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 "rcu-string.h"
  13 #include "disk-io.h"
  14 #include "block-group.h"
  15 #include "transaction.h"
  16 #include "dev-replace.h"
  17 #include "space-info.h"
  18 #include "fs.h"
  19 #include "accessors.h"
  20 #include "bio.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        const_ilog2(BTRFS_SB_LOG_FIRST_OFFSET)
  41 #define BTRFS_SB_LOG_SECOND_SHIFT       const_ilog2(BTRFS_SB_LOG_SECOND_OFFSET)
  42
  43 /* Number of superblock log zones */
  44 #define BTRFS_NR_SB_LOG_ZONES 2
  45
  46 /*
  47  * Minimum of active zones we need:
  48  *
  49  * - BTRFS_SUPER_MIRROR_MAX zones for superblock mirrors
  50  * - 3 zones to ensure at least one zone per SYSTEM, META and DATA block group
  51  * - 1 zone for tree-log dedicated block group
  52  * - 1 zone for relocation
  53  */
  54 #define BTRFS_MIN_ACTIVE_ZONES          (BTRFS_SUPER_MIRROR_MAX + 5)
  55
  56 /*
  57  * Minimum / maximum supported zone size. Currently, SMR disks have a zone
  58  * size of 256MiB, and we are expecting ZNS drives to be in the 1-4GiB range.
  59  * We do not expect the zone size to become larger than 8GiB or smaller than
  60  * 4MiB in the near future.
  61  */
  62 #define BTRFS_MAX_ZONE_SIZE             SZ_8G
  63 #define BTRFS_MIN_ZONE_SIZE             SZ_4M
  64
  65 #define SUPER_INFO_SECTORS      ((u64)BTRFS_SUPER_INFO_SIZE >> SECTOR_SHIFT)
  66
  67 static inline bool sb_zone_is_full(const struct blk_zone *zone)
  68 {
  69         return (zone->cond == BLK_ZONE_COND_FULL) ||
  70                 (zone->wp + SUPER_INFO_SECTORS > zone->start + zone->capacity);
  71 }
  72
  73 static int copy_zone_info_cb(struct blk_zone *zone, unsigned int idx, void *data)
  74 {
  75         struct blk_zone *zones = data;
  76
  77         memcpy(&zones[idx], zone, sizeof(*zone));
  78
  79         return 0;
  80 }
  81
  82 static int sb_write_pointer(struct block_device *bdev, struct blk_zone *zones,
  83                             u64 *wp_ret)
  84 {
  85         bool empty[BTRFS_NR_SB_LOG_ZONES];
  86         bool full[BTRFS_NR_SB_LOG_ZONES];
  87         sector_t sector;
  88         int i;
  89
  90         for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
  91                 ASSERT(zones[i].type != BLK_ZONE_TYPE_CONVENTIONAL);
  92                 empty[i] = (zones[i].cond == BLK_ZONE_COND_EMPTY);
  93                 full[i] = sb_zone_is_full(&zones[i]);
  94         }
  95
  96         /*
  97          * Possible states of log buffer zones
  98          *
  99          *           Empty[0]  In use[0]  Full[0]
 100          * Empty[1]         *          0        1
 101          * In use[1]        x          x        1
 102          * Full[1]          0          0        C
 103          *
 104          * Log position:
 105          *   *: Special case, no superblock is written
 106          *   0: Use write pointer of zones[0]
 107          *   1: Use write pointer of zones[1]
 108          *   C: Compare super blocks from zones[0] and zones[1], use the latest
 109          *      one determined by generation
 110          *   x: Invalid state
 111          */
 112
 113         if (empty[0] && empty[1]) {
 114                 /* Special case to distinguish no superblock to read */
 115                 *wp_ret = zones[0].start << SECTOR_SHIFT;
 116                 return -ENOENT;
 117         } else if (full[0] && full[1]) {
 118                 /* Compare two super blocks */
 119                 struct address_space *mapping = bdev->bd_inode->i_mapping;
 120                 struct page *page[BTRFS_NR_SB_LOG_ZONES];
 121                 struct btrfs_super_block *super[BTRFS_NR_SB_LOG_ZONES];
 122                 int i;
 123
 124                 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
 125                         u64 bytenr;
 126
 127                         bytenr = ((zones[i].start + zones[i].len)
 128                                    << SECTOR_SHIFT) - BTRFS_SUPER_INFO_SIZE;
 129
 130                         page[i] = read_cache_page_gfp(mapping,
 131                                         bytenr >> PAGE_SHIFT, GFP_NOFS);
 132                         if (IS_ERR(page[i])) {
 133                                 if (i == 1)
 134                                         btrfs_release_disk_super(super[0]);
 135                                 return PTR_ERR(page[i]);
 136                         }
 137                         super[i] = page_address(page[i]);
 138                 }
 139
 140                 if (btrfs_super_generation(super[0]) >
 141                     btrfs_super_generation(super[1]))
 142                         sector = zones[1].start;
 143                 else
 144                         sector = zones[0].start;
 145
 146                 for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++)
 147                         btrfs_release_disk_super(super[i]);
 148         } else if (!full[0] && (empty[1] || full[1])) {
 149                 sector = zones[0].wp;
 150         } else if (full[0]) {
 151                 sector = zones[1].wp;
 152         } else {
 153                 return -EUCLEAN;
 154         }
 155         *wp_ret = sector << SECTOR_SHIFT;
 156         return 0;
 157 }
 158
 159 /*
 160  * Get the first zone number of the superblock mirror
 161  */
 162 static inline u32 sb_zone_number(int shift, int mirror)
 163 {
 164         u64 zone = U64_MAX;
 165
 166         ASSERT(mirror < BTRFS_SUPER_MIRROR_MAX);
 167         switch (mirror) {
 168         case 0: zone = 0; break;
 169         case 1: zone = 1ULL << (BTRFS_SB_LOG_FIRST_SHIFT - shift); break;
 170         case 2: zone = 1ULL << (BTRFS_SB_LOG_SECOND_SHIFT - shift); break;
 171         }
 172
 173         ASSERT(zone <= U32_MAX);
 174
 175         return (u32)zone;
 176 }
 177
 178 static inline sector_t zone_start_sector(u32 zone_number,
 179                                          struct block_device *bdev)
 180 {
 181         return (sector_t)zone_number << ilog2(bdev_zone_sectors(bdev));
 182 }
 183
 184 static inline u64 zone_start_physical(u32 zone_number,
 185                                       struct btrfs_zoned_device_info *zone_info)
 186 {
 187         return (u64)zone_number << zone_info->zone_size_shift;
 188 }
 189
 190 /*
 191  * Emulate blkdev_report_zones() for a non-zoned device. It slices up the block
 192  * device into static sized chunks and fake a conventional zone on each of
 193  * them.
 194  */
 195 static int emulate_report_zones(struct btrfs_device *device, u64 pos,
 196                                 struct blk_zone *zones, unsigned int nr_zones)
 197 {
 198         const sector_t zone_sectors = device->fs_info->zone_size >> SECTOR_SHIFT;
 199         sector_t bdev_size = bdev_nr_sectors(device->bdev);
 200         unsigned int i;
 201
 202         pos >>= SECTOR_SHIFT;
 203         for (i = 0; i < nr_zones; i++) {
 204                 zones[i].start = i * zone_sectors + pos;
 205                 zones[i].len = zone_sectors;
 206                 zones[i].capacity = zone_sectors;
 207                 zones[i].wp = zones[i].start + zone_sectors;
 208                 zones[i].type = BLK_ZONE_TYPE_CONVENTIONAL;
 209                 zones[i].cond = BLK_ZONE_COND_NOT_WP;
 210
 211                 if (zones[i].wp >= bdev_size) {
 212                         i++;
 213                         break;
 214                 }
 215         }
 216
 217         return i;
 218 }
 219
 220 static int btrfs_get_dev_zones(struct btrfs_device *device, u64 pos,
 221                                struct blk_zone *zones, unsigned int *nr_zones)
 222 {
 223         struct btrfs_zoned_device_info *zinfo = device->zone_info;
 224         int ret;
 225
 226         if (!*nr_zones)
 227                 return 0;
 228
 229         if (!bdev_is_zoned(device->bdev)) {
 230                 ret = emulate_report_zones(device, pos, zones, *nr_zones);
 231                 *nr_zones = ret;
 232                 return 0;
 233         }
 234
 235         /* Check cache */
 236         if (zinfo->zone_cache) {
 237                 unsigned int i;
 238                 u32 zno;
 239
 240                 ASSERT(IS_ALIGNED(pos, zinfo->zone_size));
 241                 zno = pos >> zinfo->zone_size_shift;
 242                 /*
 243                  * We cannot report zones beyond the zone end. So, it is OK to
 244                  * cap *nr_zones to at the end.
