfo->zone_size_shift; 1177 unsigned long begin = start >> shift; 1178 unsigned long nbits = size >> shift; 1179 u64 pos; 1180 int ret; 1181 1182 ASSERT(IS_ALIGNED(start, zinfo->zone_size), 1183 "start=%llu, zinfo->zone_size=%llu", start, zinfo->zone_size); 1184 ASSERT(IS_ALIGNED(size, zinfo->zone_size), 1185 "size=%llu, zinfo->zone_size=%llu", size, zinfo->zone_size); 1186 1187 if (begin + nbits > zinfo->nr_zones) 1188 return -ERANGE; 1189 1190 /* All the zones are conventional */ 1191 if (bitmap_test_range_all_zero(zinfo->seq_zones, begin, nbits)) 1192 return 0; 1193 1194 /* All the zones are sequential and empty */ 1195 if (bitmap_test_range_all_set(zinfo->seq_zones, begin, nbits) && 1196 bitmap_test_range_all_set(zinfo->empty_zones, begin, nbits)) 1197 return 0; 1198 1199 for (pos = start; pos < start + size; pos += zinfo->zone_size) { 1200 u64 reset_bytes; 1201 1202 if (!btrfs_dev_is_sequential(device, pos) || 1203 btrfs_dev_is_empty_zone(device, pos)) 1204 continue; 1205 1206 /* Free regions should be empty */ 1207 btrfs_warn( 1208 device->fs_info, 1209 "zoned: resetting device %s (devid %llu) zone %llu for allocation", 1210 rcu_dereference(device->name), device->devid, pos >> shift); 1211 WARN_ON_ONCE(1); 1212 1213 ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size, 1214 &reset_bytes); 1215 if (ret) 1216 return ret; 1217 } 1218 1219 return 0; 1220} 1221 1222/* 1223 * Calculate an allocation pointer from the extent allocation information 1224 * for a block group consist of conventional zones. It is pointed to the 1225 * end of the highest addressed extent in the block group as an allocation 1226 * offset. 1227 */ 1228static int calculate_alloc_pointer(struct btrfs_block_group *cache, 1229 u64 *offset_ret, bool new) 1230{ 1231 struct btrfs_fs_info *fs_info = cache->fs_info; 1232 struct btrfs_root *root; 1233 BTRFS_PATH_AUTO_FREE(path); 1234 struct btrfs_key key; 1235 struct btrfs_key found_key; 1236 int ret; 1237 u64 length; 1238 1239 /* 1240 * Avoid tree lookups for a new block group, there's no use for it. 1241 * It must always be 0. 1242 * 1243 * Also, we have a lock chain of extent buffer lock -> chunk mutex. 1244 * For new a block group, this function is called from 1245 * btrfs_make_block_group() which is already taking the chunk mutex. 1246 * Thus, we cannot call calculate_alloc_pointer() which takes extent 1247 * buffer locks to avoid deadlock. 1248 */ 1249 if (new) { 1250 *offset_ret = 0; 1251 return 0; 1252 } 1253 1254 path = btrfs_alloc_path(); 1255 if (!path) 1256 return -ENOMEM; 1257 1258 key.objectid = cache->start + cache->length; 1259 key.type = 0; 1260 key.offset = 0; 1261 1262 root = btrfs_extent_root(fs_info, key.objectid); 1263 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 1264 /* We should not find the exact match */ 1265 if (unlikely(!ret)) 1266 ret = -EUCLEAN; 1267 if (ret < 0) 1268 return ret; 1269 1270 ret = btrfs_previous_extent_item(root, path, cache->start); 1271 if (ret) { 1272 if (ret == 1) { 1273 ret = 0; 1274 *offset_ret = 0; 1275 } 1276 return ret; 1277 } 1278 1279 btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]); 1280 1281 if (found_key.type == BTRFS_EXTENT_ITEM_KEY) 1282 length = found_key.offset; 1283 else 1284 length = fs_info->nodesize; 1285 1286 if (unlikely(!(found_key.objectid >= cache->start && 1287 found_key.objectid + length <= cache->start + cache->length))) { 1288 return -EUCLEAN; 1289 } 1290 *offset_ret = found_key.objectid + length - cache->start; 1291 return 0; 1292} 1293 1294struct zone_info { 1295 u64 physical; 1296 u64 capacity; 1297 u64 alloc_offset; 1298}; 1299 1300static int btrfs_load_zone_info(struct btrfs_fs_info *fs_info, int zone_idx, 1301 struct zone_info *info, unsigned long *active, 1302 struct btrfs_chunk_map *map, bool new) 1303{ 1304 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 1305 struct btrfs_device *device; 1306 int dev_replace_is_ongoing = 0; 1307 unsigned int nofs_flag; 1308 struct blk_zone zone; 1309 int ret; 1310 1311 info->physical = map->stripes[zone_idx].physical; 1312 1313 down_read(&dev_replace->rwsem); 1314 device = map->stripes[zone_idx].dev; 1315 1316 if (!device->bdev) { 1317 up_read(&dev_replace->rwsem); 1318 info->alloc_offset = WP_MISSING_DEV; 1319 return 0; 1320 } 1321 1322 /* Consider a zone as active if we can allow any number of active zones. */ 1323 if (!device->zone_info->max_active_zones) 1324 __set_bit(zone_idx, active); 1325 1326 if (!btrfs_dev_is_sequential(device, info->physical)) { 1327 up_read(&dev_replace->rwsem); 1328 info->alloc_offset = WP_CONVENTIONAL; 1329 info->capacity = device->zone_info->zone_size; 1330 return 0; 1331 } 1332 1333 ASSERT(!new || btrfs_dev_is_empty_zone(device, info->physical)); 1334 1335 /* This zone will be used for allocation, so mark this zone non-empty. */ 1336 btrfs_dev_clear_zone_empty(device, info->physical); 1337 1338 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); 1339 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL) 1340 btrfs_dev_clear_zone_empty(dev_replace->tgtdev, info->physical); 1341 1342 /* 1343 * The group is mapped to a sequential zone. Get the zone write pointer 1344 * to determine the allocation offset within the zone. 1345 */ 1346 WARN_ON(!IS_ALIGNED(info->physical, fs_info->zone_size)); 1347 1348 if (new) { 1349 sector_t capacity; 1350 1351 capacity = bdev_zone_capacity(device->bdev, info->physical >> SECTOR_SHIFT); 1352 up_read(&dev_replace->rwsem); 1353 info->alloc_offset = 0; 1354 info->capacity = capacity << SECTOR_SHIFT; 1355 1356 return 0; 1357 } 1358 1359 nofs_flag = memalloc_nofs_save(); 1360 ret = btrfs_get_dev_zone(device, info->physical, &zone); 1361 memalloc_nofs_restore(nofs_flag); 1362 if (ret) { 1363 up_read(&dev_replace->rwsem); 1364 if (ret != -EIO && ret != -EOPNOTSUPP) 1365 return ret; 1366 info->alloc_offset = WP_MISSING_DEV; 1367 return 0; 1368 } 1369 1370 if (unlikely(zone.type == BLK_ZONE_TYPE_CONVENTIONAL)) { 1371 btrfs_err(fs_info, 1372 "zoned: unexpected conventional zone %llu on device %s (devid %llu)", 1373 zone.start << SECTOR_SHIFT, rcu_dereference(device->name), 1374 device->devid); 1375 up_read(&dev_replace->rwsem); 1376 return -EIO; 1377 } 1378 1379 info->capacity = (zone.capacity << SECTOR_SHIFT); 1380 1381 switch (zone.cond) { 1382 case BLK_ZONE_COND_OFFLINE: 1383 case BLK_ZONE_COND_READONLY: 1384 btrfs_err(fs_info, 1385 "zoned: offline/readonly zone %llu on device %s (devid %llu)", 1386 (info->physical >> device->zone_info->zone_size_shift), 1387 rcu_dereference(device->name), device->devid); 1388 info->alloc_offset = WP_MISSING_DEV; 1389 break; 1390 case BLK_ZONE_COND_EMPTY: 1391 info->alloc_offset = 0; 1392 break; 1393 case BLK_ZONE_COND_FULL: 1394 info->alloc_offset = info->capacity; 1395 break; 1396 default: 1397 /* Partially used zone. */ 1398 info->alloc_offset = ((zone.wp - zone.start) << SECTOR_SHIFT); 1399 __set_bit(zone_idx, active); 1400 break; 1401 } 1402 1403 up_read(&dev_replace->rwsem); 1404 1405 return 0; 1406} 1407 1408static int btrfs_load_block_group_single(struct btrfs_block_group *bg, 1409 struct zone_info *info, 1410 unsigned long *active) 1411{ 1412 if (unlikely(info->alloc_offset == WP_MISSING_DEV)) { 1413 btrfs_err(bg->fs_info, 1414 "zoned: cannot recover write pointer for zone %llu", 1415 info->physical); 1416 return -EIO; 1417 } 1418 1419 bg->alloc_offset = info->alloc_offset; 1420 bg->zone_capacity = info->capacity; 1421 if (test_bit(0, active)) 1422 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &bg->runtime_flags); 1423 return 0; 1424} 1425 1426static int btrfs_load_block_group_dup(struct btrfs_block_group *bg, 1427 struct btrfs_chunk_map *map, 1428 struct zone_info *zone_info, 1429 unsigned long *active, 1430 u64 last_alloc) 1431{ 1432 struct btrfs_fs_info *fs_info = bg->fs_info; 1433 1434 if ((map->type & BTRFS_BLOCK_GROUP_DATA) && !fs_info->stripe_root) { 1435 btrfs_err(fs_info, "zoned: data DUP profile needs raid-stripe-tree"); 1436 return -EINVAL; 1437 } 1438 1439 bg->zone_capacity = min_not_zero(zone_info[0].capacity, zone_info[1].capacity); 1440 1441 if (unlikely(zone_info[0].alloc_offset == WP_MISSING_DEV)) { 1442 btrfs_err(bg->fs_info, 1443 "zoned: cannot recover write pointer for zone %llu", 1444 zone_info[0].physical); 1445 return -EIO; 1446 } 1447 if (unlikely(zone_info[1].alloc_offset == WP_MISSING_DEV)) { 1448 btrfs_err(bg->fs_info, 1449 "zoned: cannot recover write pointer for zone %llu", 1450 zone_info[1].physical); 1451 return -EIO; 1452 } 1453 1454 /* 1455 * When the last extent is removed, last_alloc can be smaller than the other write 1456 * pointer. In that case, last_alloc should be moved to the corresponding write 1457 * pointer position. 