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 "bkey_on_stack.h"
   6 #include "btree_gc.h"
   7 #include "btree_update.h"
   8 #include "btree_update_interior.h"
   9 #include "buckets.h"
  10 #include "disk_groups.h"
  11 #include "inode.h"
  12 #include "io.h"
  13 #include "journal_reclaim.h"
  14 #include "move.h"
  15 #include "replicas.h"
  16 #include "super-io.h"
  17 #include "keylist.h"
  18
  19 #include <linux/ioprio.h>
  20 #include <linux/kthread.h>
  21
  22 #include <trace/events/bcachefs.h>
  23
  24 #define SECTORS_IN_FLIGHT_PER_DEVICE    2048
  25
  26 struct moving_io {
  27         struct list_head        list;
  28         struct closure          cl;
  29         bool                    read_completed;
  30
  31         unsigned                read_sectors;
  32         unsigned                write_sectors;
  33
  34         struct bch_read_bio     rbio;
  35
  36         struct migrate_write    write;
  37         /* Must be last since it is variable size */
  38         struct bio_vec          bi_inline_vecs[0];
  39 };
  40
  41 struct moving_context {
  42         /* Closure for waiting on all reads and writes to complete */
  43         struct closure          cl;
  44
  45         struct bch_move_stats   *stats;
  46
  47         struct list_head        reads;
  48
  49         /* in flight sectors: */
  50         atomic_t                read_sectors;
  51         atomic_t                write_sectors;
  52
  53         wait_queue_head_t       wait;
  54 };
  55
  56 static int bch2_migrate_index_update(struct bch_write_op *op)
  57 {
  58         struct bch_fs *c = op->c;
  59         struct btree_trans trans;
  60         struct btree_iter *iter;
  61         struct migrate_write *m =
  62                 container_of(op, struct migrate_write, op);
  63         struct keylist *keys = &op->insert_keys;
  64         int ret = 0;
  65
  66         bch2_trans_init(&trans, c, BTREE_ITER_MAX, 0);
  67
  68         iter = bch2_trans_get_iter(&trans, m->btree_id,
  69                                    bkey_start_pos(&bch2_keylist_front(keys)->k),
  70                                    BTREE_ITER_SLOTS|BTREE_ITER_INTENT);
  71
  72         while (1) {
  73                 struct bkey_s_c k = bch2_btree_iter_peek_slot(iter);
  74                 struct bkey_i *insert;
  75                 struct bkey_i_extent *new =
  76                         bkey_i_to_extent(bch2_keylist_front(keys));
  77                 BKEY_PADDED(k) _new, _insert;
  78                 const union bch_extent_entry *entry;
  79                 struct extent_ptr_decoded p;
  80                 bool did_work = false;
  81                 int nr;
  82
  83                 ret = bkey_err(k);
  84                 if (ret)
  85                         break;
  86
  87                 if (bversion_cmp(k.k->version, new->k.version) ||
  88                     !bch2_bkey_matches_ptr(c, k, m->ptr, m->offset))
  89                         goto nomatch;
  90
  91                 if (m->data_cmd == DATA_REWRITE &&
  92                     !bch2_bkey_has_device(k, m->data_opts.rewrite_dev))
  93                         goto nomatch;
  94
  95                 bkey_reassemble(&_insert.k, k);
  96                 insert = &_insert.k;
  97
  98                 bkey_copy(&_new.k, bch2_keylist_front(keys));
  99                 new = bkey_i_to_extent(&_new.k);
 100                 bch2_cut_front(iter->pos, &new->k_i);
 101
 102                 bch2_cut_front(iter->pos,       insert);
 103                 bch2_cut_back(new->k.p,         insert);
 104                 bch2_cut_back(insert->k.p,      &new->k_i);
 105
 106                 if (m->data_cmd == DATA_REWRITE)
 107                         bch2_bkey_drop_device(bkey_i_to_s(insert),
 108                                               m->data_opts.rewrite_dev);
 109
 110                 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) {
 111                         if (bch2_bkey_has_device(bkey_i_to_s_c(insert), p.ptr.dev)) {
 112                                 /*
 113                                  * raced with another move op? extent already
 114                                  * has a pointer to the device we just wrote
 115                                  * data to
 116                                  */
 117                                 continue;
 118                         }
 119
 120                         bch2_extent_ptr_decoded_append(insert, &p);
 121                         did_work = true;
 122                 }
 123
 124                 if (!did_work)
 125                         goto nomatch;
 126
