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_foreground.h"
   5 #include "btree_gc.h"
   6 #include "btree_update.h"
   7 #include "btree_update_interior.h"
   8 #include "buckets.h"
   9 #include "disk_groups.h"
  10 #include "inode.h"
  11 #include "io.h"
  12 #include "journal_reclaim.h"
  13 #include "move.h"
  14 #include "replicas.h"
  15 #include "super-io.h"
  16 #include "keylist.h"
  17
  18 #include <linux/ioprio.h>
  19 #include <linux/kthread.h>
  20
  21 #include <trace/events/bcachefs.h>
  22
  23 #define SECTORS_IN_FLIGHT_PER_DEVICE    2048
  24
  25 struct moving_io {
  26         struct list_head        list;
  27         struct closure          cl;
  28         bool                    read_completed;
  29
  30         unsigned                read_sectors;
  31         unsigned                write_sectors;
  32
  33         struct bch_read_bio     rbio;
  34
  35         struct migrate_write    write;
  36         /* Must be last since it is variable size */
  37         struct bio_vec          bi_inline_vecs[0];
  38 };
  39
  40 struct moving_context {
  41         /* Closure for waiting on all reads and writes to complete */
  42         struct closure          cl;
  43
  44         struct bch_move_stats   *stats;
  45
  46         struct list_head        reads;
  47
  48         /* in flight sectors: */
  49         atomic_t                read_sectors;
  50         atomic_t                write_sectors;
  51
  52         wait_queue_head_t       wait;
  53 };
  54
  55 static int bch2_migrate_index_update(struct bch_write_op *op)
  56 {
  57         struct bch_fs *c = op->c;
  58         struct btree_trans trans;
  59         struct btree_iter *iter;
  60         struct migrate_write *m =
  61                 container_of(op, struct migrate_write, op);
  62         struct keylist *keys = &op->insert_keys;
  63         int ret = 0;
  64
  65         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
  66
  67         iter = bch2_trans_get_iter(&trans, m->btree_id,
  68                                    bkey_start_pos(&bch2_keylist_front(keys)->k),
  69                                    BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
  70
  71         while (1) {
  72                 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
  73                 struct bkey_i *insert;
  74                 struct bkey_i_extent *new =
  75                         bkey_i_to_extent(bch2_keylist_front(keys));
  76                 BKEY_PADDED(k) _new, _insert;
  77                 const union bch_extent_entry *entry;
  78                 struct extent_ptr_decoded p;
  79                 bool did_work = false;
  80                 int nr;
  81
  82                 ret = bkey_err(k);
  83                 if (ret)
  84                         break;
  85
  86                 if (bversion_cmp(k.k->version, new->k.version) ||
  87                     !bch2_bkey_matches_ptr(c, k, m->ptr, m->offset))
  88                         goto nomatch;
  89
  90                 if (m->data_cmd == DATA_REWRITE &&
  91                     !bch2_bkey_has_device(k, m->data_opts.rewrite_dev))
  92                         goto nomatch;
  93
  94                 bkey_reassemble(&_insert.k, k);
  95                 insert = &_insert.k;
  96
  97                 bkey_copy(&_new.k, bch2_keylist_front(keys));
  98                 new = bkey_i_to_extent(&_new.k);
  99
 100                 bch2_cut_front(iter->pos, insert);
 101                 bch2_cut_back(new->k.p, &insert->k);
 102                 bch2_cut_back(insert->k.p, &new->k);
 103
 104                 if (m->data_cmd == DATA_REWRITE)
 105                         bch2_bkey_drop_device(bkey_i_to_s(insert),
 106                                               m->data_opts.rewrite_dev);
 107
 108                 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) {
 109                         if (bch2_bkey_has_device(bkey_i_to_s_c(insert), p.ptr.dev)) {
 110                                 /*
 111                                  * raced with another move op? extent already
 112                                  * has a pointer to the device we just wrote
 113                                  * data to
 114                                  */
 115                                 continue;
 116                         }
 117
 118                         bch2_extent_ptr_decoded_append(insert, &p);
 119                         did_work = true;
 120                 }
 121
 122                 if (!did_work)
 123                         goto nomatch;
