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;
  74                 struct bkey_i *insert;
  75                 struct bkey_i_extent *new;
  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                 bch2_trans_reset(&trans, 0);
  83
  84                 k = bch2_btree_iter_peek_slot(iter);
  85                 ret = bkey_err(k);
  86                 if (ret) {
  87                         if (ret == -EINTR)
  88                                 continue;
  89                         break;
  90                 }
  91
  92                 new = bkey_i_to_extent(bch2_keylist_front(keys));
  93
  94                 if (bversion_cmp(k.k->version, new->k.version) ||
  95                     !bch2_bkey_matches_ptr(c, k, m->ptr, m->offset))
  96                         goto nomatch;
  97
  98                 if (m->data_cmd == DATA_REWRITE &&
  99                     !bch2_bkey_has_device(k, m->data_opts.rewrite_dev))
 100                         goto nomatch;
 101
 102                 bkey_reassemble(&_insert.k, k);
 103                 insert = &_insert.k;
 104
 105                 bkey_copy(&_new.k, bch2_keylist_front(keys));
 106                 new = bkey_i_to_extent(&_new.k);
 107                 bch2_cut_front(iter->pos, &new->k_i);
 108
 109                 bch2_cut_front(iter->pos,       insert);
 110                 bch2_cut_back(new->k.p,         insert);
 111                 bch2_cut_back(insert->k.p,      &new->k_i);
 112
 113                 if (m->data_cmd == DATA_REWRITE)
 114                         bch2_bkey_drop_device(bkey_i_to_s(insert),
 115                                               m->data_opts.rewrite_dev);
 116
 117                 extent_for_each_ptr_decode(extent_i_to_s(new), p, entry) {
 118                         if (bch2_bkey_has_device(bkey_i_to_s_c(insert), p.ptr.dev)) {
 119                                 /*
 120                                  * raced with another move op? extent already
 121                                  * has a pointer to the device we just wrote
 122                                  * data to
 123                                  */
 124                                 continue;
 125                         }
 126
 127                         bch2_extent_ptr_decoded_append(insert, &p);
 128                         did_work = true;
 129                 }
 130
 131                 if (!did_work)
 132                         goto nomatch;
 133
 134                 bch2_bkey_narrow_crcs(insert,
 135                                 (struct bch_extent_crc_unpacked) { 0 });
 136                 bch2_extent_normalize(c, bkey_i_to_s(insert));
 137                 bch2_bkey_mark_replicas_cached(c, bkey_i_to_s(insert),
 138                                                op->opts.background_target,
 139                                                op->opts.data_replicas);
 140
 141                 /*
 142                  * If we're not fully overwriting @k, and it's compressed, we
 143                  * need a reservation for all the pointers in @insert
 144                  */
 145                 nr = bch2_bkey_nr_ptrs_allocated(bkey_i_to_s_c(insert)) -
 146                          m->nr_ptrs_reserved;
 147
 148                 if (insert->k.size < k.k->size &&
 149                     bch2_bkey_sectors_compressed(k) &&
 150                     nr > 0) {
 151                         ret = bch2_disk_reservation_add(c, &op->res,
 152                                         keylist_sectors(keys) * nr, 0);
 153                         if (ret)
 154                                 goto out;
 155
 156                         m->nr_ptrs_reserved += nr;
 157                         goto next;
 158                 }
 159
 160                 bch2_trans_update(&trans, iter, insert, 0);
 161
 162                 ret = bch2_trans_commit(&trans, &op->res,
 163                                 op_journal_seq(op),
 164                                 BTREE_INSERT_NOFAIL|
 165                                 BTREE_INSERT_USE_RESERVE|
 166                                 m->data_opts.btree_insert_flags);
 167                 if (!ret)
 168                         atomic_long_inc(&c->extent_migrate_done);