 245                  */
 246                 *nr_zones = min_t(u32, *nr_zones, zinfo->nr_zones - zno);
 247
 248                 for (i = 0; i < *nr_zones; i++) {
 249                         struct blk_zone *zone_info;
 250
 251                         zone_info = &zinfo->zone_cache[zno + i];
 252                         if (!zone_info->len)
 253                                 break;
 254                 }
 255
 256                 if (i == *nr_zones) {
 257                         /* Cache hit on all the zones */
 258                         memcpy(zones, zinfo->zone_cache + zno,
 259                                sizeof(*zinfo->zone_cache) * *nr_zones);
 260                         return 0;
 261                 }
 262         }
 263
 264         ret = blkdev_report_zones(device->bdev, pos >> SECTOR_SHIFT, *nr_zones,
 265                                   copy_zone_info_cb, zones);
 266         if (ret < 0) {
 267                 btrfs_err_in_rcu(device->fs_info,
 268                                  "zoned: failed to read zone %llu on %s (devid %llu)",
 269                                  pos, rcu_str_deref(device->name),
 270                                  device->devid);
 271                 return ret;
 272         }
 273         *nr_zones = ret;
 274         if (!ret)
 275                 return -EIO;
 276
 277         /* Populate cache */
 278         if (zinfo->zone_cache) {
 279                 u32 zno = pos >> zinfo->zone_size_shift;
 280
 281                 memcpy(zinfo->zone_cache + zno, zones,
 282                        sizeof(*zinfo->zone_cache) * *nr_zones);
 283         }
 284
 285         return 0;
 286 }
 287
 288 /* The emulated zone size is determined from the size of device extent */
 289 static int calculate_emulated_zone_size(struct btrfs_fs_info *fs_info)
 290 {
 291         struct btrfs_path *path;
 292         struct btrfs_root *root = fs_info->dev_root;
 293         struct btrfs_key key;
 294         struct extent_buffer *leaf;
 295         struct btrfs_dev_extent *dext;
 296         int ret = 0;
 297
 298         key.objectid = 1;
 299         key.type = BTRFS_DEV_EXTENT_KEY;
 300         key.offset = 0;
 301
 302         path = btrfs_alloc_path();
 303         if (!path)
 304                 return -ENOMEM;
 305
 306         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 307         if (ret < 0)
 308                 goto out;
 309
 310         if (path->slots[0] >= btrfs_header_nritems(path->nodes[0])) {
 311                 ret = btrfs_next_leaf(root, path);
 312                 if (ret < 0)
 313                         goto out;
 314                 /* No dev extents at all? Not good */
 315                 if (ret > 0) {
 316                         ret = -EUCLEAN;
 317                         goto out;
 318                 }
 319         }
 320
 321         leaf = path->nodes[0];
 322         dext = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_extent);
 323         fs_info->zone_size = btrfs_dev_extent_length(leaf, dext);
 324         ret = 0;
 325
 326 out:
 327         btrfs_free_path(path);
 328
 329         return ret;
 330 }
 331
 332 int btrfs_get_dev_zone_info_all_devices(struct btrfs_fs_info *fs_info)
 333 {
 334         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
 335         struct btrfs_device *device;
 336         int ret = 0;
 337
 338         /* fs_info->zone_size might not set yet. Use the incomapt flag here. */
 339         if (!btrfs_fs_incompat(fs_info, ZONED))
 340                 return 0;
 341
 342         mutex_lock(&fs_devices->device_list_mutex);
 343         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 344                 /* We can skip reading of zone info for missing devices */
 345                 if (!device->bdev)
 346                         continue;
 347
 348                 ret = btrfs_get_dev_zone_info(device, true);
 349                 if (ret)
 350                         break;
 351         }
 352         mutex_unlock(&fs_devices->device_list_mutex);
 353
 354         return ret;
 355 }
 356
 357 int 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_in_rcu(fs_info,
 407                 "zoned: %s: zone size %llu larger than supported maximum %llu",
 408                                  rcu_str_deref(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_in_rcu(fs_info,
 414                 "zoned: %s: zone size %llu smaller than supported minimum %u",
 415                                  rcu_str_deref(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 = bdev_max_active_zones(bdev);
 428         if (max_active_zones && max_active_zones < BTRFS_MIN_ACTIVE_ZONES) {
 429                 btrfs_err_in_rcu(fs_info,
 430 "zoned: %s: max active zones %u is too small, need at least %u active zones",
 431                                  rcu_str_deref(device->name), max_active_zones,
 432                                  BTRFS_MIN_ACTIVE_ZONES);
 433                 ret = -EINVAL;
 434                 goto out;
 435         }
 436         zone_info->max_active_zones = max_active_zones;
 437
 438         zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
 439         if (!zone_info->seq_zones) {
 440                 ret = -ENOMEM;
 441                 goto out;
 442         }
 443
 444         zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
 445         if (!zone_info->empty_zones) {
 446                 ret = -ENOMEM;
 447                 goto out;
 448         }
 449
 450         zone_info->active_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
 451         if (!zone_info->active_zones) {
 452                 ret = -ENOMEM;
 453                 goto out;
 454         }
 455
 456         zones = kvcalloc(BTRFS_REPORT_NR_ZONES, sizeof(struct blk_zone), GFP_KERNEL);
 457         if (!zones) {
 458                 ret = -ENOMEM;
 459                 goto out;
 460         }
 461
 462         /*
 463          * Enable zone cache only for a zoned device. On a non-zoned device, we
 464          * fill the zone info with emulated CONVENTIONAL zones, so no need to
 465          * use the cache.
 466          */
 467         if (populate_cache && bdev_is_zoned(device->bdev)) {
 468                 zone_info->zone_cache = vzalloc(sizeof(struct blk_zone) *
 469                                                 zone_info->nr_zones);
 470                 if (!zone_info->zone_cache) {
 471                         btrfs_err_in_rcu(device->fs_info,
 472                                 "zoned: failed to allocate zone cache for %s",
 473                                 rcu_str_deref(device->name));
 474                         ret = -ENOMEM;
 475                         goto out;
 476                 }
 477         }
 478
 479         /* Get zones type */
 480         nactive = 0;
 481         while (sector < nr_sectors) {
 482                 nr_zones = BTRFS_REPORT_NR_ZONES;
 483                 ret = btrfs_get_dev_zones(device, sector << SECTOR_SHIFT, zones,
 484                                           &nr_zones);
 485                 if (ret)
 486                         goto out;
 487
 488                 for (i = 0; i < nr_zones; i++) {
 489                         if (zones[i].type == BLK_ZONE_TYPE_SEQWRITE_REQ)
 490                                 __set_bit(nreported, zone_info->seq_zones);
 491                         switch (zones[i].cond) {
 492                         case BLK_ZONE_COND_EMPTY:
 493                                 __set_bit(nreported, zone_info->empty_zones);
 494                                 break;
 495                         case BLK_ZONE_COND_IMP_OPEN:
 496                         case BLK_ZONE_COND_EXP_OPEN:
 497                         case BLK_ZONE_COND_CLOSED:
 498                                 __set_bit(nreported, zone_info->active_zones);
 499                                 nactive++;
 500                                 break;
 501                         }
 502                         nreported++;
 503                 }
 504                 sector = zones[nr_zones - 1].start + zones[nr_zones - 1].len;
 505         }
 506
 507         if (nreported != zone_info->nr_zones) {
 508                 btrfs_err_in_rcu(device->fs_info,
 509                                  "inconsistent number of zones on %s (%u/%u)",
 510                                  rcu_str_deref(device->name), nreported,
 511                                  zone_info->nr_zones);
 512                 ret = -EIO;
 513                 goto out;
 514         }
 515
 516         if (max_active_zones) {
 517                 if (nactive > max_active_zones) {
 518                         btrfs_err_in_rcu(device->fs_info,
 519                         "zoned: %u active zones on %s exceeds max_active_zones %u",
 520                                          nactive, rcu_str_deref(device->name),
 521                                          max_active_zones);
 522                         ret = -EIO;
 523                         goto out;
 524                 }
 525                 atomic_set(&zone_info->active_zones_left,
 526                            max_active_zones - nactive);
 527                 set_bit(BTRFS_FS_ACTIVE_ZONE_TRACKING, &fs_info->flags);
 528         }
 529
 530         /* Validate superblock log */
 531         nr_zones = BTRFS_NR_SB_LOG_ZONES;
 532         for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
 533                 u32 sb_zone;
 534                 u64 sb_wp;
 535                 int sb_pos = BTRFS_NR_SB_LOG_ZONES * i;
 536
 537                 sb_zone = sb_zone_number(zone_info->zone_size_shift, i);
 538                 if (sb_zone + 1 >= zone_info->nr_zones)
 539                         continue;
 540
 541                 ret = btrfs_get_dev_zones(device,
 542                                           zone_start_physical(sb_zone, zone_info),
 543                                           &zone_info->sb_zones[sb_pos],
 544                                           &nr_zones);
 545                 if (ret)
 546                         goto out;
 547
 548                 if (nr_zones != BTRFS_NR_SB_LOG_ZONES) {
 549                         btrfs_err_in_rcu(device->fs_info,
 550         "zoned: failed to read super block log zone info at devid %llu zone %u",
 551                                          device->devid, sb_zone);
 552                         ret = -EUCLEAN;
 553                         goto out;
 554                 }
 555
 556                 /*
 557                  * If zones[0] is conventional, always use the beginning of the
 558                  * zone to record superblock. No need to validate in that case.
 559                  */
 560                 if (zone_info->sb_zones[BTRFS_NR_SB_LOG_ZONES * i].type ==
 561                     BLK_ZONE_TYPE_CONVENTIONAL)
 562                         continue;
 563
 564                 ret = sb_write_pointer(device->bdev,
 565                                        &zone_info->sb_zones[sb_pos], &sb_wp);
 566                 if (ret != -ENOENT && ret) {
 567                         btrfs_err_in_rcu(device->fs_info,
 568                         "zoned: super block log zone corrupted devid %llu zone %u",
 569                                          device->devid, sb_zone);
 570                         ret = -EUCLEAN;
 571                         goto out;
 572                 }
 573         }
 574
 575
 576         kvfree(zones);
 577
 578         switch (bdev_zoned_model(bdev)) {
 579         case BLK_ZONED_HM:
 580                 model = "host-managed zoned";
 581                 emulated = "";
 582                 break;
 583         case BLK_ZONED_HA:
 584                 model = "host-aware zoned";
 585                 emulated = "";
 586                 break;
 587         case BLK_ZONED_NONE:
 588                 model = "regular";
 589                 emulated = "emulated ";
 590                 break;
 591         default:
 592                 /* Just in case */
 593                 btrfs_err_in_rcu(fs_info, "zoned: unsupported model %d on %s",
 594                                  bdev_zoned_model(bdev),
 595                                  rcu_str_deref(device->name));
 596                 ret = -EOPNOTSUPP;
 597                 goto out_free_zone_info;
 598         }
 599
 600         btrfs_info_in_rcu(fs_info,
 601                 "%s block device %s, %u %szones of %llu bytes",
 602                 model, rcu_str_deref(device->name), zone_info->nr_zones,
 603                 emulated, zone_info->zone_size);
 604
 605         return 0;
 606
 607 out:
 608         kvfree(zones);
 609 out_free_zone_info:
 610         btrfs_destroy_dev_zone_info(device);
 611
 612         return ret;
 613 }
 614
 615 void btrfs_destroy_dev_zone_info(struct btrfs_device *device)
 616 {
 617         struct btrfs_zoned_device_info *zone_info = device->zone_info;
 618
 619         if (!zone_info)
 620                 return;
 621
 622         bitmap_free(zone_info->active_zones);
 623         bitmap_free(zone_info->seq_zones);
 624         bitmap_free(zone_info->empty_zones);
 625         vfree(zone_info->zone_cache);
 626         kfree(zone_info);
 627         device->zone_info = NULL;
 628 }
 629
 630 struct btrfs_zoned_device_info *btrfs_clone_dev_zone_info(struct btrfs_device *orig_dev)
 631 {
 632         struct btrfs_zoned_device_info *zone_info;
 633
 634         zone_info = kmemdup(orig_dev->zone_info, sizeof(*zone_info), GFP_KERNEL);
 635         if (!zone_info)
 636                 return NULL;
 637
 638         zone_info->seq_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
 639         if (!zone_info->seq_zones)
 640                 goto out;
 641
 642         bitmap_copy(zone_info->seq_zones, orig_dev->zone_info->seq_zones,
 643                     zone_info->nr_zones);
 644
 645         zone_info->empty_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
 646         if (!zone_info->empty_zones)
 647                 goto out;
 648
 649         bitmap_copy(zone_info->empty_zones, orig_dev->zone_info->empty_zones,
 650                     zone_info->nr_zones);
 651
 652         zone_info->active_zones = bitmap_zalloc(zone_info->nr_zones, GFP_KERNEL);
 653         if (!zone_info->active_zones)
 654                 goto out;
 655
 656         bitmap_copy(zone_info->active_zones, orig_dev->zone_info->active_zones,
 657                     zone_info->nr_zones);
 658         zone_info->zone_cache = NULL;
 659
 660         return zone_info;
 661
 662 out:
 663         bitmap_free(zone_info->seq_zones);
 664         bitmap_free(zone_info->empty_zones);
 665         bitmap_free(zone_info->active_zones);
 666         kfree(zone_info);
 667         return NULL;
 668 }
 669
 670 int btrfs_get_dev_zone(struct btrfs_device *device, u64 pos,
 671                        struct blk_zone *zone)
 672 {
 673         unsigned int nr_zones = 1;
 674         int ret;
 675
 676         ret = btrfs_get_dev_zones(device, pos, zone, &nr_zones);
 677         if (ret != 0 || !nr_zones)
 678                 return ret ? ret : -EIO;
 679
 680         return 0;
 681 }
 682
 683 static int btrfs_check_for_zoned_device(struct btrfs_fs_info *fs_info)
 684 {
 685         struct btrfs_device *device;
 686
 687         list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
 688                 if (device->bdev &&
 689                     bdev_zoned_model(device->bdev) == BLK_ZONED_HM) {
 690                         btrfs_err(fs_info,
 691                                 "zoned: mode not enabled but zoned device found: %pg",
 692                                 device->bdev);
 693                         return -EINVAL;
 694                 }
 695         }
 696
 697         return 0;
 698 }
 699
 700 int btrfs_check_zoned_mode(struct btrfs_fs_info *fs_info)
 701 {
 702         struct queue_limits *lim = &fs_info->limits;
 703         struct btrfs_device *device;
 704         u64 zone_size = 0;
 705         int ret;
 706
 707         /*
 708          * Host-Managed devices can't be used without the ZONED flag.  With the
 709          * ZONED all devices can be used, using zone emulation if required.