 127                 bch2_bkey_narrow_crcs(insert,
 128                                 (struct bch_extent_crc_unpacked) { 0 });
 129                 bch2_extent_normalize(c, bkey_i_to_s(insert));
 130                 bch2_bkey_mark_replicas_cached(c, bkey_i_to_s(insert),
 131                                                op->opts.background_target,
 132                                                op->opts.data_replicas);
 133
 134                 /*
 135                  * If we're not fully overwriting @k, and it's compressed, we
 136                  * need a reservation for all the pointers in @insert
 137                  */
 138                 nr = bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(insert)) -
 139                          m->nr_ptrs_reserved;
 140
 141                 if (insert->k.size < k.k->size &&
 142                     bch2_bkey_sectors_compressed(k) &&
 143                     nr > 0) {
 144                         ret = bch2_disk_reservation_add(c, &op->res,
 145                                         keylist_sectors(keys) * nr, 0);
 146                         if (ret)
 147                                 goto out;
 148
 149                         m->nr_ptrs_reserved += nr;
 150                         goto next;
 151                 }
 152
 153                 bch2_trans_update(&trans, iter, insert);
 154
 155                 ret = bch2_trans_commit(&trans, &op->res,
 156                                 op_journal_seq(op),
 157                                 BTREE_INSERT_NOFAIL|
 158                                 BTREE_INSERT_USE_RESERVE|
 159                                 m->data_opts.btree_insert_flags);
 160                 if (!ret)
 161                         atomic_long_inc(&c->extent_migrate_done);
 162                 if (ret == -EINTR)
 163                         ret = 0;
 164                 if (ret)
 165                         break;
 166 next:
 167                 while (bkey_cmp(iter->pos, bch2_keylist_front(keys)->k.p) >= 0) {
 168                         bch2_keylist_pop_front(keys);
 169                         if (bch2_keylist_empty(keys))
 170                                 goto out;
 171                 }
 172                 continue;
 173 nomatch:
 174                 if (m->ctxt)
 175                         atomic64_add(k.k->p.offset - iter->pos.offset,
 176                                      &m->ctxt->stats->sectors_raced);
 177                 atomic_long_inc(&c->extent_migrate_raced);
 178                 trace_move_race(&new->k);
 179                 bch2_btree_iter_next_slot(iter);
 180                 goto next;
 181         }
 182 out:
 183         bch2_trans_exit(&trans);
 184         BUG_ON(ret == -EINTR);
 185         return ret;
 186 }
 187
 188 void bch2_migrate_read_done(struct migrate_write *m, struct bch_read_bio *rbio)
 189 {
 190         /* write bio must own pages: */
 191         BUG_ON(!m->op.wbio.bio.bi_vcnt);
 192
 193         m->ptr          = rbio->pick.ptr;
 194         m->offset       = rbio->pos.offset - rbio->pick.crc.offset;
 195         m->op.devs_have = rbio->devs_have;
 196         m->op.pos       = rbio->pos;
 197         m->op.version   = rbio->version;
 198         m->op.crc       = rbio->pick.crc;
 199         m->op.wbio.bio.bi_iter.bi_size = m->op.crc.compressed_size << 9;
 200
 201         if (bch2_csum_type_is_encryption(m->op.crc.csum_type)) {
 202                 m->op.nonce     = m->op.crc.nonce + m->op.crc.offset;
 203                 m->op.csum_type = m->op.crc.csum_type;
 204         }
 205
 206         if (m->data_cmd == DATA_REWRITE)
 207                 bch2_dev_list_drop_dev(&m->op.devs_have, m->data_opts.rewrite_dev);
 208 }
 209
 210 int bch2_migrate_write_init(struct bch_fs *c, struct migrate_write *m,
 211                             struct write_point_specifier wp,
 212                             struct bch_io_opts io_opts,
 213                             enum data_cmd data_cmd,
 214                             struct data_opts data_opts,
 215                             enum btree_id btree_id,
 216                             struct bkey_s_c k)
 217 {
 218         int ret;
 219
 220         m->btree_id     = btree_id;
 221         m->data_cmd     = data_cmd;
 222         m->data_opts    = data_opts;
 223         m->nr_ptrs_reserved = 0;
 224
 225         bch2_write_op_init(&m->op, c, io_opts);
 226         m->op.compression_type =
 227                 bch2_compression_opt_to_type[io_opts.background_compression ?:
 228                                              io_opts.compression];
 229         m->op.target    = data_opts.target,
 230         m->op.write_point = wp;
