git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/move.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include "bcachefs.h"
   4 #include "alloc_background.h"
   5 #include "alloc_foreground.h"
   6 #include "backpointers.h"
   7 #include "bkey_buf.h"
   8 #include "btree_gc.h"
   9 #include "btree_update.h"
  10 #include "btree_update_interior.h"
  11 #include "buckets.h"
  12 #include "disk_groups.h"
  13 #include "ec.h"
  14 #include "error.h"
  15 #include "inode.h"
  16 #include "io.h"
  17 #include "journal_reclaim.h"
  18 #include "move.h"
  19 #include "replicas.h"
  20 #include "subvolume.h"
  21 #include "super-io.h"
  22 #include "keylist.h"
  23
  24 #include <linux/ioprio.h>
  25 #include <linux/kthread.h>
  26
  27 #include <trace/events/bcachefs.h>
  28
  29 #define SECTORS_IN_FLIGHT_PER_DEVICE    2048
  30
  31 struct moving_io {
  32         struct list_head        list;
  33         struct closure          cl;
  34         bool                    read_completed;
  35
  36         unsigned                read_sectors;
  37         unsigned                write_sectors;
  38
  39         struct bch_read_bio     rbio;
  40
  41         struct migrate_write    write;
  42         /* Must be last since it is variable size */
  43         struct bio_vec          bi_inline_vecs[0];
  44 };
  45
  46 struct moving_context {
  47         /* Closure for waiting on all reads and writes to complete */
  48         struct closure          cl;
  49
  50         struct bch_move_stats   *stats;
  51
  52         struct list_head        reads;
  53
  54         /* in flight sectors: */
  55         atomic_t                read_sectors;
  56         atomic_t                write_sectors;
  57
  58         wait_queue_head_t       wait;
  59 };
  60
  61 static int insert_snapshot_whiteouts(struct btree_trans *trans,
  62                                      enum btree_id id,
  63                                      struct bpos old_pos,
  64                                      struct bpos new_pos)
  65 {
  66         struct bch_fs *c = trans->c;
  67         struct btree_iter iter, update_iter;
  68         struct bkey_s_c k;
  69         struct snapshots_seen s;
  70         int ret;
  71
  72         if (!btree_type_has_snapshots(id))
  73                 return 0;
  74
  75         snapshots_seen_init(&s);
  76
  77         if (!bkey_cmp(old_pos, new_pos))
  78                 return 0;
  79
  80         if (!snapshot_t(c, old_pos.snapshot)->children[0])
  81                 return 0;
  82
  83         bch2_trans_iter_init(trans, &iter, id, old_pos,
  84                              BTREE_ITER_NOT_EXTENTS|
  85                              BTREE_ITER_ALL_SNAPSHOTS);
  86         while (1) {
  87 next:
  88                 k = bch2_btree_iter_prev(&iter);
  89                 ret = bkey_err(k);
  90                 if (ret)
  91                         break;
  92
  93                 if (bkey_cmp(old_pos, k.k->p))
  94                         break;
  95
  96                 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, old_pos.snapshot)) {
  97                         struct bkey_i *update;
  98                         u32 *i;
  99
 100                         darray_for_each(s.ids, i)
 101                                 if (bch2_snapshot_is_ancestor(c, k.k->p.snapshot, *i))
 102                                         goto next;
 103
 104                         update = bch2_trans_kmalloc(trans, sizeof(struct bkey_i));
 105
 106                         ret = PTR_ERR_OR_ZERO(update);
 107                         if (ret)
 108                                 break;
 109
 110                         bkey_init(&update->k);
 111                         update->k.p = new_pos;
 112                         update->k.p.snapshot = k.k->p.snapshot;
 113
 114                         bch2_trans_iter_init(trans, &update_iter, id, update->k.p,
 115                                              BTREE_ITER_NOT_EXTENTS|
 116                                              BTREE_ITER_ALL_SNAPSHOTS|
 117                                              BTREE_ITER_INTENT);
 118                         ret   = bch2_btree_iter_traverse(&update_iter) ?:
 119                                 bch2_trans_update(trans, &update_iter, update,
 120                                           BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE);
 121                         bch2_trans_iter_exit(trans, &update_iter);
 122                         if (ret)
 123                                 break;
 124
 125                         ret = snapshots_seen_add(c, &s, k.k->p.snapshot);
 126                         if (ret)
 127                                 break;
 128                 }
 129         }
 130         bch2_trans_iter_exit(trans, &iter);
 131         darray_exit(&s.ids);
 132
 133         return ret;
 134 }
 135
 136 static int bch2_migrate_index_update(struct bch_write_op *op)
 137 {
 138         struct bch_fs *c = op->c;
 139         struct btree_trans trans;
 140         struct btree_iter iter;
 141         struct migrate_write *m =
 142                 container_of(op, struct migrate_write, op);
 143         struct open_bucket *ec_ob = ec_open_bucket(c, &op->open_buckets);
 144         struct keylist *keys = &op->insert_keys;
 145         struct bkey_buf _new, _insert;
 146         int ret = 0;
 147
 148         bch2_bkey_buf_init(&_new);
 149         bch2_bkey_buf_init(&_insert);
 150         bch2_bkey_buf_realloc(&_insert, c, U8_MAX);
 151
 152         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 1024);
 153
 154         bch2_trans_iter_init(&trans, &iter, m->btree_id,
 155                              bkey_start_pos(&bch2_keylist_front(keys)->k),
 156                              BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
 157
 158         while (1) {
 159                 struct bkey_s_c k;
 160                 struct bkey_i *insert;
 161                 struct bkey_i_extent *new;
 162                 const union bch_extent_entry *entry;
 163                 struct extent_ptr_decoded p;
 164                 struct bpos next_pos;
 165                 bool did_work = false;
 166                 bool should_check_enospc;
 167                 s64 i_sectors_delta = 0, disk_sectors_delta = 0;
 168
 169                 bch2_trans_begin(&trans);
 170
 171                 k = bch2_btree_iter_peek_slot(&iter);
 172                 ret = bkey_err(k);
 173                 if (ret)
 174                         goto err;
 175
 176                 new = bkey_i_to_extent(bch2_keylist_front(keys));