 710          */
 711         if (!btrfs_fs_incompat(fs_info, ZONED))
 712                 return btrfs_check_for_zoned_device(fs_info);
 713
 714         blk_set_stacking_limits(lim);
 715
 716         list_for_each_entry(device, &fs_info->fs_devices->devices, dev_list) {
 717                 struct btrfs_zoned_device_info *zone_info = device->zone_info;
 718
 719                 if (!device->bdev)
 720                         continue;
 721
 722                 if (!zone_size) {
 723                         zone_size = zone_info->zone_size;
 724                 } else if (zone_info->zone_size != zone_size) {
 725                         btrfs_err(fs_info,
 726                 "zoned: unequal block device zone sizes: have %llu found %llu",
 727                                   zone_info->zone_size, zone_size);
 728                         return -EINVAL;
 729                 }
 730
 731                 /*
 732                  * With the zoned emulation, we can have non-zoned device on the
 733                  * zoned mode. In this case, we don't have a valid max zone
 734                  * append size.
 735                  */
 736                 if (bdev_is_zoned(device->bdev)) {
 737                         blk_stack_limits(lim,
 738                                          &bdev_get_queue(device->bdev)->limits,
 739                                          0);
 740                 }
 741         }
 742
 743         /*
 744          * stripe_size is always aligned to BTRFS_STRIPE_LEN in
 745          * btrfs_create_chunk(). Since we want stripe_len == zone_size,
 746          * check the alignment here.
 747          */
 748         if (!IS_ALIGNED(zone_size, BTRFS_STRIPE_LEN)) {
 749                 btrfs_err(fs_info,
 750                           "zoned: zone size %llu not aligned to stripe %u",
 751                           zone_size, BTRFS_STRIPE_LEN);
 752                 return -EINVAL;
 753         }
 754
 755         if (btrfs_fs_incompat(fs_info, MIXED_GROUPS)) {
 756                 btrfs_err(fs_info, "zoned: mixed block groups not supported");
 757                 return -EINVAL;
 758         }
 759
 760         fs_info->zone_size = zone_size;
 761         /*
 762          * Also limit max_zone_append_size by max_segments * PAGE_SIZE.
 763          * Technically, we can have multiple pages per segment. But, since
 764          * we add the pages one by one to a bio, and cannot increase the
 765          * metadata reservation even if it increases the number of extents, it
 766          * is safe to stick with the limit.
 767          */
 768         fs_info->max_zone_append_size = ALIGN_DOWN(
 769                 min3((u64)lim->max_zone_append_sectors << SECTOR_SHIFT,
 770                      (u64)lim->max_sectors << SECTOR_SHIFT,
 771                      (u64)lim->max_segments << PAGE_SHIFT),
 772                 fs_info->sectorsize);
 773         fs_info->fs_devices->chunk_alloc_policy = BTRFS_CHUNK_ALLOC_ZONED;
 774         if (fs_info->max_zone_append_size < fs_info->max_extent_size)
 775                 fs_info->max_extent_size = fs_info->max_zone_append_size;
 776
 777         /*
 778          * Check mount options here, because we might change fs_info->zoned
 779          * from fs_info->zone_size.
 780          */
 781         ret = btrfs_check_mountopts_zoned(fs_info);
 782         if (ret)
 783                 return ret;
 784
 785         btrfs_info(fs_info, "zoned mode enabled with zone size %llu", zone_size);
 786         return 0;
 787 }
 788
 789 int btrfs_check_mountopts_zoned(struct btrfs_fs_info *info)
 790 {
 791         if (!btrfs_is_zoned(info))
 792                 return 0;
 793
 794         /*
 795          * Space cache writing is not COWed. Disable that to avoid write errors
 796          * in sequential zones.
 797          */
 798         if (btrfs_test_opt(info, SPACE_CACHE)) {
 799                 btrfs_err(info, "zoned: space cache v1 is not supported");
 800                 return -EINVAL;
 801         }
 802
 803         if (btrfs_test_opt(info, NODATACOW)) {
 804                 btrfs_err(info, "zoned: NODATACOW not supported");
 805                 return -EINVAL;
 806         }
 807
 808         return 0;
 809 }
 810
 811 static int sb_log_location(struct block_device *bdev, struct blk_zone *zones,
 812                            int rw, u64 *bytenr_ret)
 813 {
 814         u64 wp;
 815         int ret;
 816
 817         if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) {
 818                 *bytenr_ret = zones[0].start << SECTOR_SHIFT;
 819                 return 0;
 820         }
 821
 822         ret = sb_write_pointer(bdev, zones, &wp);
 823         if (ret != -ENOENT && ret < 0)
 824                 return ret;
 825
 826         if (rw == WRITE) {
 827                 struct blk_zone *reset = NULL;
 828
 829                 if (wp == zones[0].start << SECTOR_SHIFT)
 830                         reset = &zones[0];
 831                 else if (wp == zones[1].start << SECTOR_SHIFT)
 832                         reset = &zones[1];
 833
 834                 if (reset && reset->cond != BLK_ZONE_COND_EMPTY) {
 835                         ASSERT(sb_zone_is_full(reset));
 836
 837                         ret = blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
 838                                                reset->start, reset->len,
 839                                                GFP_NOFS);
 840                         if (ret)
 841                                 return ret;
 842
 843                         reset->cond = BLK_ZONE_COND_EMPTY;
 844                         reset->wp = reset->start;
 845                 }
 846         } else if (ret != -ENOENT) {
 847                 /*
 848                  * For READ, we want the previous one. Move write pointer to
 849                  * the end of a zone, if it is at the head of a zone.
 850                  */
 851                 u64 zone_end = 0;
 852
 853                 if (wp == zones[0].start << SECTOR_SHIFT)
 854                         zone_end = zones[1].start + zones[1].capacity;
 855                 else if (wp == zones[1].start << SECTOR_SHIFT)
 856                         zone_end = zones[0].start + zones[0].capacity;
 857                 if (zone_end)
 858                         wp = ALIGN_DOWN(zone_end << SECTOR_SHIFT,
 859                                         BTRFS_SUPER_INFO_SIZE);
 860
 861                 wp -= BTRFS_SUPER_INFO_SIZE;
 862         }
 863
 864         *bytenr_ret = wp;
 865         return 0;
 866
 867 }
 868
 869 int btrfs_sb_log_location_bdev(struct block_device *bdev, int mirror, int rw,
 870                                u64 *bytenr_ret)
 871 {
 872         struct blk_zone zones[BTRFS_NR_SB_LOG_ZONES];
 873         sector_t zone_sectors;
 874         u32 sb_zone;
 875         int ret;
 876         u8 zone_sectors_shift;
 877         sector_t nr_sectors;
 878         u32 nr_zones;
 879
 880         if (!bdev_is_zoned(bdev)) {
 881                 *bytenr_ret = btrfs_sb_offset(mirror);
 882                 return 0;
 883         }
 884
 885         ASSERT(rw == READ || rw == WRITE);
 886
 887         zone_sectors = bdev_zone_sectors(bdev);
 888         if (!is_power_of_2(zone_sectors))
 889                 return -EINVAL;
 890         zone_sectors_shift = ilog2(zone_sectors);
 891         nr_sectors = bdev_nr_sectors(bdev);
 892         nr_zones = nr_sectors >> zone_sectors_shift;
 893
 894         sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
 895         if (sb_zone + 1 >= nr_zones)
 896                 return -ENOENT;
 897
 898         ret = blkdev_report_zones(bdev, zone_start_sector(sb_zone, bdev),
 899                                   BTRFS_NR_SB_LOG_ZONES, copy_zone_info_cb,
 900                                   zones);
 901         if (ret < 0)
 902                 return ret;
 903         if (ret != BTRFS_NR_SB_LOG_ZONES)
 904                 return -EIO;
 905
 906         return sb_log_location(bdev, zones, rw, bytenr_ret);
 907 }
 908
 909 int btrfs_sb_log_location(struct btrfs_device *device, int mirror, int rw,
 910                           u64 *bytenr_ret)
 911 {
 912         struct btrfs_zoned_device_info *zinfo = device->zone_info;
 913         u32 zone_num;
 914
 915         /*
 916          * For a zoned filesystem on a non-zoned block device, use the same
 917          * super block locations as regular filesystem. Doing so, the super
 918          * block can always be retrieved and the zoned flag of the volume
 919          * detected from the super block information.
