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, 0);
 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         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 219         const union bch_extent_entry *entry;
 220         struct extent_ptr_decoded p;
 221         int ret;
 222
 223         m->btree_id     = btree_id;
 224         m->data_cmd     = data_cmd;
 225         m->data_opts    = data_opts;
 226         m->nr_ptrs_reserved = 0;
 227
 228         bch2_write_op_init(&m->op, c, io_opts);
 229
 230         if (!bch2_bkey_is_incompressible(k))
 231                 m->op.compression_type =
 232                         bch2_compression_opt_to_type[io_opts.background_compression ?:
 233                                                      io_opts.compression];
 234         else
 235                 m->op.incompressible = true;
 236
 237         m->op.target    = data_opts.target,
 238         m->op.write_point = wp;
 239
 240         if (m->data_opts.btree_insert_flags & BTREE_INSERT_USE_RESERVE)
 241                 m->op.alloc_reserve = RESERVE_MOVINGGC;
 242
 243         m->op.flags |= BCH_WRITE_ONLY_SPECIFIED_DEVS|
 244                 BCH_WRITE_PAGES_STABLE|
 245                 BCH_WRITE_PAGES_OWNED|
 246                 BCH_WRITE_DATA_ENCODED|
 247                 BCH_WRITE_FROM_INTERNAL;
 248
 249         m->op.nr_replicas       = 1;
 250         m->op.nr_replicas_required = 1;
 251         m->op.index_update_fn   = bch2_migrate_index_update;
 252
 253         switch (data_cmd) {
 254         case DATA_ADD_REPLICAS: {
 255                 /*
 256                  * DATA_ADD_REPLICAS is used for moving data to a different
 257                  * device in the background, and due to compression the new copy
 258                  * might take up more space than the old copy:
 259                  */
 260 #if 0
 261                 int nr = (int) io_opts.data_replicas -
 262                         bch2_bkey_nr_ptrs_allocated(k);
 263 #endif
 264                 int nr = (int) io_opts.data_replicas;
 265
 266                 if (nr > 0) {
 267                         m->op.nr_replicas = m->nr_ptrs_reserved = nr;
 268
 269                         ret = bch2_disk_reservation_get(c, &m->op.res,
 270                                         k.k->size, m->op.nr_replicas, 0);
 271                         if (ret)
 272                                 return ret;
 273                 }
 274                 break;
 275         }
 276         case DATA_REWRITE: {
 277                 unsigned compressed_sectors = 0;
 278
 279                 bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 280                         if (!p.ptr.cached &&
 281                             crc_is_compressed(p.crc) &&
 282                             bch2_dev_in_target(c, p.ptr.dev, data_opts.target))
 283                                 compressed_sectors += p.crc.compressed_size;
 284
 285                 if (compressed_sectors) {
 286                         ret = bch2_disk_reservation_add(c, &m->op.res,
 287                                         compressed_sectors,
 288                                         BCH_DISK_RESERVATION_NOFAIL);
 289                         if (ret)
 290                                 return ret;
 291                 }
 292                 break;
 293         }
 294         case DATA_PROMOTE:
 295                 m->op.flags     |= BCH_WRITE_ALLOC_NOWAIT;
 296                 m->op.flags     |= BCH_WRITE_CACHED;
 297                 break;
 298         default:
 299                 BUG();
 300         }
 301
 302         return 0;
 303 }
 304
 305 static void move_free(struct closure *cl)
 306 {
 307         struct moving_io *io = container_of(cl, struct moving_io, cl);
 308         struct moving_context *ctxt = io->write.ctxt;
 309         struct bvec_iter_all iter;
 310         struct bio_vec *bv;
 311
 312         bch2_disk_reservation_put(io->write.op.c, &io->write.op.res);
 313
 314         bio_for_each_segment_all(bv, &io->write.op.wbio.bio, iter)
 315                 if (bv->bv_page)
 316                         __free_page(bv->bv_page);
 317
 318         wake_up(&ctxt->wait);
 319
 320         kfree(io);
 321 }
 322
 323 static void move_write_done(struct closure *cl)
 324 {
 325         struct moving_io *io = container_of(cl, struct moving_io, cl);
 326
 327         atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
 328         closure_return_with_destructor(cl, move_free);
 329 }
 330
 331 static void move_write(struct closure *cl)
 332 {
 333         struct moving_io *io = container_of(cl, struct moving_io, cl);
 334
 335         if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
