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 "backpointers.h"
   6 #include "bkey_buf.h"
   7 #include "btree_gc.h"
   8 #include "btree_update.h"
   9 #include "btree_update_interior.h"
  10 #include "disk_groups.h"
  11 #include "ec.h"
  12 #include "errcode.h"
  13 #include "error.h"
  14 #include "inode.h"
  15 #include "io.h"
  16 #include "journal_reclaim.h"
  17 #include "move.h"
  18 #include "replicas.h"
  19 #include "super-io.h"
  20 #include "keylist.h"
  21
  22 #include <linux/ioprio.h>
  23 #include <linux/kthread.h>
  24
  25 #include <trace/events/bcachefs.h>
  26
  27 static void progress_list_add(struct bch_fs *c, struct bch_move_stats *stats)
  28 {
  29         mutex_lock(&c->data_progress_lock);
  30         list_add(&stats->list, &c->data_progress_list);
  31         mutex_unlock(&c->data_progress_lock);
  32 }
  33
  34 static void progress_list_del(struct bch_fs *c, struct bch_move_stats *stats)
  35 {
  36         mutex_lock(&c->data_progress_lock);
  37         list_del(&stats->list);
  38         mutex_unlock(&c->data_progress_lock);
  39 }
  40
  41 struct moving_io {
  42         struct list_head        list;
  43         struct closure          cl;
  44         bool                    read_completed;
  45
  46         unsigned                read_sectors;
  47         unsigned                write_sectors;
  48
  49         struct bch_read_bio     rbio;
  50
  51         struct data_update      write;
  52         /* Must be last since it is variable size */
  53         struct bio_vec          bi_inline_vecs[0];
  54 };
  55
  56 static void move_free(struct moving_io *io)
  57 {
  58         struct moving_context *ctxt = io->write.ctxt;
  59         struct bch_fs *c = ctxt->c;
  60
  61         bch2_data_update_exit(&io->write);
  62         wake_up(&ctxt->wait);
  63         percpu_ref_put(&c->writes);
  64         kfree(io);
  65 }
  66
  67 static void move_write_done(struct bch_write_op *op)
  68 {
  69         struct moving_io *io = container_of(op, struct moving_io, write.op);
  70         struct moving_context *ctxt = io->write.ctxt;
  71
  72         if (io->write.op.error)
  73                 ctxt->write_error = true;
  74
  75         atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
  76         move_free(io);
  77         closure_put(&ctxt->cl);
  78 }
  79
  80 static void move_write(struct moving_io *io)
  81 {
  82         if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
  83                 move_free(io);
  84                 return;
  85         }
  86
  87         closure_get(&io->write.ctxt->cl);
  88         atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
  89
  90         bch2_data_update_read_done(&io->write, io->rbio.pick.crc);
  91 }
  92
  93 static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
  94 {
  95         struct moving_io *io =
  96                 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
  97
  98         return io && io->read_completed ? io : NULL;
  99 }
 100
 101 static void move_read_endio(struct bio *bio)
 102 {
 103         struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
 104         struct moving_context *ctxt = io->write.ctxt;
 105
 106         atomic_sub(io->read_sectors, &ctxt->read_sectors);
 107         io->read_completed = true;
 108
 109         wake_up(&ctxt->wait);
 110         closure_put(&ctxt->cl);
 111 }
 112
 113 static void do_pending_writes(struct moving_context *ctxt, struct btree_trans *trans)
 114 {
 115         struct moving_io *io;
 116
 117         if (trans)
 118                 bch2_trans_unlock(trans);
 119
 120         while ((io = next_pending_write(ctxt))) {
 121                 list_del(&io->list);
 122                 move_write(io);
 123         }
 124 }
 125
 126 #define move_ctxt_wait_event(_ctxt, _trans, _cond)              \
 127 do {                                                            \
 128         do_pending_writes(_ctxt, _trans);                       \
 129                                                                 \
 130         if (_cond)                                              \
 131                 break;                                          \
 132         __wait_event((_ctxt)->wait,                             \
 133                      next_pending_write(_ctxt) || (_cond));     \
 134 } while (1)
 135
 136 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt,
 137                                        struct btree_trans *trans)
 138 {
 139         unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
 140
 141         move_ctxt_wait_event(ctxt, trans,
 142                 !atomic_read(&ctxt->write_sectors) ||
 143                 atomic_read(&ctxt->write_sectors) != sectors_pending);
 144 }
 145
 146 void bch2_moving_ctxt_exit(struct moving_context *ctxt)
 147 {
 148         move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
 149         closure_sync(&ctxt->cl);
 150         EBUG_ON(atomic_read(&ctxt->write_sectors));
 151
 152         if (ctxt->stats) {
 153                 progress_list_del(ctxt->c, ctxt->stats);
 154
 155                 trace_move_data(ctxt->c,