 920          */
 921         if (!bdev_is_zoned(device->bdev)) {
 922                 *bytenr_ret = btrfs_sb_offset(mirror);
 923                 return 0;
 924         }
 925
 926         zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
 927         if (zone_num + 1 >= zinfo->nr_zones)
 928                 return -ENOENT;
 929
 930         return sb_log_location(device->bdev,
 931                                &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror],
 932                                rw, bytenr_ret);
 933 }
 934
 935 static inline bool is_sb_log_zone(struct btrfs_zoned_device_info *zinfo,
 936                                   int mirror)
 937 {
 938         u32 zone_num;
 939
 940         if (!zinfo)
 941                 return false;
 942
 943         zone_num = sb_zone_number(zinfo->zone_size_shift, mirror);
 944         if (zone_num + 1 >= zinfo->nr_zones)
 945                 return false;
 946
 947         if (!test_bit(zone_num, zinfo->seq_zones))
 948                 return false;
 949
 950         return true;
 951 }
 952
 953 int btrfs_advance_sb_log(struct btrfs_device *device, int mirror)
 954 {
 955         struct btrfs_zoned_device_info *zinfo = device->zone_info;
 956         struct blk_zone *zone;
 957         int i;
 958
 959         if (!is_sb_log_zone(zinfo, mirror))
 960                 return 0;
 961
 962         zone = &zinfo->sb_zones[BTRFS_NR_SB_LOG_ZONES * mirror];
 963         for (i = 0; i < BTRFS_NR_SB_LOG_ZONES; i++) {
 964                 /* Advance the next zone */
 965                 if (zone->cond == BLK_ZONE_COND_FULL) {
 966                         zone++;
 967                         continue;
 968                 }
 969
 970                 if (zone->cond == BLK_ZONE_COND_EMPTY)
 971                         zone->cond = BLK_ZONE_COND_IMP_OPEN;
 972
 973                 zone->wp += SUPER_INFO_SECTORS;
 974
 975                 if (sb_zone_is_full(zone)) {
 976                         /*
 977                          * No room left to write new superblock. Since
 978                          * superblock is written with REQ_SYNC, it is safe to
 979                          * finish the zone now.
 980                          *
 981                          * If the write pointer is exactly at the capacity,
 982                          * explicit ZONE_FINISH is not necessary.
 983                          */
 984                         if (zone->wp != zone->start + zone->capacity) {
 985                                 int ret;
 986
 987                                 ret = blkdev_zone_mgmt(device->bdev,
 988                                                 REQ_OP_ZONE_FINISH, zone->start,
 989                                                 zone->len, GFP_NOFS);
 990                                 if (ret)
 991                                         return ret;
 992                         }
 993
 994                         zone->wp = zone->start + zone->len;
 995                         zone->cond = BLK_ZONE_COND_FULL;
 996                 }
 997                 return 0;
 998         }
 999
1000         /* All the zones are FULL. Should not reach here. */
1001         ASSERT(0);
1002         return -EIO;
1003 }
1004
1005 int btrfs_reset_sb_log_zones(struct block_device *bdev, int mirror)
1006 {
1007         sector_t zone_sectors;
1008         sector_t nr_sectors;
1009         u8 zone_sectors_shift;
1010         u32 sb_zone;
1011         u32 nr_zones;
1012
1013         zone_sectors = bdev_zone_sectors(bdev);
1014         zone_sectors_shift = ilog2(zone_sectors);
1015         nr_sectors = bdev_nr_sectors(bdev);
1016         nr_zones = nr_sectors >> zone_sectors_shift;
1017
1018         sb_zone = sb_zone_number(zone_sectors_shift + SECTOR_SHIFT, mirror);
1019         if (sb_zone + 1 >= nr_zones)
1020                 return -ENOENT;
1021
1022         return blkdev_zone_mgmt(bdev, REQ_OP_ZONE_RESET,
1023                                 zone_start_sector(sb_zone, bdev),
1024                                 zone_sectors * BTRFS_NR_SB_LOG_ZONES, GFP_NOFS);
1025 }
1026
1027 /*
1028  * Find allocatable zones within a given region.
1029  *
1030  * @device:     the device to allocate a region on
1031  * @hole_start: the position of the hole to allocate the region
1032  * @num_bytes:  size of wanted region
1033  * @hole_end:   the end of the hole
1034  * @return:     position of allocatable zones
1035  *
1036  * Allocatable region should not contain any superblock locations.
1037  */
1038 u64 btrfs_find_allocatable_zones(struct btrfs_device *device, u64 hole_start,
1039                                  u64 hole_end, u64 num_bytes)
1040 {
1041         struct btrfs_zoned_device_info *zinfo = device->zone_info;
1042         const u8 shift = zinfo->zone_size_shift;
1043         u64 nzones = num_bytes >> shift;
1044         u64 pos = hole_start;
1045         u64 begin, end;
1046         bool have_sb;
1047         int i;
1048
1049         ASSERT(IS_ALIGNED(hole_start, zinfo->zone_size));
1050         ASSERT(IS_ALIGNED(num_bytes, zinfo->zone_size));
1051
1052         while (pos < hole_end) {
1053                 begin = pos >> shift;
1054                 end = begin + nzones;
1055
1056                 if (end > zinfo->nr_zones)
1057                         return hole_end;
1058
1059                 /* Check if zones in the region are all empty */
1060                 if (btrfs_dev_is_sequential(device, pos) &&
1061                     find_next_zero_bit(zinfo->empty_zones, end, begin) != end) {
1062                         pos += zinfo->zone_size;
1063                         continue;
1064                 }
1065
1066                 have_sb = false;
1067                 for (i = 0; i < BTRFS_SUPER_MIRROR_MAX; i++) {
1068                         u32 sb_zone;
1069                         u64 sb_pos;
1070
1071                         sb_zone = sb_zone_number(shift, i);
1072                         if (!(end <= sb_zone ||
1073                               sb_zone + BTRFS_NR_SB_LOG_ZONES <= begin)) {
1074                                 have_sb = true;
1075                                 pos = zone_start_physical(
1076                                         sb_zone + BTRFS_NR_SB_LOG_ZONES, zinfo);
1077                                 break;
1078                         }
1079
1080                         /* We also need to exclude regular superblock positions */
1081                         sb_pos = btrfs_sb_offset(i);
1082                         if (!(pos + num_bytes <= sb_pos ||
1083                               sb_pos + BTRFS_SUPER_INFO_SIZE <= pos)) {
1084                                 have_sb = true;
1085                                 pos = ALIGN(sb_pos + BTRFS_SUPER_INFO_SIZE,
1086                                             zinfo->zone_size);
1087                                 break;
1088                         }
1089                 }
1090                 if (!have_sb)
1091                         break;
1092         }
1093
1094         return pos;
1095 }
1096
1097 static bool btrfs_dev_set_active_zone(struct btrfs_device *device, u64 pos)
1098 {
1099         struct btrfs_zoned_device_info *zone_info = device->zone_info;
1100         unsigned int zno = (pos >> zone_info->zone_size_shift);
1101
1102         /* We can use any number of zones */
1103         if (zone_info->max_active_zones == 0)
1104                 return true;
1105
1106         if (!test_bit(zno, zone_info->active_zones)) {
1107                 /* Active zone left? */
1108                 if (atomic_dec_if_positive(&zone_info->active_zones_left) < 0)
1109                         return false;
1110                 if (test_and_set_bit(zno, zone_info->active_zones)) {
1111                         /* Someone already set the bit */
1112                         atomic_inc(&zone_info->active_zones_left);
1113                 }
1114         }
1115
1116         return true;
1117 }
1118
1119 static void btrfs_dev_clear_active_zone(struct btrfs_device *device, u64 pos)
1120 {
1121         struct btrfs_zoned_device_info *zone_info = device->zone_info;
1122         unsigned int zno = (pos >> zone_info->zone_size_shift);
1123
1124         /* We can use any number of zones */
1125         if (zone_info->max_active_zones == 0)
1126                 return;
1127
1128         if (test_and_clear_bit(zno, zone_info->active_zones))
1129                 atomic_inc(&zone_info->active_zones_left);
1130 }
1131
1132 int btrfs_reset_device_zone(struct btrfs_device *device, u64 physical,
1133                             u64 length, u64 *bytes)
1134 {
1135         int ret;
1136
1137         *bytes = 0;
1138         ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_RESET,
1139                                physical >> SECTOR_SHIFT, length >> SECTOR_SHIFT,
1140                                GFP_NOFS);
1141         if (ret)
1142                 return ret;
1143
1144         *bytes = length;
1145         while (length) {
1146                 btrfs_dev_set_zone_empty(device, physical);
1147                 btrfs_dev_clear_active_zone(device, physical);
1148                 physical += device->zone_info->zone_size;
1149                 length -= device->zone_info->zone_size;
1150         }
1151
1152         return 0;
1153 }
1154
1155 int btrfs_ensure_empty_zones(struct btrfs_device *device, u64 start, u64 size)
1156 {
1157         struct btrfs_zoned_device_info *zinfo = device->zone_info;
1158         const u8 shift = zinfo->zone_size_shift;
1159         unsigned long begin = start >> shift;
1160         unsigned long end = (start + size) >> shift;
1161         u64 pos;
1162         int ret;
1163
1164         ASSERT(IS_ALIGNED(start, zinfo->zone_size));
1165         ASSERT(IS_ALIGNED(size, zinfo->zone_size));
1166
1167         if (end > zinfo->nr_zones)
1168                 return -ERANGE;
1169
1170         /* All the zones are conventional */
1171         if (find_next_bit(zinfo->seq_zones, begin, end) == end)
1172                 return 0;
1173
1174         /* All the zones are sequential and empty */
1175         if (find_next_zero_bit(zinfo->seq_zones, begin, end) == end &&
1176             find_next_zero_bit(zinfo->empty_zones, begin, end) == end)
1177                 return 0;
1178
1179         for (pos = start; pos < start + size; pos += zinfo->zone_size) {
1180                 u64 reset_bytes;
1181
1182                 if (!btrfs_dev_is_sequential(device, pos) ||
1183                     btrfs_dev_is_empty_zone(device, pos))
1184                         continue;
1185
1186                 /* Free regions should be empty */
1187                 btrfs_warn_in_rcu(
1188                         device->fs_info,
1189                 "zoned: resetting device %s (devid %llu) zone %llu for allocation",
1190                         rcu_str_deref(device->name), device->devid, pos >> shift);
1191                 WARN_ON_ONCE(1);
1192
1193                 ret = btrfs_reset_device_zone(device, pos, zinfo->zone_size,
1194                                               &reset_bytes);
1195                 if (ret)
1196                         return ret;
1197         }
1198
1199         return 0;
1200 }
1201
1202 /*
1203  * Calculate an allocation pointer from the extent allocation information
1204  * for a block group consist of conventional zones. It is pointed to the
1205  * end of the highest addressed extent in the block group as an allocation
1206  * offset.
1207  */
1208 static int calculate_alloc_pointer(struct btrfs_block_group *cache,
1209                                    u64 *offset_ret, bool new)
1210 {
1211         struct btrfs_fs_info *fs_info = cache->fs_info;
1212         struct btrfs_root *root;
1213         struct btrfs_path *path;
1214         struct btrfs_key key;
1215         struct btrfs_key found_key;
1216         int ret;
1217         u64 length;
1218
1219         /*
1220          * Avoid  tree lookups for a new block group, there's no use for it.