 336                 closure_return_with_destructor(cl, move_free);
 337                 return;
 338         }
 339
 340         bch2_migrate_read_done(&io->write, &io->rbio);
 341
 342         atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
 343         closure_call(&io->write.op.cl, bch2_write, NULL, cl);
 344         continue_at(cl, move_write_done, NULL);
 345 }
 346
 347 static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
 348 {
 349         struct moving_io *io =
 350                 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
 351
 352         return io && io->read_completed ? io : NULL;
 353 }
 354
 355 static void move_read_endio(struct bio *bio)
 356 {
 357         struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
 358         struct moving_context *ctxt = io->write.ctxt;
 359
 360         atomic_sub(io->read_sectors, &ctxt->read_sectors);
 361         io->read_completed = true;
 362
 363         if (next_pending_write(ctxt))
 364                 wake_up(&ctxt->wait);
 365
 366         closure_put(&ctxt->cl);
 367 }
 368
 369 static void do_pending_writes(struct moving_context *ctxt)
 370 {
 371         struct moving_io *io;
 372
 373         while ((io = next_pending_write(ctxt))) {
 374                 list_del(&io->list);
 375                 closure_call(&io->cl, move_write, NULL, &ctxt->cl);
 376         }
 377 }
 378
 379 #define move_ctxt_wait_event(_ctxt, _cond)                      \
 380 do {                                                            \
 381         do_pending_writes(_ctxt);                               \
 382                                                                 \
 383         if (_cond)                                              \
 384                 break;                                          \
 385         __wait_event((_ctxt)->wait,                             \
 386                      next_pending_write(_ctxt) || (_cond));     \
 387 } while (1)
 388
 389 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt)
 390 {
 391         unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
 392
 393         move_ctxt_wait_event(ctxt,
 394                 !atomic_read(&ctxt->write_sectors) ||
 395                 atomic_read(&ctxt->write_sectors) != sectors_pending);
 396 }
 397
 398 static int bch2_move_extent(struct bch_fs *c,
 399                             struct moving_context *ctxt,
 400                             struct write_point_specifier wp,
 401                             struct bch_io_opts io_opts,
 402                             enum btree_id btree_id,
 403                             struct bkey_s_c k,
 404                             enum data_cmd data_cmd,
 405                             struct data_opts data_opts)
 406 {
 407         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 408         struct moving_io *io;
 409         const union bch_extent_entry *entry;
 410         struct extent_ptr_decoded p;
 411         unsigned sectors = k.k->size, pages;
 412         int ret = -ENOMEM;
 413
 414         move_ctxt_wait_event(ctxt,
 415                 atomic_read(&ctxt->write_sectors) <
 416                 SECTORS_IN_FLIGHT_PER_DEVICE);
 417
 418         move_ctxt_wait_event(ctxt,
 419                 atomic_read(&ctxt->read_sectors) <
 420                 SECTORS_IN_FLIGHT_PER_DEVICE);
 421
 422         /* write path might have to decompress data: */
 423         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 424                 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
 425
 426         pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
 427         io = kzalloc(sizeof(struct moving_io) +
 428                      sizeof(struct bio_vec) * pages, GFP_KERNEL);
 429         if (!io)
 430                 goto err;
 431
 432         io->write.ctxt          = ctxt;
 433         io->read_sectors        = k.k->size;
 434         io->write_sectors       = k.k->size;
 435
 436         bio_init(&io->write.op.wbio.bio, io->bi_inline_vecs, pages);
 437         bio_set_prio(&io->write.op.wbio.bio,
 438                      IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 439
 440         if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
 441                                  GFP_KERNEL))
 442                 goto err_free;
 443
 444         io->rbio.c              = c;
 445         io->rbio.opts           = io_opts;
 446         bio_init(&io->rbio.bio, io->bi_inline_vecs, pages);
 447         io->rbio.bio.bi_vcnt = pages;
 448         bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 449         io->rbio.bio.bi_iter.bi_size = sectors << 9;
 450
 451         bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