 177
 178                 if (bversion_cmp(k.k->version, new->k.version) ||
 179                     !bch2_bkey_matches_ptr(c, k, m->ptr, m->offset))
 180                         goto nomatch;
 181
 182                 bkey_reassemble(_insert.k, k);
 183                 insert = _insert.k;
 184
 185                 bch2_bkey_buf_copy(&_new, c, bch2_keylist_front(keys));
 186                 new = bkey_i_to_extent(_new.k);
 187                 bch2_cut_front(iter.pos, &new->k_i);
 188
 189                 bch2_cut_front(iter.pos,        insert);
 190                 bch2_cut_back(new->k.p,         insert);
 191                 bch2_cut_back(insert->k.p,      &new->k_i);
 192
 193                 if (m->data_cmd == DATA_REWRITE) {
 194                         struct bch_extent_ptr *new_ptr, *old_ptr = (void *)
 195                                 bch2_bkey_has_device(bkey_i_to_s_c(insert),
 196                                                      m->data_opts.rewrite_dev);
 197                         if (!old_ptr)
 198                                 goto nomatch;
 199
 200                         if (old_ptr->cached)
 201                                 extent_for_each_ptr(extent_i_to_s(new), new_ptr)
 202                                         new_ptr->cached = true;
 203
 204                         __bch2_bkey_drop_ptr(bkey_i_to_s(insert), old_ptr);
 205                 }
 206
 207                 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) {
 208                         if (bch2_bkey_has_device(bkey_i_to_s_c(insert), p.ptr.dev)) {
 209                                 /*
 210                                  * raced with another move op? extent already
 211                                  * has a pointer to the device we just wrote
 212                                  * data to
 213                                  */
 214                                 continue;
 215                         }
 216
 217                         bch2_extent_ptr_decoded_append(insert, &p);
 218                         did_work = true;
 219                 }
 220
 221                 if (!did_work)
 222                         goto nomatch;
 223
 224                 bch2_bkey_narrow_crcs(insert,
 225                                 (struct bch_extent_crc_unpacked) { 0 });
 226                 bch2_extent_normalize(c, bkey_i_to_s(insert));
 227                 bch2_bkey_mark_replicas_cached(c, bkey_i_to_s(insert),
 228                                                op->opts.background_target,
 229                                                op->opts.data_replicas);
 230
 231                 ret = bch2_sum_sector_overwrites(&trans, &iter, insert,
 232                                                  &should_check_enospc,
 233                                                  &i_sectors_delta,
 234                                                  &disk_sectors_delta);
 235                 if (ret)
 236                         goto err;
 237
 238                 if (disk_sectors_delta > (s64) op->res.sectors) {
 239                         ret = bch2_disk_reservation_add(c, &op->res,
 240                                                 disk_sectors_delta - op->res.sectors,
 241                                                 !should_check_enospc
 242                                                 ? BCH_DISK_RESERVATION_NOFAIL : 0);
 243                         if (ret)
 244                                 goto out;
 245                 }
 246
 247                 next_pos = insert->k.p;
 248
 249                 ret   = insert_snapshot_whiteouts(&trans, m->btree_id,
 250                                                   k.k->p, insert->k.p) ?:
 251                         bch2_trans_update(&trans, &iter, insert,
 252                                 BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 253                         bch2_trans_commit(&trans, &op->res,
 254                                 op_journal_seq(op),
 255                                 BTREE_INSERT_NOFAIL|
 256                                 m->data_opts.btree_insert_flags);
 257                 if (!ret) {
 258                         bch2_btree_iter_set_pos(&iter, next_pos);
 259                         atomic_long_inc(&c->extent_migrate_done);
 260                         if (ec_ob)
 261                                 bch2_ob_add_backpointer(c, ec_ob, &insert->k);
 262                 }
 263 err:
 264                 if (ret == -EINTR)
 265                         ret = 0;
 266                 if (ret)
 267                         break;
 268 next:
 269                 while (bkey_cmp(iter.pos, bch2_keylist_front(keys)->k.p) >= 0) {
 270                         bch2_keylist_pop_front(keys);
 271                         if (bch2_keylist_empty(keys))
 272                                 goto out;
 273                 }
 274                 continue;
 275 nomatch:
 276                 if (m->ctxt) {
 277                         BUG_ON(k.k->p.offset <= iter.pos.offset);
 278                         atomic64_inc(&m->ctxt->stats->keys_raced);
 279                         atomic64_add(k.k->p.offset - iter.pos.offset,
 280                                      &m->ctxt->stats->sectors_raced);
 281                 }
 282                 atomic_long_inc(&c->extent_migrate_raced);
 283                 trace_move_race(&new->k);
 284                 bch2_btree_iter_advance(&iter);
 285                 goto next;
 286         }
 287 out:
 288         bch2_trans_iter_exit(&trans, &iter);
 289         bch2_trans_exit(&trans);
 290         bch2_bkey_buf_exit(&_insert, c);
 291         bch2_bkey_buf_exit(&_new, c);
 292         BUG_ON(ret == -EINTR);
 293         return ret;
 294 }
 295
 296 void bch2_migrate_read_done(struct migrate_write *m, struct bch_read_bio *rbio)
 297 {
 298         /* write bio must own pages: */
 299         BUG_ON(!m->op.wbio.bio.bi_vcnt);
 300
 301         m->ptr          = rbio->pick.ptr;
 302         m->offset       = rbio->data_pos.offset - rbio->pick.crc.offset;
 303         m->op.devs_have = rbio->devs_have;
 304         m->op.pos       = rbio->data_pos;
 305         m->op.version   = rbio->version;
 306         m->op.crc       = rbio->pick.crc;
 307         m->op.wbio.bio.bi_iter.bi_size = m->op.crc.compressed_size << 9;
 308
 309         if (m->data_cmd == DATA_REWRITE)
 310                 bch2_dev_list_drop_dev(&m->op.devs_have, m->data_opts.rewrite_dev);
 311 }
 312
 313 int bch2_migrate_write_init(struct bch_fs *c, struct migrate_write *m,
 314                             struct write_point_specifier wp,