 156                                 atomic64_read(&ctxt->stats->sectors_moved),
 157                                 atomic64_read(&ctxt->stats->keys_moved));
 158         }
 159 }
 160
 161 void bch2_moving_ctxt_init(struct moving_context *ctxt,
 162                            struct bch_fs *c,
 163                            struct bch_ratelimit *rate,
 164                            struct bch_move_stats *stats,
 165                            struct write_point_specifier wp,
 166                            bool wait_on_copygc)
 167 {
 168         memset(ctxt, 0, sizeof(*ctxt));
 169
 170         ctxt->c         = c;
 171         ctxt->rate      = rate;
 172         ctxt->stats     = stats;
 173         ctxt->wp        = wp;
 174         ctxt->wait_on_copygc = wait_on_copygc;
 175
 176         closure_init_stack(&ctxt->cl);
 177         INIT_LIST_HEAD(&ctxt->reads);
 178         init_waitqueue_head(&ctxt->wait);
 179
 180         if (stats) {
 181                 progress_list_add(c, stats);
 182                 stats->data_type = BCH_DATA_user;
 183         }
 184 }
 185
 186 void bch2_move_stats_init(struct bch_move_stats *stats, char *name)
 187 {
 188         memset(stats, 0, sizeof(*stats));
 189         scnprintf(stats->name, sizeof(stats->name), "%s", name);
 190 }
 191
 192 static int bch2_extent_drop_ptrs(struct btree_trans *trans,
 193                                  struct btree_iter *iter,
 194                                  struct bkey_s_c k,
 195                                  struct data_update_opts data_opts)
 196 {
 197         struct bch_fs *c = trans->c;
 198         struct bkey_i *n;
 199         int ret;
 200
 201         n = bch2_bkey_make_mut(trans, k);
 202         ret = PTR_ERR_OR_ZERO(n);
 203         if (ret)
 204                 return ret;
 205
 206         while (data_opts.kill_ptrs) {
 207                 unsigned i = 0, drop = __fls(data_opts.kill_ptrs);
 208                 struct bch_extent_ptr *ptr;
 209
 210                 bch2_bkey_drop_ptrs(bkey_i_to_s(n), ptr, i++ == drop);
 211                 data_opts.kill_ptrs ^= 1U << drop;
 212         }
 213
 214         /*
 215          * If the new extent no longer has any pointers, bch2_extent_normalize()
 216          * will do the appropriate thing with it (turning it into a
 217          * KEY_TYPE_error key, or just a discard if it was a cached extent)
 218          */
 219         bch2_extent_normalize(c, bkey_i_to_s(n));
 220
 221         /*
 222          * Since we're not inserting through an extent iterator
 223          * (BTREE_ITER_ALL_SNAPSHOTS iterators aren't extent iterators),
 224          * we aren't using the extent overwrite path to delete, we're
 225          * just using the normal key deletion path:
 226          */
 227         if (bkey_deleted(&n->k))
 228                 n->k.size = 0;
 229
 230         return bch2_trans_update(trans, iter, n, BTREE_UPDATE_INTERNAL_SNAPSHOT_NODE) ?:
 231                 bch2_trans_commit(trans, NULL, NULL, BTREE_INSERT_NOFAIL);
 232 }
 233
 234 static int bch2_move_extent(struct btree_trans *trans,
 235                             struct btree_iter *iter,
 236                             struct moving_context *ctxt,
 237                             struct bch_io_opts io_opts,
 238                             enum btree_id btree_id,
 239                             struct bkey_s_c k,
 240                             struct data_update_opts data_opts)
 241 {
 242         struct bch_fs *c = trans->c;
 243         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 244         struct moving_io *io;
 245         const union bch_extent_entry *entry;
 246         struct extent_ptr_decoded p;
 247         unsigned sectors = k.k->size, pages;
 248         int ret = -ENOMEM;
 249
 250         bch2_data_update_opts_normalize(k, &data_opts);
 251
 252         if (!data_opts.rewrite_ptrs &&
 253             !data_opts.extra_replicas) {
 254                 if (data_opts.kill_ptrs)
 255                         return bch2_extent_drop_ptrs(trans, iter, k, data_opts);
 256                 return 0;
 257         }
 258
 259         if (!percpu_ref_tryget_live(&c->writes))
 260                 return -EROFS;
 261
 262         /*
 263          * Before memory allocations & taking nocow locks in
 264          * bch2_data_update_init():
 265          */
 266         bch2_trans_unlock(trans);
 267
 268         /* write path might have to decompress data: */
 269         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 270                 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
 271
 272         pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
 273         io = kzalloc(sizeof(struct moving_io) +
 274                      sizeof(struct bio_vec) * pages, GFP_KERNEL);
 275         if (!io)
 276                 goto err;
 277
 278         io->write.ctxt          = ctxt;
 279         io->read_sectors        = k.k->size;
 280         io->write_sectors       = k.k->size;
 281
 282         bio_init(&io->write.op.wbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 283         bio_set_prio(&io->write.op.wbio.bio,
 284                      IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 285