 124
 125                 bch2_bkey_narrow_crcs(insert,
 126                                 (struct bch_extent_crc_unpacked) { 0 });
 127                 bch2_extent_normalize(c, bkey_i_to_s(insert));
 128                 bch2_bkey_mark_replicas_cached(c, bkey_i_to_s(insert),
 129                                                op->opts.background_target,
 130                                                op->opts.data_replicas);
 131
 132                 /*
 133                  * If we're not fully overwriting @k, and it's compressed, we
 134                  * need a reservation for all the pointers in @insert
 135                  */
 136                 nr = bch2_bkey_nr_dirty_ptrs(bkey_i_to_s_c(insert)) -
 137                          m->nr_ptrs_reserved;
 138
 139                 if (insert->k.size < k.k->size &&
 140                     bch2_extent_is_compressed(k) &&
 141                     nr > 0) {
 142                         ret = bch2_disk_reservation_add(c, &op->res,
 143                                         keylist_sectors(keys) * nr, 0);
 144                         if (ret)
 145                                 goto out;
 146
 147                         m->nr_ptrs_reserved += nr;
 148                         goto next;
 149                 }
 150
 151                 bch2_trans_update(&trans, iter, insert);
 152
 153                 ret = bch2_trans_commit(&trans, &op->res,
 154                                 op_journal_seq(op),
 155                                 BTREE_INSERT_ATOMIC|
 156                                 BTREE_INSERT_NOFAIL|
 157                                 BTREE_INSERT_USE_RESERVE|
 158                                 m->data_opts.btree_insert_flags);
 159                 if (!ret)
 160                         atomic_long_inc(&c->extent_migrate_done);
 161                 if (ret == -EINTR)
 162                         ret = 0;
 163                 if (ret)
 164                         break;
 165 next:
 166                 while (bkey_cmp(iter->pos, bch2_keylist_front(keys)->k.p) >= 0) {
 167                         bch2_keylist_pop_front(keys);
 168                         if (bch2_keylist_empty(keys))
 169                                 goto out;
 170                 }
 171
 172                 bch2_cut_front(iter->pos, bch2_keylist_front(keys));
 173                 continue;
 174 nomatch:
 175                 if (m->ctxt)
 176                         atomic64_add(k.k->p.offset - iter->pos.offset,
 177                                      &m->ctxt->stats->sectors_raced);
 178                 atomic_long_inc(&c->extent_migrate_raced);
 179                 trace_move_race(&new->k);
 180                 bch2_btree_iter_next_slot(iter);
 181                 goto next;
 182         }
 183 out:
 184         bch2_trans_exit(&trans);
 185         BUG_ON(ret == -EINTR);
 186         return ret;
 187 }
 188
 189 void bch2_migrate_read_done(struct migrate_write *m, struct bch_read_bio *rbio)
 190 {
 191         /* write bio must own pages: */
 192         BUG_ON(!m->op.wbio.bio.bi_vcnt);
 193
 194         m->ptr          = rbio->pick.ptr;
 195         m->offset       = rbio->pos.offset - rbio->pick.crc.offset;
 196         m->op.devs_have = rbio->devs_have;
 197         m->op.pos       = rbio->pos;
 198         m->op.version   = rbio->version;
 199         m->op.crc       = rbio->pick.crc;
 200         m->op.wbio.bio.bi_iter.bi_size = m->op.crc.compressed_size << 9;
 201
 202         if (bch2_csum_type_is_encryption(m->op.crc.csum_type)) {
 203                 m->op.nonce     = m->op.crc.nonce + m->op.crc.offset;
 204                 m->op.csum_type = m->op.crc.csum_type;
 205         }
 206
 207         if (m->data_cmd == DATA_REWRITE)
 208                 bch2_dev_list_drop_dev(&m->op.devs_have, m->data_opts.rewrite_dev);
 209 }
 210
 211 int bch2_migrate_write_init(struct bch_fs *c, struct migrate_write *m,
 212                             struct write_point_specifier wp,
 213                             struct bch_io_opts io_opts,
 214                             enum data_cmd data_cmd,
 215                             struct data_opts data_opts,
 216                             enum btree_id btree_id,
 217                             struct bkey_s_c k)
 218 {
 219         int ret;
 220
 221         m->btree_id     = btree_id;
 222         m->data_cmd     = data_cmd;
 223         m->data_opts    = data_opts;
 224         m->nr_ptrs_reserved = 0;
 225
 226         bch2_write_op_init(&m->op, c, io_opts);
 227         m->op.compression_type =
