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 closure *cl)
  57 {
  58         struct moving_io *io = container_of(cl, struct moving_io, cl);
  59         struct moving_context *ctxt = io->write.ctxt;
  60         struct bch_fs *c = ctxt->c;
  61
  62         bch2_data_update_exit(&io->write);
  63         wake_up(&ctxt->wait);
  64         percpu_ref_put(&c->writes);
  65         kfree(io);
  66 }
  67
  68 static void move_write_done(struct closure *cl)
  69 {
  70         struct moving_io *io = container_of(cl, struct moving_io, cl);
  71         struct moving_context *ctxt = io->write.ctxt;
  72
  73         if (io->write.op.error)
  74                 ctxt->write_error = true;
  75
  76         atomic_sub(io->write_sectors, &io->write.ctxt->write_sectors);
  77         closure_return_with_destructor(cl, move_free);
  78 }
  79
  80 static void move_write(struct closure *cl)
  81 {
  82         struct moving_io *io = container_of(cl, struct moving_io, cl);
  83
  84         if (unlikely(io->rbio.bio.bi_status || io->rbio.hole)) {
  85                 closure_return_with_destructor(cl, move_free);
  86                 return;
  87         }
  88
  89         atomic_add(io->write_sectors, &io->write.ctxt->write_sectors);
  90
  91         bch2_data_update_read_done(&io->write, io->rbio.pick.crc, cl);
  92         continue_at(cl, move_write_done, NULL);
  93 }
  94
  95 static inline struct moving_io *next_pending_write(struct moving_context *ctxt)
  96 {
  97         struct moving_io *io =
  98                 list_first_entry_or_null(&ctxt->reads, struct moving_io, list);
  99
 100         return io && io->read_completed ? io : NULL;
 101 }
 102
 103 static void move_read_endio(struct bio *bio)
 104 {
 105         struct moving_io *io = container_of(bio, struct moving_io, rbio.bio);
 106         struct moving_context *ctxt = io->write.ctxt;
 107
 108         atomic_sub(io->read_sectors, &ctxt->read_sectors);
 109         io->read_completed = true;
 110
 111         wake_up(&ctxt->wait);
 112         closure_put(&ctxt->cl);
 113 }
 114
 115 static void do_pending_writes(struct moving_context *ctxt, struct btree_trans *trans)
 116 {
 117         struct moving_io *io;
 118
 119         if (trans)
 120                 bch2_trans_unlock(trans);
 121
 122         while ((io = next_pending_write(ctxt))) {
 123                 list_del(&io->list);
 124                 closure_call(&io->cl, move_write, NULL, &ctxt->cl);
 125         }
 126 }
 127
 128 #define move_ctxt_wait_event(_ctxt, _trans, _cond)              \
 129 do {                                                            \
 130         do_pending_writes(_ctxt, _trans);                       \
 131                                                                 \
 132         if (_cond)                                              \
 133                 break;                                          \
 134         __wait_event((_ctxt)->wait,                             \
 135                      next_pending_write(_ctxt) || (_cond));     \
 136 } while (1)
 137
 138 static void bch2_move_ctxt_wait_for_io(struct moving_context *ctxt,
 139                                        struct btree_trans *trans)
 140 {
 141         unsigned sectors_pending = atomic_read(&ctxt->write_sectors);
 142
 143         move_ctxt_wait_event(ctxt, trans,
 144                 !atomic_read(&ctxt->write_sectors) ||
 145                 atomic_read(&ctxt->write_sectors) != sectors_pending);
 146 }
 147
 148 void bch2_moving_ctxt_exit(struct moving_context *ctxt)
 149 {
 150         move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
 151         closure_sync(&ctxt->cl);
 152         EBUG_ON(atomic_read(&ctxt->write_sectors));
 153
 154         if (ctxt->stats) {
 155                 progress_list_del(ctxt->c, ctxt->stats);
 156
 157                 trace_move_data(ctxt->c,
 158                                 atomic64_read(&ctxt->stats->sectors_moved),
 159                                 atomic64_read(&ctxt->stats->keys_moved));
 160         }
 161 }
 162
 163 void bch2_moving_ctxt_init(struct moving_context *ctxt,
 164                            struct bch_fs *c,
