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 "error.h"
  13 #include "inode.h"
  14 #include "io.h"
  15 #include "journal_reclaim.h"
  16 #include "move.h"
  17 #include "replicas.h"
  18 #include "super-io.h"
  19 #include "keylist.h"
  20
  21 #include <linux/ioprio.h>
  22 #include <linux/kthread.h>
  23
  24 #include <trace/events/bcachefs.h>
  25
  26 static void progress_list_add(struct bch_fs *c, struct bch_move_stats *stats)
  27 {
  28         mutex_lock(&c->data_progress_lock);
  29         list_add(&stats->list, &c->data_progress_list);
  30         mutex_unlock(&c->data_progress_lock);
  31 }
  32
  33 static void progress_list_del(struct bch_fs *c, struct bch_move_stats *stats)
  34 {
  35         mutex_lock(&c->data_progress_lock);
  36         list_del(&stats->list);
  37         mutex_unlock(&c->data_progress_lock);
  38 }
  39
  40 struct moving_io {
  41         struct list_head        list;
  42         struct closure          cl;
  43         bool                    read_completed;
  44
  45         unsigned                read_sectors;
  46         unsigned                write_sectors;
  47
  48         struct bch_read_bio     rbio;
  49
  50         struct data_update      write;
  51         /* Must be last since it is variable size */
  52         struct bio_vec          bi_inline_vecs[0];
  53 };
  54
  55 static void move_free(struct closure *cl)
  56 {
  57         struct moving_io *io = container_of(cl, struct moving_io, cl);
  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 closure *cl)
  68 {
  69         struct moving_io *io = container_of(cl, struct moving_io, cl);
  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         closure_return_with_destructor(cl, move_free);
  77 }
  78
  79 static void move_write(struct closure *cl)
  80 {
  81         struct moving_io *io = container_of(cl, struct moving_io, cl);
  82
  83         if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
  84                 closure_return_with_destructor(cl, move_free);
  85                 return;
  86         }
  87
  88         atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
  89
  90         bch2_data_update_read_done(&io->write, io->rbio.pick.crc, cl);
  91         continue_at(cl, move_write_done, NULL);
  92 }
  93
  94 static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
  95 {
  96         struct moving_io *io =
  97                 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
  98
  99         return io && io->read_completed ? io : NULL;
 100 }
 101
 102 static void move_read_endio(struct bio *bio)
 103 {
 104         struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
 105         struct moving_context *ctxt = io->write.ctxt;
 106
 107         atomic_sub(io->read_sectors, &ctxt->read_sectors);
 108         io->read_completed = true;
 109
 110         wake_up(&ctxt->wait);
 111         closure_put(&ctxt->cl);
 112 }
 113
 114 static void do_pending_writes(struct moving_context *ctxt, struct btree_trans *trans)
 115 {
 116         struct moving_io *io;
 117
 118         if (trans)
 119                 bch2_trans_unlock(trans);
 120
 121         while ((io = next_pending_write(ctxt))) {
 122                 list_del(&io->list);
 123                 closure_call(&io->cl, move_write, NULL, &ctxt->cl);
 124         }
 125 }
 126
 127 #define move_ctxt_wait_event(_ctxt, _trans, _cond)              \
 128 do {                                                            \
 129         do_pending_writes(_ctxt, _trans);                       \
 130                                                                 \
 131         if (_cond)                                              \
 132                 break;                                          \
 133         __wait_event((_ctxt)->wait,                             \
 134                      next_pending_write(_ctxt) || (_cond));     \
 135 } while (1)
 136
 137 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt,
 138                                        struct btree_trans *trans)
 139 {
 140         unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
 141
 142         move_ctxt_wait_event(ctxt, trans,
 143                 !atomic_read(&ctxt->write_sectors) ||
 144                 atomic_read(&ctxt->write_sectors) != sectors_pending);
 145 }
 146
 147 void bch2_moving_ctxt_exit(struct moving_context *ctxt)
 148 {
 149         move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
 150         closure_sync(&ctxt->cl);
 151         EBUG_ON(atomic_read(&ctxt->write_sectors));