 228                 bch2_compression_opt_to_type[io_opts.background_compression ?:
 229                                              io_opts.compression];
 230         m->op.target    = data_opts.target,
 231         m->op.write_point = wp;
 232
 233         if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE)
 234                 m->op.alloc_reserve = RESERVE_MOVINGGC;
 235
 236         m->op.flags |= BCH_WRITE_ONLY_SPECIFIED_DEVS|
 237                 BCH_WRITE_PAGES_STABLE|
 238                 BCH_WRITE_PAGES_OWNED|
 239                 BCH_WRITE_DATA_ENCODED;
 240
 241         m->op.nr_replicas       = 1;
 242         m->op.nr_replicas_required = 1;
 243         m->op.index_update_fn   = bch2_migrate_index_update;
 244
 245         switch (data_cmd) {
 246         case DATA_ADD_REPLICAS: {
 247                 /*
 248                  * DATA_ADD_REPLICAS is used for moving data to a different
 249                  * device in the background, and due to compression the new copy
 250                  * might take up more space than the old copy:
 251                  */
 252 #if 0
 253                 int nr = (int) io_opts.data_replicas -
 254                         bch2_bkey_nr_dirty_ptrs(k);
 255 #endif
 256                 int nr = (int) io_opts.data_replicas;
 257
 258                 if (nr > 0) {
 259                         m->op.nr_replicas = m->nr_ptrs_reserved = nr;
 260
 261                         ret = bch2_disk_reservation_get(c, &m->op.res,
 262                                         k.k->size, m->op.nr_replicas, 0);
 263                         if (ret)
 264                                 return ret;
 265                 }
 266                 break;
 267         }
 268         case DATA_REWRITE: {
 269                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 270                 const union bch_extent_entry *entry;
 271                 struct extent_ptr_decoded p;
 272                 unsigned compressed_sectors = 0;
 273
 274                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 275                         if (!p.ptr.cached &&
 276                             p.crc.compression_type != BCH_COMPRESSION_NONE &&
 277                             bch2_dev_in_target(c, p.ptr.dev, data_opts.target))
 278                                 compressed_sectors += p.crc.compressed_size;
 279
 280                 if (compressed_sectors) {
 281                         ret = bch2_disk_reservation_add(c, &m->op.res,
 282                                         compressed_sectors,
 283                                         BCH_DISK_RESERVATION_NOFAIL);
 284                         if (ret)
 285                                 return ret;
 286                 }
 287                 break;
 288         }
 289         case DATA_PROMOTE:
 290                 m->op.flags     |= BCH_WRITE_ALLOC_NOWAIT;
 291                 m->op.flags     |= BCH_WRITE_CACHED;
 292                 break;
 293         default:
 294                 BUG();
 295         }
 296
 297         return 0;
 298 }
 299
 300 static void move_free(struct closure *cl)
 301 {
 302         struct moving_io *io = container_of(cl, struct moving_io, cl);
 303         struct moving_context *ctxt = io->write.ctxt;
 304         struct bvec_iter_all iter;
 305         struct bio_vec *bv;
 306
 307         bch2_disk_reservation_put(io->write.op.c, &io->write.op.res);
 308
 309         bio_for_each_segment_all(bv, &io->write.op.wbio.bio, iter)
 310                 if (bv->bv_page)
 311                         __free_page(bv->bv_page);
 312
 313         wake_up(&ctxt->wait);
 314
 315         kfree(io);
 316 }
 317
 318 static void move_write_done(struct closure *cl)
 319 {
 320         struct moving_io *io = container_of(cl, struct moving_io, cl);
 321
 322         atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
 323         closure_return_with_destructor(cl, move_free);
 324 }
 325
 326 static void move_write(struct closure *cl)
 327 {
 328         struct moving_io *io = container_of(cl, struct moving_io, cl);
 329
 330         if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
 331                 closure_return_with_destructor(cl, move_free);
 332                 return;
 333         }
 334
 335         bch2_migrate_read_done(&io->write, &io->rbio);
 336
 337         atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
 338         closure_call(&io->write.op.cl, bch2_write, NULL, cl);
 339         continue_at(cl, move_write_done, NULL);
 340 }
 341
 342 static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