 315                             struct bch_io_opts io_opts,
 316                             enum data_cmd data_cmd,
 317                             struct data_opts data_opts,
 318                             enum btree_id btree_id,
 319                             struct bkey_s_c k)
 320 {
 321         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 322         const union bch_extent_entry *entry;
 323         struct bch_extent_crc_unpacked crc;
 324         struct extent_ptr_decoded p;
 325         int ret;
 326
 327         m->btree_id     = btree_id;
 328         m->data_cmd     = data_cmd;
 329         m->data_opts    = data_opts;
 330         m->nr_ptrs_reserved = 0;
 331
 332         bch2_write_op_init(&m->op, c, io_opts);
 333
 334         if (!bch2_bkey_is_incompressible(k))
 335                 m->op.compression_type =
 336                         bch2_compression_opt_to_type[io_opts.background_compression ?:
 337                                                      io_opts.compression];
 338         else
 339                 m->op.incompressible = true;
 340
 341         m->op.target    = data_opts.target,
 342         m->op.write_point = wp;
 343
 344         /*
 345          * op->csum_type is normally initialized from the fs/file's current
 346          * options - but if an extent is encrypted, we require that it stays
 347          * encrypted:
 348          */
 349         bkey_for_each_crc(k.k, ptrs, crc, entry)
 350                 if (bch2_csum_type_is_encryption(crc.csum_type)) {
 351                         m->op.nonce     = crc.nonce + crc.offset;
 352                         m->op.csum_type = crc.csum_type;
 353                         break;
 354                 }
 355
 356         if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE) {
 357                 m->op.alloc_reserve = RESERVE_movinggc;
 358         } else {
 359                 /* XXX: this should probably be passed in */
 360                 m->op.flags |= BCH_WRITE_ONLY_SPECIFIED_DEVS;
 361         }
 362
 363         m->op.flags |= BCH_WRITE_PAGES_STABLE|
 364                 BCH_WRITE_PAGES_OWNED|
 365                 BCH_WRITE_DATA_ENCODED|
 366                 BCH_WRITE_FROM_INTERNAL;
 367
 368         m->op.nr_replicas       = data_opts.nr_replicas;
 369         m->op.nr_replicas_required = data_opts.nr_replicas;
 370         m->op.index_update_fn   = bch2_migrate_index_update;
 371
 372         switch (data_cmd) {
 373         case DATA_ADD_REPLICAS: {
 374                 /*
 375                  * DATA_ADD_REPLICAS is used for moving data to a different
 376                  * device in the background, and due to compression the new copy
 377                  * might take up more space than the old copy:
 378                  */
 379 #if 0
 380                 int nr = (int) io_opts.data_replicas -
 381                         bch2_bkey_nr_ptrs_allocated(k);
 382 #endif
 383                 int nr = (int) io_opts.data_replicas;
 384
 385                 if (nr > 0) {
 386                         m->op.nr_replicas = m->nr_ptrs_reserved = nr;
 387
 388                         ret = bch2_disk_reservation_get(c, &m->op.res,
 389                                         k.k->size, m->op.nr_replicas, 0);
 390                         if (ret)
 391                                 return ret;
 392                 }
 393                 break;
 394         }
 395         case DATA_REWRITE: {
 396                 unsigned compressed_sectors = 0;
 397
 398                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 399                         if (p.ptr.dev == data_opts.rewrite_dev) {
 400                                 if (p.ptr.cached)
 401                                         m->op.flags |= BCH_WRITE_CACHED;
 402
 403                                 if (!p.ptr.cached &&
 404                                     crc_is_compressed(p.crc))
 405                                         compressed_sectors += p.crc.compressed_size;
 406                         }
 407
 408                 if (compressed_sectors) {
 409                         ret = bch2_disk_reservation_add(c, &m->op.res,
 410                                         k.k->size * m->op.nr_replicas,
 411                                         BCH_DISK_RESERVATION_NOFAIL);
 412                         if (ret)
 413                                 return ret;
 414                 }
 415                 break;
 416         }
 417         case DATA_PROMOTE:
 418                 m->op.flags     |= BCH_WRITE_ALLOC_NOWAIT;
 419                 m->op.flags     |= BCH_WRITE_CACHED;
 420                 break;
 421         default:
 422                 BUG();
 423         }
 424
 425         return 0;
 426 }
 427
 428 static void move_free(struct closure *cl)
 429 {
 430         struct moving_io *io = container_of(cl, struct moving_io, cl);
 431         struct moving_context *ctxt = io->write.ctxt;
 432         struct bvec_iter_all iter;
 433         struct bio_vec *bv;
 434
 435         bch2_disk_reservation_put(io->write.op.c, &io->write.op.res);
 436
 437         bio_for_each_segment_all(bv, &io->write.op.wbio.bio, iter)
 438                 if (bv->bv_page)
 439                         __free_page(bv->bv_page);
 440
 441         wake_up(&ctxt->wait);
 442
 443         kfree(io);
 444 }
 445
 446 static void move_write_done(struct closure *cl)
 447 {
 448         struct moving_io *io = container_of(cl, struct moving_io, cl);
 449
 450         atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
 451         closure_return_with_destructor(cl, move_free);
 452 }
 453
 454 static void move_write(struct closure *cl)
 455 {
 456         struct moving_io *io = container_of(cl, struct moving_io, cl);
 457
 458         if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
 459                 closure_return_with_destructor(cl, move_free);
 460                 return;
 461         }
 462
 463         bch2_migrate_read_done(&io->write, &io->rbio);
 464
 465         atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
 466         closure_call(&io->write.op.cl, bch2_write, NULL, cl);
 467         continue_at(cl, move_write_done, NULL);
 468 }
 469
 470 static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
 471 {
 472         struct moving_io *io =
 473                 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
 474
 475         return io && io->read_completed ? io : NULL;
 476 }
 477
 478 static void move_read_endio(struct bio *bio)
 479 {
 480         struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
 481         struct moving_context *ctxt = io->write.ctxt;