 152
 153         if (ctxt->stats) {
 154                 progress_list_del(ctxt->c, ctxt->stats);
 155
 156                 trace_move_data(ctxt->c,
 157                                 atomic64_read(&ctxt->stats->sectors_moved),
 158                                 atomic64_read(&ctxt->stats->keys_moved));
 159         }
 160 }
 161
 162 void bch2_moving_ctxt_init(struct moving_context *ctxt,
 163                            struct bch_fs *c,
 164                            struct bch_ratelimit *rate,
 165                            struct bch_move_stats *stats,
 166                            struct write_point_specifier wp,
 167                            bool wait_on_copygc)
 168 {
 169         memset(ctxt, 0, sizeof(*ctxt));
 170
 171         ctxt->c         = c;
 172         ctxt->rate      = rate;
 173         ctxt->stats     = stats;
 174         ctxt->wp        = wp;
 175         ctxt->wait_on_copygc = wait_on_copygc;
 176
 177         closure_init_stack(&ctxt->cl);
 178         INIT_LIST_HEAD(&ctxt->reads);
 179         init_waitqueue_head(&ctxt->wait);
 180
 181         if (stats) {
 182                 progress_list_add(c, stats);
 183                 stats->data_type = BCH_DATA_user;
 184         }
 185 }
 186
 187 void bch_move_stats_init(struct bch_move_stats *stats, char *name)
 188 {
 189         memset(stats, 0, sizeof(*stats));
 190         scnprintf(stats->name, sizeof(stats->name), "%s", name);
 191 }
 192
 193 static int bch2_move_extent(struct btree_trans *trans,
 194                             struct moving_context *ctxt,
 195                             struct bch_io_opts io_opts,
 196                             enum btree_id btree_id,
 197                             struct bkey_s_c k,
 198                             struct data_update_opts data_opts)
 199 {
 200         struct bch_fs *c = trans->c;
 201         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 202         struct moving_io *io;
 203         const union bch_extent_entry *entry;
 204         struct extent_ptr_decoded p;
 205         unsigned sectors = k.k->size, pages;
 206         int ret = -ENOMEM;
 207
 208         if (!percpu_ref_tryget_live(&c->writes))
 209                 return -EROFS;
 210
 211         /* write path might have to decompress data: */
 212         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 213                 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
 214
 215         pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
 216         io = kzalloc(sizeof(struct moving_io) +
 217                      sizeof(struct bio_vec) * pages, GFP_KERNEL);
 218         if (!io)
 219                 goto err;
 220
 221         io->write.ctxt          = ctxt;
 222         io->read_sectors        = k.k->size;
 223         io->write_sectors       = k.k->size;
 224
 225         bio_init(&io->write.op.wbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 226         bio_set_prio(&io->write.op.wbio.bio,
 227                      IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 228
 229         if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
 230                                  GFP_KERNEL))
 231                 goto err_free;
 232
 233         io->rbio.c              = c;
 234         io->rbio.opts           = io_opts;
 235         bio_init(&io->rbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 236         io->rbio.bio.bi_vcnt = pages;
 237         bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 238         io->rbio.bio.bi_iter.bi_size = sectors << 9;
 239
 240         bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
 241         io->rbio.bio.bi_iter.bi_sector  = bkey_start_offset(k.k);
 242         io->rbio.bio.bi_end_io          = move_read_endio;
 243
 244         ret = bch2_data_update_init(c, &io->write, ctxt->wp, io_opts,
 245                                     data_opts, btree_id, k);
 246         if (ret)
 247                 goto err_free_pages;
 248
 249         io->write.ctxt = ctxt;
 250
 251         atomic64_inc(&ctxt->stats->keys_moved);
 252         atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
 253         this_cpu_add(c->counters[BCH_COUNTER_io_move], k.k->size);
 254
 255         trace_move_extent(k.k);
 256
 257         atomic_add(io->read_sectors, &ctxt->read_sectors);
 258         list_add_tail(&io->list, &ctxt->reads);
 259
 260         /*
 261          * dropped by move_read_endio() - guards against use after free of
 262          * ctxt when doing wakeup
 263          */