 231
 232         if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE)
 233                 m->op.alloc_reserve = RESERVE_MOVINGGC;
 234
 235         m->op.flags |= BCH_WRITE_ONLY_SPECIFIED_DEVS|
 236                 BCH_WRITE_PAGES_STABLE|
 237                 BCH_WRITE_PAGES_OWNED|
 238                 BCH_WRITE_DATA_ENCODED;
 239
 240         m->op.nr_replicas       = 1;
 241         m->op.nr_replicas_required = 1;
 242         m->op.index_update_fn   = bch2_migrate_index_update;
 243
 244         switch (data_cmd) {
 245         case DATA_ADD_REPLICAS: {
 246                 /*
 247                  * DATA_ADD_REPLICAS is used for moving data to a different
 248                  * device in the background, and due to compression the new copy
 249                  * might take up more space than the old copy:
 250                  */
 251 #if 0
 252                 int nr = (int) io_opts.data_replicas -
 253                         bch2_bkey_nr_ptrs_allocated(k);
 254 #endif
 255                 int nr = (int) io_opts.data_replicas;
 256
 257                 if (nr > 0) {
 258                         m->op.nr_replicas = m->nr_ptrs_reserved = nr;
 259
 260                         ret = bch2_disk_reservation_get(c, &m->op.res,
 261                                         k.k->size, m->op.nr_replicas, 0);
 262                         if (ret)
 263                                 return ret;
 264                 }
 265                 break;
 266         }
 267         case DATA_REWRITE: {
 268                 struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 269                 const union bch_extent_entry *entry;
 270                 struct extent_ptr_decoded p;
 271                 unsigned compressed_sectors = 0;
 272
 273                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 274                         if (!p.ptr.cached &&
 275                             p.crc.compression_type != BCH_COMPRESSION_TYPE_none &&
 276                             bch2_dev_in_target(c, p.ptr.dev, data_opts.target))
 277                                 compressed_sectors += p.crc.compressed_size;
 278
 279                 if (compressed_sectors) {
 280                         ret = bch2_disk_reservation_add(c, &m->op.res,
 281                                         compressed_sectors,
 282                                         BCH_DISK_RESERVATION_NOFAIL);
 283                         if (ret)
 284                                 return ret;
 285                 }
 286                 break;
 287         }
 288         case DATA_PROMOTE:
 289                 m->op.flags     |= BCH_WRITE_ALLOC_NOWAIT;
 290                 m->op.flags     |= BCH_WRITE_CACHED;
 291                 break;
 292         default:
 293                 BUG();
 294         }
 295
 296         return 0;
 297 }
 298
 299 static void move_free(struct closure *cl)
 300 {
 301         struct moving_io *io = container_of(cl, struct moving_io, cl);
 302         struct moving_context *ctxt = io->write.ctxt;
 303         struct bvec_iter_all iter;
 304         struct bio_vec *bv;
 305
 306         bch2_disk_reservation_put(io->write.op.c, &io->write.op.res);
 307
 308         bio_for_each_segment_all(bv, &io->write.op.wbio.bio, iter)
 309                 if (bv->bv_page)
 310                         __free_page(bv->bv_page);
 311
 312         wake_up(&ctxt->wait);
 313
 314         kfree(io);
 315 }
 316
 317 static void move_write_done(struct closure *cl)
 318 {
 319         struct moving_io *io = container_of(cl, struct moving_io, cl);
 320
 321         atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
 322         closure_return_with_destructor(cl, move_free);
 323 }
 324
 325 static void move_write(struct closure *cl)
 326 {
 327         struct moving_io *io = container_of(cl, struct moving_io, cl);
 328
 329         if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
 330                 closure_return_with_destructor(cl, move_free);
 331                 return;
 332         }
 333
 334         bch2_migrate_read_done(&io->write, &io->rbio);
 335
 336         atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
 337         closure_call(&io->write.op.cl, bch2_write, NULL, cl);
 338         continue_at(cl, move_write_done, NULL);
 339 }
 340
 341 static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
 342 {
 343         struct moving_io *io =
 344                 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
 345
 346         return io && io->read_completed ? io : NULL;
 347 }
 348
 349 static void move_read_endio(struct bio *bio)
 350 {