 169                 if (ret == -EINTR)
 170                         ret = 0;
 171                 if (ret)
 172                         break;
 173 next:
 174                 while (bkey_cmp(iter->pos, bch2_keylist_front(keys)->k.p) >= 0) {
 175                         bch2_keylist_pop_front(keys);
 176                         if (bch2_keylist_empty(keys))
 177                                 goto out;
 178                 }
 179                 continue;
 180 nomatch:
 181                 if (m->ctxt)
 182                         atomic64_add(k.k->p.offset - iter->pos.offset,
 183                                      &m->ctxt->stats->sectors_raced);
 184                 atomic_long_inc(&c->extent_migrate_raced);
 185                 trace_move_race(&new->k);
 186                 bch2_btree_iter_next_slot(iter);
 187                 goto next;
 188         }
 189 out:
 190         bch2_trans_exit(&trans);
 191         BUG_ON(ret == -EINTR);
 192         return ret;
 193 }
 194
 195 void bch2_migrate_read_done(struct migrate_write *m, struct bch_read_bio *rbio)
 196 {
 197         /* write bio must own pages: */
 198         BUG_ON(!m->op.wbio.bio.bi_vcnt);
 199
 200         m->ptr          = rbio->pick.ptr;
 201         m->offset       = rbio->pos.offset - rbio->pick.crc.offset;
 202         m->op.devs_have = rbio->devs_have;
 203         m->op.pos       = rbio->pos;
 204         m->op.version   = rbio->version;
 205         m->op.crc       = rbio->pick.crc;
 206         m->op.wbio.bio.bi_iter.bi_size = m->op.crc.compressed_size << 9;
 207
 208         if (bch2_csum_type_is_encryption(m->op.crc.csum_type)) {
 209                 m->op.nonce     = m->op.crc.nonce + m->op.crc.offset;
 210                 m->op.csum_type = m->op.crc.csum_type;
 211         }
 212
 213         if (m->data_cmd == DATA_REWRITE)
 214                 bch2_dev_list_drop_dev(&m->op.devs_have, m->data_opts.rewrite_dev);
 215 }
 216
 217 int bch2_migrate_write_init(struct bch_fs *c, struct migrate_write *m,
 218                             struct write_point_specifier wp,
 219                             struct bch_io_opts io_opts,
 220                             enum data_cmd data_cmd,
 221                             struct data_opts data_opts,
 222                             enum btree_id btree_id,
 223                             struct bkey_s_c k)
 224 {
 225         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 226         const union bch_extent_entry *entry;
 227         struct extent_ptr_decoded p;
 228         int ret;
 229
 230         m->btree_id     = btree_id;
 231         m->data_cmd     = data_cmd;
 232         m->data_opts    = data_opts;
 233         m->nr_ptrs_reserved = 0;
 234
 235         bch2_write_op_init(&m->op, c, io_opts);
 236
 237         if (!bch2_bkey_is_incompressible(k))
 238                 m->op.compression_type =
 239                         bch2_compression_opt_to_type[io_opts.background_compression ?:
 240                                                      io_opts.compression];
 241         else
 242                 m->op.incompressible = true;
 243
 244         m->op.target    = data_opts.target,
 245         m->op.write_point = wp;
 246
 247         if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE)
 248                 m->op.alloc_reserve = RESERVE_MOVINGGC;
 249
 250         m->op.flags |= BCH_WRITE_ONLY_SPECIFIED_DEVS|
 251                 BCH_WRITE_PAGES_STABLE|
 252                 BCH_WRITE_PAGES_OWNED|
 253                 BCH_WRITE_DATA_ENCODED|
 254                 BCH_WRITE_FROM_INTERNAL;
 255
 256         m->op.nr_replicas       = 1;
 257         m->op.nr_replicas_required = 1;
 258         m->op.index_update_fn   = bch2_migrate_index_update;
 259
 260         switch (data_cmd) {
 261         case DATA_ADD_REPLICAS: {
 262                 /*
 263                  * DATA_ADD_REPLICAS is used for moving data to a different
 264                  * device in the background, and due to compression the new copy
 265                  * might take up more space than the old copy:
 266                  */
 267 #if 0
 268                 int nr = (int) io_opts.data_replicas -
 269                         bch2_bkey_nr_ptrs_allocated(k);
 270 #endif