 264         closure_get(&ctxt->cl);
 265         bch2_read_extent(trans, &io->rbio,
 266                          bkey_start_pos(k.k),
 267                          btree_id, k, 0,
 268                          BCH_READ_NODECODE|
 269                          BCH_READ_LAST_FRAGMENT);
 270         return 0;
 271 err_free_pages:
 272         bio_free_pages(&io->write.op.wbio.bio);
 273 err_free:
 274         kfree(io);
 275 err:
 276         percpu_ref_put(&c->writes);
 277         trace_move_alloc_mem_fail(k.k);
 278         return ret;
 279 }
 280
 281 static int lookup_inode(struct btree_trans *trans, struct bpos pos,
 282                         struct bch_inode_unpacked *inode)
 283 {
 284         struct btree_iter iter;
 285         struct bkey_s_c k;
 286         int ret;
 287
 288         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
 289                              BTREE_ITER_ALL_SNAPSHOTS);
 290         k = bch2_btree_iter_peek(&iter);
 291         ret = bkey_err(k);
 292         if (ret)
 293                 goto err;
 294
 295         if (!k.k || bkey_cmp(k.k->p, pos)) {
 296                 ret = -ENOENT;
 297                 goto err;
 298         }
 299
 300         ret = bkey_is_inode(k.k) ? 0 : -EIO;
 301         if (ret)
 302                 goto err;
 303
 304         ret = bch2_inode_unpack(k, inode);
 305         if (ret)
 306                 goto err;
 307 err:
 308         bch2_trans_iter_exit(trans, &iter);
 309         return ret;
 310 }
 311
 312 static int move_ratelimit(struct btree_trans *trans,
 313                           struct moving_context *ctxt)
 314 {
 315         struct bch_fs *c = trans->c;
 316         u64 delay;
 317
 318         if (ctxt->wait_on_copygc) {
 319                 bch2_trans_unlock(trans);
 320                 wait_event_killable(c->copygc_running_wq,
 321                                     !c->copygc_running ||
 322                                     kthread_should_stop());
 323         }
 324
 325         do {
 326                 delay = ctxt->rate ? bch2_ratelimit_delay(ctxt->rate) : 0;
 327
 328                 if (delay) {
 329                         bch2_trans_unlock(trans);
 330                         set_current_state(TASK_INTERRUPTIBLE);
 331                 }
 332
 333                 if ((current->flags & PF_KTHREAD) && kthread_should_stop()) {
 334                         __set_current_state(TASK_RUNNING);
 335                         return 1;
 336                 }
 337
 338                 if (delay)
 339                         schedule_timeout(delay);
 340
 341                 if (unlikely(freezing(current))) {
 342                         move_ctxt_wait_event(ctxt, trans, list_empty(&ctxt->reads));
 343                         try_to_freeze();
 344                 }
 345         } while (delay);
 346
 347         move_ctxt_wait_event(ctxt, trans,
 348                 atomic_read(&ctxt->write_sectors) <
 349                 c->opts.move_bytes_in_flight >> 9);
 350
 351         move_ctxt_wait_event(ctxt, trans,
 352                 atomic_read(&ctxt->read_sectors) <
 353                 c->opts.move_bytes_in_flight >> 9);
 354
 355         return 0;
 356 }
 357
 358 static int move_get_io_opts(struct btree_trans *trans,
 359                             struct bch_io_opts *io_opts,
 360                             struct bkey_s_c k, u64 *cur_inum)
 361 {
 362         struct bch_inode_unpacked inode;
 363         int ret;
 364
 365         if (*cur_inum == k.k->p.inode)
 366                 return 0;
 367
 368         *io_opts = bch2_opts_to_inode_opts(trans->c->opts);
 369
 370         ret = lookup_inode(trans,
 371                            SPOS(0, k.k->p.inode, k.k->p.snapshot),
 372                            &inode);
 373         if (ret == -EINTR)
 374                 return ret;
 375
 376         if (!ret)
 377                 bch2_io_opts_apply(io_opts, bch2_inode_opts_get(&inode));
 378
 379         *cur_inum = k.k->p.inode;
 380         return 0;
 381 }
 382
 383 static int __bch2_move_data(struct moving_context *ctxt,
 384                             struct bpos start,
 385                             struct bpos end,
 386                             move_pred_fn pred, void *arg,
 387                             enum btree_id btree_id)
 388 {
 389         struct bch_fs *c = ctxt->c;
 390         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 391         struct bkey_buf sk;
 392         struct btree_trans trans;
 393         struct btree_iter iter;
 394         struct bkey_s_c k;