 343 {
 344         struct moving_io *io =
 345                 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
 346
 347         return io && io->read_completed ? io : NULL;
 348 }
 349
 350 static void move_read_endio(struct bio *bio)
 351 {
 352         struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
 353         struct moving_context *ctxt = io->write.ctxt;
 354
 355         atomic_sub(io->read_sectors, &ctxt->read_sectors);
 356         io->read_completed = true;
 357
 358         if (next_pending_write(ctxt))
 359                 wake_up(&ctxt->wait);
 360
 361         closure_put(&ctxt->cl);
 362 }
 363
 364 static void do_pending_writes(struct moving_context *ctxt)
 365 {
 366         struct moving_io *io;
 367
 368         while ((io = next_pending_write(ctxt))) {
 369                 list_del(&io->list);
 370                 closure_call(&io->cl, move_write, NULL, &ctxt->cl);
 371         }
 372 }
 373
 374 #define move_ctxt_wait_event(_ctxt, _cond)                      \
 375 do {                                                            \
 376         do_pending_writes(_ctxt);                               \
 377                                                                 \
 378         if (_cond)                                              \
 379                 break;                                          \
 380         __wait_event((_ctxt)->wait,                             \
 381                      next_pending_write(_ctxt) || (_cond));     \
 382 } while (1)
 383
 384 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt)
 385 {
 386         unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
 387
 388         move_ctxt_wait_event(ctxt,
 389                 !atomic_read(&ctxt->write_sectors) ||
 390                 atomic_read(&ctxt->write_sectors) != sectors_pending);
 391 }
 392
 393 static int bch2_move_extent(struct bch_fs *c,
 394                             struct moving_context *ctxt,
 395                             struct write_point_specifier wp,
 396                             struct bch_io_opts io_opts,
 397                             enum btree_id btree_id,
 398                             struct bkey_s_c k,
 399                             enum data_cmd data_cmd,
 400                             struct data_opts data_opts)
 401 {
 402         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 403         struct moving_io *io;
 404         const union bch_extent_entry *entry;
 405         struct extent_ptr_decoded p;
 406         unsigned sectors = k.k->size, pages;
 407         int ret = -ENOMEM;
 408
 409         move_ctxt_wait_event(ctxt,
 410                 atomic_read(&ctxt->write_sectors) <
 411                 SECTORS_IN_FLIGHT_PER_DEVICE);
 412
 413         move_ctxt_wait_event(ctxt,
 414                 atomic_read(&ctxt->read_sectors) <
 415                 SECTORS_IN_FLIGHT_PER_DEVICE);
 416
 417         /* write path might have to decompress data: */
 418         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 419                 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
 420
 421         pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
 422         io = kzalloc(sizeof(struct moving_io) +
 423                      sizeof(struct bio_vec) * pages, GFP_KERNEL);
 424         if (!io)
 425                 goto err;
 426
 427         io->write.ctxt          = ctxt;
 428         io->read_sectors        = k.k->size;
 429         io->write_sectors       = k.k->size;
 430
 431         bio_init(&io->write.op.wbio.bio, io->bi_inline_vecs, pages);
 432         bio_set_prio(&io->write.op.wbio.bio,
 433                      IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 434
 435         if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
 436                                  GFP_KERNEL))
 437                 goto err_free;
 438
 439         io->rbio.c              = c;
 440         io->rbio.opts           = io_opts;
 441         bio_init(&io->rbio.bio, io->bi_inline_vecs, pages);
 442         io->rbio.bio.bi_vcnt = pages;
 443         bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 444         io->rbio.bio.bi_iter.bi_size = sectors << 9;
 445
 446         bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
 447         io->rbio.bio.bi_iter.bi_sector  = bkey_start_offset(k.k);
 448         io->rbio.bio.bi_end_io          = move_read_endio;
 449
 450         ret = bch2_migrate_write_init(c, &io->write, wp, io_opts,
 451                                       data_cmd, data_opts, btree_id, k);