 286         if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
 287                                  GFP_KERNEL))
 288                 goto err_free;
 289
 290         io->rbio.c              = c;
 291         io->rbio.opts           = io_opts;
 292         bio_init(&io->rbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 293         io->rbio.bio.bi_vcnt = pages;
 294         bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 295         io->rbio.bio.bi_iter.bi_size = sectors << 9;
 296
 297         bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
 298         io->rbio.bio.bi_iter.bi_sector  = bkey_start_offset(k.k);
 299         io->rbio.bio.bi_end_io          = move_read_endio;
 300
 301         ret = bch2_data_update_init(c, &io->write, ctxt->wp, io_opts,
 302                                     data_opts, btree_id, k);
 303         if (ret && ret != -BCH_ERR_unwritten_extent_update)
 304                 goto err_free_pages;
 305
 306         io->write.ctxt = ctxt;
 307         io->write.op.end_io = move_write_done;
 308
 309         atomic64_inc(&ctxt->stats->keys_moved);
 310         atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
 311
 312         if (ret == -BCH_ERR_unwritten_extent_update) {
 313                 bch2_update_unwritten_extent(trans, &io->write);
 314                 move_free(io);
 315                 return 0;
 316         }
 317
 318         BUG_ON(ret);
 319
 320         this_cpu_add(c->counters[BCH_COUNTER_io_move], k.k->size);
 321         this_cpu_add(c->counters[BCH_COUNTER_move_extent_read], k.k->size);
 322         trace_move_extent_read(k.k);
 323
 324         atomic_add(io->read_sectors, &ctxt->read_sectors);
 325         list_add_tail(&io->list, &ctxt->reads);
 326
 327         /*
 328          * dropped by move_read_endio() - guards against use after free of
 329          * ctxt when doing wakeup
 330          */
 331         closure_get(&ctxt->cl);
 332         bch2_read_extent(trans, &io->rbio,
 333                          bkey_start_pos(k.k),
 334                          btree_id, k, 0,
 335                          BCH_READ_NODECODE|
 336                          BCH_READ_LAST_FRAGMENT);
 337         return 0;
 338 err_free_pages:
 339         bio_free_pages(&io->write.op.wbio.bio);
 340 err_free:
 341         kfree(io);
 342 err:
 343         percpu_ref_put(&c->writes);
 344         trace_and_count(c, move_extent_alloc_mem_fail, k.k);
 345         return ret;
 346 }
 347
 348 static int lookup_inode(struct btree_trans *trans, struct bpos pos,
 349                         struct bch_inode_unpacked *inode)
 350 {
 351         struct btree_iter iter;
 352         struct bkey_s_c k;
 353         int ret;
 354
 355         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
 356                              BTREE_ITER_ALL_SNAPSHOTS);
 357         k = bch2_btree_iter_peek(&iter);
 358         ret = bkey_err(k);
 359         if (ret)
 360                 goto err;
 361
 362         if (!k.k || !bkey_eq(k.k->p, pos)) {
 363                 ret = -ENOENT;
 364                 goto err;
 365         }
 366
 367         ret = bkey_is_inode(k.k) ? 0 : -EIO;
 368         if (ret)
 369                 goto err;
 370
 371         ret = bch2_inode_unpack(k, inode);
 372         if (ret)
 373                 goto err;
 374 err:
 375         bch2_trans_iter_exit(trans, &iter);
 376         return ret;
 377 }
 378
 379 static int move_ratelimit(struct btree_trans *trans,
 380                           struct moving_context *ctxt)
 381 {
 382         struct bch_fs *c = trans->c;
 383         u64 delay;
 384
 385         if (ctxt->wait_on_copygc) {
 386                 bch2_trans_unlock(trans);
 387                 wait_event_killable(c->copygc_running_wq,
 388                                     !c->copygc_running ||
 389                                     kthread_should_stop());
 390         }
 391
 392         do {
 393                 delay = ctxt->rate ? bch2_ratelimit_delay(ctxt->rate) : 0;
 394
 395                 if (delay) {
 396                         bch2_trans_unlock(trans);
 397                         set_current_state(TASK_INTERRUPTIBLE);
 398                 }
 399
 400                 if ((current->flags & PF_KTHREAD) && kthread_should_stop()) {
 401                         __set_current_state(TASK_RUNNING);
 402                         return 1;
 403                 }
 404
 405                 if (delay)
 406                         schedule_timeout(delay);
 407
 408                 if (unlikely(freezing(current))) {
 409                         move_ctxt_wait_event(ctxt, trans, list_empty(&ctxt->reads));
 410                         try_to_freeze();
 411                 }
 412         } while (delay);
 413
 414         move_ctxt_wait_event(ctxt, trans,
 415                 atomic_read(&ctxt->write_sectors) <
 416                 c->opts.move_bytes_in_flight >> 9);
 417
 418         move_ctxt_wait_event(ctxt, trans,
 419                 atomic_read(&ctxt->read_sectors) <
 420                 c->opts.move_bytes_in_flight >> 9);
 421
 422         return 0;
 423 }
 424
 425 static int move_get_io_opts(struct btree_trans *trans,