 395         struct data_update_opts data_opts;
 396         u64 cur_inum = U64_MAX;
 397         int ret = 0, ret2;
 398
 399         bch2_bkey_buf_init(&sk);
 400         bch2_trans_init(&trans, c, 0, 0);
 401
 402         ctxt->stats->data_type  = BCH_DATA_user;
 403         ctxt->stats->btree_id   = btree_id;
 404         ctxt->stats->pos        = start;
 405
 406         bch2_trans_iter_init(&trans, &iter, btree_id, start,
 407                              BTREE_ITER_PREFETCH|
 408                              BTREE_ITER_ALL_SNAPSHOTS);
 409
 410         if (ctxt->rate)
 411                 bch2_ratelimit_reset(ctxt->rate);
 412
 413         while (!move_ratelimit(&trans, ctxt)) {
 414                 bch2_trans_begin(&trans);
 415
 416                 k = bch2_btree_iter_peek(&iter);
 417                 if (!k.k)
 418                         break;
 419
 420                 ret = bkey_err(k);
 421                 if (ret == -EINTR)
 422                         continue;
 423                 if (ret)
 424                         break;
 425
 426                 if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
 427                         break;
 428
 429                 ctxt->stats->pos = iter.pos;
 430
 431                 if (!bkey_extent_is_direct_data(k.k))
 432                         goto next_nondata;
 433
 434                 ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 435                 if (ret)
 436                         continue;
 437
 438                 memset(&data_opts, 0, sizeof(data_opts));
 439                 if (!pred(c, arg, k, &io_opts, &data_opts))
 440                         goto next;
 441
 442                 /*
 443                  * The iterator gets unlocked by __bch2_read_extent - need to
 444                  * save a copy of @k elsewhere:
 445                   */
 446                 bch2_bkey_buf_reassemble(&sk, c, k);
 447                 k = bkey_i_to_s_c(sk.k);
 448
 449                 ret2 = bch2_move_extent(&trans, ctxt, io_opts,
 450                                         btree_id, k, data_opts);
 451                 if (ret2) {
 452                         if (ret2 == -EINTR)
 453                                 continue;
 454
 455                         if (ret2 == -ENOMEM) {
 456                                 /* memory allocation failure, wait for some IO to finish */
 457                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 458                                 continue;
 459                         }
 460
 461                         /* XXX signal failure */
 462                         goto next;
 463                 }
 464
 465                 if (ctxt->rate)
 466                         bch2_ratelimit_increment(ctxt->rate, k.k->size);
 467 next:
 468                 atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 469 next_nondata:
 470                 bch2_btree_iter_advance(&iter);
 471         }
 472
 473         bch2_trans_iter_exit(&trans, &iter);
 474         bch2_trans_exit(&trans);
 475         bch2_bkey_buf_exit(&sk, c);
 476
 477         return ret;
 478 }
 479
 480 int bch2_move_data(struct bch_fs *c,
 481                    enum btree_id start_btree_id, struct bpos start_pos,
 482                    enum btree_id end_btree_id,   struct bpos end_pos,
 483                    struct bch_ratelimit *rate,
 484                    struct bch_move_stats *stats,
 485                    struct write_point_specifier wp,
 486                    bool wait_on_copygc,
 487                    move_pred_fn pred, void *arg)
 488 {
 489         struct moving_context ctxt;
 490         enum btree_id id;
 491         int ret;
 492
 493         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 494
 495         for (id = start_btree_id;
 496              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 497              id++) {
 498                 stats->btree_id = id;
 499
 500                 if (id != BTREE_ID_extents &&
 501                     id != BTREE_ID_reflink)
 502                         continue;
 503
 504                 ret = __bch2_move_data(&ctxt,
 505                                        id == start_btree_id ? start_pos : POS_MIN,
 506                                        id == end_btree_id   ? end_pos   : POS_MAX,
 507                                        pred, arg, id);
 508                 if (ret)
 509                         break;
 510         }
 511
 512         bch2_moving_ctxt_exit(&ctxt);
 513
 514         return ret;
 515 }
 516