 482
 483         atomic_sub(io->read_sectors, &ctxt->read_sectors);
 484         io->read_completed = true;
 485
 486         wake_up(&ctxt->wait);
 487         closure_put(&ctxt->cl);
 488 }
 489
 490 static void do_pending_writes(struct moving_context *ctxt, struct btree_trans *trans)
 491 {
 492         struct moving_io *io;
 493
 494         if (trans)
 495                 bch2_trans_unlock(trans);
 496
 497         while ((io = next_pending_write(ctxt))) {
 498                 list_del(&io->list);
 499                 closure_call(&io->cl, move_write, NULL, &ctxt->cl);
 500         }
 501 }
 502
 503 #define move_ctxt_wait_event(_ctxt, _trans, _cond)              \
 504 do {                                                            \
 505         do_pending_writes(_ctxt, _trans);                       \
 506                                                                 \
 507         if (_cond)                                              \
 508                 break;                                          \
 509         __wait_event((_ctxt)->wait,                             \
 510                      next_pending_write(_ctxt) || (_cond));     \
 511 } while (1)
 512
 513 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt,
 514                                        struct btree_trans *trans)
 515 {
 516         unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
 517
 518         move_ctxt_wait_event(ctxt, trans,
 519                 !atomic_read(&ctxt->write_sectors) ||
 520                 atomic_read(&ctxt->write_sectors) != sectors_pending);
 521 }
 522
 523 static int bch2_move_extent(struct btree_trans *trans,
 524                             struct moving_context *ctxt,
 525                             struct write_point_specifier wp,
 526                             struct bch_io_opts io_opts,
 527                             enum btree_id btree_id,
 528                             struct bkey_s_c k,
 529                             enum data_cmd data_cmd,
 530                             struct data_opts data_opts)
 531 {
 532         struct bch_fs *c = trans->c;
 533         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 534         struct moving_io *io;
 535         const union bch_extent_entry *entry;
 536         struct extent_ptr_decoded p;
 537         unsigned sectors = k.k->size, pages;
 538         int ret = -ENOMEM;
 539
 540         /* write path might have to decompress data: */
 541         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 542                 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
 543
 544         pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
 545         io = kzalloc(sizeof(struct moving_io) +
 546                      sizeof(struct bio_vec) * pages, GFP_KERNEL);
 547         if (!io)
 548                 goto err;
 549
 550         io->write.ctxt          = ctxt;
 551         io->read_sectors        = k.k->size;
 552         io->write_sectors       = k.k->size;
 553
 554         bio_init(&io->write.op.wbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 555         bio_set_prio(&io->write.op.wbio.bio,
 556                      IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 557
 558         if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
 559                                  GFP_KERNEL))
 560                 goto err_free;
 561
 562         io->rbio.c              = c;
 563         io->rbio.opts           = io_opts;
 564         bio_init(&io->rbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 565         io->rbio.bio.bi_vcnt = pages;
 566         bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 567         io->rbio.bio.bi_iter.bi_size = sectors << 9;
 568
 569         bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
 570         io->rbio.bio.bi_iter.bi_sector  = bkey_start_offset(k.k);
 571         io->rbio.bio.bi_end_io          = move_read_endio;
 572
 573         ret = bch2_migrate_write_init(c, &io->write, wp, io_opts,
 574                                       data_cmd, data_opts, btree_id, k);
 575         if (ret)
 576                 goto err_free_pages;
 577
 578         atomic64_inc(&ctxt->stats->keys_moved);
 579         atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
 580         this_cpu_add(c->counters[BCH_COUNTER_io_move], k.k->size);
 581
 582         trace_move_extent(k.k);
 583
 584         atomic_add(io->read_sectors, &ctxt->read_sectors);
 585         list_add_tail(&io->list, &ctxt->reads);
 586
 587         /*
 588          * dropped by move_read_endio() - guards against use after free of
 589          * ctxt when doing wakeup
 590          */
 591         closure_get(&ctxt->cl);
 592         bch2_read_extent(trans, &io->rbio,
 593                          bkey_start_pos(k.k),
 594                          btree_id, k, 0,
 595                          BCH_READ_NODECODE|
 596                          BCH_READ_LAST_FRAGMENT);
 597         return 0;
 598 err_free_pages:
 599         bio_free_pages(&io->write.op.wbio.bio);
 600 err_free:
 601         kfree(io);
 602 err:
 603         trace_move_alloc_mem_fail(k.k);
 604         return ret;
 605 }
 606
 607 static int lookup_inode(struct btree_trans *trans, struct bpos pos,
 608                         struct bch_inode_unpacked *inode)
 609 {
 610         struct btree_iter iter;
 611         struct bkey_s_c k;
 612         int ret;
 613
 614         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
 615                              BTREE_ITER_ALL_SNAPSHOTS);
 616         k = bch2_btree_iter_peek(&iter);
 617         ret = bkey_err(k);
 618         if (ret)
 619                 goto err;
 620
 621         if (!k.k || bkey_cmp(k.k->p, pos)) {
 622                 ret = -ENOENT;
 623                 goto err;
 624         }
 625
 626         ret = bkey_is_inode(k.k) ? 0 : -EIO;
 627         if (ret)
 628                 goto err;
 629
 630         ret = bch2_inode_unpack(k, inode);
 631         if (ret)
 632                 goto err;
 633 err:
 634         bch2_trans_iter_exit(trans, &iter);
 635         return ret;
 636 }
 637
 638 static int move_ratelimit(struct btree_trans *trans,
 639                           struct moving_context *ctxt,
 640                           struct bch_ratelimit *rate)
 641 {
 642         u64 delay;
 643
 644         do {
 645                 delay = rate ? bch2_ratelimit_delay(rate) : 0;
 646
 647                 if (delay) {
 648                         bch2_trans_unlock(trans);
 649                         set_current_state(TASK_INTERRUPTIBLE);