 426                             struct bch_io_opts *io_opts,
 427                             struct bkey_s_c k, u64 *cur_inum)
 428 {
 429         struct bch_inode_unpacked inode;
 430         int ret;
 431
 432         if (*cur_inum == k.k->p.inode)
 433                 return 0;
 434
 435         ret = lookup_inode(trans,
 436                            SPOS(0, k.k->p.inode, k.k->p.snapshot),
 437                            &inode);
 438         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 439                 return ret;
 440
 441         if (!ret)
 442                 bch2_inode_opts_get(io_opts, trans->c, &inode);
 443         else
 444                 *io_opts = bch2_opts_to_inode_opts(trans->c->opts);
 445         *cur_inum = k.k->p.inode;
 446         return 0;
 447 }
 448
 449 static int __bch2_move_data(struct moving_context *ctxt,
 450                             struct bpos start,
 451                             struct bpos end,
 452                             move_pred_fn pred, void *arg,
 453                             enum btree_id btree_id)
 454 {
 455         struct bch_fs *c = ctxt->c;
 456         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 457         struct bkey_buf sk;
 458         struct btree_trans trans;
 459         struct btree_iter iter;
 460         struct bkey_s_c k;
 461         struct data_update_opts data_opts;
 462         u64 cur_inum = U64_MAX;
 463         int ret = 0, ret2;
 464
 465         bch2_bkey_buf_init(&sk);
 466         bch2_trans_init(&trans, c, 0, 0);
 467
 468         ctxt->stats->data_type  = BCH_DATA_user;
 469         ctxt->stats->btree_id   = btree_id;
 470         ctxt->stats->pos        = start;
 471
 472         bch2_trans_iter_init(&trans, &iter, btree_id, start,
 473                              BTREE_ITER_PREFETCH|
 474                              BTREE_ITER_ALL_SNAPSHOTS);
 475
 476         if (ctxt->rate)
 477                 bch2_ratelimit_reset(ctxt->rate);
 478
 479         while (!move_ratelimit(&trans, ctxt)) {
 480                 bch2_trans_begin(&trans);
 481
 482                 k = bch2_btree_iter_peek(&iter);
 483                 if (!k.k)
 484                         break;
 485
 486                 ret = bkey_err(k);
 487                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 488                         continue;
 489                 if (ret)
 490                         break;
 491
 492                 if (bkey_ge(bkey_start_pos(k.k), end))
 493                         break;
 494
 495                 ctxt->stats->pos = iter.pos;
 496
 497                 if (!bkey_extent_is_direct_data(k.k))
 498                         goto next_nondata;
 499
 500                 ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 501                 if (ret)
 502                         continue;
 503
 504                 memset(&data_opts, 0, sizeof(data_opts));
 505                 if (!pred(c, arg, k, &io_opts, &data_opts))
 506                         goto next;
 507
 508                 /*
 509                  * The iterator gets unlocked by __bch2_read_extent - need to
 510                  * save a copy of @k elsewhere:
 511                  */
 512                 bch2_bkey_buf_reassemble(&sk, c, k);
 513                 k = bkey_i_to_s_c(sk.k);
 514                 bch2_trans_unlock(&trans);
 515
 516                 ret2 = bch2_move_extent(&trans, &iter, ctxt, io_opts,
 517                                         btree_id, k, data_opts);
 518                 if (ret2) {
 519                         if (bch2_err_matches(ret2, BCH_ERR_transaction_restart))
 520                                 continue;
 521
 522                         if (ret2 == -ENOMEM) {
 523                                 /* memory allocation failure, wait for some IO to finish */
 524                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 525                                 continue;
 526                         }
 527
 528                         /* XXX signal failure */
 529                         goto next;
 530                 }
 531
 532                 if (ctxt->rate)
 533                         bch2_ratelimit_increment(ctxt->rate, k.k->size);
 534 next:
 535                 atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 536 next_nondata:
 537                 bch2_btree_iter_advance(&iter);
 538         }
 539
 540         bch2_trans_iter_exit(&trans, &iter);
 541         bch2_trans_exit(&trans);
 542         bch2_bkey_buf_exit(&sk, c);
 543
 544         return ret;
 545 }
 546
 547 int bch2_move_data(struct bch_fs *c,
 548                    enum btree_id start_btree_id, struct bpos start_pos,
 549                    enum btree_id end_btree_id,   struct bpos end_pos,
 550                    struct bch_ratelimit *rate,
 551                    struct bch_move_stats *stats,
 552                    struct write_point_specifier wp,
 553                    bool wait_on_copygc,
 554                    move_pred_fn pred, void *arg)
 555 {
 556         struct moving_context ctxt;
 557         enum btree_id id;
 558         int ret;
 559
 560         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 561