 517 static int verify_bucket_evacuated(struct btree_trans *trans, struct bpos bucket, int gen)
 518 {
 519         struct bch_fs *c = trans->c;
 520         struct btree_iter iter;
 521         struct bkey_s_c k;
 522         int ret;
 523
 524         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
 525                              bucket, BTREE_ITER_CACHED);
 526 again:
 527         k = bch2_btree_iter_peek_slot(&iter);
 528         ret = bkey_err(k);
 529
 530         if (!ret && k.k->type == KEY_TYPE_alloc_v4) {
 531                 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
 532
 533                 if (a.v->gen == gen &&
 534                     a.v->dirty_sectors) {
 535                         struct printbuf buf = PRINTBUF;
 536
 537                         if (a.v->data_type == BCH_DATA_btree) {
 538                                 bch2_trans_unlock(trans);
 539                                 if (bch2_btree_interior_updates_flush(c))
 540                                         goto again;
 541                         }
 542
 543                         prt_str(&buf, "failed to evacuate bucket ");
 544                         bch2_bkey_val_to_text(&buf, c, k);
 545
 546                         bch2_trans_inconsistent(trans, "%s", buf.buf);
 547                         printbuf_exit(&buf);
 548                 }
 549         }
 550
 551         bch2_trans_iter_exit(trans, &iter);
 552         return ret;
 553 }
 554
 555 int __bch2_evacuate_bucket(struct moving_context *ctxt,
 556                            struct bpos bucket, int gen,
 557                            struct data_update_opts _data_opts)
 558 {
 559         struct bch_fs *c = ctxt->c;
 560         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 561         struct btree_trans trans;
 562         struct btree_iter iter;
 563         struct bkey_buf sk;
 564         struct bch_backpointer bp;
 565         struct data_update_opts data_opts;
 566         u64 bp_offset = 0, cur_inum = U64_MAX;
 567         int ret = 0;
 568
 569         bch2_bkey_buf_init(&sk);
 570         bch2_trans_init(&trans, c, 0, 0);
 571
 572         while (!(ret = move_ratelimit(&trans, ctxt))) {
 573                 bch2_trans_begin(&trans);
 574
 575                 ret = bch2_get_next_backpointer(&trans, bucket, gen,
 576                                                 &bp_offset, &bp);
 577                 if (ret == -EINTR)
 578                         continue;
 579                 if (ret)
 580                         goto err;
 581                 if (bp_offset == U64_MAX)
 582                         break;
 583
 584                 if (!bp.level) {
 585                         const struct bch_extent_ptr *ptr;
 586                         struct bkey_s_c k;
 587                         unsigned i = 0;
 588
 589                         k = bch2_backpointer_get_key(&trans, &iter,
 590                                                 bucket, bp_offset, bp);
 591                         ret = bkey_err(k);
 592                         if (ret == -EINTR)
 593                                 continue;
 594                         if (ret)
 595                                 goto err;
 596                         if (!k.k)
 597                                 continue;
 598
 599                         bch2_bkey_buf_reassemble(&sk, c, k);
 600                         k = bkey_i_to_s_c(sk.k);
 601                         bch2_trans_iter_exit(&trans, &iter);
 602
 603                         ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 604                         if (ret)
 605                                 continue;
 606
 607                         data_opts = _data_opts;
 608                         data_opts.target        = io_opts.background_target;
 609                         data_opts.rewrite_ptrs = 0;
 610
 611                         bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
 612                                 if (ptr->dev == bucket.inode)
 613                                         data_opts.rewrite_ptrs |= 1U << i;
 614                                 i++;
 615                         }
 616
 617                         ret = bch2_move_extent(&trans, ctxt, io_opts,
 618                                                bp.btree_id, k, data_opts);
 619                         if (ret == -EINTR)
 620                                 continue;
 621                         if (ret == -ENOMEM) {
 622                                 /* memory allocation failure, wait for some IO to finish */