1458 */ 1459 for (int i = 0; i < map->num_stripes; i++) { 1460 if (zone_info[i].alloc_offset == WP_CONVENTIONAL) 1461 continue; 1462 if (last_alloc <= zone_info[i].alloc_offset) { 1463 last_alloc = zone_info[i].alloc_offset; 1464 break; 1465 } 1466 } 1467 1468 if (zone_info[0].alloc_offset == WP_CONVENTIONAL) 1469 zone_info[0].alloc_offset = last_alloc; 1470 1471 if (zone_info[1].alloc_offset == WP_CONVENTIONAL) 1472 zone_info[1].alloc_offset = last_alloc; 1473 1474 if (unlikely(zone_info[0].alloc_offset != zone_info[1].alloc_offset)) { 1475 btrfs_err(bg->fs_info, 1476 "zoned: write pointer offset mismatch of zones in DUP profile"); 1477 return -EIO; 1478 } 1479 1480 if (test_bit(0, active) != test_bit(1, active)) { 1481 if (unlikely(!btrfs_zone_activate(bg))) 1482 return -EIO; 1483 } else if (test_bit(0, active)) { 1484 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &bg->runtime_flags); 1485 } 1486 1487 bg->alloc_offset = zone_info[0].alloc_offset; 1488 return 0; 1489} 1490 1491static int btrfs_load_block_group_raid1(struct btrfs_block_group *bg, 1492 struct btrfs_chunk_map *map, 1493 struct zone_info *zone_info, 1494 unsigned long *active, 1495 u64 last_alloc) 1496{ 1497 struct btrfs_fs_info *fs_info = bg->fs_info; 1498 int i; 1499 1500 if ((map->type & BTRFS_BLOCK_GROUP_DATA) && !fs_info->stripe_root) { 1501 btrfs_err(fs_info, "zoned: data %s needs raid-stripe-tree", 1502 btrfs_bg_type_to_raid_name(map->type)); 1503 return -EINVAL; 1504 } 1505 1506 /* In case a device is missing we have a cap of 0, so don't use it. */ 1507 bg->zone_capacity = min_not_zero(zone_info[0].capacity, zone_info[1].capacity); 1508 1509 /* 1510 * When the last extent is removed, last_alloc can be smaller than the other write 1511 * pointer. In that case, last_alloc should be moved to the corresponding write 1512 * pointer position. 1513 */ 1514 for (i = 0; i < map->num_stripes; i++) { 1515 if (zone_info[i].alloc_offset == WP_MISSING_DEV || 1516 zone_info[i].alloc_offset == WP_CONVENTIONAL) 1517 continue; 1518 if (last_alloc <= zone_info[i].alloc_offset) { 1519 last_alloc = zone_info[i].alloc_offset; 1520 break; 1521 } 1522 } 1523 1524 for (i = 0; i < map->num_stripes; i++) { 1525 if (zone_info[i].alloc_offset == WP_MISSING_DEV) 1526 continue; 1527 1528 if (zone_info[i].alloc_offset == WP_CONVENTIONAL) 1529 zone_info[i].alloc_offset = last_alloc; 1530 1531 if (unlikely((zone_info[0].alloc_offset != zone_info[i].alloc_offset) && 1532 !btrfs_test_opt(fs_info, DEGRADED))) { 1533 btrfs_err(fs_info, 1534 "zoned: write pointer offset mismatch of zones in %s profile", 1535 btrfs_bg_type_to_raid_name(map->type)); 1536 return -EIO; 1537 } 1538 if (test_bit(0, active) != test_bit(i, active)) { 1539 if (unlikely(!btrfs_test_opt(fs_info, DEGRADED) && 1540 !btrfs_zone_activate(bg))) { 1541 return -EIO; 1542 } 1543 } else { 1544 if (test_bit(0, active)) 1545 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &bg->runtime_flags); 1546 } 1547 } 1548 1549 if (zone_info[0].alloc_offset != WP_MISSING_DEV) 1550 bg->alloc_offset = zone_info[0].alloc_offset; 1551 else 1552 bg->alloc_offset = zone_info[i - 1].alloc_offset; 1553 1554 return 0; 1555} 1556 1557static int btrfs_load_block_group_raid0(struct btrfs_block_group *bg, 1558 struct btrfs_chunk_map *map, 1559 struct zone_info *zone_info, 1560 unsigned long *active, 1561 u64 last_alloc) 1562{ 1563 struct btrfs_fs_info *fs_info = bg->fs_info; 1564 u64 stripe_nr = 0, stripe_offset = 0; 1565 u64 prev_offset = 0; 1566 u32 stripe_index = 0; 1567 bool has_partial = false, has_conventional = false; 1568 1569 if ((map->type & BTRFS_BLOCK_GROUP_DATA) && !fs_info->stripe_root) { 1570 btrfs_err(fs_info, "zoned: data %s needs raid-stripe-tree", 1571 btrfs_bg_type_to_raid_name(map->type)); 1572 return -EINVAL; 1573 } 1574 1575 /* 1576 * When the last extent is removed, last_alloc can be smaller than the other write 1577 * pointer. In that case, last_alloc should be moved to the corresponding write 1578 * pointer position. 1579 */ 1580 for (int i = 0; i < map->num_stripes; i++) { 1581 u64 alloc; 1582 1583 if (zone_info[i].alloc_offset == WP_MISSING_DEV || 1584 zone_info[i].alloc_offset == WP_CONVENTIONAL) 1585 continue; 1586 1587 stripe_nr = zone_info[i].alloc_offset >> BTRFS_STRIPE_LEN_SHIFT; 1588 stripe_offset = zone_info[i].alloc_offset & BTRFS_STRIPE_LEN_MASK; 1589 if (stripe_offset == 0 && stripe_nr > 0) { 1590 stripe_nr--; 1591 stripe_offset = BTRFS_STRIPE_LEN; 1592 } 1593 alloc = ((stripe_nr * map->num_stripes + i) << BTRFS_STRIPE_LEN_SHIFT) + 1594 stripe_offset; 1595 last_alloc = max(last_alloc, alloc); 1596 1597 /* Partially written stripe found. It should be last. */ 1598 if (zone_info[i].alloc_offset & BTRFS_STRIPE_LEN_MASK) 1599 break; 1600 } 1601 stripe_nr = 0; 1602 stripe_offset = 0; 1603 1604 if (last_alloc) { 1605 u32 factor = map->num_stripes; 1606 1607 stripe_nr = last_alloc >> BTRFS_STRIPE_LEN_SHIFT; 1608 stripe_offset = last_alloc & BTRFS_STRIPE_LEN_MASK; 1609 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 1610 } 1611 1612 for (int i = 0; i < map->num_stripes; i++) { 1613 if (zone_info[i].alloc_offset == WP_MISSING_DEV) 1614 continue; 1615 1616 if (zone_info[i].alloc_offset == WP_CONVENTIONAL) { 1617 has_conventional = true; 1618 zone_info[i].alloc_offset = btrfs_stripe_nr_to_offset(stripe_nr); 1619 1620 if (stripe_index > i) 1621 zone_info[i].alloc_offset += BTRFS_STRIPE_LEN; 1622 else if (stripe_index == i) 1623 zone_info[i].alloc_offset += stripe_offset; 1624 } 1625 1626 /* Verification */ 1627 if (i != 0) { 1628 if (unlikely(prev_offset < zone_info[i].alloc_offset)) { 1629 btrfs_err(fs_info, 1630 "zoned: stripe position disorder found in block group %llu", 1631 bg->start); 1632 return -EIO; 1633 } 1634 1635 if (unlikely(has_partial && 1636 (zone_info[i].alloc_offset & BTRFS_STRIPE_LEN_MASK))) { 1637 btrfs_err(fs_info, 1638 "zoned: multiple partial written stripe found in block group %llu", 1639 bg->start); 1640 return -EIO; 1641 } 1642 } 1643 prev_offset = zone_info[i].alloc_offset; 1644 1645 if ((zone_info[i].alloc_offset & BTRFS_STRIPE_LEN_MASK) != 0) 1646 has_partial = true; 1647 1648 if (test_bit(0, active) != test_bit(i, active)) { 1649 if (unlikely(!btrfs_zone_activate(bg))) 1650 return -EIO; 1651 } else { 1652 if (test_bit(0, active)) 1653 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &bg->runtime_flags); 1654 } 1655 bg->zone_capacity += zone_info[i].capacity; 1656 bg->alloc_offset += zone_info[i].alloc_offset; 1657 } 1658 1659 /* Check if all devices stay in the same stripe row. */ 1660 if (unlikely(zone_info[0].alloc_offset - 1661 zone_info[map->num_stripes - 1].alloc_offset > BTRFS_STRIPE_LEN)) { 1662 btrfs_err(fs_info, "zoned: stripe gap too large in block group %llu", bg->start); 1663 return -EIO; 1664 } 1665 1666 if (unlikely(has_conventional && bg->alloc_offset < last_alloc)) { 1667 btrfs_err(fs_info, "zoned: allocated extent stays beyond write pointers %llu %llu", 1668 bg->alloc_offset, last_alloc); 1669 return -EIO; 1670 } 1671 1672 return 0; 1673} 1674 1675static int btrfs_load_block_group_raid10(struct btrfs_block_group *bg, 1676 struct btrfs_chunk_map *map, 1677 struct zone_info *zone_info, 1678 unsigned long *active, 1679 u64 last_alloc) 1680{ 1681 struct btrfs_fs_info *fs_info = bg->fs_info; 1682 u64 AUTO_KFREE(raid0_allocs); 1683 u64 stripe_nr = 0, stripe_offset = 0; 1684 u32 stripe_index = 0; 1685 bool has_partial = false, has_conventional = false; 1686 u64 prev_offset = 0; 1687 1688 if ((map->type & BTRFS_BLOCK_GROUP_DATA) && !fs_info->stripe_root) { 1689 btrfs_err(fs_info, "zoned: data %s needs raid-stripe-tree", 1690 btrfs_bg_type_to_raid_name(map->type)); 1691 return -EINVAL; 1692 } 1693 1694 raid0_allocs = kcalloc(map->num_stripes / map->sub_stripes, sizeof(*raid0_allocs), 1695 GFP_NOFS); 1696 if (!raid0_allocs) 1697 return -ENOMEM; 1698 1699 /* 1700 * When the last extent is removed, last_alloc can be smaller than the other write 1701 * pointer. In that case, last_alloc should be moved to the corresponding write 1702 * pointer position. 