 452         if (ret)
 453                 goto err_free_pages;
 454
 455         atomic64_inc(&ctxt->stats->keys_moved);
 456         atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
 457
 458         trace_move_extent(k.k);
 459
 460         atomic_add(io->read_sectors, &ctxt->read_sectors);
 461         list_add_tail(&io->list, &ctxt->reads);
 462
 463         /*
 464          * dropped by move_read_endio() - guards against use after free of
 465          * ctxt when doing wakeup
 466          */
 467         closure_get(&ctxt->cl);
 468         bch2_read_extent(c, &io->rbio, k, 0,
 469                          BCH_READ_NODECODE|
 470                          BCH_READ_LAST_FRAGMENT);
 471         return 0;
 472 err_free_pages:
 473         bio_free_pages(&io->write.op.wbio.bio);
 474 err_free:
 475         kfree(io);
 476 err:
 477         trace_move_alloc_fail(k.k);
 478         return ret;
 479 }
 480
 481 static int __bch2_move_data(struct bch_fs *c,
 482                 struct moving_context *ctxt,
 483                 struct bch_ratelimit *rate,
 484                 struct write_point_specifier wp,
 485                 struct bpos start,
 486                 struct bpos end,
 487                 move_pred_fn pred, void *arg,
 488                 struct bch_move_stats *stats,
 489                 enum btree_id btree_id)
 490 {
 491         bool kthread = (current->flags & PF_KTHREAD) != 0;
 492         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 493         BKEY_PADDED(k) tmp;
 494         struct btree_trans trans;
 495         struct btree_iter *iter;
 496         struct bkey_s_c k;
 497         struct data_opts data_opts;
 498         enum data_cmd data_cmd;
 499         u64 delay, cur_inum = U64_MAX;
 500         int ret = 0, ret2;
 501
 502         bch2_trans_init(&trans, c, 0, 0);
 503
 504         stats->data_type = BCH_DATA_USER;
 505         stats->btree_id = btree_id;
 506         stats->pos      = POS_MIN;
 507
 508         iter = bch2_trans_get_iter(&trans, btree_id, start,
 509                                    BTREE_ITER_PREFETCH);
 510
 511         if (rate)
 512                 bch2_ratelimit_reset(rate);
 513
 514         while (1) {
 515                 do {
 516                         delay = rate ? bch2_ratelimit_delay(rate) : 0;
 517
 518                         if (delay) {
 519                                 bch2_trans_unlock(&trans);
 520                                 set_current_state(TASK_INTERRUPTIBLE);
 521                         }
 522
 523                         if (kthread && (ret = kthread_should_stop())) {
 524                                 __set_current_state(TASK_RUNNING);
 525                                 goto out;
 526                         }
 527
 528                         if (delay)
 529                                 schedule_timeout(delay);
 530
 531                         if (unlikely(freezing(current))) {
 532                                 bch2_trans_unlock(&trans);
 533                                 move_ctxt_wait_event(ctxt, list_empty(&ctxt->reads));
 534                                 try_to_freeze();
 535                         }
 536                 } while (delay);
 537 peek:
 538                 k = bch2_btree_iter_peek(iter);
 539
 540                 stats->pos = iter->pos;
 541
 542                 if (!k.k)
 543                         break;
 544                 ret = bkey_err(k);
 545                 if (ret)
 546                         break;
 547                 if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
 548                         break;
 549
 550                 if (!bkey_extent_is_direct_data(k.k))
 551                         goto next_nondata;
 552
 553                 if (btree_id == BTREE_ID_EXTENTS &&
 554                     cur_inum != k.k->p.inode) {
 555                         struct bch_inode_unpacked inode;
 556
 557                         /* don't hold btree locks while looking up inode: */
 558                         bch2_trans_unlock(&trans);
 559
 560                         io_opts = bch2_opts_to_inode_opts(c->opts);
 561                         if (!bch2_inode_find_by_inum(c, k.k->p.inode, &inode))
 562                                 bch2_io_opts_apply(&io_opts, bch2_inode_opts_get(&inode));
 563                         cur_inum = k.k->p.inode;
 564                         goto peek;
 565                 }
 566
 567                 switch ((data_cmd = pred(c, arg, k, &io_opts, &data_opts))) {
 568                 case DATA_SKIP:
 569                         goto next;