 623                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 624                                 continue;
 625                         }
 626                         if (ret)
 627                                 goto err;
 628
 629                         if (ctxt->rate)
 630                                 bch2_ratelimit_increment(ctxt->rate, k.k->size);
 631                         atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 632                 } else {
 633                         struct btree *b;
 634
 635                         b = bch2_backpointer_get_node(&trans, &iter,
 636                                                 bucket, bp_offset, bp);
 637                         ret = PTR_ERR_OR_ZERO(b);
 638                         if (ret == -EINTR)
 639                                 continue;
 640                         if (ret)
 641                                 goto err;
 642                         if (!b)
 643                                 continue;
 644
 645                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0);
 646                         bch2_trans_iter_exit(&trans, &iter);
 647
 648                         if (ret == -EINTR)
 649                                 continue;
 650                         if (ret)
 651                                 goto err;
 652
 653                         if (ctxt->rate)
 654                                 bch2_ratelimit_increment(ctxt->rate,
 655                                                          c->opts.btree_node_size >> 9);
 656                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_seen);
 657                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_moved);
 658                 }
 659
 660                 bp_offset++;
 661         }
 662
 663         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && gen >= 0) {
 664                 bch2_trans_unlock(&trans);
 665                 move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
 666                 closure_sync(&ctxt->cl);
 667                 if (!ctxt->write_error)
 668                         lockrestart_do(&trans, verify_bucket_evacuated(&trans, bucket, gen));
 669         }
 670 err:
 671         bch2_trans_exit(&trans);
 672         bch2_bkey_buf_exit(&sk, c);
 673         return ret;
 674 }
 675
 676 int bch2_evacuate_bucket(struct bch_fs *c,
 677                          struct bpos bucket, int gen,
 678                          struct data_update_opts data_opts,
 679                          struct bch_ratelimit *rate,
 680                          struct bch_move_stats *stats,
 681                          struct write_point_specifier wp,
 682                          bool wait_on_copygc)
 683 {
 684         struct moving_context ctxt;
 685         int ret;
 686
 687         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 688         ret = __bch2_evacuate_bucket(&ctxt, bucket, gen, data_opts);
 689         bch2_moving_ctxt_exit(&ctxt);
 690
 691         return ret;
 692 }
 693
 694 typedef bool (*move_btree_pred)(struct bch_fs *, void *,
 695                                 struct btree *, struct bch_io_opts *,
 696                                 struct data_update_opts *);
 697
 698 static int bch2_move_btree(struct bch_fs *c,
 699                            enum btree_id start_btree_id, struct bpos start_pos,
 700                            enum btree_id end_btree_id,   struct bpos end_pos,
 701                            move_btree_pred pred, void *arg,
 702                            struct bch_move_stats *stats)
 703 {
 704         bool kthread = (current->flags & PF_KTHREAD) != 0;
 705         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 706         struct btree_trans trans;
 707         struct btree_iter iter;
 708         struct btree *b;
 709         enum btree_id id;
 710         struct data_update_opts data_opts;
 711         int ret = 0;
 712
 713         bch2_trans_init(&trans, c, 0, 0);
 714         progress_list_add(c, stats);
 715
 716         stats->data_type = BCH_DATA_btree;
 717
 718         for (id = start_btree_id;
 719              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 720              id++) {
 721                 stats->btree_id = id;
 722
 723                 bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
 724                                           BTREE_ITER_PREFETCH);
 725 retry:
 726                 ret = 0;
 727                 while (bch2_trans_begin(&trans),
 728                        (b = bch2_btree_iter_peek_node(&iter)) &&