 271                 int nr = (int) io_opts.data_replicas;
 272
 273                 if (nr > 0) {
 274                         m->op.nr_replicas = m->nr_ptrs_reserved = nr;
 275
 276                         ret = bch2_disk_reservation_get(c, &m->op.res,
 277                                         k.k->size, m->op.nr_replicas, 0);
 278                         if (ret)
 279                                 return ret;
 280                 }
 281                 break;
 282         }
 283         case DATA_REWRITE: {
 284                 unsigned compressed_sectors = 0;
 285
 286                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 287                         if (!p.ptr.cached &&
 288                             crc_is_compressed(p.crc) &&
 289                             bch2_dev_in_target(c, p.ptr.dev, data_opts.target))
 290                                 compressed_sectors += p.crc.compressed_size;
 291
 292                 if (compressed_sectors) {
 293                         ret = bch2_disk_reservation_add(c, &m->op.res,
 294                                         compressed_sectors,
 295                                         BCH_DISK_RESERVATION_NOFAIL);
 296                         if (ret)
 297                                 return ret;
 298                 }
 299                 break;
 300         }
 301         case DATA_PROMOTE:
 302                 m->op.flags     |= BCH_WRITE_ALLOC_NOWAIT;
 303                 m->op.flags     |= BCH_WRITE_CACHED;
 304                 break;
 305         default:
 306                 BUG();
 307         }
 308
 309         return 0;
 310 }
 311
 312 static void move_free(struct closure *cl)
 313 {
 314         struct moving_io *io = container_of(cl, struct moving_io, cl);
 315         struct moving_context *ctxt = io->write.ctxt;
 316         struct bvec_iter_all iter;
 317         struct bio_vec *bv;
 318
 319         bch2_disk_reservation_put(io->write.op.c, &io->write.op.res);
 320
 321         bio_for_each_segment_all(bv, &io->write.op.wbio.bio, iter)
 322                 if (bv->bv_page)
 323                         __free_page(bv->bv_page);
 324
 325         wake_up(&ctxt->wait);
 326
 327         kfree(io);
 328 }
 329
 330 static void move_write_done(struct closure *cl)
 331 {
 332         struct moving_io *io = container_of(cl, struct moving_io, cl);
 333
 334         atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
 335         closure_return_with_destructor(cl, move_free);
 336 }
 337
 338 static void move_write(struct closure *cl)
 339 {
 340         struct moving_io *io = container_of(cl, struct moving_io, cl);
 341
 342         if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
 343                 closure_return_with_destructor(cl, move_free);
 344                 return;
 345         }
 346
 347         bch2_migrate_read_done(&io->write, &io->rbio);
 348
 349         atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
 350         closure_call(&io->write.op.cl, bch2_write, NULL, cl);
 351         continue_at(cl, move_write_done, NULL);
 352 }
 353
 354 static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
 355 {
 356         struct moving_io *io =
 357                 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
 358
 359         return io && io->read_completed ? io : NULL;
 360 }
 361
 362 static void move_read_endio(struct bio *bio)
 363 {
 364         struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
 365         struct moving_context *ctxt = io->write.ctxt;
 366
 367         atomic_sub(io->read_sectors, &ctxt->read_sectors);
 368         io->read_completed = true;
 369
 370         if (next_pending_write(ctxt))
 371                 wake_up(&ctxt->wait);
 372
 373         closure_put(&ctxt->cl);
 374 }
 375
 376 static void do_pending_writes(struct moving_context *ctxt)
 377 {
 378         struct moving_io *io;
 379
 380         while ((io = next_pending_write(ctxt))) {
 381                 list_del(&io->list);
 382                 closure_call(&io->cl, move_write, NULL, &ctxt->cl);
 383         }
 384 }
 385
 386 #define move_ctxt_wait_event(_ctxt, _cond)                      \
 387 do {                                                            \