 351         struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
 352         struct moving_context *ctxt = io->write.ctxt;
 353
 354         atomic_sub(io->read_sectors, &ctxt->read_sectors);
 355         io->read_completed = true;
 356
 357         if (next_pending_write(ctxt))
 358                 wake_up(&ctxt->wait);
 359
 360         closure_put(&ctxt->cl);
 361 }
 362
 363 static void do_pending_writes(struct moving_context *ctxt)
 364 {
 365         struct moving_io *io;
 366
 367         while ((io = next_pending_write(ctxt))) {
 368                 list_del(&io->list);
 369                 closure_call(&io->cl, move_write, NULL, &ctxt->cl);
 370         }
 371 }
 372
 373 #define move_ctxt_wait_event(_ctxt, _cond)                      \
 374 do {                                                            \
 375         do_pending_writes(_ctxt);                               \
 376                                                                 \
 377         if (_cond)                                              \
 378                 break;                                          \
 379         __wait_event((_ctxt)->wait,                             \
 380                      next_pending_write(_ctxt) || (_cond));     \
 381 } while (1)
 382
 383 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt)
 384 {
 385         unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
 386
 387         move_ctxt_wait_event(ctxt,
 388                 !atomic_read(&ctxt->write_sectors) ||
 389                 atomic_read(&ctxt->write_sectors) != sectors_pending);
 390 }
 391
 392 static int bch2_move_extent(struct bch_fs *c,
 393                             struct moving_context *ctxt,
 394                             struct write_point_specifier wp,
 395                             struct bch_io_opts io_opts,
 396                             enum btree_id btree_id,
 397                             struct bkey_s_c k,
 398                             enum data_cmd data_cmd,
 399                             struct data_opts data_opts)
 400 {
 401         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 402         struct moving_io *io;
 403         const union bch_extent_entry *entry;
 404         struct extent_ptr_decoded p;
 405         unsigned sectors = k.k->size, pages;
 406         int ret = -ENOMEM;
 407
 408         move_ctxt_wait_event(ctxt,
 409                 atomic_read(&ctxt->write_sectors) <
 410                 SECTORS_IN_FLIGHT_PER_DEVICE);
 411
 412         move_ctxt_wait_event(ctxt,
 413                 atomic_read(&ctxt->read_sectors) <
 414                 SECTORS_IN_FLIGHT_PER_DEVICE);
 415
 416         /* write path might have to decompress data: */
 417         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 418                 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
 419
 420         pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
 421         io = kzalloc(sizeof(struct moving_io) +
 422                      sizeof(struct bio_vec) * pages, GFP_KERNEL);
 423         if (!io)
 424                 goto err;
 425
 426         io->write.ctxt          = ctxt;
 427         io->read_sectors        = k.k->size;
 428         io->write_sectors       = k.k->size;
 429
 430         bio_init(&io->write.op.wbio.bio, io->bi_inline_vecs, pages);
 431         bio_set_prio(&io->write.op.wbio.bio,
 432                      IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 433
 434         if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
 435                                  GFP_KERNEL))
 436                 goto err_free;
 437
 438         io->rbio.c              = c;
 439         io->rbio.opts           = io_opts;
 440         bio_init(&io->rbio.bio, io->bi_inline_vecs, pages);
 441         io->rbio.bio.bi_vcnt = pages;
 442         bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 443         io->rbio.bio.bi_iter.bi_size = sectors << 9;
 444
 445         bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
 446         io->rbio.bio.bi_iter.bi_sector  = bkey_start_offset(k.k);
 447         io->rbio.bio.bi_end_io          = move_read_endio;
 448
 449         ret = bch2_migrate_write_init(c, &io->write, wp, io_opts,
 450                                       data_cmd, data_opts, btree_id, k);
 451         if (ret)
 452                 goto err_free_pages;
 453
 454         atomic64_inc(&ctxt->stats->keys_moved);
 455         atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
 456
 457         trace_move_extent(k.k);
 458
 459         atomic_add(io->read_sectors, &ctxt->read_sectors);
 460         list_add_tail(&io->list, &ctxt->reads);
 461
 462         /*