1703 */ 1704 for (int i = 0; i < map->num_stripes; i += map->sub_stripes) { 1705 u64 alloc = zone_info[i].alloc_offset; 1706 1707 for (int j = 1; j < map->sub_stripes; j++) { 1708 int idx = i + j; 1709 1710 if (zone_info[idx].alloc_offset == WP_MISSING_DEV || 1711 zone_info[idx].alloc_offset == WP_CONVENTIONAL) 1712 continue; 1713 if (alloc == WP_MISSING_DEV || alloc == WP_CONVENTIONAL) { 1714 alloc = zone_info[idx].alloc_offset; 1715 } else if (unlikely(zone_info[idx].alloc_offset != alloc)) { 1716 btrfs_err(fs_info, 1717 "zoned: write pointer mismatch found in block group %llu", 1718 bg->start); 1719 return -EIO; 1720 } 1721 } 1722 1723 raid0_allocs[i / map->sub_stripes] = alloc; 1724 if (alloc == WP_CONVENTIONAL) 1725 continue; 1726 if (unlikely(alloc == WP_MISSING_DEV)) { 1727 btrfs_err(fs_info, 1728 "zoned: cannot recover write pointer of block group %llu due to missing device", 1729 bg->start); 1730 return -EIO; 1731 } 1732 1733 stripe_nr = alloc >> BTRFS_STRIPE_LEN_SHIFT; 1734 stripe_offset = alloc & BTRFS_STRIPE_LEN_MASK; 1735 if (stripe_offset == 0 && stripe_nr > 0) { 1736 stripe_nr--; 1737 stripe_offset = BTRFS_STRIPE_LEN; 1738 } 1739 1740 alloc = ((stripe_nr * (map->num_stripes / map->sub_stripes) + 1741 (i / map->sub_stripes)) << 1742 BTRFS_STRIPE_LEN_SHIFT) + stripe_offset; 1743 last_alloc = max(last_alloc, alloc); 1744 } 1745 stripe_nr = 0; 1746 stripe_offset = 0; 1747 1748 if (last_alloc) { 1749 u32 factor = map->num_stripes / map->sub_stripes; 1750 1751 stripe_nr = last_alloc >> BTRFS_STRIPE_LEN_SHIFT; 1752 stripe_offset = last_alloc & BTRFS_STRIPE_LEN_MASK; 1753 stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); 1754 } 1755 1756 for (int i = 0; i < map->num_stripes; i++) { 1757 int idx = i / map->sub_stripes; 1758 1759 if (raid0_allocs[idx] == WP_CONVENTIONAL) { 1760 has_conventional = true; 1761 raid0_allocs[idx] = btrfs_stripe_nr_to_offset(stripe_nr); 1762 1763 if (stripe_index > idx) 1764 raid0_allocs[idx] += BTRFS_STRIPE_LEN; 1765 else if (stripe_index == idx) 1766 raid0_allocs[idx] += stripe_offset; 1767 } 1768 1769 if ((i % map->sub_stripes) == 0) { 1770 /* Verification */ 1771 if (i != 0) { 1772 if (unlikely(prev_offset < raid0_allocs[idx])) { 1773 btrfs_err(fs_info, 1774 "zoned: stripe position disorder found in block group %llu", 1775 bg->start); 1776 return -EIO; 1777 } 1778 1779 if (unlikely(has_partial && 1780 (raid0_allocs[idx] & BTRFS_STRIPE_LEN_MASK))) { 1781 btrfs_err(fs_info, 1782 "zoned: multiple partial written stripe found in block group %llu", 1783 bg->start); 1784 return -EIO; 1785 } 1786 } 1787 prev_offset = raid0_allocs[idx]; 1788 1789 if ((raid0_allocs[idx] & BTRFS_STRIPE_LEN_MASK) != 0) 1790 has_partial = true; 1791 } 1792 1793 if (zone_info[i].alloc_offset == WP_MISSING_DEV || 1794 zone_info[i].alloc_offset == WP_CONVENTIONAL) 1795 zone_info[i].alloc_offset = raid0_allocs[idx]; 1796 1797 if (test_bit(0, active) != test_bit(i, active)) { 1798 if (unlikely(!btrfs_zone_activate(bg))) 1799 return -EIO; 1800 } else if (test_bit(0, active)) { 1801 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &bg->runtime_flags); 1802 } 1803 1804 if ((i % map->sub_stripes) == 0) { 1805 bg->zone_capacity += zone_info[i].capacity; 1806 bg->alloc_offset += zone_info[i].alloc_offset; 1807 } 1808 } 1809 1810 /* Check if all devices stay in the same stripe row. */ 1811 if (unlikely(zone_info[0].alloc_offset - 1812 zone_info[map->num_stripes - 1].alloc_offset > BTRFS_STRIPE_LEN)) { 1813 btrfs_err(fs_info, "zoned: stripe gap too large in block group %llu", 1814 bg->start); 1815 return -EIO; 1816 } 1817 1818 if (unlikely(has_conventional && bg->alloc_offset < last_alloc)) { 1819 btrfs_err(fs_info, "zoned: allocated extent stays beyond write pointers %llu %llu", 1820 bg->alloc_offset, last_alloc); 1821 return -EIO; 1822 } 1823 1824 return 0; 1825} 1826 1827int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new) 1828{ 1829 struct btrfs_fs_info *fs_info = cache->fs_info; 1830 struct btrfs_chunk_map *map; 1831 u64 logical = cache->start; 1832 u64 length = cache->length; 1833 struct zone_info AUTO_KFREE(zone_info); 1834 int ret; 1835 int i; 1836 unsigned long *active = NULL; 1837 u64 last_alloc = 0; 1838 u32 num_sequential = 0, num_conventional = 0; 1839 u64 profile; 1840 1841 if (!btrfs_is_zoned(fs_info)) 1842 return 0; 1843 1844 /* Sanity check */ 1845 if (unlikely(!IS_ALIGNED(length, fs_info->zone_size))) { 1846 btrfs_err(fs_info, 1847 "zoned: block group %llu len %llu unaligned to zone size %llu", 1848 logical, length, fs_info->zone_size); 1849 return -EIO; 1850 } 1851 1852 map = btrfs_find_chunk_map(fs_info, logical, length); 1853 if (!map) 1854 return -EINVAL; 1855 1856 cache->physical_map = map; 1857 1858 zone_info = kcalloc(map->num_stripes, sizeof(*zone_info), GFP_NOFS); 1859 if (!zone_info) { 1860 ret = -ENOMEM; 1861 goto out; 1862 } 1863 1864 active = bitmap_zalloc(map->num_stripes, GFP_NOFS); 1865 if (!active) { 1866 ret = -ENOMEM; 1867 goto out; 1868 } 1869 1870 for (i = 0; i < map->num_stripes; i++) { 1871 ret = btrfs_load_zone_info(fs_info, i, &zone_info[i], active, map, new); 1872 if (ret) 1873 goto out; 1874 1875 if (zone_info[i].alloc_offset == WP_CONVENTIONAL) 1876 num_conventional++; 1877 else 1878 num_sequential++; 1879 } 1880 1881 if (num_sequential > 0) 1882 set_bit(BLOCK_GROUP_FLAG_SEQUENTIAL_ZONE, &cache->runtime_flags); 1883 1884 if (num_conventional > 0) { 1885 ret = calculate_alloc_pointer(cache, &last_alloc, new); 1886 if (ret) { 1887 btrfs_err(fs_info, 1888 "zoned: failed to determine allocation offset of bg %llu", 1889 cache->start); 1890 goto out; 1891 } else if (map->num_stripes == num_conventional) { 1892 cache->alloc_offset = last_alloc; 1893 cache->zone_capacity = cache->length; 1894 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags); 1895 goto out; 1896 } 1897 } 1898 1899 profile = map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK; 1900 switch (profile) { 1901 case 0: /* single */ 1902 ret = btrfs_load_block_group_single(cache, &zone_info[0], active); 1903 break; 1904 case BTRFS_BLOCK_GROUP_DUP: 1905 ret = btrfs_load_block_group_dup(cache, map, zone_info, active, 1906 last_alloc); 1907 break; 1908 case BTRFS_BLOCK_GROUP_RAID1: 1909 case BTRFS_BLOCK_GROUP_RAID1C3: 1910 case BTRFS_BLOCK_GROUP_RAID1C4: 1911 ret = btrfs_load_block_group_raid1(cache, map, zone_info, 1912 active, last_alloc); 1913 break; 1914 case BTRFS_BLOCK_GROUP_RAID0: 1915 ret = btrfs_load_block_group_raid0(cache, map, zone_info, 1916 active, last_alloc); 1917 break; 1918 case BTRFS_BLOCK_GROUP_RAID10: 1919 ret = btrfs_load_block_group_raid10(cache, map, zone_info, 1920 active, last_alloc); 1921 break; 1922 case BTRFS_BLOCK_GROUP_RAID5: 1923 case BTRFS_BLOCK_GROUP_RAID6: 1924 default: 1925 btrfs_err(fs_info, "zoned: profile %s not yet supported", 1926 btrfs_bg_type_to_raid_name(map->type)); 1927 ret = -EINVAL; 1928 goto out; 1929 } 1930 1931 if (ret == -EIO && profile != 0 && profile != BTRFS_BLOCK_GROUP_RAID0 && 1932 profile != BTRFS_BLOCK_GROUP_RAID10) { 1933 /* 1934 * Detected broken write pointer. Make this block group 1935 * unallocatable by setting the allocation pointer at the end of 1936 * allocatable region. Relocating this block group will fix the 1937 * mismatch. 