 570                 case DATA_SCRUB:
 571                         BUG();
 572                 case DATA_ADD_REPLICAS:
 573                 case DATA_REWRITE:
 574                 case DATA_PROMOTE:
 575                         break;
 576                 default:
 577                         BUG();
 578                 }
 579
 580                 /* unlock before doing IO: */
 581                 bkey_reassemble(&tmp.k, k);
 582                 k = bkey_i_to_s_c(&tmp.k);
 583                 bch2_trans_unlock(&trans);
 584
 585                 ret2 = bch2_move_extent(c, ctxt, wp, io_opts, btree_id, k,
 586                                         data_cmd, data_opts);
 587                 if (ret2) {
 588                         if (ret2 == -ENOMEM) {
 589                                 /* memory allocation failure, wait for some IO to finish */
 590                                 bch2_move_ctxt_wait_for_io(ctxt);
 591                                 continue;
 592                         }
 593
 594                         /* XXX signal failure */
 595                         goto next;
 596                 }
 597
 598                 if (rate)
 599                         bch2_ratelimit_increment(rate, k.k->size);
 600 next:
 601                 atomic64_add(k.k->size * bch2_bkey_nr_dirty_ptrs(k),
 602                              &stats->sectors_seen);
 603 next_nondata:
 604                 bch2_btree_iter_next(iter);
 605                 bch2_trans_cond_resched(&trans);
 606         }
 607 out:
 608         ret = bch2_trans_exit(&trans) ?: ret;
 609
 610         return ret;
 611 }
 612
 613 int bch2_move_data(struct bch_fs *c,
 614                    struct bch_ratelimit *rate,
 615                    struct write_point_specifier wp,
 616                    struct bpos start,
 617                    struct bpos end,
 618                    move_pred_fn pred, void *arg,
 619                    struct bch_move_stats *stats)
 620 {
 621         struct moving_context ctxt = { .stats = stats };
 622         int ret;
 623
 624         closure_init_stack(&ctxt.cl);
 625         INIT_LIST_HEAD(&ctxt.reads);
 626         init_waitqueue_head(&ctxt.wait);
 627
 628         stats->data_type = BCH_DATA_USER;
 629
 630         ret =   __bch2_move_data(c, &ctxt, rate, wp, start, end,
 631                                  pred, arg, stats, BTREE_ID_EXTENTS) ?:
 632                 __bch2_move_data(c, &ctxt, rate, wp, start, end,
 633                                  pred, arg, stats, BTREE_ID_REFLINK);
 634
 635         move_ctxt_wait_event(&ctxt, list_empty(&ctxt.reads));
 636         closure_sync(&ctxt.cl);
 637
 638         EBUG_ON(atomic_read(&ctxt.write_sectors));
 639
 640         trace_move_data(c,
 641                         atomic64_read(&stats->sectors_moved),
 642                         atomic64_read(&stats->keys_moved));
 643
 644         return ret;
 645 }
 646
 647 static int bch2_move_btree(struct bch_fs *c,
 648                            move_pred_fn pred,
 649                            void *arg,
 650                            struct bch_move_stats *stats)
 651 {
 652         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 653         struct btree_trans trans;
 654         struct btree_iter *iter;
 655         struct btree *b;
 656         unsigned id;
 657         struct data_opts data_opts;
 658         enum data_cmd cmd;
 659         int ret = 0;
 660
 661         bch2_trans_init(&trans, c, 0, 0);
 662
 663         stats->data_type = BCH_DATA_BTREE;
 664
 665         for (id = 0; id < BTREE_ID_NR; id++) {
 666                 stats->btree_id = id;
 667
 668                 for_each_btree_node(&trans, iter, id, POS_MIN,
 669                                     BTREE_ITER_PREFETCH, b) {
 670                         stats->pos = iter->pos;
 671
 672                         switch ((cmd = pred(c, arg,
 673                                             bkey_i_to_s_c(&b->key),
 674                                             &io_opts, &data_opts))) {
 675                         case DATA_SKIP:
 676                                 goto next;
 677                         case DATA_SCRUB:
 678                                 BUG();
 679                         case DATA_ADD_REPLICAS:
 680                         case DATA_REWRITE:
 681                                 break;
 682                         default:
 683                                 BUG();
 684                         }
 685
 686                         ret = bch2_btree_node_rewrite(c, iter,