 562         for (id = start_btree_id;
 563              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 564              id++) {
 565                 stats->btree_id = id;
 566
 567                 if (id != BTREE_ID_extents &&
 568                     id != BTREE_ID_reflink)
 569                         continue;
 570
 571                 ret = __bch2_move_data(&ctxt,
 572                                        id == start_btree_id ? start_pos : POS_MIN,
 573                                        id == end_btree_id   ? end_pos   : POS_MAX,
 574                                        pred, arg, id);
 575                 if (ret)
 576                         break;
 577         }
 578
 579         bch2_moving_ctxt_exit(&ctxt);
 580
 581         return ret;
 582 }
 583
 584 static int verify_bucket_evacuated(struct btree_trans *trans, struct bpos bucket, int gen)
 585 {
 586         struct bch_fs *c = trans->c;
 587         struct btree_iter iter;
 588         struct bkey_s_c k;
 589         struct printbuf buf = PRINTBUF;
 590         struct bch_backpointer bp;
 591         u64 bp_offset = 0;
 592         int ret;
 593
 594         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
 595                              bucket, BTREE_ITER_CACHED);
 596 again:
 597         k = bch2_btree_iter_peek_slot(&iter);
 598         ret = bkey_err(k);
 599
 600         if (!ret && k.k->type == KEY_TYPE_alloc_v4) {
 601                 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
 602
 603                 if (a.v->gen == gen &&
 604                     a.v->dirty_sectors) {
 605                         if (a.v->data_type == BCH_DATA_btree) {
 606                                 bch2_trans_unlock(trans);
 607                                 if (bch2_btree_interior_updates_flush(c))
 608                                         goto again;
 609                                 goto failed_to_evacuate;
 610                         }
 611                 }
 612         }
 613
 614         bch2_trans_iter_exit(trans, &iter);
 615         return ret;
 616 failed_to_evacuate:
 617         bch2_trans_iter_exit(trans, &iter);
 618
 619         prt_printf(&buf, bch2_log_msg(c, "failed to evacuate bucket "));
 620         bch2_bkey_val_to_text(&buf, c, k);
 621
 622         while (1) {
 623                 bch2_trans_begin(trans);
 624
 625                 ret = bch2_get_next_backpointer(trans, bucket, gen,
 626                                                 &bp_offset, &bp,
 627                                                 BTREE_ITER_CACHED);
 628                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 629                         continue;
 630                 if (ret)
 631                         break;
 632                 if (bp_offset == U64_MAX)
 633                         break;
 634
 635                 k = bch2_backpointer_get_key(trans, &iter,
 636                                              bucket, bp_offset, bp);
 637                 ret = bkey_err(k);
 638                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 639                         continue;
 640                 if (ret)
 641                         break;
 642                 if (!k.k)
 643                         continue;
 644                 prt_newline(&buf);
 645                 bch2_bkey_val_to_text(&buf, c, k);
 646                 bch2_trans_iter_exit(trans, &iter);
 647         }
 648
 649         bch2_print_string_as_lines(KERN_ERR, buf.buf);
 650         printbuf_exit(&buf);
 651         return 0;
 652 }
 653
 654 int __bch2_evacuate_bucket(struct moving_context *ctxt,
 655                            struct bpos bucket, int gen,
 656                            struct data_update_opts _data_opts)
 657 {
 658         struct bch_fs *c = ctxt->c;
 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 bkey_buf sk;
 663         struct bch_backpointer bp;
 664         struct data_update_opts data_opts;
 665         u64 bp_offset = 0, cur_inum = U64_MAX;
 666         int ret = 0;
 667
 668         bch2_bkey_buf_init(&sk);
 669         bch2_trans_init(&trans, c, 0, 0);
 670
 671         while (!(ret = move_ratelimit(&trans, ctxt))) {
 672                 bch2_trans_begin(&trans);
 673
 674                 ret = bch2_get_next_backpointer(&trans, bucket, gen,
 675                                                 &bp_offset, &bp,
 676                                                 BTREE_ITER_CACHED);
 677                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 678                         continue;
 679                 if (ret)
 680                         goto err;
 681                 if (bp_offset == U64_MAX)
 682                         break;
 683
 684                 if (!bp.level) {
 685                         const struct bch_extent_ptr *ptr;
 686                         struct bkey_s_c k;
 687                         unsigned i = 0;
 688
 689                         k = bch2_backpointer_get_key(&trans, &iter,
 690                                                 bucket, bp_offset, bp);
 691                         ret = bkey_err(k);