1221          * It must always be 0.
1222          *
1223          * Also, we have a lock chain of extent buffer lock -> chunk mutex.
1224          * For new a block group, this function is called from
1225          * btrfs_make_block_group() which is already taking the chunk mutex.
1226          * Thus, we cannot call calculate_alloc_pointer() which takes extent
1227          * buffer locks to avoid deadlock.
1228          */
1229         if (new) {
1230                 *offset_ret = 0;
1231                 return 0;
1232         }
1233
1234         path = btrfs_alloc_path();
1235         if (!path)
1236                 return -ENOMEM;
1237
1238         key.objectid = cache->start + cache->length;
1239         key.type = 0;
1240         key.offset = 0;
1241
1242         root = btrfs_extent_root(fs_info, key.objectid);
1243         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1244         /* We should not find the exact match */
1245         if (!ret)
1246                 ret = -EUCLEAN;
1247         if (ret < 0)
1248                 goto out;
1249
1250         ret = btrfs_previous_extent_item(root, path, cache->start);
1251         if (ret) {
1252                 if (ret == 1) {
1253                         ret = 0;
1254                         *offset_ret = 0;
1255                 }
1256                 goto out;
1257         }
1258
1259         btrfs_item_key_to_cpu(path->nodes[0], &found_key, path->slots[0]);
1260
1261         if (found_key.type == BTRFS_EXTENT_ITEM_KEY)
1262                 length = found_key.offset;
1263         else
1264                 length = fs_info->nodesize;
1265
1266         if (!(found_key.objectid >= cache->start &&
1267                found_key.objectid + length <= cache->start + cache->length)) {
1268                 ret = -EUCLEAN;
1269                 goto out;
1270         }
1271         *offset_ret = found_key.objectid + length - cache->start;
1272         ret = 0;
1273
1274 out:
1275         btrfs_free_path(path);
1276         return ret;
1277 }
1278
1279 int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new)
1280 {
1281         struct btrfs_fs_info *fs_info = cache->fs_info;
1282         struct extent_map_tree *em_tree = &fs_info->mapping_tree;
1283         struct extent_map *em;
1284         struct map_lookup *map;
1285         struct btrfs_device *device;
1286         u64 logical = cache->start;
1287         u64 length = cache->length;
1288         int ret;
1289         int i;
1290         unsigned int nofs_flag;
1291         u64 *alloc_offsets = NULL;
1292         u64 *caps = NULL;
1293         u64 *physical = NULL;
1294         unsigned long *active = NULL;
1295         u64 last_alloc = 0;
1296         u32 num_sequential = 0, num_conventional = 0;
1297
1298         if (!btrfs_is_zoned(fs_info))
1299                 return 0;
1300
1301         /* Sanity check */
1302         if (!IS_ALIGNED(length, fs_info->zone_size)) {
1303                 btrfs_err(fs_info,
1304                 "zoned: block group %llu len %llu unaligned to zone size %llu",
1305                           logical, length, fs_info->zone_size);
1306                 return -EIO;
1307         }
1308
1309         /* Get the chunk mapping */
1310         read_lock(&em_tree->lock);
1311         em = lookup_extent_mapping(em_tree, logical, length);
1312         read_unlock(&em_tree->lock);
1313
1314         if (!em)
1315                 return -EINVAL;
1316
1317         map = em->map_lookup;
1318
1319         cache->physical_map = kmemdup(map, map_lookup_size(map->num_stripes), GFP_NOFS);
1320         if (!cache->physical_map) {
1321                 ret = -ENOMEM;
1322                 goto out;
1323         }
1324
1325         alloc_offsets = kcalloc(map->num_stripes, sizeof(*alloc_offsets), GFP_NOFS);
1326         if (!alloc_offsets) {
1327                 ret = -ENOMEM;
1328                 goto out;
1329         }
1330
1331         caps = kcalloc(map->num_stripes, sizeof(*caps), GFP_NOFS);
1332         if (!caps) {
1333                 ret = -ENOMEM;
1334                 goto out;
1335         }
1336
1337         physical = kcalloc(map->num_stripes, sizeof(*physical), GFP_NOFS);
1338         if (!physical) {
1339                 ret = -ENOMEM;
1340                 goto out;
1341         }
1342
1343         active = bitmap_zalloc(map->num_stripes, GFP_NOFS);
1344         if (!active) {
1345                 ret = -ENOMEM;
1346                 goto out;
1347         }
1348
1349         for (i = 0; i < map->num_stripes; i++) {
1350                 bool is_sequential;
1351                 struct blk_zone zone;
1352                 struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
1353                 int dev_replace_is_ongoing = 0;
1354
1355                 device = map->stripes[i].dev;
1356                 physical[i] = map->stripes[i].physical;
1357
1358                 if (device->bdev == NULL) {
1359                         alloc_offsets[i] = WP_MISSING_DEV;
1360                         continue;
1361                 }
1362
1363                 is_sequential = btrfs_dev_is_sequential(device, physical[i]);
1364                 if (is_sequential)
1365                         num_sequential++;
1366                 else
1367                         num_conventional++;
1368
1369                 /*
1370                  * Consider a zone as active if we can allow any number of
1371                  * active zones.
1372                  */
1373                 if (!device->zone_info->max_active_zones)
1374                         __set_bit(i, active);
1375
1376                 if (!is_sequential) {
1377                         alloc_offsets[i] = WP_CONVENTIONAL;
1378                         continue;
1379                 }
1380
1381                 /*
1382                  * This zone will be used for allocation, so mark this zone
1383                  * non-empty.
1384                  */
1385                 btrfs_dev_clear_zone_empty(device, physical[i]);
1386
1387                 down_read(&dev_replace->rwsem);
1388                 dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
1389                 if (dev_replace_is_ongoing && dev_replace->tgtdev != NULL)
1390                         btrfs_dev_clear_zone_empty(dev_replace->tgtdev, physical[i]);
1391                 up_read(&dev_replace->rwsem);
1392
1393                 /*
1394                  * The group is mapped to a sequential zone. Get the zone write
1395                  * pointer to determine the allocation offset within the zone.
1396                  */
1397                 WARN_ON(!IS_ALIGNED(physical[i], fs_info->zone_size));
1398                 nofs_flag = memalloc_nofs_save();
1399                 ret = btrfs_get_dev_zone(device, physical[i], &zone);
1400                 memalloc_nofs_restore(nofs_flag);
1401                 if (ret == -EIO || ret == -EOPNOTSUPP) {
1402                         ret = 0;
1403                         alloc_offsets[i] = WP_MISSING_DEV;
1404                         continue;
1405                 } else if (ret) {
1406                         goto out;
1407                 }
1408
1409                 if (zone.type == BLK_ZONE_TYPE_CONVENTIONAL) {
1410                         btrfs_err_in_rcu(fs_info,
1411         "zoned: unexpected conventional zone %llu on device %s (devid %llu)",
1412                                 zone.start << SECTOR_SHIFT,
1413                                 rcu_str_deref(device->name), device->devid);
1414                         ret = -EIO;
1415                         goto out;
1416                 }
1417
1418                 caps[i] = (zone.capacity << SECTOR_SHIFT);
1419
1420                 switch (zone.cond) {
1421                 case BLK_ZONE_COND_OFFLINE:
1422                 case BLK_ZONE_COND_READONLY:
1423                         btrfs_err(fs_info,
1424                 "zoned: offline/readonly zone %llu on device %s (devid %llu)",
1425                                   physical[i] >> device->zone_info->zone_size_shift,
1426                                   rcu_str_deref(device->name), device->devid);
1427                         alloc_offsets[i] = WP_MISSING_DEV;
1428                         break;
1429                 case BLK_ZONE_COND_EMPTY:
1430                         alloc_offsets[i] = 0;
1431                         break;
1432                 case BLK_ZONE_COND_FULL:
1433                         alloc_offsets[i] = caps[i];
1434                         break;
1435                 default:
1436                         /* Partially used zone */
1437                         alloc_offsets[i] =
1438                                         ((zone.wp - zone.start) << SECTOR_SHIFT);
1439                         __set_bit(i, active);
1440                         break;
1441                 }
1442         }
1443
1444         if (num_sequential > 0)
1445                 set_bit(BLOCK_GROUP_FLAG_SEQUENTIAL_ZONE, &cache->runtime_flags);
1446
1447         if (num_conventional > 0) {
1448                 /* Zone capacity is always zone size in emulation */
1449                 cache->zone_capacity = cache->length;
1450                 ret = calculate_alloc_pointer(cache, &last_alloc, new);
1451                 if (ret) {
1452                         btrfs_err(fs_info,
1453                         "zoned: failed to determine allocation offset of bg %llu",
1454                                   cache->start);
1455                         goto out;
1456                 } else if (map->num_stripes == num_conventional) {
1457                         cache->alloc_offset = last_alloc;
1458                         set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags);
1459                         goto out;
1460                 }
1461         }
1462
1463         switch (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
1464         case 0: /* single */
1465                 if (alloc_offsets[0] == WP_MISSING_DEV) {
1466                         btrfs_err(fs_info,
1467                         "zoned: cannot recover write pointer for zone %llu",
1468                                 physical[0]);
1469                         ret = -EIO;
1470                         goto out;
1471                 }
1472                 cache->alloc_offset = alloc_offsets[0];
1473                 cache->zone_capacity = caps[0];
1474                 if (test_bit(0, active))
1475                         set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags);
1476                 break;
1477         case BTRFS_BLOCK_GROUP_DUP:
1478                 if (map->type & BTRFS_BLOCK_GROUP_DATA) {
1479                         btrfs_err(fs_info, "zoned: profile DUP not yet supported on data bg");
1480                         ret = -EINVAL;
1481                         goto out;
1482                 }
1483                 if (alloc_offsets[0] == WP_MISSING_DEV) {
1484                         btrfs_err(fs_info,
1485                         "zoned: cannot recover write pointer for zone %llu",
1486                                 physical[0]);
1487                         ret = -EIO;
1488                         goto out;
1489                 }
1490                 if (alloc_offsets[1] == WP_MISSING_DEV) {
1491                         btrfs_err(fs_info,
1492                         "zoned: cannot recover write pointer for zone %llu",
1493                                 physical[1]);
1494                         ret = -EIO;
1495                         goto out;
1496                 }
1497                 if (alloc_offsets[0] != alloc_offsets[1]) {
1498                         btrfs_err(fs_info,
1499                         "zoned: write pointer offset mismatch of zones in DUP profile");
1500                         ret = -EIO;
1501                         goto out;
1502                 }
1503                 if (test_bit(0, active) != test_bit(1, active)) {
1504                         if (!btrfs_zone_activate(cache)) {
1505                                 ret = -EIO;
1506                                 goto out;
1507                         }
1508                 } else {
1509                         if (test_bit(0, active))
1510                                 set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
1511                                         &cache->runtime_flags);
1512                 }
1513                 cache->alloc_offset = alloc_offsets[0];
1514                 cache->zone_capacity = min(caps[0], caps[1]);
1515                 break;
1516         case BTRFS_BLOCK_GROUP_RAID1:
1517         case BTRFS_BLOCK_GROUP_RAID0:
1518         case BTRFS_BLOCK_GROUP_RAID10:
1519         case BTRFS_BLOCK_GROUP_RAID5:
1520         case BTRFS_BLOCK_GROUP_RAID6:
1521                 /* non-single profiles are not supported yet */
1522         default:
1523                 btrfs_err(fs_info, "zoned: profile %s not yet supported",
1524                           btrfs_bg_type_to_raid_name(map->type));
1525                 ret = -EINVAL;
1526                 goto out;
1527         }
1528
1529 out:
1530         if (cache->alloc_offset > fs_info->zone_size) {
1531                 btrfs_err(fs_info,
1532                         "zoned: invalid write pointer %llu in block group %llu",
1533                         cache->alloc_offset, cache->start);
1534                 ret = -EIO;
1535         }
1536
1537         if (cache->alloc_offset > cache->zone_capacity) {
1538                 btrfs_err(fs_info,
1539 "zoned: invalid write pointer %llu (larger than zone capacity %llu) in block group %llu",
1540                           cache->alloc_offset, cache->zone_capacity,
1541                           cache->start);
1542                 ret = -EIO;
1543         }
1544
1545         /* An extent is allocated after the write pointer */
1546         if (!ret && num_conventional && last_alloc > cache->alloc_offset) {
1547                 btrfs_err(fs_info,
1548                           "zoned: got wrong write pointer in BG %llu: %llu > %llu",
1549                           logical, last_alloc, cache->alloc_offset);
1550                 ret = -EIO;
1551         }
1552
1553         if (!ret) {
1554                 cache->meta_write_pointer = cache->alloc_offset + cache->start;
1555                 if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags)) {
1556                         btrfs_get_block_group(cache);
1557                         spin_lock(&fs_info->zone_active_bgs_lock);
1558                         list_add_tail(&cache->active_bg_list,
1559                                       &fs_info->zone_active_bgs);
1560                         spin_unlock(&fs_info->zone_active_bgs_lock);
1561                 }
1562         } else {
1563                 kfree(cache->physical_map);
1564                 cache->physical_map = NULL;
1565         }
1566         bitmap_free(active);
1567         kfree(physical);
1568         kfree(caps);
1569         kfree(alloc_offsets);
1570         free_extent_map(em);
1571
1572         return ret;
1573 }
1574
1575 void btrfs_calc_zone_unusable(struct btrfs_block_group *cache)
1576 {
1577         u64 unusable, free;
1578
1579         if (!btrfs_is_zoned(cache->fs_info))
1580                 return;
1581
1582         WARN_ON(cache->bytes_super != 0);
1583
1584         /* Check for block groups never get activated */
1585         if (test_bit(BTRFS_FS_ACTIVE_ZONE_TRACKING, &cache->fs_info->flags) &&
1586             cache->flags & (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM) &&
1587             !test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &cache->runtime_flags) &&
1588             cache->alloc_offset == 0) {
1589                 unusable = cache->length;
1590                 free = 0;
1591         } else {
1592                 unusable = (cache->alloc_offset - cache->used) +
1593                            (cache->length - cache->zone_capacity);
1594                 free = cache->zone_capacity - cache->alloc_offset;
1595         }
1596
1597         /* We only need ->free_space in ALLOC_SEQ block groups */
1598         cache->cached = BTRFS_CACHE_FINISHED;
1599         cache->free_space_ctl->free_space = free;
1600         cache->zone_unusable = unusable;
1601 }
1602
1603 void btrfs_redirty_list_add(struct btrfs_transaction *trans,
1604                             struct extent_buffer *eb)
1605 {
1606         struct btrfs_fs_info *fs_info = eb->fs_info;
1607
1608         if (!btrfs_is_zoned(fs_info) ||
1609             btrfs_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN) ||
1610             !list_empty(&eb->release_list))
1611                 return;
1612
1613         set_extent_buffer_dirty(eb);
1614         set_extent_bits_nowait(&trans->dirty_pages, eb->start,
1615                                eb->start + eb->len - 1, EXTENT_DIRTY);
1616         memzero_extent_buffer(eb, 0, eb->len);
1617         set_bit(EXTENT_BUFFER_NO_CHECK, &eb->bflags);
1618
1619         spin_lock(&trans->releasing_ebs_lock);
1620         list_add_tail(&eb->release_list, &trans->releasing_ebs);
1621         spin_unlock(&trans->releasing_ebs_lock);
1622         atomic_inc(&eb->refs);
1623 }
1624
1625 void btrfs_free_redirty_list(struct btrfs_transaction *trans)
1626 {
1627         spin_lock(&trans->releasing_ebs_lock);
1628         while (!list_empty(&trans->releasing_ebs)) {
1629                 struct extent_buffer *eb;
1630
1631                 eb = list_first_entry(&trans->releasing_ebs,
1632                                       struct extent_buffer, release_list);
1633                 list_del_init(&eb->release_list);
1634                 free_extent_buffer(eb);
1635         }
1636         spin_unlock(&trans->releasing_ebs_lock);
1637 }
1638
1639 bool btrfs_use_zone_append(struct btrfs_bio *bbio)
1640 {
1641         u64 start = (bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT);
1642         struct btrfs_inode *inode = bbio->inode;
1643         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1644         struct btrfs_block_group *cache;
1645         bool ret = false;
1646
1647         if (!btrfs_is_zoned(fs_info))
1648                 return false;
1649
1650         if (!is_data_inode(&inode->vfs_inode))
1651                 return false;
1652
1653         if (btrfs_op(&bbio->bio) != BTRFS_MAP_WRITE)
1654                 return false;
1655
1656         /*
1657          * Using REQ_OP_ZONE_APPNED for relocation can break assumptions on the
1658          * extent layout the relocation code has.
1659          * Furthermore we have set aside own block-group from which only the
1660          * relocation "process" can allocate and make sure only one process at a
1661          * time can add pages to an extent that gets relocated, so it's safe to
1662          * use regular REQ_OP_WRITE for this special case.
1663          */
1664         if (btrfs_is_data_reloc_root(inode->root))
1665                 return false;
1666
1667         cache = btrfs_lookup_block_group(fs_info, start);
1668         ASSERT(cache);
1669         if (!cache)
1670                 return false;
1671
1672         ret = !!test_bit(BLOCK_GROUP_FLAG_SEQUENTIAL_ZONE, &cache->runtime_flags);
1673         btrfs_put_block_group(cache);
1674
1675         return ret;
1676 }
1677
1678 void btrfs_record_physical_zoned(struct btrfs_bio *bbio)
1679 {
1680         const u64 physical = bbio->bio.bi_iter.bi_sector << SECTOR_SHIFT;
1681         struct btrfs_ordered_extent *ordered;
1682
1683         ordered = btrfs_lookup_ordered_extent(bbio->inode, bbio->file_offset);
1684         if (WARN_ON(!ordered))
1685                 return;
1686
1687         ordered->physical = physical;
1688         btrfs_put_ordered_extent(ordered);
1689 }
1690
1691 void btrfs_rewrite_logical_zoned(struct btrfs_ordered_extent *ordered)
1692 {
1693         struct btrfs_inode *inode = BTRFS_I(ordered->inode);
1694         struct btrfs_fs_info *fs_info = inode->root->fs_info;
1695         struct extent_map_tree *em_tree;
1696         struct extent_map *em;
1697         struct btrfs_ordered_sum *sum;
1698         u64 orig_logical = ordered->disk_bytenr;
1699         struct map_lookup *map;
1700         u64 physical = ordered->physical;
1701         u64 chunk_start_phys;
1702         u64 logical;
1703
1704         em = btrfs_get_chunk_map(fs_info, orig_logical, 1);
1705         if (IS_ERR(em))
1706                 return;
1707         map = em->map_lookup;
1708         chunk_start_phys = map->stripes[0].physical;
1709
1710         if (WARN_ON_ONCE(map->num_stripes > 1) ||
1711             WARN_ON_ONCE((map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) != 0) ||
1712             WARN_ON_ONCE(physical < chunk_start_phys) ||
1713             WARN_ON_ONCE(physical > chunk_start_phys + em->orig_block_len)) {
1714                 free_extent_map(em);
1715                 return;
1716         }
1717         logical = em->start + (physical - map->stripes[0].physical);
1718         free_extent_map(em);
1719
1720         if (orig_logical == logical)
1721                 return;
1722
1723         ordered->disk_bytenr = logical;
1724
1725         em_tree = &inode->extent_tree;
1726         write_lock(&em_tree->lock);
1727         em = search_extent_mapping(em_tree, ordered->file_offset,
1728                                    ordered->num_bytes);
1729         em->block_start = logical;
1730         free_extent_map(em);
1731         write_unlock(&em_tree->lock);
1732
1733         list_for_each_entry(sum, &ordered->list, list) {
1734                 if (logical < orig_logical)
1735                         sum->bytenr -= orig_logical - logical;
1736                 else
1737                         sum->bytenr += logical - orig_logical;
1738         }
1739 }
1740
1741 bool btrfs_check_meta_write_pointer(struct btrfs_fs_info *fs_info,
1742                                     struct extent_buffer *eb,
1743                                     struct btrfs_block_group **cache_ret)
1744 {
1745         struct btrfs_block_group *cache;
1746         bool ret = true;
1747
1748         if (!btrfs_is_zoned(fs_info))
1749                 return true;
1750
1751         cache = btrfs_lookup_block_group(fs_info, eb->start);
1752         if (!cache)
1753                 return true;
1754
1755         if (cache->meta_write_pointer != eb->start) {
1756                 btrfs_put_block_group(cache);
1757                 cache = NULL;
1758                 ret = false;
1759         } else {
1760                 cache->meta_write_pointer = eb->start + eb->len;
1761         }
1762
1763         *cache_ret = cache;
1764
1765         return ret;
1766 }
1767
1768 void btrfs_revert_meta_write_pointer(struct btrfs_block_group *cache,
1769                                      struct extent_buffer *eb)
1770 {
1771         if (!btrfs_is_zoned(eb->fs_info) || !cache)
1772                 return;
1773
1774         ASSERT(cache->meta_write_pointer == eb->start + eb->len);
1775         cache->meta_write_pointer = eb->start;
1776 }
1777
1778 int btrfs_zoned_issue_zeroout(struct btrfs_device *device, u64 physical, u64 length)
1779 {
1780         if (!btrfs_dev_is_sequential(device, physical))
1781                 return -EOPNOTSUPP;
1782
1783         return blkdev_issue_zeroout(device->bdev, physical >> SECTOR_SHIFT,
1784                                     length >> SECTOR_SHIFT, GFP_NOFS, 0);
1785 }
1786
1787 static int read_zone_info(struct btrfs_fs_info *fs_info, u64 logical,
1788                           struct blk_zone *zone)
1789 {
1790         struct btrfs_io_context *bioc = NULL;
1791         u64 mapped_length = PAGE_SIZE;
1792         unsigned int nofs_flag;
1793         int nmirrors;
1794         int i, ret;
1795
1796         ret = btrfs_map_sblock(fs_info, BTRFS_MAP_GET_READ_MIRRORS, logical,
1797                                &mapped_length, &bioc);
1798         if (ret || !bioc || mapped_length < PAGE_SIZE) {
1799                 ret = -EIO;
1800                 goto out_put_bioc;
1801         }
1802
1803         if (bioc->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) {
1804                 ret = -EINVAL;
1805                 goto out_put_bioc;
1806         }
1807
1808         nofs_flag = memalloc_nofs_save();
1809         nmirrors = (int)bioc->num_stripes;
1810         for (i = 0; i < nmirrors; i++) {
1811                 u64 physical = bioc->stripes[i].physical;
1812                 struct btrfs_device *dev = bioc->stripes[i].dev;
1813
1814                 /* Missing device */
1815                 if (!dev->bdev)
1816                         continue;
1817
1818                 ret = btrfs_get_dev_zone(dev, physical, zone);
1819                 /* Failing device */
1820                 if (ret == -EIO || ret == -EOPNOTSUPP)
1821                         continue;
1822                 break;
1823         }
1824         memalloc_nofs_restore(nofs_flag);
1825 out_put_bioc:
1826         btrfs_put_bioc(bioc);
1827         return ret;
1828 }
1829
1830 /*
1831  * Synchronize write pointer in a zone at @physical_start on @tgt_dev, by
1832  * filling zeros between @physical_pos to a write pointer of dev-replace
1833  * source device.