1938 * 1939 * Currently, we cannot handle RAID0 or RAID10 case like this 1940 * because we don't have a proper zone_capacity value. But, 1941 * reading from this block group won't work anyway by a missing 1942 * stripe. 1943 */ 1944 cache->alloc_offset = cache->zone_capacity; 1945 } 1946 1947out: 1948 /* Reject non SINGLE data profiles without RST */ 1949 if ((map->type & BTRFS_BLOCK_GROUP_DATA) && 1950 (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) && 1951 !fs_info->stripe_root) { 1952 btrfs_err(fs_info, "zoned: data %s needs raid-stripe-tree", 1953 btrfs_bg_type_to_raid_name(map->type)); 1954 ret = -EINVAL; 1955 } 1956 1957 if (unlikely(cache->alloc_offset > cache->zone_capacity)) { 1958 btrfs_err(fs_info, 1959"zoned: invalid write pointer %llu (larger than zone capacity %llu) in block group %llu", 1960 cache->alloc_offset, cache->zone_capacity, 1961 cache->start); 1962 ret = -EIO; 1963 } 1964 1965 /* An extent is allocated after the write pointer */ 1966 if (!ret && num_conventional && last_alloc > cache->alloc_offset) { 1967 btrfs_err(fs_info, 1968 "zoned: got wrong write pointer in BG %llu: %llu > %llu", 1969 logical, last_alloc, cache->alloc_offset); 1970 ret = -EIO; 1971 } 1972 1973 if (!ret) { 1974 cache->meta_write_pointer = cache->alloc_offset + cache->start; 1975 if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags)) { 1976 btrfs_get_block_group(cache); 1977 spin_lock(&fs_info->zone_active_bgs_lock); 1978 list_add_tail(&cache->active_bg_list, 1979 &fs_info->zone_active_bgs); 1980 spin_unlock(&fs_info->zone_active_bgs_lock); 1981 } 1982 } else { 1983 btrfs_free_chunk_map(cache->physical_map); 1984 cache->physical_map = NULL; 1985 } 1986 bitmap_free(active); 1987 1988 return ret; 1989} 1990 1991void btrfs_calc_zone_unusable(struct btrfs_block_group *cache) 1992{ 1993 u64 unusable, free; 1994 1995 if (!btrfs_is_zoned(cache->fs_info)) 1996 return; 1997 1998 WARN_ON(cache->bytes_super != 0); 1999 unusable = (cache->alloc_offset - cache->used) + 2000 (cache->length - cache->zone_capacity); 2001 free = cache->zone_capacity - cache->alloc_offset; 2002 2003 /* We only need ->free_space in ALLOC_SEQ block groups */ 2004 cache->cached = BTRFS_CACHE_FINISHED; 2005 cache->free_space_ctl->free_space = free; 2006 cache->zone_unusable = unusable; 2007} 2008 2009bool btrfs_use_zone_append(struct btrfs_bio *bbio) 2010{ 2011 u64 start = (bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT); 2012 struct btrfs_inode *inode = bbio->inode; 2013 struct btrfs_fs_info *fs_info = inode->root->fs_info; 2014 struct btrfs_block_group *cache; 2015 bool ret = false; 2016 2017 if (!btrfs_is_zoned(fs_info)) 2018 return false; 2019 2020 if (!is_data_inode(inode)) 2021 return false; 2022 2023 if (btrfs_op(&bbio->bio) != BTRFS_MAP_WRITE) 2024 return false; 2025 2026 /* 2027 * Using REQ_OP_ZONE_APPEND for relocation can break assumptions on the 2028 * extent layout the relocation code has. 2029 * Furthermore we have set aside own block-group from which only the 2030 * relocation "process" can allocate and make sure only one process at a 2031 * time can add pages to an extent that gets relocated, so it's safe to 2032 * use regular REQ_OP_WRITE for this special case. 2033 */ 2034 if (btrfs_is_data_reloc_root(inode->root)) 2035 return false; 2036 2037 cache = btrfs_lookup_block_group(fs_info, start); 2038 ASSERT(cache); 2039 if (!cache) 2040 return false; 2041 2042 ret = !!test_bit(BLOCK_GROUP_FLAG_SEQUENTIAL_ZONE, &cache->runtime_flags); 2043 btrfs_put_block_group(cache); 2044 2045 return ret; 2046} 2047 2048void btrfs_record_physical_zoned(struct btrfs_bio *bbio) 2049{ 2050 const u64 physical = bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT; 2051 struct btrfs_ordered_sum *sum = bbio->sums; 2052 2053 if (physical < bbio->orig_physical) 2054 sum->logical -= bbio->orig_physical - physical; 2055 else 2056 sum->logical += physical - bbio->orig_physical; 2057} 2058 2059static void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered, 2060 u64 logical) 2061{ 2062 struct extent_map_tree *em_tree = &ordered->inode->extent_tree; 2063 struct extent_map *em; 2064 2065 ordered->disk_bytenr = logical; 2066 2067 write_lock(&em_tree->lock); 2068 em = btrfs_search_extent_mapping(em_tree, ordered->file_offset, 2069 ordered->num_bytes); 2070 /* The em should be a new COW extent, thus it should not have an offset. */ 2071 ASSERT(em->offset == 0, "em->offset=%llu", em->offset); 2072 em->disk_bytenr = logical; 2073 btrfs_free_extent_map(em); 2074 write_unlock(&em_tree->lock); 2075} 2076 2077static bool btrfs_zoned_split_ordered(struct btrfs_ordered_extent *ordered, 2078 u64 logical, u64 len) 2079{ 2080 struct btrfs_ordered_extent *new; 2081 2082 if (!test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags) && 2083 btrfs_split_extent_map(ordered->inode, ordered->file_offset, 2084 ordered->num_bytes, len, logical)) 2085 return false; 2086 2087 new = btrfs_split_ordered_extent(ordered, len); 2088 if (IS_ERR(new)) 2089 return false; 2090 new->disk_bytenr = logical; 2091 btrfs_finish_one_ordered(new); 2092 return true; 2093} 2094 2095void btrfs_finish_ordered_zoned(struct btrfs_ordered_extent *ordered) 2096{ 2097 struct btrfs_inode *inode = ordered->inode; 2098 struct btrfs_fs_info *fs_info = inode->root->fs_info; 2099 struct btrfs_ordered_sum *sum; 2100 u64 logical, len; 2101 2102 /* 2103 * Write to pre-allocated region is for the data relocation, and so 2104 * it should use WRITE operation. No split/rewrite are necessary. 2105 */ 2106 if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags)) 2107 return; 2108 2109 ASSERT(!list_empty(&ordered->list)); 2110 /* The ordered->list can be empty in the above pre-alloc case. */ 2111 sum = list_first_entry(&ordered->list, struct btrfs_ordered_sum, list); 2112 logical = sum->logical; 2113 len = sum->len; 2114 2115 while (len < ordered->disk_num_bytes) { 2116 sum = list_next_entry(sum, list); 2117 if (sum->logical == logical + len) { 2118 len += sum->len; 2119 continue; 2120 } 2121 if (!btrfs_zoned_split_ordered(ordered, logical, len)) { 2122 set_bit(BTRFS_ORDERED_IOERR, &ordered->flags); 2123 btrfs_err(fs_info, "failed to split ordered extent"); 2124 goto out; 2125 } 2126 logical = sum->logical; 2127 len = sum->len; 2128 } 2129 2130 if (ordered->disk_bytenr != logical) 2131 btrfs_rewrite_logical_zoned(ordered, logical); 2132 2133out: 2134 /* 2135 * If we end up here for nodatasum I/O, the btrfs_ordered_sum structures 2136 * were allocated by btrfs_alloc_dummy_sum only to record the logical 2137 * addresses and don't contain actual checksums. We thus must free them 2138 * here so that we don't attempt to log the csums later. 2139 */ 2140 if ((inode->flags & BTRFS_INODE_NODATASUM) || 2141 test_bit(BTRFS_FS_STATE_NO_DATA_CSUMS, &fs_info->fs_state)) { 2142 while ((sum = list_first_entry_or_null(&ordered->list, 2143 typeof(*sum), list))) { 2144 list_del(&sum->list); 2145 kfree(sum); 2146 } 2147 } 2148} 2149 2150static bool check_bg_is_active(struct btrfs_eb_write_context *ctx, 2151 struct btrfs_block_group **active_bg) 2152{ 2153 const struct writeback_control *wbc = ctx->wbc; 2154 struct btrfs_block_group *block_group = ctx->zoned_bg; 2155 struct btrfs_fs_info *fs_info = block_group->fs_info; 2156 2157 if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) 2158 return true; 2159 2160 if (fs_info->treelog_bg == block_group->start) { 2161 if (!btrfs_zone_activate(block_group)) { 2162 int ret_fin = btrfs_zone_finish_one_bg(fs_info); 2163 2164 if (ret_fin != 1 || !btrfs_zone_activate(block_group)) 2165 return false; 2166 } 2167 } else if (*active_bg != block_group) { 2168 struct btrfs_block_group *tgt = *active_bg; 2169 2170 /* zoned_meta_io_lock protects fs_info->active_{meta,system}_bg. */ 2171 lockdep_assert_held(&fs_info->zoned_meta_io_lock); 2172 2173 if (tgt) { 2174 /* 2175 * If there is an unsent IO left in the allocated area, 2176 * we cannot wait for them as it may cause a deadlock. 