 463          * dropped by move_read_endio() - guards against use after free of
 464          * ctxt when doing wakeup
 465          */
 466         closure_get(&ctxt->cl);
 467         bch2_read_extent(c, &io->rbio, k, 0,
 468                          BCH_READ_NODECODE|
 469                          BCH_READ_LAST_FRAGMENT);
 470         return 0;
 471 err_free_pages:
 472         bio_free_pages(&io->write.op.wbio.bio);
 473 err_free:
 474         kfree(io);
 475 err:
 476         trace_move_alloc_fail(k.k);
 477         return ret;
 478 }
 479
 480 static int __bch2_move_data(struct bch_fs *c,
 481                 struct moving_context *ctxt,
 482                 struct bch_ratelimit *rate,
 483                 struct write_point_specifier wp,
 484                 struct bpos start,
 485                 struct bpos end,
 486                 move_pred_fn pred, void *arg,
 487                 struct bch_move_stats *stats,
 488                 enum btree_id btree_id)
 489 {
 490         bool kthread = (current->flags & PF_KTHREAD) != 0;
 491         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 492         struct bkey_on_stack sk;
 493         struct btree_trans trans;
 494         struct btree_iter *iter;
 495         struct bkey_s_c k;
 496         struct data_opts data_opts;
 497         enum data_cmd data_cmd;
 498         u64 delay, cur_inum = U64_MAX;
 499         int ret = 0, ret2;
 500
 501         bkey_on_stack_init(&sk);
 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_on_stack_reassemble(&sk, c, k);
 582                 k = bkey_i_to_s_c(sk.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_ptrs_allocated(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         bkey_on_stack_exit(&sk, c);
 610
 611         return ret;
 612 }
 613
 614 int bch2_move_data(struct bch_fs *c,
 615                    struct bch_ratelimit *rate,
 616                    struct write_point_specifier wp,
 617                    struct bpos start,
 618                    struct bpos end,
 619                    move_pred_fn pred, void *arg,
 620                    struct bch_move_stats *stats)
 621 {
 622         struct moving_context ctxt = { .stats = stats };
 623         int ret;
 624
 625         closure_init_stack(&ctxt.cl);
 626         INIT_LIST_HEAD(&ctxt.reads);
 627         init_waitqueue_head(&ctxt.wait);
 628
 629         stats->data_type = BCH_DATA_USER;
 630
 631         ret =   __bch2_move_data(c, &ctxt, rate, wp, start, end,
 632                                  pred, arg, stats, BTREE_ID_EXTENTS) ?:
 633                 __bch2_move_data(c, &ctxt, rate, wp, start, end,
 634                                  pred, arg, stats, BTREE_ID_REFLINK);
 635
 636         move_ctxt_wait_event(&ctxt, list_empty(&ctxt.reads));
 637         closure_sync(&ctxt.cl);
 638
 639         EBUG_ON(atomic_read(&ctxt.write_sectors));
 640
 641         trace_move_data(c,
 642                         atomic64_read(&stats->sectors_moved),
 643                         atomic64_read(&stats->keys_moved));
 644
 645         return ret;
 646 }
 647
 648 static int bch2_move_btree(struct bch_fs *c,
 649                            move_pred_fn pred,
 650                            void *arg,
 651                            struct bch_move_stats *stats)
 652 {
 653         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 654         struct btree_trans trans;
 655         struct btree_iter *iter;
 656         struct btree *b;
 657         unsigned id;
 658         struct data_opts data_opts;
 659         enum data_cmd cmd;
 660         int ret = 0;
 661
 662         bch2_trans_init(&trans, c, 0, 0);
 663
 664         stats->data_type = BCH_DATA_BTREE;
 665
 666         for (id = 0; id < BTREE_ID_NR; id++) {
 667                 stats->btree_id = id;
 668
 669                 for_each_btree_node(&trans, iter, id, POS_MIN,
 670                                     BTREE_ITER_PREFETCH, b) {
 671                         stats->pos = iter->pos;
 672
 673                         switch ((cmd = pred(c, arg,
 674                                             bkey_i_to_s_c(&b->key),
 675                                             &io_opts, &data_opts))) {
 676                         case DATA_SKIP:
 677                                 goto next;
 678                         case DATA_SCRUB:
 679                                 BUG();
 680                         case DATA_ADD_REPLICAS:
 681                         case DATA_REWRITE:
 682                                 break;
 683                         default:
 684                                 BUG();
 685                         }
 686