1834  */
1835 int btrfs_sync_zone_write_pointer(struct btrfs_device *tgt_dev, u64 logical,
1836                                     u64 physical_start, u64 physical_pos)
1837 {
1838         struct btrfs_fs_info *fs_info = tgt_dev->fs_info;
1839         struct blk_zone zone;
1840         u64 length;
1841         u64 wp;
1842         int ret;
1843
1844         if (!btrfs_dev_is_sequential(tgt_dev, physical_pos))
1845                 return 0;
1846
1847         ret = read_zone_info(fs_info, logical, &zone);
1848         if (ret)
1849                 return ret;
1850
1851         wp = physical_start + ((zone.wp - zone.start) << SECTOR_SHIFT);
1852
1853         if (physical_pos == wp)
1854                 return 0;
1855
1856         if (physical_pos > wp)
1857                 return -EUCLEAN;
1858
1859         length = wp - physical_pos;
1860         return btrfs_zoned_issue_zeroout(tgt_dev, physical_pos, length);
1861 }
1862
1863 /*
1864  * Activate block group and underlying device zones
1865  *
1866  * @block_group: the block group to activate
1867  *
1868  * Return: true on success, false otherwise
1869  */
1870 bool btrfs_zone_activate(struct btrfs_block_group *block_group)
1871 {
1872         struct btrfs_fs_info *fs_info = block_group->fs_info;
1873         struct btrfs_space_info *space_info = block_group->space_info;
1874         struct map_lookup *map;
1875         struct btrfs_device *device;
1876         u64 physical;
1877         bool ret;
1878         int i;
1879
1880         if (!btrfs_is_zoned(block_group->fs_info))
1881                 return true;
1882
1883         map = block_group->physical_map;
1884
1885         spin_lock(&space_info->lock);
1886         spin_lock(&block_group->lock);
1887         if (test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) {
1888                 ret = true;
1889                 goto out_unlock;
1890         }
1891
1892         /* No space left */
1893         if (btrfs_zoned_bg_is_full(block_group)) {
1894                 ret = false;
1895                 goto out_unlock;
1896         }
1897
1898         for (i = 0; i < map->num_stripes; i++) {
1899                 device = map->stripes[i].dev;
1900                 physical = map->stripes[i].physical;
1901
1902                 if (device->zone_info->max_active_zones == 0)
1903                         continue;
1904
1905                 if (!btrfs_dev_set_active_zone(device, physical)) {
1906                         /* Cannot activate the zone */
1907                         ret = false;
1908                         goto out_unlock;
1909                 }
1910         }
1911
1912         /* Successfully activated all the zones */
1913         set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags);
1914         WARN_ON(block_group->alloc_offset != 0);
1915         if (block_group->zone_unusable == block_group->length) {
1916                 block_group->zone_unusable = block_group->length - block_group->zone_capacity;
1917                 space_info->bytes_zone_unusable -= block_group->zone_capacity;
1918         }
1919         spin_unlock(&block_group->lock);
1920         btrfs_try_granting_tickets(fs_info, space_info);
1921         spin_unlock(&space_info->lock);
1922
1923         /* For the active block group list */
1924         btrfs_get_block_group(block_group);
1925
1926         spin_lock(&fs_info->zone_active_bgs_lock);
1927         list_add_tail(&block_group->active_bg_list, &fs_info->zone_active_bgs);
1928         spin_unlock(&fs_info->zone_active_bgs_lock);
1929
1930         return true;
1931
1932 out_unlock:
1933         spin_unlock(&block_group->lock);
1934         spin_unlock(&space_info->lock);
1935         return ret;
1936 }
1937
1938 static void wait_eb_writebacks(struct btrfs_block_group *block_group)
1939 {
1940         struct btrfs_fs_info *fs_info = block_group->fs_info;
1941         const u64 end = block_group->start + block_group->length;
1942         struct radix_tree_iter iter;
1943         struct extent_buffer *eb;
1944         void __rcu **slot;
1945
1946         rcu_read_lock();
1947         radix_tree_for_each_slot(slot, &fs_info->buffer_radix, &iter,
1948                                  block_group->start >> fs_info->sectorsize_bits) {
1949                 eb = radix_tree_deref_slot(slot);
1950                 if (!eb)
1951                         continue;
1952                 if (radix_tree_deref_retry(eb)) {
1953                         slot = radix_tree_iter_retry(&iter);
1954                         continue;
1955                 }
1956
1957                 if (eb->start < block_group->start)
1958                         continue;
1959                 if (eb->start >= end)
1960                         break;
1961
1962                 slot = radix_tree_iter_resume(slot, &iter);
1963                 rcu_read_unlock();
1964                 wait_on_extent_buffer_writeback(eb);
1965                 rcu_read_lock();
1966         }
1967         rcu_read_unlock();
1968 }
1969
1970 static int do_zone_finish(struct btrfs_block_group *block_group, bool fully_written)
1971 {
1972         struct btrfs_fs_info *fs_info = block_group->fs_info;
1973         struct map_lookup *map;
1974         const bool is_metadata = (block_group->flags &
1975                         (BTRFS_BLOCK_GROUP_METADATA | BTRFS_BLOCK_GROUP_SYSTEM));
1976         int ret = 0;
1977         int i;
1978
1979         spin_lock(&block_group->lock);
1980         if (!test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags)) {
1981                 spin_unlock(&block_group->lock);
1982                 return 0;
1983         }
1984
1985         /* Check if we have unwritten allocated space */
1986         if (is_metadata &&
1987             block_group->start + block_group->alloc_offset > block_group->meta_write_pointer) {
1988                 spin_unlock(&block_group->lock);
1989                 return -EAGAIN;
1990         }
1991
1992         /*
1993          * If we are sure that the block group is full (= no more room left for
1994          * new allocation) and the IO for the last usable block is completed, we
1995          * don't need to wait for the other IOs. This holds because we ensure
1996          * the sequential IO submissions using the ZONE_APPEND command for data
1997          * and block_group->meta_write_pointer for metadata.
1998          */
1999         if (!fully_written) {
2000                 spin_unlock(&block_group->lock);
2001
2002                 ret = btrfs_inc_block_group_ro(block_group, false);
2003                 if (ret)
2004                         return ret;
2005
2006                 /* Ensure all writes in this block group finish */
2007                 btrfs_wait_block_group_reservations(block_group);
2008                 /* No need to wait for NOCOW writers. Zoned mode does not allow that */
2009                 btrfs_wait_ordered_roots(fs_info, U64_MAX, block_group->start,
2010                                          block_group->length);
2011                 /* Wait for extent buffers to be written. */
2012                 if (is_metadata)
2013                         wait_eb_writebacks(block_group);
2014
2015                 spin_lock(&block_group->lock);
2016
2017                 /*
2018                  * Bail out if someone already deactivated the block group, or
2019                  * allocated space is left in the block group.