 388         do_pending_writes(_ctxt);                               \
 389                                                                 \
 390         if (_cond)                                              \
 391                 break;                                          \
 392         __wait_event((_ctxt)->wait,                             \
 393                      next_pending_write(_ctxt) || (_cond));     \
 394 } while (1)
 395
 396 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt)
 397 {
 398         unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
 399
 400         move_ctxt_wait_event(ctxt,
 401                 !atomic_read(&ctxt->write_sectors) ||
 402                 atomic_read(&ctxt->write_sectors) != sectors_pending);
 403 }
 404
 405 static int bch2_move_extent(struct bch_fs *c,
 406                             struct moving_context *ctxt,
 407                             struct write_point_specifier wp,
 408                             struct bch_io_opts io_opts,
 409                             enum btree_id btree_id,
 410                             struct bkey_s_c k,
 411                             enum data_cmd data_cmd,
 412                             struct data_opts data_opts)
 413 {
 414         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 415         struct moving_io *io;
 416         const union bch_extent_entry *entry;
 417         struct extent_ptr_decoded p;
 418         unsigned sectors = k.k->size, pages;
 419         int ret = -ENOMEM;
 420
 421         move_ctxt_wait_event(ctxt,
 422                 atomic_read(&ctxt->write_sectors) <
 423                 SECTORS_IN_FLIGHT_PER_DEVICE);
 424
 425         move_ctxt_wait_event(ctxt,
 426                 atomic_read(&ctxt->read_sectors) <
 427                 SECTORS_IN_FLIGHT_PER_DEVICE);
 428
 429         /* write path might have to decompress data: */
 430         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 431                 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
 432
 433         pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
 434         io = kzalloc(sizeof(struct moving_io) +
 435                      sizeof(struct bio_vec) * pages, GFP_KERNEL);
 436         if (!io)
 437                 goto err;
 438
 439         io->write.ctxt          = ctxt;
 440         io->read_sectors        = k.k->size;
 441         io->write_sectors       = k.k->size;
 442
 443         bio_init(&io->write.op.wbio.bio, io->bi_inline_vecs, pages);
 444         bio_set_prio(&io->write.op.wbio.bio,
 445                      IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 446
 447         if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
 448                                  GFP_KERNEL))
 449                 goto err_free;
 450
 451         io->rbio.c              = c;
 452         io->rbio.opts           = io_opts;
 453         bio_init(&io->rbio.bio, io->bi_inline_vecs, pages);
 454         io->rbio.bio.bi_vcnt = pages;
 455         bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 456         io->rbio.bio.bi_iter.bi_size = sectors << 9;
 457
 458         bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
 459         io->rbio.bio.bi_iter.bi_sector  = bkey_start_offset(k.k);
 460         io->rbio.bio.bi_end_io          = move_read_endio;
 461
 462         ret = bch2_migrate_write_init(c, &io->write, wp, io_opts,
 463                                       data_cmd, data_opts, btree_id, k);
 464         if (ret)
 465                 goto err_free_pages;
 466
 467         atomic64_inc(&ctxt->stats->keys_moved);
 468         atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
 469
 470         trace_move_extent(k.k);
 471
 472         atomic_add(io->read_sectors, &ctxt->read_sectors);
 473         list_add_tail(&io->list, &ctxt->reads);
 474
 475         /*
 476          * dropped by move_read_endio() - guards against use after free of
 477          * ctxt when doing wakeup
 478          */
 479         closure_get(&ctxt->cl);
 480         bch2_read_extent(c, &io->rbio, k, 0,
 481                          BCH_READ_NODECODE|
 482                          BCH_READ_LAST_FRAGMENT);
 483         return 0;
 484 err_free_pages:
 485         bio_free_pages(&io->write.op.wbio.bio);
 486 err_free:
 487         kfree(io);
 488 err:
 489         trace_move_alloc_fail(k.k);
 490         return ret;
 491 }
 492
 493 static int __bch2_move_data(struct bch_fs *c,
 494                 struct moving_context *ctxt,
 495                 struct bch_ratelimit *rate,
 496                 struct write_point_specifier wp,
 497                 struct bpos start,
 498                 struct bpos end,
 499                 move_pred_fn pred, void *arg,
 500                 struct bch_move_stats *stats,
 501                 enum btree_id btree_id)
 502 {
 503         bool kthread = (current->flags & PF_KTHREAD) != 0;
 504         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 505         struct bkey_on_stack sk;
 506         struct btree_trans trans;
 507         struct btree_iter *iter;
 508         struct bkey_s_c k;
 509         struct data_opts data_opts;
 510         enum data_cmd data_cmd;
 511         u64 delay, cur_inum = U64_MAX;
 512         int ret = 0, ret2;
 513
 514         bkey_on_stack_init(&sk);
 515         bch2_trans_init(&trans, c, 0, 0);
 516
 517         stats->data_type = BCH_DATA_USER;
 518         stats->btree_id = btree_id;
 519         stats->pos      = POS_MIN;
 520
 521         iter = bch2_trans_get_iter(&trans, btree_id, start,
 522                                    BTREE_ITER_PREFETCH);
 523
 524         if (rate)
 525                 bch2_ratelimit_reset(rate);
 526
 527         while (1) {
 528                 do {
 529                         delay = rate ? bch2_ratelimit_delay(rate) : 0;
 530
 531                         if (delay) {
 532                                 bch2_trans_unlock(&trans);
 533                                 set_current_state(TASK_INTERRUPTIBLE);
 534                         }
 535
 536                         if (kthread && (ret = kthread_should_stop())) {
 537                                 __set_current_state(TASK_RUNNING);
 538                                 goto out;
 539                         }
 540
 541                         if (delay)
 542                                 schedule_timeout(delay);
 543
 544                         if (unlikely(freezing(current))) {
 545                                 bch2_trans_unlock(&trans);
 546                                 move_ctxt_wait_event(ctxt, list_empty(&ctxt->reads));
 547                                 try_to_freeze();
 548                         }
 549                 } while (delay);
 550 peek:
 551                 k = bch2_btree_iter_peek(iter);
 552
 553                 stats->pos = iter->pos;
 554
 555                 if (!k.k)
 556                         break;
 557                 ret = bkey_err(k);
 558                 if (ret)
 559                         break;
 560                 if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
 561                         break;
 562
 563                 if (!bkey_extent_is_direct_data(k.k))
 564                         goto next_nondata;
 565
 566                 if (btree_id == BTREE_ID_EXTENTS &&
 567                     cur_inum != k.k->p.inode) {
 568                         struct bch_inode_unpacked inode;
 569
 570                         /* don't hold btree locks while looking up inode: */
 571                         bch2_trans_unlock(&trans);
 572
 573                         io_opts = bch2_opts_to_inode_opts(c->opts);
 574                         if (!bch2_inode_find_by_inum(c, k.k->p.inode, &inode))
 575                                 bch2_io_opts_apply(&io_opts, bch2_inode_opts_get(&inode));
 576                         cur_inum = k.k->p.inode;
 577                         goto peek;
 578                 }
 579
 580                 switch ((data_cmd = pred(c, arg, k, &io_opts, &data_opts))) {
 581                 case DATA_SKIP:
 582                         goto next;
 583                 case DATA_SCRUB:
 584                         BUG();
 585                 case DATA_ADD_REPLICAS:
 586                 case DATA_REWRITE:
 587                 case DATA_PROMOTE:
 588                         break;
 589                 default:
 590                         BUG();
 591                 }
 592
 593                 /* unlock before doing IO: */