 650                 }
 651
 652                 if ((current->flags & PF_KTHREAD) && kthread_should_stop()) {
 653                         __set_current_state(TASK_RUNNING);
 654                         return 1;
 655                 }
 656
 657                 if (delay)
 658                         schedule_timeout(delay);
 659
 660                 if (unlikely(freezing(current))) {
 661                         move_ctxt_wait_event(ctxt, trans, list_empty(&ctxt->reads));
 662                         try_to_freeze();
 663                 }
 664         } while (delay);
 665
 666         move_ctxt_wait_event(ctxt, trans,
 667                 atomic_read(&ctxt->write_sectors) <
 668                 SECTORS_IN_FLIGHT_PER_DEVICE);
 669
 670         move_ctxt_wait_event(ctxt, trans,
 671                 atomic_read(&ctxt->read_sectors) <
 672                 SECTORS_IN_FLIGHT_PER_DEVICE);
 673
 674         return 0;
 675 }
 676
 677 static int move_get_io_opts(struct btree_trans *trans,
 678                             struct bch_io_opts *io_opts,
 679                             struct bkey_s_c k, u64 *cur_inum)
 680 {
 681         struct bch_inode_unpacked inode;
 682         int ret;
 683
 684         if (*cur_inum == k.k->p.inode)
 685                 return 0;
 686
 687         *io_opts = bch2_opts_to_inode_opts(trans->c->opts);
 688
 689         ret = lookup_inode(trans,
 690                            SPOS(0, k.k->p.inode, k.k->p.snapshot),
 691                            &inode);
 692         if (ret == -EINTR)
 693                 return ret;
 694
 695         if (!ret)
 696                 bch2_io_opts_apply(io_opts, bch2_inode_opts_get(&inode));
 697
 698         *cur_inum = k.k->p.inode;
 699         return 0;
 700 }
 701
 702 static int __bch2_move_data(struct bch_fs *c,
 703                 struct moving_context *ctxt,
 704                 struct bch_ratelimit *rate,
 705                 struct write_point_specifier wp,
 706                 struct bpos start,
 707                 struct bpos end,
 708                 move_pred_fn pred, void *arg,
 709                 struct bch_move_stats *stats,
 710                 enum btree_id btree_id)
 711 {
 712         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 713         struct bkey_buf sk;
 714         struct btree_trans trans;
 715         struct btree_iter iter;
 716         struct bkey_s_c k;
 717         struct data_opts data_opts;
 718         enum data_cmd data_cmd;
 719         u64 cur_inum = U64_MAX;
 720         int ret = 0, ret2;
 721
 722         bch2_bkey_buf_init(&sk);
 723         bch2_trans_init(&trans, c, 0, 0);
 724
 725         stats->data_type = BCH_DATA_user;
 726         stats->btree_id = btree_id;
 727         stats->pos      = start;
 728
 729         bch2_trans_iter_init(&trans, &iter, btree_id, start,
 730                              BTREE_ITER_PREFETCH|
 731                              BTREE_ITER_ALL_SNAPSHOTS);
 732
 733         if (rate)
 734                 bch2_ratelimit_reset(rate);
 735
 736         while (!move_ratelimit(&trans, ctxt, rate)) {
 737                 bch2_trans_begin(&trans);
 738
 739                 k = bch2_btree_iter_peek(&iter);
 740                 if (!k.k)
 741                         break;
 742
 743                 ret = bkey_err(k);
 744                 if (ret == -EINTR)
 745                         continue;
 746                 if (ret)
 747                         break;
 748
 749                 if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
 750                         break;
 751
 752                 stats->pos = iter.pos;
 753
 754                 if (!bkey_extent_is_direct_data(k.k))
 755                         goto next_nondata;
 756
 757                 ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 758                 if (ret)
 759                         continue;
 760
 761                 switch ((data_cmd = pred(c, arg, k, &io_opts, &data_opts))) {
 762                 case DATA_SKIP:
 763                         goto next;
 764                 case DATA_SCRUB:
 765                         BUG();
 766                 case DATA_ADD_REPLICAS:
 767                 case DATA_REWRITE:
 768                 case DATA_PROMOTE:
 769                         break;
 770                 default:
 771                         BUG();
 772                 }
 773
 774                 /*
 775                  * The iterator gets unlocked by __bch2_read_extent - need to
 776                  * save a copy of @k elsewhere:
 777                   */
 778                 bch2_bkey_buf_reassemble(&sk, c, k);
 779                 k = bkey_i_to_s_c(sk.k);
 780
 781                 ret2 = bch2_move_extent(&trans, ctxt, wp, io_opts, btree_id, k,
 782                                         data_cmd, data_opts);
 783                 if (ret2) {
 784                         if (ret2 == -EINTR)
 785                                 continue;
 786
 787                         if (ret2 == -ENOMEM) {
 788                                 /* memory allocation failure, wait for some IO to finish */
 789                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 790                                 continue;
 791                         }
 792
 793                         /* XXX signal failure */
 794                         goto next;
 795                 }
 796
 797                 if (rate)
 798                         bch2_ratelimit_increment(rate, k.k->size);
 799 next:
 800                 atomic64_add(k.k->size, &stats->sectors_seen);
 801 next_nondata:
 802                 bch2_btree_iter_advance(&iter);
 803         }
 804
 805         bch2_trans_iter_exit(&trans, &iter);
 806         bch2_trans_exit(&trans);
 807         bch2_bkey_buf_exit(&sk, c);
 808
 809         return ret;
 810 }
 811
 812 inline void bch_move_stats_init(struct bch_move_stats *stats, char *name)
 813 {
 814         memset(stats, 0, sizeof(*stats));
 815
 816         scnprintf(stats->name, sizeof(stats->name),
 817                         "%s", name);
 818 }
 819
 820 static inline void progress_list_add(struct bch_fs *c,
 821                                      struct bch_move_stats *stats)
 822 {
 823         mutex_lock(&c->data_progress_lock);
 824         list_add(&stats->list, &c->data_progress_list);
 825         mutex_unlock(&c->data_progress_lock);
 826 }
 827
 828 static inline void progress_list_del(struct bch_fs *c,
 829                                      struct bch_move_stats *stats)
 830 {
 831         mutex_lock(&c->data_progress_lock);
 832         list_del(&stats->list);