2020                  */
2021                 if (!test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
2022                               &block_group->runtime_flags)) {
2023                         spin_unlock(&block_group->lock);
2024                         btrfs_dec_block_group_ro(block_group);
2025                         return 0;
2026                 }
2027
2028                 if (block_group->reserved) {
2029                         spin_unlock(&block_group->lock);
2030                         btrfs_dec_block_group_ro(block_group);
2031                         return -EAGAIN;
2032                 }
2033         }
2034
2035         clear_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &block_group->runtime_flags);
2036         block_group->alloc_offset = block_group->zone_capacity;
2037         block_group->free_space_ctl->free_space = 0;
2038         btrfs_clear_treelog_bg(block_group);
2039         btrfs_clear_data_reloc_bg(block_group);
2040         spin_unlock(&block_group->lock);
2041
2042         map = block_group->physical_map;
2043         for (i = 0; i < map->num_stripes; i++) {
2044                 struct btrfs_device *device = map->stripes[i].dev;
2045                 const u64 physical = map->stripes[i].physical;
2046
2047                 if (device->zone_info->max_active_zones == 0)
2048                         continue;
2049
2050                 ret = blkdev_zone_mgmt(device->bdev, REQ_OP_ZONE_FINISH,
2051                                        physical >> SECTOR_SHIFT,
2052                                        device->zone_info->zone_size >> SECTOR_SHIFT,
2053                                        GFP_NOFS);
2054
2055                 if (ret)
2056                         return ret;
2057
2058                 btrfs_dev_clear_active_zone(device, physical);
2059         }
2060
2061         if (!fully_written)
2062                 btrfs_dec_block_group_ro(block_group);
2063
2064         spin_lock(&fs_info->zone_active_bgs_lock);
2065         ASSERT(!list_empty(&block_group->active_bg_list));
2066         list_del_init(&block_group->active_bg_list);
2067         spin_unlock(&fs_info->zone_active_bgs_lock);
2068
2069         /* For active_bg_list */
2070         btrfs_put_block_group(block_group);
2071
2072         clear_and_wake_up_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags);
2073
2074         return 0;
2075 }
2076
2077 int btrfs_zone_finish(struct btrfs_block_group *block_group)
2078 {
2079         if (!btrfs_is_zoned(block_group->fs_info))
2080                 return 0;
2081
2082         return do_zone_finish(block_group, false);
2083 }
2084
2085 bool btrfs_can_activate_zone(struct btrfs_fs_devices *fs_devices, u64 flags)
2086 {
2087         struct btrfs_fs_info *fs_info = fs_devices->fs_info;
2088         struct btrfs_device *device;
2089         bool ret = false;
2090
2091         if (!btrfs_is_zoned(fs_info))
2092                 return true;
2093
2094         /* Check if there is a device with active zones left */
2095         mutex_lock(&fs_info->chunk_mutex);
2096         list_for_each_entry(device, &fs_devices->alloc_list, dev_alloc_list) {
2097                 struct btrfs_zoned_device_info *zinfo = device->zone_info;
2098
2099                 if (!device->bdev)
2100                         continue;
2101
2102                 if (!zinfo->max_active_zones) {
2103                         ret = true;
2104                         break;
2105                 }
2106
2107                 switch (flags & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
2108                 case 0: /* single */
2109                         ret = (atomic_read(&zinfo->active_zones_left) >= 1);
2110                         break;
2111                 case BTRFS_BLOCK_GROUP_DUP:
2112                         ret = (atomic_read(&zinfo->active_zones_left) >= 2);
2113                         break;
2114                 }
2115                 if (ret)
2116                         break;
2117         }
2118         mutex_unlock(&fs_info->chunk_mutex);
2119
2120         if (!ret)
2121                 set_bit(BTRFS_FS_NEED_ZONE_FINISH, &fs_info->flags);
2122
2123         return ret;
2124 }
2125
2126 void btrfs_zone_finish_endio(struct btrfs_fs_info *fs_info, u64 logical, u64 length)
2127 {
2128         struct btrfs_block_group *block_group;
2129         u64 min_alloc_bytes;
2130
2131         if (!btrfs_is_zoned(fs_info))
2132                 return;
2133
2134         block_group = btrfs_lookup_block_group(fs_info, logical);
2135         ASSERT(block_group);
2136
2137         /* No MIXED_BG on zoned btrfs. */
2138         if (block_group->flags & BTRFS_BLOCK_GROUP_DATA)
2139                 min_alloc_bytes = fs_info->sectorsize;
2140         else
2141                 min_alloc_bytes = fs_info->nodesize;
2142
2143         /* Bail out if we can allocate more data from this block group. */
2144         if (logical + length + min_alloc_bytes <=
2145             block_group->start + block_group->zone_capacity)
2146                 goto out;
2147
2148         do_zone_finish(block_group, true);
2149
2150 out:
2151         btrfs_put_block_group(block_group);
2152 }
2153
2154 static void btrfs_zone_finish_endio_workfn(struct work_struct *work)
2155 {
2156         struct btrfs_block_group *bg =
2157                 container_of(work, struct btrfs_block_group, zone_finish_work);
2158
2159         wait_on_extent_buffer_writeback(bg->last_eb);
2160         free_extent_buffer(bg->last_eb);
2161         btrfs_zone_finish_endio(bg->fs_info, bg->start, bg->length);
2162         btrfs_put_block_group(bg);
2163 }
2164
2165 void btrfs_schedule_zone_finish_bg(struct btrfs_block_group *bg,
2166                                    struct extent_buffer *eb)
2167 {
2168         if (!test_bit(BLOCK_GROUP_FLAG_SEQUENTIAL_ZONE, &bg->runtime_flags) ||
2169             eb->start + eb->len * 2 <= bg->start + bg->zone_capacity)
2170                 return;
2171
2172         if (WARN_ON(bg->zone_finish_work.func == btrfs_zone_finish_endio_workfn)) {
2173                 btrfs_err(bg->fs_info, "double scheduling of bg %llu zone finishing",
2174                           bg->start);
2175                 return;
2176         }
2177
2178         /* For the work */
2179         btrfs_get_block_group(bg);
2180         atomic_inc(&eb->refs);
2181         bg->last_eb = eb;
2182         INIT_WORK(&bg->zone_finish_work, btrfs_zone_finish_endio_workfn);
2183         queue_work(system_unbound_wq, &bg->zone_finish_work);
2184 }
2185
2186 void btrfs_clear_data_reloc_bg(struct btrfs_block_group *bg)
2187 {
2188         struct btrfs_fs_info *fs_info = bg->fs_info;
2189
2190         spin_lock(&fs_info->relocation_bg_lock);
2191         if (fs_info->data_reloc_bg == bg->start)
2192                 fs_info->data_reloc_bg = 0;
2193         spin_unlock(&fs_info->relocation_bg_lock);
2194 }
2195
2196 void btrfs_free_zone_cache(struct btrfs_fs_info *fs_info)
2197 {
2198         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2199         struct btrfs_device *device;
2200
2201         if (!btrfs_is_zoned(fs_info))
2202                 return;
2203
2204         mutex_lock(&fs_devices->device_list_mutex);
2205         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2206                 if (device->zone_info) {
2207                         vfree(device->zone_info->zone_cache);
2208                         device->zone_info->zone_cache = NULL;
2209                 }
2210         }
2211         mutex_unlock(&fs_devices->device_list_mutex);
2212 }
2213
2214 bool btrfs_zoned_should_reclaim(struct btrfs_fs_info *fs_info)
2215 {
2216         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
2217         struct btrfs_device *device;
2218         u64 used = 0;
2219         u64 total = 0;
2220         u64 factor;
2221
2222         ASSERT(btrfs_is_zoned(fs_info));
2223
2224         if (fs_info->bg_reclaim_threshold == 0)
2225                 return false;
2226
2227         mutex_lock(&fs_devices->device_list_mutex);
2228         list_for_each_entry(device, &fs_devices->devices, dev_list) {
2229                 if (!device->bdev)
2230                         continue;
2231
2232                 total += device->disk_total_bytes;
2233                 used += device->bytes_used;
2234         }
2235         mutex_unlock(&fs_devices->device_list_mutex);
2236
2237         factor = div64_u64(used * 100, total);
2238         return factor >= fs_info->bg_reclaim_threshold;
2239 }
2240
2241 void btrfs_zoned_release_data_reloc_bg(struct btrfs_fs_info *fs_info, u64 logical,
2242                                        u64 length)
2243 {
2244         struct btrfs_block_group *block_group;
2245
2246         if (!btrfs_is_zoned(fs_info))
2247                 return;
2248
2249         block_group = btrfs_lookup_block_group(fs_info, logical);
2250         /* It should be called on a previous data relocation block group. */
2251         ASSERT(block_group && (block_group->flags & BTRFS_BLOCK_GROUP_DATA));
2252
2253         spin_lock(&block_group->lock);
2254         if (!test_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC, &block_group->runtime_flags))
2255                 goto out;
2256
2257         /* All relocation extents are written. */
2258         if (block_group->start + block_group->alloc_offset == logical + length) {
2259                 /* Now, release this block group for further allocations. */
2260                 clear_bit(BLOCK_GROUP_FLAG_ZONED_DATA_RELOC,
2261                           &block_group->runtime_flags);
2262         }
2263
2264 out:
2265         spin_unlock(&block_group->lock);
2266         btrfs_put_block_group(block_group);
2267 }
2268
2269 int btrfs_zone_finish_one_bg(struct btrfs_fs_info *fs_info)
2270 {
2271         struct btrfs_block_group *block_group;
2272         struct btrfs_block_group *min_bg = NULL;
2273         u64 min_avail = U64_MAX;
2274         int ret;
2275
2276         spin_lock(&fs_info->zone_active_bgs_lock);
2277         list_for_each_entry(block_group, &fs_info->zone_active_bgs,
2278                             active_bg_list) {
2279                 u64 avail;
2280
2281                 spin_lock(&block_group->lock);
2282                 if (block_group->reserved || block_group->alloc_offset == 0 ||
2283                     (block_group->flags & BTRFS_BLOCK_GROUP_SYSTEM)) {
2284                         spin_unlock(&block_group->lock);
2285                         continue;
2286                 }
2287
2288                 avail = block_group->zone_capacity - block_group->alloc_offset;
2289                 if (min_avail > avail) {
2290                         if (min_bg)
2291                                 btrfs_put_block_group(min_bg);
2292                         min_bg = block_group;
2293                         min_avail = avail;
2294                         btrfs_get_block_group(min_bg);
2295                 }
2296                 spin_unlock(&block_group->lock);
2297         }
2298         spin_unlock(&fs_info->zone_active_bgs_lock);
2299
2300         if (!min_bg)
2301                 return 0;
2302
2303         ret = btrfs_zone_finish(min_bg);
2304         btrfs_put_block_group(min_bg);
2305
2306         return ret < 0 ? ret : 1;
2307 }
2308
2309 int btrfs_zoned_activate_one_bg(struct btrfs_fs_info *fs_info,
2310                                 struct btrfs_space_info *space_info,
2311                                 bool do_finish)
2312 {
2313         struct btrfs_block_group *bg;
2314         int index;
2315
2316         if (!btrfs_is_zoned(fs_info) || (space_info->flags & BTRFS_BLOCK_GROUP_DATA))
2317                 return 0;
2318
2319         for (;;) {
2320                 int ret;
2321                 bool need_finish = false;
2322
2323                 down_read(&space_info->groups_sem);
2324                 for (index = 0; index < BTRFS_NR_RAID_TYPES; index++) {
2325                         list_for_each_entry(bg, &space_info->block_groups[index],
2326                                             list) {
2327                                 if (!spin_trylock(&bg->lock))
2328                                         continue;
2329                                 if (btrfs_zoned_bg_is_full(bg) ||
2330                                     test_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE,
2331                                              &bg->runtime_flags)) {
2332                                         spin_unlock(&bg->lock);
2333                                         continue;
2334                                 }
2335                                 spin_unlock(&bg->lock);
2336
2337                                 if (btrfs_zone_activate(bg)) {
2338                                         up_read(&space_info->groups_sem);
2339                                         return 1;
2340                                 }
2341
2342                                 need_finish = true;
2343                         }
2344                 }
2345                 up_read(&space_info->groups_sem);
2346
2347                 if (!do_finish || !need_finish)
2348                         break;
2349
2350                 ret = btrfs_zone_finish_one_bg(fs_info);
2351                 if (ret == 0)
2352                         break;
2353                 if (ret < 0)
2354                         return ret;
2355         }
2356
2357         return 0;
2358 }