 452         io->rbio.bio.bi_iter.bi_sector  = bkey_start_offset(k.k);
 453         io->rbio.bio.bi_end_io          = move_read_endio;
 454
 455         ret = bch2_migrate_write_init(c, &io->write, wp, io_opts,
 456                                       data_cmd, data_opts, btree_id, k);
 457         if (ret)
 458                 goto err_free_pages;
 459
 460         atomic64_inc(&ctxt->stats->keys_moved);
 461         atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
 462
 463         trace_move_extent(k.k);
 464
 465         atomic_add(io->read_sectors, &ctxt->read_sectors);
 466         list_add_tail(&io->list, &ctxt->reads);
 467
 468         /*
 469          * dropped by move_read_endio() - guards against use after free of
 470          * ctxt when doing wakeup
 471          */
 472         closure_get(&ctxt->cl);
 473         bch2_read_extent(c, &io->rbio, k, 0,
 474                          BCH_READ_NODECODE|
 475                          BCH_READ_LAST_FRAGMENT);
 476         return 0;
 477 err_free_pages:
 478         bio_free_pages(&io->write.op.wbio.bio);
 479 err_free:
 480         kfree(io);
 481 err:
 482         trace_move_alloc_fail(k.k);
 483         return ret;
 484 }
 485
 486 static int __bch2_move_data(struct bch_fs *c,
 487                 struct moving_context *ctxt,
 488                 struct bch_ratelimit *rate,
 489                 struct write_point_specifier wp,
 490                 struct bpos start,
 491                 struct bpos end,
 492                 move_pred_fn pred, void *arg,
 493                 struct bch_move_stats *stats,
 494                 enum btree_id btree_id)
 495 {
 496         bool kthread = (current->flags & PF_KTHREAD) != 0;
 497         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 498         struct bkey_on_stack sk;
 499         struct btree_trans trans;
 500         struct btree_iter *iter;
 501         struct bkey_s_c k;
 502         struct data_opts data_opts;
 503         enum data_cmd data_cmd;
 504         u64 delay, cur_inum = U64_MAX;
 505         int ret = 0, ret2;
 506
 507         bkey_on_stack_init(&sk);
 508         bch2_trans_init(&trans, c, 0, 0);
 509
 510         stats->data_type = BCH_DATA_USER;
 511         stats->btree_id = btree_id;
 512         stats->pos      = POS_MIN;
 513
 514         iter = bch2_trans_get_iter(&trans, btree_id, start,
 515                                    BTREE_ITER_PREFETCH);
 516
 517         if (rate)
 518                 bch2_ratelimit_reset(rate);
 519
 520         while (1) {
 521                 do {
 522                         delay = rate ? bch2_ratelimit_delay(rate) : 0;
 523
 524                         if (delay) {
 525                                 bch2_trans_unlock(&trans);
 526                                 set_current_state(TASK_INTERRUPTIBLE);
 527                         }
 528
 529                         if (kthread && (ret = kthread_should_stop())) {
 530                                 __set_current_state(TASK_RUNNING);
 531                                 goto out;
 532                         }
 533
 534                         if (delay)
 535                                 schedule_timeout(delay);
 536
 537                         if (unlikely(freezing(current))) {
 538                                 bch2_trans_unlock(&trans);
 539                                 move_ctxt_wait_event(ctxt, list_empty(&ctxt->reads));
 540                                 try_to_freeze();
 541                         }
 542                 } while (delay);
 543 peek:
 544                 k = bch2_btree_iter_peek(iter);
 545
 546                 stats->pos = iter->pos;
 547
 548                 if (!k.k)
 549                         break;
 550                 ret = bkey_err(k);
 551                 if (ret)
 552                         break;
 553                 if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
 554                         break;
 555
 556                 if (!bkey_extent_is_direct_data(k.k))
 557                         goto next_nondata;
 558
 559                 if (btree_id == BTREE_ID_EXTENTS &&
 560                     cur_inum != k.k->p.inode) {
 561                         struct bch_inode_unpacked inode;
 562
 563                         /* don't hold btree locks while looking up inode: */
 564                         bch2_trans_unlock(&trans);
 565
 566                         io_opts = bch2_opts_to_inode_opts(c->opts);
 567                         if (!bch2_inode_find_by_inum(c, k.k->p.inode, &inode))
 568                                 bch2_io_opts_apply(&io_opts, bch2_inode_opts_get(&inode));
 569                         cur_inum = k.k->p.inode;