 833         mutex_unlock(&c->data_progress_lock);
 834 }
 835
 836 int bch2_move_data(struct bch_fs *c,
 837                    enum btree_id start_btree_id, struct bpos start_pos,
 838                    enum btree_id end_btree_id,   struct bpos end_pos,
 839                    struct bch_ratelimit *rate,
 840                    struct write_point_specifier wp,
 841                    move_pred_fn pred, void *arg,
 842                    struct bch_move_stats *stats)
 843 {
 844         struct moving_context ctxt = { .stats = stats };
 845         enum btree_id id;
 846         int ret;
 847
 848         progress_list_add(c, stats);
 849         closure_init_stack(&ctxt.cl);
 850         INIT_LIST_HEAD(&ctxt.reads);
 851         init_waitqueue_head(&ctxt.wait);
 852
 853         stats->data_type = BCH_DATA_user;
 854
 855         for (id = start_btree_id;
 856              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 857              id++) {
 858                 stats->btree_id = id;
 859
 860                 if (id != BTREE_ID_extents &&
 861                     id != BTREE_ID_reflink)
 862                         continue;
 863
 864                 ret = __bch2_move_data(c, &ctxt, rate, wp,
 865                                        id == start_btree_id ? start_pos : POS_MIN,
 866                                        id == end_btree_id   ? end_pos   : POS_MAX,
 867                                        pred, arg, stats, id);
 868                 if (ret)
 869                         break;
 870         }
 871
 872         move_ctxt_wait_event(&ctxt, NULL, list_empty(&ctxt.reads));
 873         closure_sync(&ctxt.cl);
 874
 875         EBUG_ON(atomic_read(&ctxt.write_sectors));
 876
 877         trace_move_data(c,
 878                         atomic64_read(&stats->sectors_moved),
 879                         atomic64_read(&stats->keys_moved));
 880
 881         progress_list_del(c, stats);
 882         return ret;
 883 }
 884
 885 static int verify_bucket_evacuated(struct btree_trans *trans, struct bpos bucket, int gen)
 886 {
 887         struct bch_fs *c = trans->c;
 888         struct btree_iter iter;
 889         struct bkey_s_c k;
 890         int ret;
 891
 892         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
 893                              bucket, BTREE_ITER_CACHED);
 894         k = bch2_btree_iter_peek_slot(&iter);
 895         ret = bkey_err(k);
 896
 897         if (!ret && k.k->type == KEY_TYPE_alloc_v4) {
 898                 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
 899
 900                 if (a.v->gen == gen &&
 901                     a.v->dirty_sectors) {
 902                         struct printbuf buf = PRINTBUF;
 903
 904                         prt_str(&buf, "failed to evacuate bucket ");
 905                         bch2_bkey_val_to_text(&buf, c, k);
 906
 907                         bch_err_ratelimited(c, "%s", buf.buf);
 908                         printbuf_exit(&buf);
 909                 }
 910         }
 911
 912         bch2_trans_iter_exit(trans, &iter);
 913         return ret;
 914 }
 915
 916 int bch2_evacuate_bucket(struct bch_fs *c,
 917                          struct bpos bucket, int gen,
 918                          struct bch_ratelimit *rate,
 919                          struct write_point_specifier wp,
 920                          enum data_cmd data_cmd,
 921                          struct data_opts *data_opts,
 922                          struct bch_move_stats *stats)
 923 {
 924         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 925         struct moving_context ctxt = { .stats = stats };
 926         struct btree_trans trans;
 927         struct btree_iter iter;
 928         struct bkey_buf sk;
 929         struct bch_backpointer bp;
 930         u64 bp_offset = 0, cur_inum = U64_MAX;
 931         int ret = 0;
 932
 933         bch2_bkey_buf_init(&sk);
 934         bch2_trans_init(&trans, c, 0, 0);
 935         progress_list_add(c, stats);
 936         closure_init_stack(&ctxt.cl);
 937         INIT_LIST_HEAD(&ctxt.reads);
 938         init_waitqueue_head(&ctxt.wait);
 939
 940         stats->data_type = BCH_DATA_user;
 941
 942         while (!(ret = move_ratelimit(&trans, &ctxt, rate))) {
 943                 bch2_trans_begin(&trans);
 944
 945                 ret = bch2_get_next_backpointer(&trans, bucket, gen,
 946                                                 &bp_offset, &bp);
 947                 if (ret == -EINTR)
 948                         continue;
 949                 if (ret)
 950                         goto err;
 951                 if (bp_offset == U64_MAX)
 952                         break;
 953
 954                 if (!bp.level) {
 955                         struct bkey_s_c k;
 956
 957                         k = bch2_backpointer_get_key(&trans, &iter,
 958                                                 bucket, bp_offset, bp);
 959                         ret = bkey_err(k);
 960                         if (ret == -EINTR)
 961                                 continue;
 962                         if (ret)
 963                                 goto err;
 964                         if (!k.k)
 965                                 continue;
 966
 967                         bch2_bkey_buf_reassemble(&sk, c, k);
 968                         k = bkey_i_to_s_c(sk.k);
 969                         bch2_trans_iter_exit(&trans, &iter);
 970
 971                         ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 972                         if (ret)
 973                                 continue;
 974
 975                         data_opts->target       = io_opts.background_target;
 976                         data_opts->rewrite_dev  = bucket.inode;
 977
 978                         ret = bch2_move_extent(&trans, &ctxt, wp, io_opts, bp.btree_id, k,
 979                                                 data_cmd, *data_opts);
 980                         if (ret == -EINTR)
 981                                 continue;
 982                         if (ret == -ENOMEM) {
 983                                 /* memory allocation failure, wait for some IO to finish */
 984                                 bch2_move_ctxt_wait_for_io(&ctxt, &trans);
 985                                 continue;
 986                         }
 987                         if (ret)
 988                                 goto err;
 989
 990                         if (rate)
 991                                 bch2_ratelimit_increment(rate, k.k->size);
 992                         atomic64_add(k.k->size, &stats->sectors_seen);