 165                            struct bch_ratelimit *rate,
 166                            struct bch_move_stats *stats,
 167                            struct write_point_specifier wp,
 168                            bool wait_on_copygc)
 169 {
 170         memset(ctxt, 0, sizeof(*ctxt));
 171
 172         ctxt->c         = c;
 173         ctxt->rate      = rate;
 174         ctxt->stats     = stats;
 175         ctxt->wp        = wp;
 176         ctxt->wait_on_copygc = wait_on_copygc;
 177
 178         closure_init_stack(&ctxt->cl);
 179         INIT_LIST_HEAD(&ctxt->reads);
 180         init_waitqueue_head(&ctxt->wait);
 181
 182         if (stats) {
 183                 progress_list_add(c, stats);
 184                 stats->data_type = BCH_DATA_user;
 185         }
 186 }
 187
 188 void bch_move_stats_init(struct bch_move_stats *stats, char *name)
 189 {
 190         memset(stats, 0, sizeof(*stats));
 191         scnprintf(stats->name, sizeof(stats->name), "%s", name);
 192 }
 193
 194 static int bch2_move_extent(struct btree_trans *trans,
 195                             struct moving_context *ctxt,
 196                             struct bch_io_opts io_opts,
 197                             enum btree_id btree_id,
 198                             struct bkey_s_c k,
 199                             struct data_update_opts data_opts)
 200 {
 201         struct bch_fs *c = trans->c;
 202         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 203         struct moving_io *io;
 204         const union bch_extent_entry *entry;
 205         struct extent_ptr_decoded p;
 206         unsigned sectors = k.k->size, pages;
 207         int ret = -ENOMEM;
 208
 209         if (!percpu_ref_tryget_live(&c->writes))
 210                 return -EROFS;
 211
 212         /* write path might have to decompress data: */
 213         bkey_for_each_ptr_decode(k.k, ptrs, p, entry)
 214                 sectors = max_t(unsigned, sectors, p.crc.uncompressed_size);
 215
 216         pages = DIV_ROUND_UP(sectors, PAGE_SECTORS);
 217         io = kzalloc(sizeof(struct moving_io) +
 218                      sizeof(struct bio_vec) * pages, GFP_KERNEL);
 219         if (!io)
 220                 goto err;
 221
 222         io->write.ctxt          = ctxt;
 223         io->read_sectors        = k.k->size;
 224         io->write_sectors       = k.k->size;
 225
 226         bio_init(&io->write.op.wbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 227         bio_set_prio(&io->write.op.wbio.bio,
 228                      IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 229
 230         if (bch2_bio_alloc_pages(&io->write.op.wbio.bio, sectors << 9,
 231                                  GFP_KERNEL))
 232                 goto err_free;
 233
 234         io->rbio.c              = c;
 235         io->rbio.opts           = io_opts;
 236         bio_init(&io->rbio.bio, NULL, io->bi_inline_vecs, pages, 0);
 237         io->rbio.bio.bi_vcnt = pages;
 238         bio_set_prio(&io->rbio.bio, IOPRIO_PRIO_VALUE(IOPRIO_CLASS_IDLE, 0));
 239         io->rbio.bio.bi_iter.bi_size = sectors << 9;
 240
 241         bio_set_op_attrs(&io->rbio.bio, REQ_OP_READ, 0);
 242         io->rbio.bio.bi_iter.bi_sector  = bkey_start_offset(k.k);
 243         io->rbio.bio.bi_end_io          = move_read_endio;
 244
 245         ret = bch2_data_update_init(c, &io->write, ctxt->wp, io_opts,
 246                                     data_opts, btree_id, k);
 247         if (ret)
 248                 goto err_free_pages;
 249
 250         io->write.ctxt = ctxt;
 251
 252         atomic64_inc(&ctxt->stats->keys_moved);
 253         atomic64_add(k.k->size, &ctxt->stats->sectors_moved);
 254         this_cpu_add(c->counters[BCH_COUNTER_io_move], k.k->size);
 255         this_cpu_add(c->counters[BCH_COUNTER_move_extent_read], k.k->size);
 256         trace_move_extent_read(k.k);
 257
 258         atomic_add(io->read_sectors, &ctxt->read_sectors);
 259         list_add_tail(&io->list, &ctxt->reads);
 260
 261         /*
 262          * dropped by move_read_endio() - guards against use after free of
 263          * ctxt when doing wakeup