2177 */ 2178 if (tgt->meta_write_pointer < tgt->start + tgt->alloc_offset) { 2179 if (wbc->sync_mode == WB_SYNC_NONE || 2180 (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync)) 2181 return false; 2182 } 2183 2184 /* Pivot active metadata/system block group. */ 2185 btrfs_zoned_meta_io_unlock(fs_info); 2186 wait_eb_writebacks(tgt); 2187 do_zone_finish(tgt, true); 2188 btrfs_zoned_meta_io_lock(fs_info); 2189 if (*active_bg == tgt) { 2190 btrfs_put_block_group(tgt); 2191 *active_bg = NULL; 2192 } 2193 } 2194 if (!btrfs_zone_activate(block_group)) 2195 return false; 2196 if (*active_bg != block_group) { 2197 ASSERT(*active_bg == NULL); 2198 *active_bg = block_group; 2199 btrfs_get_block_group(block_group); 2200 } 2201 } 2202 2203 return true; 2204} 2205 2206/* 2207 * Check if @ctx->eb is aligned to the write pointer. 2208 * 2209 * Return: 2210 * 0: @ctx->eb is at the write pointer. You can write it. 2211 * -EAGAIN: There is a hole. The caller should handle the case. 2212 * -EBUSY: There is a hole, but the caller can just bail out. 2213 */ 2214int btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info, 2215 struct btrfs_eb_write_context *ctx) 2216{ 2217 const struct writeback_control *wbc = ctx->wbc; 2218 const struct extent_buffer *eb = ctx->eb; 2219 struct btrfs_block_group *block_group = ctx->zoned_bg; 2220 2221 if (!btrfs_is_zoned(fs_info)) 2222 return 0; 2223 2224 if (block_group) { 2225 if (block_group->start > eb->start || 2226 block_group->start + block_group->length <= eb->start) { 2227 btrfs_put_block_group(block_group); 2228 block_group = NULL; 2229 ctx->zoned_bg = NULL; 2230 } 2231 } 2232 2233 if (!block_group) { 2234 block_group = btrfs_lookup_block_group(fs_info, eb->start); 2235 if (!block_group) 2236 return 0; 2237 ctx->zoned_bg = block_group; 2238 } 2239 2240 if (block_group->meta_write_pointer == eb->start) { 2241 struct btrfs_block_group **tgt; 2242 2243 if (!test_bit(BTRFS_FS_ACTIVE_ZONE_TRACKING, &fs_info->flags)) 2244 return 0; 2245 2246 if (block_group->flags & BTRFS_BLOCK_GROUP_SYSTEM) 2247 tgt = &fs_info->active_system_bg; 2248 else 2249 tgt = &fs_info->active_meta_bg; 2250 if (check_bg_is_active(ctx, tgt)) 2251 return 0; 2252 } 2253 2254 /* 2255 * Since we may release fs_info->zoned_meta_io_lock, someone can already 2256 * start writing this eb. In that case, we can just bail out. 2257 */ 2258 if (block_group->meta_write_pointer > eb->start) 2259 return -EBUSY; 2260 2261 /* If for_sync, this hole will be filled with transaction commit. */ 2262 if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) 2263 return -EAGAIN; 2264 return -EBUSY; 2265} 2266 2267int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length) 2268{ 2269 if (!btrfs_dev_is_sequential(device, physical)) 2270 return -EOPNOTSUPP; 2271 2272 return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT, 2273 length >> SECTOR_SHIFT, GFP_NOFS, 0); 2274} 2275 2276static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical, 2277 struct blk_zone *zone) 2278{ 2279 struct btrfs_io_context *bioc = NULL; 2280 u64 mapped_length = PAGE_SIZE; 2281 unsigned int nofs_flag; 2282 int nmirrors; 2283 int i, ret; 2284 2285 ret = btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical, 2286 &mapped_length, &bioc, NULL, NULL); 2287 if (unlikely(ret || !bioc || mapped_length < PAGE_SIZE)) { 2288 ret = -EIO; 2289 goto out_put_bioc; 2290 } 2291 2292 if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) { 2293 ret = -EINVAL; 2294 goto out_put_bioc; 2295 } 2296 2297 nofs_flag = memalloc_nofs_save(); 2298 nmirrors = (int)bioc->num_stripes; 2299 for (i = 0; i < nmirrors; i++) { 2300 u64 physical = bioc->stripes[i].physical; 2301 struct btrfs_device *dev = bioc->stripes[i].dev; 2302 2303 /* Missing device */ 2304 if (!dev->bdev) 2305 continue; 2306 2307 ret = btrfs_get_dev_zone(dev, physical, zone); 2308 /* Failing device */ 2309 if (ret == -EIO || ret == -EOPNOTSUPP) 2310 continue; 2311 break; 2312 } 2313 memalloc_nofs_restore(nofs_flag); 2314out_put_bioc: 2315 btrfs_put_bioc(bioc); 2316 return ret; 2317} 2318 2319/* 2320 * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by 2321 * filling zeros between @physical_pos to a write pointer of dev-replace 2322 * source device. 2323 */ 2324int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical, 2325 u64 physical_start, u64 physical_pos) 2326{ 2327 struct btrfs_fs_info *fs_info = tgt_dev->fs_info; 2328 struct blk_zone zone; 2329 u64 length; 2330 u64 wp; 2331 int ret; 2332 2333 if (!btrfs_dev_is_sequential(tgt_dev, physical_pos)) 2334 return 0; 2335 2336 ret = read_zone_info(fs_info, logical, &zone); 2337 if (ret) 2338 return ret; 2339 2340 wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT); 2341 2342 if (physical_pos == wp) 2343 return 0; 2344 2345 if (unlikely(physical_pos > wp)) 2346 return -EUCLEAN; 2347 2348 length = wp - physical_pos; 2349 return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length); 2350} 2351 2352/* 2353 * Activate block group and underlying device zones 2354 * 2355 * @block_group: the block group to activate 2356 * 2357 * Return: true on success, false otherwise 2358 */ 2359bool btrfs_zone_activate(struct btrfs_block_group *block_group) 2360{ 2361 struct btrfs_fs_info *fs_info = block_group->fs_info; 2362 struct btrfs_chunk_map *map; 2363 struct btrfs_device *device; 2364 u64 physical; 2365 const bool is_data = (block_group->flags & BTRFS_BLOCK_GROUP_DATA); 2366 bool ret; 2367 int i; 2368 2369 if (!btrfs_is_zoned(block_group->fs_info)) 2370 return true; 2371 2372 map = block_group->physical_map; 2373 2374 spin_lock(&fs_info->zone_active_bgs_lock); 2375 spin_lock(&block_group->lock); 2376 if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) { 2377 ret = true; 2378 goto out_unlock; 2379 } 2380 2381 if (block_group->flags & BTRFS_BLOCK_GROUP_DATA) { 2382 /* The caller should check if the block group is full. */ 2383 if (WARN_ON_ONCE(btrfs_zoned_bg_is_full(block_group))) { 2384 ret = false; 2385 goto out_unlock; 2386 } 2387 } else { 2388 /* Since it is already written, it should have been active. */ 2389 WARN_ON_ONCE(block_group->meta_write_pointer != block_group->start); 2390 } 2391 2392 for (i = 0; i < map->num_stripes; i++) { 2393 struct btrfs_zoned_device_info *zinfo; 2394 int reserved = 0; 2395 2396 device = map->stripes[i].dev; 2397 physical = map->stripes[i].physical; 2398 zinfo = device->zone_info; 2399 2400 if (!device->bdev) 2401 continue; 2402 2403 if (zinfo->max_active_zones == 0) 2404 continue; 2405 2406 if (is_data) 2407 reserved = zinfo->reserved_active_zones; 2408 /* 2409 * For the data block group, leave active zones for one 2410 * metadata block group and one system block group. 