 692                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 693                                 continue;
 694                         if (ret)
 695                                 goto err;
 696                         if (!k.k)
 697                                 continue;
 698
 699                         bch2_bkey_buf_reassemble(&sk, c, k);
 700                         k = bkey_i_to_s_c(sk.k);
 701
 702                         ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 703                         if (ret) {
 704                                 bch2_trans_iter_exit(&trans, &iter);
 705                                 continue;
 706                         }
 707
 708                         data_opts = _data_opts;
 709                         data_opts.target        = io_opts.background_target;
 710                         data_opts.rewrite_ptrs = 0;
 711
 712                         bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
 713                                 if (ptr->dev == bucket.inode)
 714                                         data_opts.rewrite_ptrs |= 1U << i;
 715                                 i++;
 716                         }
 717
 718                         ret = bch2_move_extent(&trans, &iter, ctxt, io_opts,
 719                                                bp.btree_id, k, data_opts);
 720                         bch2_trans_iter_exit(&trans, &iter);
 721
 722                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 723                                 continue;
 724                         if (ret == -ENOMEM) {
 725                                 /* memory allocation failure, wait for some IO to finish */
 726                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 727                                 continue;
 728                         }
 729                         if (ret)
 730                                 goto err;
 731
 732                         if (ctxt->rate)
 733                                 bch2_ratelimit_increment(ctxt->rate, k.k->size);
 734                         atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 735                 } else {
 736                         struct btree *b;
 737
 738                         b = bch2_backpointer_get_node(&trans, &iter,
 739                                                 bucket, bp_offset, bp);
 740                         ret = PTR_ERR_OR_ZERO(b);
 741                         if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node)
 742                                 continue;
 743                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 744                                 continue;
 745                         if (ret)
 746                                 goto err;
 747                         if (!b)
 748                                 continue;
 749
 750                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0);
 751                         bch2_trans_iter_exit(&trans, &iter);
 752
 753                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 754                                 continue;
 755                         if (ret)
 756                                 goto err;
 757
 758                         if (ctxt->rate)
 759                                 bch2_ratelimit_increment(ctxt->rate,
 760                                                          c->opts.btree_node_size >> 9);
 761                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_seen);
 762                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_moved);
 763                 }
 764
 765                 bp_offset++;
 766         }
 767
 768         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && gen >= 0) {
 769                 bch2_trans_unlock(&trans);
 770                 move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
 771                 closure_sync(&ctxt->cl);
 772                 if (!ctxt->write_error)
 773                         lockrestart_do(&trans, verify_bucket_evacuated(&trans, bucket, gen));
 774         }
 775 err:
 776         bch2_trans_exit(&trans);
 777         bch2_bkey_buf_exit(&sk, c);
 778         return ret;
 779 }
 780
 781 int bch2_evacuate_bucket(struct bch_fs *c,
 782                          struct bpos bucket, int gen,
 783                          struct data_update_opts data_opts,
 784                          struct bch_ratelimit *rate,
 785                          struct bch_move_stats *stats,
 786                          struct write_point_specifier wp,
 787                          bool wait_on_copygc)
 788 {
 789         struct moving_context ctxt;
 790         int ret;
 791
 792         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 793         ret = __bch2_evacuate_bucket(&ctxt, bucket, gen, data_opts);
 794         bch2_moving_ctxt_exit(&ctxt);
 795
 796         return ret;
 797 }
 798
 799 typedef bool (*move_btree_pred)(struct bch_fs *, void *,
 800                                 struct btree *, struct bch_io_opts *,
 801                                 struct data_update_opts *);
 802
 803 static int bch2_move_btree(struct bch_fs *c,
 804                            enum btree_id start_btree_id, struct bpos start_pos,