 687                         ret = bch2_btree_node_rewrite(c, iter,
 688                                         b->data->keys.seq, 0) ?: ret;
 689 next:
 690                         bch2_trans_cond_resched(&trans);
 691                 }
 692
 693                 ret = bch2_trans_iter_free(&trans, iter) ?: ret;
 694         }
 695
 696         bch2_trans_exit(&trans);
 697
 698         return ret;
 699 }
 700
 701 #if 0
 702 static enum data_cmd scrub_pred(struct bch_fs *c, void *arg,
 703                                 struct bkey_s_c k,
 704                                 struct bch_io_opts *io_opts,
 705                                 struct data_opts *data_opts)
 706 {
 707         return DATA_SCRUB;
 708 }
 709 #endif
 710
 711 static enum data_cmd rereplicate_pred(struct bch_fs *c, void *arg,
 712                                       struct bkey_s_c k,
 713                                       struct bch_io_opts *io_opts,
 714                                       struct data_opts *data_opts)
 715 {
 716         unsigned nr_good = bch2_bkey_durability(c, k);
 717         unsigned replicas = 0;
 718
 719         switch (k.k->type) {
 720         case KEY_TYPE_btree_ptr:
 721                 replicas = c->opts.metadata_replicas;
 722                 break;
 723         case KEY_TYPE_extent:
 724                 replicas = io_opts->data_replicas;
 725                 break;
 726         }
 727
 728         if (!nr_good || nr_good >= replicas)
 729                 return DATA_SKIP;
 730
 731         data_opts->target               = 0;
 732         data_opts->btree_insert_flags   = 0;
 733         return DATA_ADD_REPLICAS;
 734 }
 735
 736 static enum data_cmd migrate_pred(struct bch_fs *c, void *arg,
 737                                   struct bkey_s_c k,
 738                                   struct bch_io_opts *io_opts,
 739                                   struct data_opts *data_opts)
 740 {
 741         struct bch_ioctl_data *op = arg;
 742
 743         if (!bch2_bkey_has_device(k, op->migrate.dev))
 744                 return DATA_SKIP;
 745
 746         data_opts->target               = 0;
 747         data_opts->btree_insert_flags   = 0;
 748         data_opts->rewrite_dev          = op->migrate.dev;
 749         return DATA_REWRITE;
 750 }
 751
 752 int bch2_data_job(struct bch_fs *c,
 753                   struct bch_move_stats *stats,
 754                   struct bch_ioctl_data op)
 755 {
 756         int ret = 0;
 757
 758         switch (op.op) {
 759         case BCH_DATA_OP_REREPLICATE:
 760                 stats->data_type = BCH_DATA_JOURNAL;
 761                 ret = bch2_journal_flush_device_pins(&c->journal, -1);
 762
 763                 ret = bch2_move_btree(c, rereplicate_pred, c, stats) ?: ret;
 764
 765                 while (1) {
 766                         closure_wait_event(&c->btree_interior_update_wait,
 767                                            !bch2_btree_interior_updates_nr_pending(c) ||
 768                                            c->btree_roots_dirty);
 769                         if (!bch2_btree_interior_updates_nr_pending(c))
 770                                 break;
 771                         bch2_journal_meta(&c->journal);
 772                 }
 773
 774                 ret = bch2_replicas_gc2(c) ?: ret;
 775
 776                 ret = bch2_move_data(c, NULL,
 777                                      writepoint_hashed((unsigned long) current),
 778                                      op.start,
 779                                      op.end,
 780                                      rereplicate_pred, c, stats) ?: ret;
 781                 ret = bch2_replicas_gc2(c) ?: ret;
 782                 break;
 783         case BCH_DATA_OP_MIGRATE:
 784                 if (op.migrate.dev >= c->sb.nr_devices)
 785                         return -EINVAL;
 786
 787                 stats->data_type = BCH_DATA_JOURNAL;
 788                 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
 789
 790                 ret = bch2_move_btree(c, migrate_pred, &op, stats) ?: ret;
 791                 ret = bch2_replicas_gc2(c) ?: ret;
 792
 793                 ret = bch2_move_data(c, NULL,
 794                                      writepoint_hashed((unsigned long) current),
 795                                      op.start,
 796                                      op.end,
 797                                      migrate_pred, &op, stats) ?: ret;
 798                 ret = bch2_replicas_gc2(c) ?: ret;
 799                 break;
 800         default:
 801                 ret = -EINVAL;
 802         }
 803
 804         return ret;
 805 }