2411 */ 2412 if (atomic_read(&zinfo->active_zones_left) <= reserved) { 2413 ret = false; 2414 goto out_unlock; 2415 } 2416 2417 if (!btrfs_dev_set_active_zone(device, physical)) { 2418 /* Cannot activate the zone */ 2419 ret = false; 2420 goto out_unlock; 2421 } 2422 if (!is_data) 2423 zinfo->reserved_active_zones--; 2424 } 2425 2426 /* Successfully activated all the zones */ 2427 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags); 2428 spin_unlock(&block_group->lock); 2429 2430 /* For the active block group list */ 2431 btrfs_get_block_group(block_group); 2432 list_add_tail(&block_group->active_bg_list, &fs_info->zone_active_bgs); 2433 spin_unlock(&fs_info->zone_active_bgs_lock); 2434 2435 return true; 2436 2437out_unlock: 2438 spin_unlock(&block_group->lock); 2439 spin_unlock(&fs_info->zone_active_bgs_lock); 2440 return ret; 2441} 2442 2443static void wait_eb_writebacks(struct btrfs_block_group *block_group) 2444{ 2445 struct btrfs_fs_info *fs_info = block_group->fs_info; 2446 const u64 end = block_group->start + block_group->length; 2447 struct extent_buffer *eb; 2448 unsigned long index, start = (block_group->start >> fs_info->nodesize_bits); 2449 2450 rcu_read_lock(); 2451 xa_for_each_start(&fs_info->buffer_tree, index, eb, start) { 2452 if (eb->start < block_group->start) 2453 continue; 2454 if (eb->start >= end) 2455 break; 2456 rcu_read_unlock(); 2457 wait_on_extent_buffer_writeback(eb); 2458 rcu_read_lock(); 2459 } 2460 rcu_read_unlock(); 2461} 2462 2463static int call_zone_finish(struct btrfs_block_group *block_group, 2464 struct btrfs_io_stripe *stripe) 2465{ 2466 struct btrfs_device *device = stripe->dev; 2467 const u64 physical = stripe->physical; 2468 struct btrfs_zoned_device_info *zinfo = device->zone_info; 2469 int ret; 2470 2471 if (!device->bdev) 2472 return 0; 2473 2474 if (zinfo->max_active_zones == 0) 2475 return 0; 2476 2477 if (btrfs_dev_is_sequential(device, physical)) { 2478 unsigned int nofs_flags; 2479 2480 nofs_flags = memalloc_nofs_save(); 2481 ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_FINISH, 2482 physical >> SECTOR_SHIFT, 2483 zinfo->zone_size >> SECTOR_SHIFT); 2484 memalloc_nofs_restore(nofs_flags); 2485 2486 if (ret) 2487 return ret; 2488 } 2489 2490 if (!(block_group->flags & BTRFS_BLOCK_GROUP_DATA)) 2491 zinfo->reserved_active_zones++; 2492 btrfs_dev_clear_active_zone(device, physical); 2493 2494 return 0; 2495} 2496 2497static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_written) 2498{ 2499 struct btrfs_fs_info *fs_info = block_group->fs_info; 2500 struct btrfs_chunk_map *map; 2501 const bool is_metadata = (block_group->flags & 2502 (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM)); 2503 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; 2504 int ret = 0; 2505 int i; 2506 2507 spin_lock(&block_group->lock); 2508 if (!test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) { 2509 spin_unlock(&block_group->lock); 2510 return 0; 2511 } 2512 2513 /* Check if we have unwritten allocated space */ 2514 if (is_metadata && 2515 block_group->start + block_group->alloc_offset > block_group->meta_write_pointer) { 2516 spin_unlock(&block_group->lock); 2517 return -EAGAIN; 2518 } 2519 2520 /* 2521 * If we are sure that the block group is full (= no more room left for 2522 * new allocation) and the IO for the last usable block is completed, we 2523 * don't need to wait for the other IOs. This holds because we ensure 2524 * the sequential IO submissions using the ZONE_APPEND command for data 2525 * and block_group->meta_write_pointer for metadata. 2526 */ 2527 if (!fully_written) { 2528 if (test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags)) { 2529 spin_unlock(&block_group->lock); 2530 return -EAGAIN; 2531 } 2532 spin_unlock(&block_group->lock); 2533 2534 ret = btrfs_inc_block_group_ro(block_group, false); 2535 if (ret) 2536 return ret; 2537 2538 /* Ensure all writes in this block group finish */ 2539 btrfs_wait_block_group_reservations(block_group); 2540 /* No need to wait for NOCOW writers. Zoned mode does not allow that */ 2541 btrfs_wait_ordered_roots(fs_info, U64_MAX, block_group); 2542 /* Wait for extent buffers to be written. */ 2543 if (is_metadata) 2544 wait_eb_writebacks(block_group); 2545 2546 spin_lock(&block_group->lock); 2547 2548 /* 2549 * Bail out if someone already deactivated the block group, or 2550 * allocated space is left in the block group. 2551 */ 2552 if (!test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, 2553 &block_group->runtime_flags)) { 2554 spin_unlock(&block_group->lock); 2555 btrfs_dec_block_group_ro(block_group); 2556 return 0; 2557 } 2558 2559 if (block_group->reserved || 2560 test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, 2561 &block_group->runtime_flags)) { 2562 spin_unlock(&block_group->lock); 2563 btrfs_dec_block_group_ro(block_group); 2564 return -EAGAIN; 2565 } 2566 } 2567 2568 clear_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags); 2569 block_group->alloc_offset = block_group->zone_capacity; 2570 if (block_group->flags & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM)) 2571 block_group->meta_write_pointer = block_group->start + 2572 block_group->zone_capacity; 2573 block_group->free_space_ctl->free_space = 0; 2574 btrfs_clear_treelog_bg(block_group); 2575 btrfs_clear_data_reloc_bg(block_group); 2576 spin_unlock(&block_group->lock); 2577 2578 down_read(&dev_replace->rwsem); 2579 map = block_group->physical_map; 2580 for (i = 0; i < map->num_stripes; i++) { 2581 2582 ret = call_zone_finish(block_group, &map->stripes[i]); 2583 if (ret) { 2584 up_read(&dev_replace->rwsem); 2585 return ret; 2586 } 2587 } 2588 up_read(&dev_replace->rwsem); 2589 2590 if (!fully_written) 2591 btrfs_dec_block_group_ro(block_group); 2592 2593 spin_lock(&fs_info->zone_active_bgs_lock); 2594 ASSERT(!list_empty(&block_group->active_bg_list)); 2595 list_del_init(&block_group->active_bg_list); 2596 spin_unlock(&fs_info->zone_active_bgs_lock); 2597 2598 /* For active_bg_list */ 2599 btrfs_put_block_group(block_group); 2600 2601 clear_and_wake_up_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags); 2602 2603 return 0; 2604} 2605 2606int btrfs_zone_finish(struct btrfs_block_group *block_group) 2607{ 2608 if (!btrfs_is_zoned(block_group->fs_info)) 2609 return 0; 2610 2611 return do_zone_finish(block_group, false); 2612} 2613 2614bool btrfs_can_activate_zone(struct btrfs_fs_devices *fs_devices, u64 flags) 2615{ 2616 struct btrfs_fs_info *fs_info = fs_devices->fs_info; 2617 struct btrfs_device *device; 2618 bool ret = false; 2619 2620 if (!btrfs_is_zoned(fs_info)) 2621 return true; 2622 2623 if (test_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags)) 2624 return false; 2625 2626 /* Check if there is a device with active zones left */ 2627 mutex_lock(&fs_info->chunk_mutex); 2628 spin_lock(&fs_info->zone_active_bgs_lock); 2629 list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) { 2630 struct btrfs_zoned_device_info *zinfo = device->zone_info; 2631 int reserved = 0; 2632 2633 if (!device->bdev) 2634 continue; 2635 2636 if (!zinfo->max_active_zones) { 2637 ret = true; 2638 break; 2639 } 2640 2641 if (flags & BTRFS_BLOCK_GROUP_DATA) 2642 reserved = zinfo->reserved_active_zones; 2643 2644 switch (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) { 2645 case 0: /* single */ 2646 ret = (atomic_read(&zinfo->active_zones_left) >= (1 + reserved)); 2647 break; 2648 case BTRFS_BLOCK_GROUP_DUP: 2649 ret = (atomic_read(&zinfo->active_zones_left) >= (2 + reserved)); 2650 break; 2651 } 2652 if (ret) 2653 break; 2654 } 2655 spin_unlock(&fs_info->zone_active_bgs_lock); 2656 mutex_unlock(&fs_info->chunk_mutex); 2657 2658 if (!ret) 2659 set_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags); 2660 2661 return ret; 2662} 2663 2664int btrfs_zone_finish_endio(struct btrfs_fs_info *fs_info, u64 logical, u64 length) 2665{ 2666 struct btrfs_block_group *block_group; 2667 u64 min_alloc_bytes; 2668 2669 if (!btrfs_is_zoned(fs_info)) 2670 return 0; 2671 2672 block_group = btrfs_lookup_block_group(fs_info, logical); 2673 if (WARN_ON_ONCE(!block_group)) 2674 return -ENOENT; 2675 2676 /* No MIXED_BG on zoned btrfs. */ 2677 if (block_group->flags & BTRFS_BLOCK_GROUP_DATA) 2678 min_alloc_bytes = fs_info->sectorsize; 2679 else 2680 min_alloc_bytes = fs_info->nodesize; 2681 2682 /* Bail out if we can