 264          */
 265         closure_get(&ctxt->cl);
 266         bch2_read_extent(trans, &io->rbio,
 267                          bkey_start_pos(k.k),
 268                          btree_id, k, 0,
 269                          BCH_READ_NODECODE|
 270                          BCH_READ_LAST_FRAGMENT);
 271         return 0;
 272 err_free_pages:
 273         bio_free_pages(&io->write.op.wbio.bio);
 274 err_free:
 275         kfree(io);
 276 err:
 277         percpu_ref_put(&c->writes);
 278         trace_and_count(c, move_extent_alloc_mem_fail, k.k);
 279         return ret;
 280 }
 281
 282 static int lookup_inode(struct btree_trans *trans, struct bpos pos,
 283                         struct bch_inode_unpacked *inode)
 284 {
 285         struct btree_iter iter;
 286         struct bkey_s_c k;
 287         int ret;
 288
 289         bch2_trans_iter_init(trans, &iter, BTREE_ID_inodes, pos,
 290                              BTREE_ITER_ALL_SNAPSHOTS);
 291         k = bch2_btree_iter_peek(&iter);
 292         ret = bkey_err(k);
 293         if (ret)
 294                 goto err;
 295
 296         if (!k.k || bkey_cmp(k.k->p, pos)) {
 297                 ret = -ENOENT;
 298                 goto err;
 299         }
 300
 301         ret = bkey_is_inode(k.k) ? 0 : -EIO;
 302         if (ret)
 303                 goto err;
 304
 305         ret = bch2_inode_unpack(k, inode);
 306         if (ret)
 307                 goto err;
 308 err:
 309         bch2_trans_iter_exit(trans, &iter);
 310         return ret;
 311 }
 312
 313 static int move_ratelimit(struct btree_trans *trans,
 314                           struct moving_context *ctxt)
 315 {
 316         struct bch_fs *c = trans->c;
 317         u64 delay;
 318
 319         if (ctxt->wait_on_copygc) {
 320                 bch2_trans_unlock(trans);
 321                 wait_event_killable(c->copygc_running_wq,
 322                                     !c->copygc_running ||
 323                                     kthread_should_stop());
 324         }
 325
 326         do {
 327                 delay = ctxt->rate ? bch2_ratelimit_delay(ctxt->rate) : 0;
 328
 329                 if (delay) {
 330                         bch2_trans_unlock(trans);
 331                         set_current_state(TASK_INTERRUPTIBLE);
 332                 }
 333
 334                 if ((current->flags & PF_KTHREAD) && kthread_should_stop()) {
 335                         __set_current_state(TASK_RUNNING);
 336                         return 1;
 337                 }
 338
 339                 if (delay)
 340                         schedule_timeout(delay);
 341
 342                 if (unlikely(freezing(current))) {
 343                         move_ctxt_wait_event(ctxt, trans, list_empty(&ctxt->reads));
 344                         try_to_freeze();
 345                 }
 346         } while (delay);
 347
 348         move_ctxt_wait_event(ctxt, trans,
 349                 atomic_read(&ctxt->write_sectors) <
 350                 c->opts.move_bytes_in_flight >> 9);
 351
 352         move_ctxt_wait_event(ctxt, trans,
 353                 atomic_read(&ctxt->read_sectors) <
 354                 c->opts.move_bytes_in_flight >> 9);
 355
 356         return 0;
 357 }
 358
 359 static int move_get_io_opts(struct btree_trans *trans,
 360                             struct bch_io_opts *io_opts,
 361                             struct bkey_s_c k, u64 *cur_inum)
 362 {
 363         struct bch_inode_unpacked inode;
 364         int ret;
 365
 366         if (*cur_inum == k.k->p.inode)
 367                 return 0;
 368
 369         *io_opts = bch2_opts_to_inode_opts(trans->c->opts);
 370
 371         ret = lookup_inode(trans,
 372                            SPOS(0, k.k->p.inode, k.k->p.snapshot),
 373                            &inode);
 374         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 375                 return ret;
 376
 377         if (!ret)
 378                 bch2_io_opts_apply(io_opts, bch2_inode_opts_get(&inode));
 379
 380         *cur_inum = k.k->p.inode;
 381         return 0;
 382 }
 383
 384 static int __bch2_move_data(struct moving_context *ctxt,