 805                            enum btree_id end_btree_id,   struct bpos end_pos,
 806                            move_btree_pred pred, void *arg,
 807                            struct bch_move_stats *stats)
 808 {
 809         bool kthread = (current->flags & PF_KTHREAD) != 0;
 810         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 811         struct btree_trans trans;
 812         struct btree_iter iter;
 813         struct btree *b;
 814         enum btree_id id;
 815         struct data_update_opts data_opts;
 816         int ret = 0;
 817
 818         bch2_trans_init(&trans, c, 0, 0);
 819         progress_list_add(c, stats);
 820
 821         stats->data_type = BCH_DATA_btree;
 822
 823         for (id = start_btree_id;
 824              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 825              id++) {
 826                 stats->btree_id = id;
 827
 828                 bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
 829                                           BTREE_ITER_PREFETCH);
 830 retry:
 831                 ret = 0;
 832                 while (bch2_trans_begin(&trans),
 833                        (b = bch2_btree_iter_peek_node(&iter)) &&
 834                        !(ret = PTR_ERR_OR_ZERO(b))) {
 835                         if (kthread && kthread_should_stop())
 836                                 break;
 837
 838                         if ((cmp_int(id, end_btree_id) ?:
 839                              bpos_cmp(b->key.k.p, end_pos)) > 0)
 840                                 break;
 841
 842                         stats->pos = iter.pos;
 843
 844                         if (!pred(c, arg, b, &io_opts, &data_opts))
 845                                 goto next;
 846
 847                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
 848                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 849                                 continue;
 850                         if (ret)
 851                                 break;
 852 next:
 853                         bch2_btree_iter_next_node(&iter);
 854                 }
 855                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 856                         goto retry;
 857
 858                 bch2_trans_iter_exit(&trans, &iter);
 859
 860                 if (kthread && kthread_should_stop())
 861                         break;
 862         }
 863
 864         bch2_trans_exit(&trans);
 865
 866         if (ret)
 867                 bch_err(c, "error in %s(): %s", __func__, bch2_err_str(ret));
 868
 869         bch2_btree_interior_updates_flush(c);
 870
 871         progress_list_del(c, stats);
 872         return ret;
 873 }
 874
 875 static bool rereplicate_pred(struct bch_fs *c, void *arg,
 876                              struct bkey_s_c k,
 877                              struct bch_io_opts *io_opts,
 878                              struct data_update_opts *data_opts)
 879 {
 880         unsigned nr_good = bch2_bkey_durability(c, k);
 881         unsigned replicas = bkey_is_btree_ptr(k.k)
 882                 ? c->opts.metadata_replicas
 883                 : io_opts->data_replicas;
 884
 885         if (!nr_good || nr_good >= replicas)
 886                 return false;
 887
 888         data_opts->target               = 0;
 889         data_opts->extra_replicas       = replicas - nr_good;
 890         data_opts->btree_insert_flags   = 0;
 891         return true;
 892 }
 893
 894 static bool migrate_pred(struct bch_fs *c, void *arg,
 895                          struct bkey_s_c k,
 896                          struct bch_io_opts *io_opts,
 897                          struct data_update_opts *data_opts)
 898 {
 899         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 900         const struct bch_extent_ptr *ptr;
 901         struct bch_ioctl_data *op = arg;
 902         unsigned i = 0;
 903
 904         data_opts->rewrite_ptrs         = 0;
 905         data_opts->target               = 0;
 906         data_opts->extra_replicas       = 0;
 907         data_opts->btree_insert_flags   = 0;
 908
 909         bkey_for_each_ptr(ptrs, ptr) {
 910                 if (ptr->dev == op->migrate.dev)
 911                         data_opts->rewrite_ptrs |= 1U << i;
 912                 i++;
 913         }
 914
 915         return data_opts->rewrite_ptrs != 0;
 916 }
 917
 918 static bool rereplicate_btree_pred(struct bch_fs *c, void *arg,
 919                                    struct btree *b,
 920                                    struct bch_io_opts *io_opts,
 921                                    struct data_update_opts *data_opts)
 922 {
 923         return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 924 }
 925
 926 static bool migrate_btree_pred(struct bch_fs *c, void *arg,
 927                                struct btree *b,
 928                                struct bch_io_opts *io_opts,
 929                                struct data_update_opts *data_opts)
 930 {
 931         return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 932 }
 933
 934 static bool bformat_needs_redo(struct bkey_format *f)
 935 {
 936         unsigned i;
 937