 993                 } else {
 994                         struct btree *b;
 995
 996                         b = bch2_backpointer_get_node(&trans, &iter,
 997                                                 bucket, bp_offset, bp);
 998                         ret = PTR_ERR_OR_ZERO(b);
 999                         if (ret == -EINTR)
1000                                 continue;
1001                         if (ret)
1002                                 goto err;
1003                         if (!b)
1004                                 continue;
1005
1006                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0);
1007                         bch2_trans_iter_exit(&trans, &iter);
1008
1009                         if (ret == -EINTR)
1010                                 continue;
1011                         if (ret)
1012                                 goto err;
1013
1014                         if (rate)
1015                                 bch2_ratelimit_increment(rate, c->opts.btree_node_size >> 9);
1016                         atomic64_add(c->opts.btree_node_size >> 9, &stats->sectors_seen);
1017                         atomic64_add(c->opts.btree_node_size >> 9, &stats->sectors_moved);
1018                 }
1019
1020                 bp_offset++;
1021         }
1022
1023         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && gen >= 0) {
1024                 bch2_trans_unlock(&trans);
1025                 move_ctxt_wait_event(&ctxt, NULL, list_empty(&ctxt.reads));
1026                 closure_sync(&ctxt.cl);
1027                 lockrestart_do(&trans, verify_bucket_evacuated(&trans, bucket, gen));
1028         }
1029 err:
1030         bch2_trans_exit(&trans);
1031         bch2_bkey_buf_exit(&sk, c);
1032
1033         move_ctxt_wait_event(&ctxt, NULL, list_empty(&ctxt.reads));
1034         closure_sync(&ctxt.cl);
1035         progress_list_del(c, stats);
1036
1037         EBUG_ON(atomic_read(&ctxt.write_sectors));
1038
1039         trace_move_data(c,
1040                         atomic64_read(&stats->sectors_moved),
1041                         atomic64_read(&stats->keys_moved));
1042
1043         return ret;
1044 }
1045
1046 typedef enum data_cmd (*move_btree_pred)(struct bch_fs *, void *,
1047                                          struct btree *, struct bch_io_opts *,
1048                                          struct data_opts *);
1049
1050 static int bch2_move_btree(struct bch_fs *c,
1051                            enum btree_id start_btree_id, struct bpos start_pos,
1052                            enum btree_id end_btree_id,   struct bpos end_pos,
1053                            move_btree_pred pred, void *arg,
1054                            struct bch_move_stats *stats)
1055 {
1056         bool kthread = (current->flags & PF_KTHREAD) != 0;
1057         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
1058         struct btree_trans trans;
1059         struct btree_iter iter;
1060         struct btree *b;
1061         enum btree_id id;
1062         struct data_opts data_opts;
1063         enum data_cmd cmd;
1064         int ret = 0;
1065
1066         bch2_trans_init(&trans, c, 0, 0);
1067         progress_list_add(c, stats);
1068
1069         stats->data_type = BCH_DATA_btree;
1070
1071         for (id = start_btree_id;
1072              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
1073              id++) {
1074                 stats->btree_id = id;
1075
1076                 bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
1077                                           BTREE_ITER_PREFETCH);
1078 retry:
1079                 ret = 0;
1080                 while (bch2_trans_begin(&trans),
1081                        (b = bch2_btree_iter_peek_node(&iter)) &&
1082                        !(ret = PTR_ERR_OR_ZERO(b))) {
1083                         if (kthread && kthread_should_stop())
1084                                 break;
1085
1086                         if ((cmp_int(id, end_btree_id) ?:
1087                              bpos_cmp(b->key.k.p, end_pos)) > 0)
1088                                 break;
1089
1090                         stats->pos = iter.pos;
1091
1092                         switch ((cmd = pred(c, arg, b, &io_opts, &data_opts))) {
1093                         case DATA_SKIP:
1094                                 goto next;
1095                         case DATA_SCRUB:
1096                                 BUG();
1097                         case DATA_ADD_REPLICAS:
1098                         case DATA_REWRITE:
1099                                 break;
1100                         default:
1101                                 BUG();
1102                         }
1103
1104                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
1105                         if (ret == -EINTR)
1106                                 continue;
1107                         if (ret)
1108                                 break;
1109 next:
1110                         bch2_btree_iter_next_node(&iter);
1111                 }
1112                 if (ret == -EINTR)
1113                         goto retry;
1114
1115                 bch2_trans_iter_exit(&trans, &iter);
1116
1117                 if (kthread && kthread_should_stop())
1118                         break;
1119         }
1120
1121         bch2_trans_exit(&trans);
1122
1123         if (ret)
1124                 bch_err(c, "error %i in bch2_move_btree", ret);
1125
1126         bch2_btree_interior_updates_flush(c);
1127
1128         progress_list_del(c, stats);
1129         return ret;
1130 }
1131
1132 #if 0
1133 static enum data_cmd scrub_pred(struct bch_fs *c, void *arg,
1134                                 struct bkey_s_c k,
1135                                 struct bch_io_opts *io_opts,
1136                                 struct data_opts *data_opts)
1137 {
1138         return DATA_SCRUB;
1139 }
1140 #endif
1141
1142 static enum data_cmd rereplicate_pred(struct bch_fs *c, void *arg,
1143                                       struct bkey_s_c k,
1144                                       struct bch_io_opts *io_opts,
1145                                       struct data_opts *data_opts)
1146 {
1147         unsigned nr_good = bch2_bkey_durability(c, k);
1148         unsigned replicas = bkey_is_btree_ptr(k.k)
1149                 ? c->opts.metadata_replicas
1150                 : io_opts->data_replicas;
1151
1152         if (!nr_good || nr_good >= replicas)
1153                 return DATA_SKIP;
1154
1155         data_opts->target               = 0;
1156         data_opts->nr_replicas          = 1;