 594                 bkey_on_stack_reassemble(&sk, c, k);
 595                 k = bkey_i_to_s_c(sk.k);
 596                 bch2_trans_unlock(&trans);
 597
 598                 ret2 = bch2_move_extent(c, ctxt, wp, io_opts, btree_id, k,
 599                                         data_cmd, data_opts);
 600                 if (ret2) {
 601                         if (ret2 == -ENOMEM) {
 602                                 /* memory allocation failure, wait for some IO to finish */
 603                                 bch2_move_ctxt_wait_for_io(ctxt);
 604                                 continue;
 605                         }
 606
 607                         /* XXX signal failure */
 608                         goto next;
 609                 }
 610
 611                 if (rate)
 612                         bch2_ratelimit_increment(rate, k.k->size);
 613 next:
 614                 atomic64_add(k.k->size * bch2_bkey_nr_ptrs_allocated(k),
 615                              &stats->sectors_seen);
 616 next_nondata:
 617                 bch2_btree_iter_next(iter);
 618                 bch2_trans_cond_resched(&trans);
 619         }
 620 out:
 621         ret = bch2_trans_exit(&trans) ?: ret;
 622         bkey_on_stack_exit(&sk, c);
 623
 624         return ret;
 625 }
 626
 627 int bch2_move_data(struct bch_fs *c,
 628                    struct bch_ratelimit *rate,
 629                    struct write_point_specifier wp,
 630                    struct bpos start,
 631                    struct bpos end,
 632                    move_pred_fn pred, void *arg,
 633                    struct bch_move_stats *stats)
 634 {
 635         struct moving_context ctxt = { .stats = stats };
 636         int ret;
 637
 638         closure_init_stack(&ctxt.cl);
 639         INIT_LIST_HEAD(&ctxt.reads);
 640         init_waitqueue_head(&ctxt.wait);
 641
 642         stats->data_type = BCH_DATA_USER;
 643
 644         ret =   __bch2_move_data(c, &ctxt, rate, wp, start, end,
 645                                  pred, arg, stats, BTREE_ID_EXTENTS) ?:
 646                 __bch2_move_data(c, &ctxt, rate, wp, start, end,
 647                                  pred, arg, stats, BTREE_ID_REFLINK);
 648
 649         move_ctxt_wait_event(&ctxt, list_empty(&ctxt.reads));
 650         closure_sync(&ctxt.cl);
 651
 652         EBUG_ON(atomic_read(&ctxt.write_sectors));
 653
 654         trace_move_data(c,
 655                         atomic64_read(&stats->sectors_moved),
 656                         atomic64_read(&stats->keys_moved));
 657
 658         return ret;
 659 }
 660
 661 static int bch2_move_btree(struct bch_fs *c,
 662                            move_pred_fn pred,
 663                            void *arg,
 664                            struct bch_move_stats *stats)
 665 {
 666         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 667         struct btree_trans trans;
 668         struct btree_iter *iter;
 669         struct btree *b;
 670         unsigned id;
 671         struct data_opts data_opts;
 672         enum data_cmd cmd;
 673         int ret = 0;
 674
 675         bch2_trans_init(&trans, c, 0, 0);
 676
 677         stats->data_type = BCH_DATA_BTREE;
 678
 679         for (id = 0; id < BTREE_ID_NR; id++) {
 680                 stats->btree_id = id;
 681
 682                 for_each_btree_node(&trans, iter, id, POS_MIN,
 683                                     BTREE_ITER_PREFETCH, b) {
 684                         stats->pos = iter->pos;
 685
 686                         switch ((cmd = pred(c, arg,
 687                                             bkey_i_to_s_c(&b->key),
 688                                             &io_opts, &data_opts))) {
 689                         case DATA_SKIP:
 690                                 goto next;
 691                         case DATA_SCRUB:
 692                                 BUG();
 693                         case DATA_ADD_REPLICAS:
 694                         case DATA_REWRITE:
 695                                 break;
 696                         default:
 697                                 BUG();
 698                         }
 699
 700                         ret = bch2_btree_node_rewrite(c, iter,