allocate more data from this block group. */ 2683 if (logical + length + min_alloc_bytes <= 2684 block_group->start + block_group->zone_capacity) 2685 goto out; 2686 2687 do_zone_finish(block_group, true); 2688 2689out: 2690 btrfs_put_block_group(block_group); 2691 return 0; 2692} 2693 2694static void btrfs_zone_finish_endio_workfn(struct work_struct *work) 2695{ 2696 int ret; 2697 struct btrfs_block_group *bg = 2698 container_of(work, struct btrfs_block_group, zone_finish_work); 2699 2700 wait_on_extent_buffer_writeback(bg->last_eb); 2701 free_extent_buffer(bg->last_eb); 2702 ret = do_zone_finish(bg, true); 2703 if (ret) 2704 btrfs_handle_fs_error(bg->fs_info, ret, 2705 "Failed to finish block-group's zone"); 2706 btrfs_put_block_group(bg); 2707} 2708 2709void btrfs_schedule_zone_finish_bg(struct btrfs_block_group *bg, 2710 struct extent_buffer *eb) 2711{ 2712 if (!test_bit(BLOCK_GROUP_FLAG_SEQUENTIAL_ZONE, &bg->runtime_flags) || 2713 eb->start + eb->len * 2 <= bg->start + bg->zone_capacity) 2714 return; 2715 2716 if (WARN_ON(bg->zone_finish_work.func == btrfs_zone_finish_endio_workfn)) { 2717 btrfs_err(bg->fs_info, "double scheduling of bg %llu zone finishing", 2718 bg->start); 2719 return; 2720 } 2721 2722 /* For the work */ 2723 btrfs_get_block_group(bg); 2724 refcount_inc(&eb->refs); 2725 bg->last_eb = eb; 2726 INIT_WORK(&bg->zone_finish_work, btrfs_zone_finish_endio_workfn); 2727 queue_work(system_dfl_wq, &bg->zone_finish_work); 2728} 2729 2730void btrfs_clear_data_reloc_bg(struct btrfs_block_group *bg) 2731{ 2732 struct btrfs_fs_info *fs_info = bg->fs_info; 2733 2734 spin_lock(&fs_info->relocation_bg_lock); 2735 if (fs_info->data_reloc_bg == bg->start) 2736 fs_info->data_reloc_bg = 0; 2737 spin_unlock(&fs_info->relocation_bg_lock); 2738} 2739 2740void btrfs_zoned_reserve_data_reloc_bg(struct btrfs_fs_info *fs_info) 2741{ 2742 struct btrfs_space_info *data_sinfo = fs_info->data_sinfo; 2743 struct btrfs_space_info *space_info = data_sinfo; 2744 struct btrfs_trans_handle *trans; 2745 struct btrfs_block_group *bg; 2746 struct list_head *bg_list; 2747 u64 alloc_flags; 2748 bool first = true; 2749 bool did_chunk_alloc = false; 2750 int index; 2751 int ret; 2752 2753 if (!btrfs_is_zoned(fs_info)) 2754 return; 2755 2756 if (fs_info->data_reloc_bg) 2757 return; 2758 2759 if (sb_rdonly(fs_info->sb)) 2760 return; 2761 2762 alloc_flags = btrfs_get_alloc_profile(fs_info, space_info->flags); 2763 index = btrfs_bg_flags_to_raid_index(alloc_flags); 2764 2765 /* Scan the data space_info to find empty block groups. Take the second one. */ 2766again: 2767 bg_list = &space_info->block_groups[index]; 2768 list_for_each_entry(bg, bg_list, list) { 2769 if (bg->alloc_offset != 0) 2770 continue; 2771 2772 if (first) { 2773 first = false; 2774 continue; 2775 } 2776 2777 if (space_info == data_sinfo) { 2778 /* Migrate the block group to the data relocation space_info. */ 2779 struct btrfs_space_info *reloc_sinfo = data_sinfo->sub_group[0]; 2780 int factor; 2781 2782 ASSERT(reloc_sinfo->subgroup_id == BTRFS_SUB_GROUP_DATA_RELOC, 2783 "reloc_sinfo->subgroup_id=%d", reloc_sinfo->subgroup_id); 2784 factor = btrfs_bg_type_to_factor(bg->flags); 2785 2786 down_write(&space_info->groups_sem); 2787 list_del_init(&bg->list); 2788 /* We can assume this as we choose the second empty one. */ 2789 ASSERT(!list_empty(&space_info->block_groups[index])); 2790 up_write(&space_info->groups_sem); 2791 2792 spin_lock(&space_info->lock); 2793 space_info->total_bytes -= bg->length; 2794 space_info->disk_total -= bg->length * factor; 2795 space_info->disk_total -= bg->zone_unusable; 2796 /* There is no allocation ever happened. */ 2797 ASSERT(bg->used == 0, "bg->used=%llu", bg->used); 2798 /* No super block in a block group on the zoned setup. */ 2799 ASSERT(bg->bytes_super == 0, "bg->bytes_super=%llu", bg->bytes_super); 2800 spin_unlock(&space_info->lock); 2801 2802 bg->space_info = reloc_sinfo; 2803 if (reloc_sinfo->block_group_kobjs[index] == NULL) 2804 btrfs_sysfs_add_block_group_type(bg); 2805 2806 btrfs_add_bg_to_space_info(fs_info, bg); 2807 } 2808 2809 fs_info->data_reloc_bg = bg->start; 2810 set_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &bg->runtime_flags); 2811 btrfs_zone_activate(bg); 2812 2813 return; 2814 } 2815 2816 if (did_chunk_alloc) 2817 return; 2818 2819 trans = btrfs_join_transaction(fs_info->tree_root); 2820 if (IS_ERR(trans)) 2821 return; 2822 2823 /* Allocate new BG in the data relocation space_info. */ 2824 space_info = data_sinfo->sub_group[0]; 2825 ASSERT(space_info->subgroup_id == BTRFS_SUB_GROUP_DATA_RELOC, 2826 "space_info->subgroup_id=%d", space_info->subgroup_id); 2827 ret = btrfs_chunk_alloc(trans, space_info, alloc_flags, CHUNK_ALLOC_FORCE); 2828 btrfs_end_transaction(trans); 2829 if (ret == 1) { 2830 /* 2831 * We allocated a new block group in the data relocation space_info. We 2832 * can take that one. 2833 */ 2834 first = false; 2835 did_chunk_alloc = true; 2836 goto again; 2837 } 2838} 2839 2840void btrfs_free_zone_cache(struct btrfs_fs_info *fs_info) 2841{ 2842 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2843 struct btrfs_device *device; 2844 2845 if (!btrfs_is_zoned(fs_info)) 2846 return; 2847 2848 mutex_lock(&fs_devices->device_list_mutex); 2849 list_for_each_entry(device, &fs_devices->devices, dev_list) { 2850 if (device->zone_info) { 2851 vfree(device->zone_info->zone_cache); 2852 device->zone_info->zone_cache = NULL; 2853 } 2854 } 2855 mutex_unlock(&fs_devices->device_list_mutex); 2856} 2857 2858bool btrfs_zoned_should_reclaim(const struct btrfs_fs_info *fs_info) 2859{ 2860 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 2861 struct btrfs_device *device; 2862 u64 total = btrfs_super_total_bytes(fs_info->super_copy); 2863 u64 used = 0; 2864 u64 factor; 2865 2866 ASSERT(btrfs_is_zoned(fs_info)); 2867 2868 if (fs_info->bg_reclaim_threshold == 0) 2869 return false; 2870 2871 mutex_lock(&fs_devices->device_list_mutex); 2872 list_for_each_entry(device, &fs_devices->devices, dev_list) { 2873 if (!device->bdev) 2874 continue; 2875 2876 used += device->bytes_used; 2877 } 2878 mutex_unlock(&fs_devices->device_list_mutex); 2879 2880 factor = div64_u64(used * 100, total); 2881 return factor >= fs_info->bg_reclaim_threshold; 2882} 2883 2884void btrfs_zoned_release_data_reloc_bg(struct btrfs_fs_info *fs_info, u64 logical, 2885 u64 length) 2886{ 2887 struct btrfs_block_group *block_group; 2888 2889 if (!btrfs_is_zoned(fs_info)) 2890 return; 2891 2892 block_group = btrfs_lookup_block_group(fs_info, logical); 2893 /* It should be called on a previous data relocation block group. */ 2894 ASSERT(block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA)); 2895 2896 spin_lock(&block_group->lock); 2897 if (!test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags)) 2898 goto out; 2899 2900 /* All relocation extents are written. */ 2901 if (block_group->start + block_group->alloc_offset == logical + length) { 2902 /* 2903 * Now, release this block group for further allocations and 2904 * zone finish. 2905 */ 2906 clear_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, 2907 &block_group->runtime_flags); 2908 } 2909 2910out: 2911 spin_unlock(&block_group->lock); 2912 btrfs_put_block_group(block_group); 2913} 2914 2915int btrfs_zone_finish_one_bg(struct btrfs_fs_info *fs_info) 2916{ 2917 struct btrfs_block_group *block_group; 2918 struct btrfs_block_group *min_bg = NULL; 2919 u64 min_avail = U64_MAX; 2920 int ret; 2921 2922 spin_lock(&fs_info->zone_active_bgs_lock); 2923 list_for_each_entry(block_group, &fs_info->zone_active_bgs, 2924 active_bg_list) { 2925 u64 avail; 2926 2927 spin_lock(&block_group->lock); 2928 if (block_group->reserved || block_group->alloc_offset == 0 || 2929 !