 938         for (i = 0; i < f->nr_fields; i++) {
 939                 unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
 940                 u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
 941                 u64 field_offset = le64_to_cpu(f->field_offset[i]);
 942
 943                 if (f->bits_per_field[i] > unpacked_bits)
 944                         return true;
 945
 946                 if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
 947                         return true;
 948
 949                 if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
 950                      unpacked_mask) <
 951                     field_offset)
 952                         return true;
 953         }
 954
 955         return false;
 956 }
 957
 958 static bool rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
 959                                    struct btree *b,
 960                                    struct bch_io_opts *io_opts,
 961                                    struct data_update_opts *data_opts)
 962 {
 963         if (b->version_ondisk != c->sb.version ||
 964             btree_node_need_rewrite(b) ||
 965             bformat_needs_redo(&b->format)) {
 966                 data_opts->target               = 0;
 967                 data_opts->extra_replicas       = 0;
 968                 data_opts->btree_insert_flags   = 0;
 969                 return true;
 970         }
 971
 972         return false;
 973 }
 974
 975 int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
 976 {
 977         int ret;
 978
 979         ret = bch2_move_btree(c,
 980                               0,                POS_MIN,
 981                               BTREE_ID_NR,      SPOS_MAX,
 982                               rewrite_old_nodes_pred, c, stats);
 983         if (!ret) {
 984                 mutex_lock(&c->sb_lock);
 985                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 986                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 987                 c->disk_sb.sb->version_min = c->disk_sb.sb->version;
 988                 bch2_write_super(c);
 989                 mutex_unlock(&c->sb_lock);
 990         }
 991
 992         return ret;
 993 }
 994
 995 int bch2_data_job(struct bch_fs *c,
 996                   struct bch_move_stats *stats,
 997                   struct bch_ioctl_data op)
 998 {
 999         int ret = 0;
1000
1001         switch (op.op) {
1002         case BCH_DATA_OP_REREPLICATE:
1003                 bch2_move_stats_init(stats, "rereplicate");
1004                 stats->data_type = BCH_DATA_journal;
1005                 ret = bch2_journal_flush_device_pins(&c->journal, -1);
1006
1007                 ret = bch2_move_btree(c,
1008                                       op.start_btree,   op.start_pos,
1009                                       op.end_btree,     op.end_pos,
1010                                       rereplicate_btree_pred, c, stats) ?: ret;
1011                 ret = bch2_replicas_gc2(c) ?: ret;
1012
1013                 ret = bch2_move_data(c,
1014                                      op.start_btree,    op.start_pos,
1015                                      op.end_btree,      op.end_pos,
1016                                      NULL,
1017                                      stats,
1018                                      writepoint_hashed((unsigned long) current),
1019                                      true,
1020                                      rereplicate_pred, c) ?: ret;
1021                 ret = bch2_replicas_gc2(c) ?: ret;
1022                 break;
1023         case BCH_DATA_OP_MIGRATE:
1024                 if (op.migrate.dev >= c->sb.nr_devices)
1025                         return -EINVAL;
1026
1027                 bch2_move_stats_init(stats, "migrate");
1028                 stats->data_type = BCH_DATA_journal;
1029                 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
1030
1031                 ret = bch2_move_btree(c,
1032                                       op.start_btree,   op.start_pos,
1033                                       op.end_btree,     op.end_pos,
1034                                       migrate_btree_pred, &op, stats) ?: ret;
1035                 ret = bch2_replicas_gc2(c) ?: ret;
1036
1037                 ret = bch2_move_data(c,
1038                                      op.start_btree,    op.start_pos,
1039                                      op.end_btree,      op.end_pos,
1040                                      NULL,
1041                                      stats,
1042                                      writepoint_hashed((unsigned long) current),
1043                                      true,
1044                                      migrate_pred, &op) ?: ret;
1045                 ret = bch2_replicas_gc2(c) ?: ret;
1046                 break;
1047         case BCH_DATA_OP_REWRITE_OLD_NODES:
1048                 bch2_move_stats_init(stats, "rewrite_old_nodes");
1049                 ret = bch2_scan_old_btree_nodes(c, stats);
1050                 break;
1051         default:
1052                 ret = -EINVAL;
1053         }
1054
1055         return ret;
1056 }