 729                        !(ret = PTR_ERR_OR_ZERO(b))) {
 730                         if (kthread && kthread_should_stop())
 731                                 break;
 732
 733                         if ((cmp_int(id, end_btree_id) ?:
 734                              bpos_cmp(b->key.k.p, end_pos)) > 0)
 735                                 break;
 736
 737                         stats->pos = iter.pos;
 738
 739                         if (!pred(c, arg, b, &io_opts, &data_opts))
 740                                 goto next;
 741
 742                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
 743                         if (ret == -EINTR)
 744                                 continue;
 745                         if (ret)
 746                                 break;
 747 next:
 748                         bch2_btree_iter_next_node(&iter);
 749                 }
 750                 if (ret == -EINTR)
 751                         goto retry;
 752
 753                 bch2_trans_iter_exit(&trans, &iter);
 754
 755                 if (kthread && kthread_should_stop())
 756                         break;
 757         }
 758
 759         bch2_trans_exit(&trans);
 760
 761         if (ret)
 762                 bch_err(c, "error %i in bch2_move_btree", ret);
 763
 764         bch2_btree_interior_updates_flush(c);
 765
 766         progress_list_del(c, stats);
 767         return ret;
 768 }
 769
 770 static bool rereplicate_pred(struct bch_fs *c, void *arg,
 771                              struct bkey_s_c k,
 772                              struct bch_io_opts *io_opts,
 773                              struct data_update_opts *data_opts)
 774 {
 775         unsigned nr_good = bch2_bkey_durability(c, k);
 776         unsigned replicas = bkey_is_btree_ptr(k.k)
 777                 ? c->opts.metadata_replicas
 778                 : io_opts->data_replicas;
 779
 780         if (!nr_good || nr_good >= replicas)
 781                 return false;
 782
 783         data_opts->target               = 0;
 784         data_opts->extra_replicas       = replicas - nr_good;
 785         data_opts->btree_insert_flags   = 0;
 786         return true;
 787 }
 788
 789 static bool migrate_pred(struct bch_fs *c, void *arg,
 790                          struct bkey_s_c k,
 791                          struct bch_io_opts *io_opts,
 792                          struct data_update_opts *data_opts)
 793 {
 794         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 795         const struct bch_extent_ptr *ptr;
 796         struct bch_ioctl_data *op = arg;
 797         unsigned i = 0;
 798
 799         data_opts->rewrite_ptrs         = 0;
 800         data_opts->target               = 0;
 801         data_opts->extra_replicas       = 0;
 802         data_opts->btree_insert_flags   = 0;
 803
 804         bkey_for_each_ptr(ptrs, ptr) {
 805                 if (ptr->dev == op->migrate.dev)
 806                         data_opts->rewrite_ptrs |= 1U << i;
 807                 i++;
 808         }
 809
 810         return data_opts->rewrite_ptrs != 0;;
 811 }
 812
 813 static bool rereplicate_btree_pred(struct bch_fs *c, void *arg,
 814                                    struct btree *b,
 815                                    struct bch_io_opts *io_opts,
 816                                    struct data_update_opts *data_opts)
 817 {
 818         return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 819 }
 820
 821 static bool migrate_btree_pred(struct bch_fs *c, void *arg,
 822                                struct btree *b,
 823                                struct bch_io_opts *io_opts,
 824                                struct data_update_opts *data_opts)
 825 {
 826         return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 827 }
 828
 829 static bool bformat_needs_redo(struct bkey_format *f)
 830 {
 831         unsigned i;
 832
 833         for (i = 0; i < f->nr_fields; i++) {
 834                 unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
 835                 u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
 836                 u64 field_offset = le64_to_cpu(f->field_offset[i]);
 837
 838                 if (f->bits_per_field[i] > unpacked_bits)
 839                         return true;
 840
 841                 if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
 842                         return true;
 843
 844                 if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
 845                      unpacked_mask) <