(block_group->flags & BTRFS_BLOCK_GROUP_DATA) || 2930 test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags)) { 2931 spin_unlock(&block_group->lock); 2932 continue; 2933 } 2934 2935 avail = block_group->zone_capacity - block_group->alloc_offset; 2936 if (min_avail > avail) { 2937 if (min_bg) 2938 btrfs_put_block_group(min_bg); 2939 min_bg = block_group; 2940 min_avail = avail; 2941 btrfs_get_block_group(min_bg); 2942 } 2943 spin_unlock(&block_group->lock); 2944 } 2945 spin_unlock(&fs_info->zone_active_bgs_lock); 2946 2947 if (!min_bg) 2948 return 0; 2949 2950 ret = btrfs_zone_finish(min_bg); 2951 btrfs_put_block_group(min_bg); 2952 2953 return ret < 0 ? ret : 1; 2954} 2955 2956int btrfs_zoned_activate_one_bg(struct btrfs_space_info *space_info, bool do_finish) 2957{ 2958 struct btrfs_fs_info *fs_info = space_info->fs_info; 2959 struct btrfs_block_group *bg; 2960 int index; 2961 2962 if (!btrfs_is_zoned(fs_info) || (space_info->flags & BTRFS_BLOCK_GROUP_DATA)) 2963 return 0; 2964 2965 for (;;) { 2966 int ret; 2967 bool need_finish = false; 2968 2969 down_read(&space_info->groups_sem); 2970 for (index = 0; index < BTRFS_NR_RAID_TYPES; index++) { 2971 list_for_each_entry(bg, &space_info->block_groups[index], 2972 list) { 2973 if (!spin_trylock(&bg->lock)) 2974 continue; 2975 if (btrfs_zoned_bg_is_full(bg) || 2976 test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, 2977 &bg->runtime_flags)) { 2978 spin_unlock(&bg->lock); 2979 continue; 2980 } 2981 spin_unlock(&bg->lock); 2982 2983 if (btrfs_zone_activate(bg)) { 2984 up_read(&space_info->groups_sem); 2985 return 1; 2986 } 2987 2988 need_finish = true; 2989 } 2990 } 2991 up_read(&space_info->groups_sem); 2992 2993 if (!do_finish || !need_finish) 2994 break; 2995 2996 ret = btrfs_zone_finish_one_bg(fs_info); 2997 if (ret == 0) 2998 break; 2999 if (ret < 0) 3000 return ret; 3001 } 3002 3003 return 0; 3004} 3005 3006/* 3007 * Reserve zones for one metadata block group, one tree-log block group, and one 3008 * system block group. 3009 */ 3010void btrfs_check_active_zone_reservation(struct btrfs_fs_info *fs_info) 3011{ 3012 struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; 3013 struct btrfs_block_group *block_group; 3014 struct btrfs_device *device; 3015 /* Reserve zones for normal SINGLE metadata and tree-log block group. */ 3016 unsigned int metadata_reserve = 2; 3017 /* Reserve a zone for SINGLE system block group. */ 3018 unsigned int system_reserve = 1; 3019 3020 if (!test_bit(BTRFS_FS_ACTIVE_ZONE_TRACKING, &fs_info->flags)) 3021 return; 3022 3023 /* 3024 * This function is called from the mount context. So, there is no 3025 * parallel process touching the bits. No need for read_seqretry(). 3026 */ 3027 if (fs_info->avail_metadata_alloc_bits & BTRFS_BLOCK_GROUP_DUP) 3028 metadata_reserve = 4; 3029 if (fs_info->avail_system_alloc_bits & BTRFS_BLOCK_GROUP_DUP) 3030 system_reserve = 2; 3031 3032 /* Apply the reservation on all the devices. */ 3033 mutex_lock(&fs_devices->device_list_mutex); 3034 list_for_each_entry(device, &fs_devices->devices, dev_list) { 3035 if (!device->bdev) 3036 continue; 3037 3038 device->zone_info->reserved_active_zones = 3039 metadata_reserve + system_reserve; 3040 } 3041 mutex_unlock(&fs_devices->device_list_mutex); 3042 3043 /* Release reservation for currently active block groups. */ 3044 spin_lock(&fs_info->zone_active_bgs_lock); 3045 list_for_each_entry(block_group, &fs_info->zone_active_bgs, active_bg_list) { 3046 struct btrfs_chunk_map *map = block_group->physical_map; 3047 3048 if (!(block_group->flags & 3049 (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM))) 3050 continue; 3051 3052 for (int i = 0; i < map->num_stripes; i++) 3053 map->stripes[i].dev->zone_info->reserved_active_zones--; 3054 } 3055 spin_unlock(&fs_info->zone_active_bgs_lock); 3056} 3057 3058/* 3059 * Reset the zones of unused block groups from @space_info->bytes_zone_unusable. 3060 * 3061 * @space_info: the space to work on 3062 * @num_bytes: targeting reclaim bytes 3063 * 3064 * This one resets the zones of a block group, so we can reuse the region 3065 * without removing the block group. On the other hand, btrfs_delete_unused_bgs() 3066 * just removes a block group and frees up the underlying zones. So, we still 3067 * need to allocate a new block group to reuse the zones. 3068 * 3069 * Resetting is faster than deleting/recreating a block group. It is similar 3070 * to freeing the logical space on the regular mode. However, we cannot change 3071 * the block group's profile with this operation. 3072 */ 3073int btrfs_reset_unused_block_groups(struct btrfs_space_info *space_info, u64 num_bytes) 3074{ 3075 struct btrfs_fs_info *fs_info = space_info->fs_info; 3076 const sector_t zone_size_sectors = fs_info->zone_size >> SECTOR_SHIFT; 3077 3078 if (!btrfs_is_zoned(fs_info)) 3079 return 0; 3080 3081 while (num_bytes > 0) { 3082 struct btrfs_chunk_map *map; 3083 struct btrfs_block_group *bg = NULL; 3084 bool found = false; 3085 u64 reclaimed = 0; 3086 3087 /* 3088 * Here, we choose a fully zone_unusable block group. It's 3089 * technically possible to reset a partly zone_unusable block 3090 * group, which still has some free space left. However, 3091 * handling that needs to cope with the allocation side, which 3092 * makes the logic more complex. So, let's handle the easy case 3093 * for now. 3094 */ 3095 spin_lock(&fs_info->unused_bgs_lock); 3096 list_for_each_entry(bg, &fs_info->unused_bgs, bg_list) { 3097 if ((bg->flags & BTRFS_BLOCK_GROUP_TYPE_MASK) != space_info->flags) 3098 continue; 3099 3100 /* 3101 * Use trylock to avoid locking order violation. In 3102 * btrfs_reclaim_bgs_work(), the lock order is 3103 * &bg->lock -> &fs_info->unused_bgs_lock. We skip a 3104 * block group if we cannot take its lock. 3105 */ 3106 if (!spin_trylock(&bg->lock)) 3107 continue; 3108 if (btrfs_is_block_group_used(bg) || bg->zone_unusable < bg->length) { 3109 spin_unlock(&bg->lock); 3110 continue; 3111 } 3112 spin_unlock(&bg->lock); 3113 found = true; 3114 break; 3115 } 3116 if (!found) { 3117 spin_unlock(&fs_info->unused_bgs_lock); 3118 return 0; 3119 } 3120 3121 list_del_init(&bg->bg_list); 3122 btrfs_put_block_group(bg); 3123 spin_unlock(&fs_info->unused_bgs_lock); 3124 3125 /* 3126 * Since the block group is fully zone_unusable and we cannot 3127 * allocate from this block group anymore, we don't need to set 3128 * this block group read-only. 3129 */ 3130 3131 down_read(&fs_info->dev_replace.rwsem); 3132 map = bg->physical_map; 3133 for (int i = 0; i < map->num_stripes; i++) { 3134 struct btrfs_io_stripe *stripe = &map->stripes[i]; 3135 unsigned int nofs_flags; 3136 int ret; 3137 3138 nofs_flags = memalloc_nofs_save(); 3139 ret = blkdev_zone_mgmt(stripe->dev->bdev, REQ_OP_ZONE_RESET, 3140 stripe->physical >> SECTOR_SHIFT, 3141 zone_size_sectors); 3142 memalloc_nofs_restore(nofs_flags); 3143 3144 if (ret) { 3145 up_read(&fs_info->dev_replace.rwsem); 3146 return ret; 3147 } 3148 } 3149 up_read(&fs_info->dev_replace.rwsem); 3150 3151 spin_lock(&space_info->lock); 3152 spin_lock(&bg->lock); 3153 ASSERT(!btrfs_is_block_group_used(bg)); 3154 if (bg->ro) { 3155 spin_unlock(&bg->lock); 3156 spin_unlock(&space_info->lock); 3157 continue; 3158 } 3159 3160 reclaimed = bg->alloc_offset; 3161 bg->zone_unusable = bg->length - bg->zone_capacity; 3162 bg->alloc_offset = 0; 3163 /* 3164 * This holds because we currently reset fully used then freed 3165 * block group. 3166 */ 3167 ASSERT(reclaimed == bg->zone_capacity, 3168 "reclaimed=%llu bg->zone_capacity=%llu", reclaimed, bg->zone_capacity); 3169 bg->free_space_ctl->free_space += reclaimed; 3170 space_info->bytes_zone_unusable -= reclaimed; 3171 spin_unlock(&bg->lock); 3172 btrfs_return_free_space(space_info, reclaimed); 3173 spin_unlock(&space_info->lock); 3174 3175 if (num_bytes <= reclaimed) 3176 break; 3177 num_bytes -= reclaimed; 3178 } 3179 3180 return 0; 3181}
~/
Read-only file preview.