1157         data_opts->btree_insert_flags   = 0;
1158         return DATA_ADD_REPLICAS;
1159 }
1160
1161 static enum data_cmd migrate_pred(struct bch_fs *c, void *arg,
1162                                   struct bkey_s_c k,
1163                                   struct bch_io_opts *io_opts,
1164                                   struct data_opts *data_opts)
1165 {
1166         struct bch_ioctl_data *op = arg;
1167
1168         if (!bch2_bkey_has_device(k, op->migrate.dev))
1169                 return DATA_SKIP;
1170
1171         data_opts->target               = 0;
1172         data_opts->nr_replicas          = 1;
1173         data_opts->btree_insert_flags   = 0;
1174         data_opts->rewrite_dev          = op->migrate.dev;
1175         return DATA_REWRITE;
1176 }
1177
1178 static enum data_cmd rereplicate_btree_pred(struct bch_fs *c, void *arg,
1179                                             struct btree *b,
1180                                             struct bch_io_opts *io_opts,
1181                                             struct data_opts *data_opts)
1182 {
1183         return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
1184 }
1185
1186 static enum data_cmd migrate_btree_pred(struct bch_fs *c, void *arg,
1187                                         struct btree *b,
1188                                         struct bch_io_opts *io_opts,
1189                                         struct data_opts *data_opts)
1190 {
1191         return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
1192 }
1193
1194 static bool bformat_needs_redo(struct bkey_format *f)
1195 {
1196         unsigned i;
1197
1198         for (i = 0; i < f->nr_fields; i++) {
1199                 unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
1200                 u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
1201                 u64 field_offset = le64_to_cpu(f->field_offset[i]);
1202
1203                 if (f->bits_per_field[i] > unpacked_bits)
1204                         return true;
1205
1206                 if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
1207                         return true;
1208
1209                 if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
1210                      unpacked_mask) <
1211                     field_offset)
1212                         return true;
1213         }
1214
1215         return false;
1216 }
1217
1218 static enum data_cmd rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
1219                                             struct btree *b,
1220                                             struct bch_io_opts *io_opts,
1221                                             struct data_opts *data_opts)
1222 {
1223         if (b->version_ondisk != c->sb.version ||
1224             btree_node_need_rewrite(b) ||
1225             bformat_needs_redo(&b->format)) {
1226                 data_opts->target               = 0;
1227                 data_opts->nr_replicas          = 1;
1228                 data_opts->btree_insert_flags   = 0;
1229                 return DATA_REWRITE;
1230         }
1231
1232         return DATA_SKIP;
1233 }
1234
1235 int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
1236 {
1237         int ret;
1238
1239         ret = bch2_move_btree(c,
1240                               0,                POS_MIN,
1241                               BTREE_ID_NR,      SPOS_MAX,
1242                               rewrite_old_nodes_pred, c, stats);
1243         if (!ret) {
1244                 mutex_lock(&c->sb_lock);
1245                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
1246                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
1247                 c->disk_sb.sb->version_min = c->disk_sb.sb->version;
1248                 bch2_write_super(c);
1249                 mutex_unlock(&c->sb_lock);
1250         }
1251
1252         return ret;
1253 }
1254
1255 int bch2_data_job(struct bch_fs *c,
1256                   struct bch_move_stats *stats,
1257                   struct bch_ioctl_data op)
1258 {
1259         int ret = 0;
1260
1261         switch (op.op) {
1262         case BCH_DATA_OP_REREPLICATE:
1263                 bch_move_stats_init(stats, "rereplicate");
1264                 stats->data_type = BCH_DATA_journal;
1265                 ret = bch2_journal_flush_device_pins(&c->journal, -1);
1266
1267                 ret = bch2_move_btree(c,
1268                                       op.start_btree,   op.start_pos,
1269                                       op.end_btree,     op.end_pos,
1270                                       rereplicate_btree_pred, c, stats) ?: ret;
1271                 ret = bch2_replicas_gc2(c) ?: ret;
1272
1273                 ret = bch2_move_data(c,
1274                                      op.start_btree,    op.start_pos,
1275                                      op.end_btree,      op.end_pos,
1276                                      NULL, writepoint_hashed((unsigned long) current),
1277                                      rereplicate_pred, c, stats) ?: ret;
1278                 ret = bch2_replicas_gc2(c) ?: ret;
1279                 break;
1280         case BCH_DATA_OP_MIGRATE:
1281                 if (op.migrate.dev >= c->sb.nr_devices)
1282                         return -EINVAL;
1283
1284                 bch_move_stats_init(stats, "migrate");
1285                 stats->data_type = BCH_DATA_journal;
1286                 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
1287
1288                 ret = bch2_move_btree(c,
1289                                       op.start_btree,   op.start_pos,
1290                                       op.end_btree,     op.end_pos,
1291                                       migrate_btree_pred, &op, stats) ?: ret;
1292                 ret = bch2_replicas_gc2(c) ?: ret;
1293
1294                 ret = bch2_move_data(c,
1295                                      op.start_btree,    op.start_pos,
1296                                      op.end_btree,      op.end_pos,
1297                                      NULL, writepoint_hashed((unsigned long) current),
1298                                      migrate_pred, &op, stats) ?: ret;
1299                 ret = bch2_replicas_gc2(c) ?: ret;
1300                 break;
1301         case BCH_DATA_OP_REWRITE_OLD_NODES:
1302                 bch_move_stats_init(stats, "rewrite_old_nodes");
1303                 ret = bch2_scan_old_btree_nodes(c, stats);
1304                 break;
1305         default:
1306                 ret = -EINVAL;
1307         }
1308
1309         return ret;
1310 }