 846                     field_offset)
 847                         return true;
 848         }
 849
 850         return false;
 851 }
 852
 853 static bool rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
 854                                    struct btree *b,
 855                                    struct bch_io_opts *io_opts,
 856                                    struct data_update_opts *data_opts)
 857 {
 858         if (b->version_ondisk != c->sb.version ||
 859             btree_node_need_rewrite(b) ||
 860             bformat_needs_redo(&b->format)) {
 861                 data_opts->target               = 0;
 862                 data_opts->extra_replicas       = 0;
 863                 data_opts->btree_insert_flags   = 0;
 864                 return true;
 865         }
 866
 867         return false;
 868 }
 869
 870 int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
 871 {
 872         int ret;
 873
 874         ret = bch2_move_btree(c,
 875                               0,                POS_MIN,
 876                               BTREE_ID_NR,      SPOS_MAX,
 877                               rewrite_old_nodes_pred, c, stats);
 878         if (!ret) {
 879                 mutex_lock(&c->sb_lock);
 880                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 881                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 882                 c->disk_sb.sb->version_min = c->disk_sb.sb->version;
 883                 bch2_write_super(c);
 884                 mutex_unlock(&c->sb_lock);
 885         }
 886
 887         return ret;
 888 }
 889
 890 int bch2_data_job(struct bch_fs *c,
 891                   struct bch_move_stats *stats,
 892                   struct bch_ioctl_data op)
 893 {
 894         int ret = 0;
 895
 896         switch (op.op) {
 897         case BCH_DATA_OP_REREPLICATE:
 898                 bch_move_stats_init(stats, "rereplicate");
 899                 stats->data_type = BCH_DATA_journal;
 900                 ret = bch2_journal_flush_device_pins(&c->journal, -1);
 901
 902                 ret = bch2_move_btree(c,
 903                                       op.start_btree,   op.start_pos,
 904                                       op.end_btree,     op.end_pos,
 905                                       rereplicate_btree_pred, c, stats) ?: ret;
 906                 ret = bch2_replicas_gc2(c) ?: ret;
 907
 908                 ret = bch2_move_data(c,
 909                                      op.start_btree,    op.start_pos,
 910                                      op.end_btree,      op.end_pos,
 911                                      NULL,
 912                                      stats,
 913                                      writepoint_hashed((unsigned long) current),
 914                                      true,
 915                                      rereplicate_pred, c) ?: ret;
 916                 ret = bch2_replicas_gc2(c) ?: ret;
 917                 break;
 918         case BCH_DATA_OP_MIGRATE:
 919                 if (op.migrate.dev >= c->sb.nr_devices)
 920                         return -EINVAL;
 921
 922                 bch_move_stats_init(stats, "migrate");
 923                 stats->data_type = BCH_DATA_journal;
 924                 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
 925
 926                 ret = bch2_move_btree(c,
 927                                       op.start_btree,   op.start_pos,
 928                                       op.end_btree,     op.end_pos,
 929                                       migrate_btree_pred, &op, stats) ?: ret;
 930                 ret = bch2_replicas_gc2(c) ?: ret;
 931
 932                 ret = bch2_move_data(c,
 933                                      op.start_btree,    op.start_pos,
 934                                      op.end_btree,      op.end_pos,
 935                                      NULL,
 936                                      stats,
 937                                      writepoint_hashed((unsigned long) current),
 938                                      true,
 939                                      migrate_pred, &op) ?: ret;
 940                 ret = bch2_replicas_gc2(c) ?: ret;
 941                 break;
 942         case BCH_DATA_OP_REWRITE_OLD_NODES:
 943                 bch_move_stats_init(stats, "rewrite_old_nodes");
 944                 ret = bch2_scan_old_btree_nodes(c, stats);
 945                 break;
 946         default:
 947                 ret = -EINVAL;
 948         }
 949
 950         return ret;
 951 }