 570                         goto peek;
 571                 }
 572
 573                 switch ((data_cmd = pred(c, arg, k, &io_opts, &data_opts))) {
 574                 case DATA_SKIP:
 575                         goto next;
 576                 case DATA_SCRUB:
 577                         BUG();
 578                 case DATA_ADD_REPLICAS:
 579                 case DATA_REWRITE:
 580                 case DATA_PROMOTE:
 581                         break;
 582                 default:
 583                         BUG();
 584                 }
 585
 586                 /* unlock before doing IO: */
 587                 bkey_on_stack_reassemble(&sk, c, k);
 588                 k = bkey_i_to_s_c(sk.k);
 589                 bch2_trans_unlock(&trans);
 590
 591                 ret2 = bch2_move_extent(c, ctxt, wp, io_opts, btree_id, k,
 592                                         data_cmd, data_opts);
 593                 if (ret2) {
 594                         if (ret2 == -ENOMEM) {
 595                                 /* memory allocation failure, wait for some IO to finish */
 596                                 bch2_move_ctxt_wait_for_io(ctxt);
 597                                 continue;
 598                         }
 599
 600                         /* XXX signal failure */
 601                         goto next;
 602                 }
 603
 604                 if (rate)
 605                         bch2_ratelimit_increment(rate, k.k->size);
 606 next:
 607                 atomic64_add(k.k->size * bch2_bkey_nr_ptrs_allocated(k),
 608                              &stats->sectors_seen);
 609 next_nondata:
 610                 bch2_btree_iter_next(iter);
 611                 bch2_trans_cond_resched(&trans);
 612         }
 613 out:
 614         ret = bch2_trans_exit(&trans) ?: ret;
 615         bkey_on_stack_exit(&sk, c);
 616
 617         return ret;
 618 }
 619
 620 int bch2_move_data(struct bch_fs *c,
 621                    struct bch_ratelimit *rate,
 622                    struct write_point_specifier wp,
 623                    struct bpos start,
 624                    struct bpos end,
 625                    move_pred_fn pred, void *arg,
 626                    struct bch_move_stats *stats)
 627 {
 628         struct moving_context ctxt = { .stats = stats };
 629         int ret;
 630
 631         closure_init_stack(&ctxt.cl);
 632         INIT_LIST_HEAD(&ctxt.reads);
 633         init_waitqueue_head(&ctxt.wait);
 634
 635         stats->data_type = BCH_DATA_USER;
 636
 637         ret =   __bch2_move_data(c, &ctxt, rate, wp, start, end,
 638                                  pred, arg, stats, BTREE_ID_EXTENTS) ?:
 639                 __bch2_move_data(c, &ctxt, rate, wp, start, end,
 640                                  pred, arg, stats, BTREE_ID_REFLINK);
 641
 642         move_ctxt_wait_event(&ctxt, list_empty(&ctxt.reads));
 643         closure_sync(&ctxt.cl);
 644
 645         EBUG_ON(atomic_read(&ctxt.write_sectors));
 646
 647         trace_move_data(c,
 648                         atomic64_read(&stats->sectors_moved),
 649                         atomic64_read(&stats->keys_moved));
 650
 651         return ret;
 652 }
 653
 654 static int bch2_move_btree(struct bch_fs *c,
 655                            move_pred_fn pred,
 656                            void *arg,
 657                            struct bch_move_stats *stats)
 658 {
 659         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 660         struct btree_trans trans;
 661         struct btree_iter *iter;
 662         struct btree *b;
 663         unsigned id;
 664         struct data_opts data_opts;
 665         enum data_cmd cmd;
 666         int ret = 0;
 667
 668         bch2_trans_init(&trans, c, 0, 0);
 669
 670         stats->data_type = BCH_DATA_BTREE;
 671
 672         for (id = 0; id < BTREE_ID_NR; id++) {
 673                 stats->btree_id = id;
 674
 675                 for_each_btree_node(&trans, iter, id, POS_MIN,
 676                                     BTREE_ITER_PREFETCH, b) {
 677                         stats->pos = iter->pos;
 678
 679                         switch ((cmd = pred(c, arg,
 680                                             bkey_i_to_s_c(&b->key),
 681                                             &io_opts, &data_opts))) {
 682                         case DATA_SKIP:
 683                                 goto next;
 684                         case DATA_SCRUB:
 685                                 BUG();
 686                         case DATA_ADD_REPLICAS:
 687                         case DATA_REWRITE:
 688                                 break;
 689                         default:
 690                                 BUG();