 687                                         b->data->keys.seq, 0) ?: ret;
 688 next:
 689                         bch2_trans_cond_resched(&trans);
 690                 }
 691
 692                 ret = bch2_trans_iter_free(&trans, iter) ?: ret;
 693         }
 694
 695         bch2_trans_exit(&trans);
 696
 697         return ret;
 698 }
 699
 700 #if 0
 701 static enum data_cmd scrub_pred(struct bch_fs *c, void *arg,
 702                                 struct bkey_s_c k,
 703                                 struct bch_io_opts *io_opts,
 704                                 struct data_opts *data_opts)
 705 {
 706         return DATA_SCRUB;
 707 }
 708 #endif
 709
 710 static enum data_cmd rereplicate_pred(struct bch_fs *c, void *arg,
 711                                       struct bkey_s_c k,
 712                                       struct bch_io_opts *io_opts,
 713                                       struct data_opts *data_opts)
 714 {
 715         unsigned nr_good = bch2_bkey_durability(c, k);
 716         unsigned replicas = 0;
 717
 718         switch (k.k->type) {
 719         case KEY_TYPE_btree_ptr:
 720                 replicas = c->opts.metadata_replicas;
 721                 break;
 722         case KEY_TYPE_extent:
 723                 replicas = io_opts->data_replicas;
 724                 break;
 725         }
 726
 727         if (!nr_good || nr_good >= replicas)
 728                 return DATA_SKIP;
 729
 730         data_opts->target               = 0;
 731         data_opts->btree_insert_flags   = 0;
 732         return DATA_ADD_REPLICAS;
 733 }
 734
 735 static enum data_cmd migrate_pred(struct bch_fs *c, void *arg,
 736                                   struct bkey_s_c k,
 737                                   struct bch_io_opts *io_opts,
 738                                   struct data_opts *data_opts)
 739 {
 740         struct bch_ioctl_data *op = arg;
 741
 742         if (!bch2_bkey_has_device(k, op->migrate.dev))
 743                 return DATA_SKIP;
 744
 745         data_opts->target               = 0;
 746         data_opts->btree_insert_flags   = 0;
 747         data_opts->rewrite_dev          = op->migrate.dev;
 748         return DATA_REWRITE;
 749 }
 750
 751 int bch2_data_job(struct bch_fs *c,
 752                   struct bch_move_stats *stats,
 753                   struct bch_ioctl_data op)
 754 {
 755         int ret = 0;
 756
 757         switch (op.op) {
 758         case BCH_DATA_OP_REREPLICATE:
 759                 stats->data_type = BCH_DATA_JOURNAL;
 760                 ret = bch2_journal_flush_device_pins(&c->journal, -1);
 761
 762                 ret = bch2_move_btree(c, rereplicate_pred, c, stats) ?: ret;
 763
 764                 while (1) {
 765                         closure_wait_event(&c->btree_interior_update_wait,
 766                                            !bch2_btree_interior_updates_nr_pending(c) ||
 767                                            c->btree_roots_dirty);
 768                         if (!bch2_btree_interior_updates_nr_pending(c))
 769                                 break;
 770                         bch2_journal_meta(&c->journal);
 771                 }
 772
 773                 ret = bch2_replicas_gc2(c) ?: ret;
 774
 775                 ret = bch2_move_data(c, NULL,
 776                                      writepoint_hashed((unsigned long) current),
 777                                      op.start,
 778                                      op.end,
 779                                      rereplicate_pred, c, stats) ?: ret;
 780                 ret = bch2_replicas_gc2(c) ?: ret;
 781                 break;
 782         case BCH_DATA_OP_MIGRATE:
 783                 if (op.migrate.dev >= c->sb.nr_devices)
 784                         return -EINVAL;
 785
 786                 stats->data_type = BCH_DATA_JOURNAL;
 787                 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
 788
 789                 ret = bch2_move_btree(c, migrate_pred, &op, stats) ?: ret;
 790                 ret = bch2_replicas_gc2(c) ?: ret;
 791
 792                 ret = bch2_move_data(c, NULL,
 793                                      writepoint_hashed((unsigned long) current),
 794                                      op.start,
 795                                      op.end,
 796                                      migrate_pred, &op, stats) ?: ret;
 797                 ret = bch2_replicas_gc2(c) ?: ret;
 798                 break;
 799         default:
 800                 ret = -EINVAL;
 801         }
 802
 803         return ret;
 804 }