 701                                         b->data->keys.seq, 0) ?: ret;
 702 next:
 703                         bch2_trans_cond_resched(&trans);
 704                 }
 705
 706                 ret = bch2_trans_iter_free(&trans, iter) ?: ret;
 707         }
 708
 709         bch2_trans_exit(&trans);
 710
 711         return ret;
 712 }
 713
 714 #if 0
 715 static enum data_cmd scrub_pred(struct bch_fs *c, void *arg,
 716                                 struct bkey_s_c k,
 717                                 struct bch_io_opts *io_opts,
 718                                 struct data_opts *data_opts)
 719 {
 720         return DATA_SCRUB;
 721 }
 722 #endif
 723
 724 static enum data_cmd rereplicate_pred(struct bch_fs *c, void *arg,
 725                                       struct bkey_s_c k,
 726                                       struct bch_io_opts *io_opts,
 727                                       struct data_opts *data_opts)
 728 {
 729         unsigned nr_good = bch2_bkey_durability(c, k);
 730         unsigned replicas = 0;
 731
 732         switch (k.k->type) {
 733         case KEY_TYPE_btree_ptr:
 734                 replicas = c->opts.metadata_replicas;
 735                 break;
 736         case KEY_TYPE_extent:
 737                 replicas = io_opts->data_replicas;
 738                 break;
 739         }
 740
 741         if (!nr_good || nr_good >= replicas)
 742                 return DATA_SKIP;
 743
 744         data_opts->target               = 0;
 745         data_opts->btree_insert_flags   = 0;
 746         return DATA_ADD_REPLICAS;
 747 }
 748
 749 static enum data_cmd migrate_pred(struct bch_fs *c, void *arg,
 750                                   struct bkey_s_c k,
 751                                   struct bch_io_opts *io_opts,
 752                                   struct data_opts *data_opts)
 753 {
 754         struct bch_ioctl_data *op = arg;
 755
 756         if (!bch2_bkey_has_device(k, op->migrate.dev))
 757                 return DATA_SKIP;
 758
 759         data_opts->target               = 0;
 760         data_opts->btree_insert_flags   = 0;
 761         data_opts->rewrite_dev          = op->migrate.dev;
 762         return DATA_REWRITE;
 763 }
 764
 765 int bch2_data_job(struct bch_fs *c,
 766                   struct bch_move_stats *stats,
 767                   struct bch_ioctl_data op)
 768 {
 769         int ret = 0;
 770
 771         switch (op.op) {
 772         case BCH_DATA_OP_REREPLICATE:
 773                 stats->data_type = BCH_DATA_JOURNAL;
 774                 ret = bch2_journal_flush_device_pins(&c->journal, -1);
 775
 776                 ret = bch2_move_btree(c, rereplicate_pred, c, stats) ?: ret;
 777
 778                 closure_wait_event(&c->btree_interior_update_wait,
 779                                    !bch2_btree_interior_updates_nr_pending(c));
 780
 781                 ret = bch2_replicas_gc2(c) ?: ret;
 782
 783                 ret = bch2_move_data(c, NULL,
 784                                      writepoint_hashed((unsigned long) current),
 785                                      op.start,
 786                                      op.end,
 787                                      rereplicate_pred, c, stats) ?: ret;
 788                 ret = bch2_replicas_gc2(c) ?: ret;
 789                 break;
 790         case BCH_DATA_OP_MIGRATE:
 791                 if (op.migrate.dev >= c->sb.nr_devices)
 792                         return -EINVAL;
 793
 794                 stats->data_type = BCH_DATA_JOURNAL;
 795                 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
 796
 797                 ret = bch2_move_btree(c, migrate_pred, &op, stats) ?: ret;
 798                 ret = bch2_replicas_gc2(c) ?: ret;
 799
 800                 ret = bch2_move_data(c, NULL,
 801                                      writepoint_hashed((unsigned long) current),
 802                                      op.start,
 803                                      op.end,
 804                                      migrate_pred, &op, stats) ?: ret;
 805                 ret = bch2_replicas_gc2(c) ?: ret;
 806                 break;
 807         default:
 808                 ret = -EINVAL;
 809         }
 810
 811         return ret;
 812 }