 691                         }
 692
 693                         ret = bch2_btree_node_rewrite(c, iter,
 694                                         b->data->keys.seq, 0) ?: ret;
 695 next:
 696                         bch2_trans_cond_resched(&trans);
 697                 }
 698
 699                 ret = bch2_trans_iter_free(&trans, iter) ?: ret;
 700         }
 701
 702         bch2_trans_exit(&trans);
 703
 704         return ret;
 705 }
 706
 707 #if 0
 708 static enum data_cmd scrub_pred(struct bch_fs *c, void *arg,
 709                                 struct bkey_s_c k,
 710                                 struct bch_io_opts *io_opts,
 711                                 struct data_opts *data_opts)
 712 {
 713         return DATA_SCRUB;
 714 }
 715 #endif
 716
 717 static enum data_cmd rereplicate_pred(struct bch_fs *c, void *arg,
 718                                       struct bkey_s_c k,
 719                                       struct bch_io_opts *io_opts,
 720                                       struct data_opts *data_opts)
 721 {
 722         unsigned nr_good = bch2_bkey_durability(c, k);
 723         unsigned replicas = 0;
 724
 725         switch (k.k->type) {
 726         case KEY_TYPE_btree_ptr:
 727                 replicas = c->opts.metadata_replicas;
 728                 break;
 729         case KEY_TYPE_extent:
 730                 replicas = io_opts->data_replicas;
 731                 break;
 732         }
 733
 734         if (!nr_good || nr_good >= replicas)
 735                 return DATA_SKIP;
 736
 737         data_opts->target               = 0;
 738         data_opts->btree_insert_flags   = 0;
 739         return DATA_ADD_REPLICAS;
 740 }
 741
 742 static enum data_cmd migrate_pred(struct bch_fs *c, void *arg,
 743                                   struct bkey_s_c k,
 744                                   struct bch_io_opts *io_opts,
 745                                   struct data_opts *data_opts)
 746 {
 747         struct bch_ioctl_data *op = arg;
 748
 749         if (!bch2_bkey_has_device(k, op->migrate.dev))
 750                 return DATA_SKIP;
 751
 752         data_opts->target               = 0;
 753         data_opts->btree_insert_flags   = 0;
 754         data_opts->rewrite_dev          = op->migrate.dev;
 755         return DATA_REWRITE;
 756 }
 757
 758 int bch2_data_job(struct bch_fs *c,
 759                   struct bch_move_stats *stats,
 760                   struct bch_ioctl_data op)
 761 {
 762         int ret = 0;
 763
 764         switch (op.op) {
 765         case BCH_DATA_OP_REREPLICATE:
 766                 stats->data_type = BCH_DATA_JOURNAL;
 767                 ret = bch2_journal_flush_device_pins(&c->journal, -1);
 768
 769                 ret = bch2_move_btree(c, rereplicate_pred, c, stats) ?: ret;
 770
 771                 while (1) {
 772                         closure_wait_event(&c->btree_interior_update_wait,
 773                                            !bch2_btree_interior_updates_nr_pending(c) ||
 774                                            c->btree_roots_dirty);
 775                         if (!bch2_btree_interior_updates_nr_pending(c))
 776                                 break;
 777                         bch2_journal_meta(&c->journal);
 778                 }
 779
 780                 ret = bch2_replicas_gc2(c) ?: ret;
 781
 782                 ret = bch2_move_data(c, NULL,
 783                                      writepoint_hashed((unsigned long) current),
 784                                      op.start,
 785                                      op.end,
 786                                      rereplicate_pred, c, stats) ?: ret;
 787                 ret = bch2_replicas_gc2(c) ?: ret;
 788                 break;
 789         case BCH_DATA_OP_MIGRATE:
 790                 if (op.migrate.dev >= c->sb.nr_devices)
 791                         return -EINVAL;
 792
 793                 stats->data_type = BCH_DATA_JOURNAL;
 794                 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
 795
 796                 ret = bch2_move_btree(c, migrate_pred, &op, stats) ?: ret;
 797                 ret = bch2_replicas_gc2(c) ?: ret;
 798
 799                 ret = bch2_move_data(c, NULL,
 800                                      writepoint_hashed((unsigned long) current),
 801                                      op.start,
 802                                      op.end,
 803                                      migrate_pred, &op, stats) ?: ret;
 804                 ret = bch2_replicas_gc2(c) ?: ret;
 805                 break;
 806         default:
 807                 ret = -EINVAL;
 808         }
 809
 810         return ret;
 811 }