 385                             struct bpos start,
 386                             struct bpos end,
 387                             move_pred_fn pred, void *arg,
 388                             enum btree_id btree_id)
 389 {
 390         struct bch_fs *c = ctxt->c;
 391         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 392         struct bkey_buf sk;
 393         struct btree_trans trans;
 394         struct btree_iter iter;
 395         struct bkey_s_c k;
 396         struct data_update_opts data_opts;
 397         u64 cur_inum = U64_MAX;
 398         int ret = 0, ret2;
 399
 400         bch2_bkey_buf_init(&sk);
 401         bch2_trans_init(&trans, c, 0, 0);
 402
 403         ctxt->stats->data_type  = BCH_DATA_user;
 404         ctxt->stats->btree_id   = btree_id;
 405         ctxt->stats->pos        = start;
 406
 407         bch2_trans_iter_init(&trans, &iter, btree_id, start,
 408                              BTREE_ITER_PREFETCH|
 409                              BTREE_ITER_ALL_SNAPSHOTS);
 410
 411         if (ctxt->rate)
 412                 bch2_ratelimit_reset(ctxt->rate);
 413
 414         while (!move_ratelimit(&trans, ctxt)) {
 415                 bch2_trans_begin(&trans);
 416
 417                 k = bch2_btree_iter_peek(&iter);
 418                 if (!k.k)
 419                         break;
 420
 421                 ret = bkey_err(k);
 422                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 423                         continue;
 424                 if (ret)
 425                         break;
 426
 427                 if (bkey_cmp(bkey_start_pos(k.k), end) >= 0)
 428                         break;
 429
 430                 ctxt->stats->pos = iter.pos;
 431
 432                 if (!bkey_extent_is_direct_data(k.k))
 433                         goto next_nondata;
 434
 435                 ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 436                 if (ret)
 437                         continue;
 438
 439                 memset(&data_opts, 0, sizeof(data_opts));
 440                 if (!pred(c, arg, k, &io_opts, &data_opts))
 441                         goto next;
 442
 443                 /*
 444                  * The iterator gets unlocked by __bch2_read_extent - need to
 445                  * save a copy of @k elsewhere:
 446                   */
 447                 bch2_bkey_buf_reassemble(&sk, c, k);
 448                 k = bkey_i_to_s_c(sk.k);
 449
 450                 ret2 = bch2_move_extent(&trans, ctxt, io_opts,
 451                                         btree_id, k, data_opts);
 452                 if (ret2) {
 453                         if (bch2_err_matches(ret2, BCH_ERR_transaction_restart))
 454                                 continue;
 455
 456                         if (ret2 == -ENOMEM) {
 457                                 /* memory allocation failure, wait for some IO to finish */
 458                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 459                                 continue;
 460                         }
 461
 462                         /* XXX signal failure */
 463                         goto next;
 464                 }
 465
 466                 if (ctxt->rate)
 467                         bch2_ratelimit_increment(ctxt->rate, k.k->size);
 468 next:
 469                 atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 470 next_nondata:
 471                 bch2_btree_iter_advance(&iter);
 472         }
 473
 474         bch2_trans_iter_exit(&trans, &iter);
 475         bch2_trans_exit(&trans);
 476         bch2_bkey_buf_exit(&sk, c);
 477
 478         return ret;
 479 }
 480
 481 int bch2_move_data(struct bch_fs *c,
 482                    enum btree_id start_btree_id, struct bpos start_pos,
 483                    enum btree_id end_btree_id,   struct bpos end_pos,
 484                    struct bch_ratelimit *rate,
 485                    struct bch_move_stats *stats,
 486                    struct write_point_specifier wp,
 487                    bool wait_on_copygc,
 488                    move_pred_fn pred, void *arg)
 489 {
 490         struct moving_context ctxt;
 491         enum btree_id id;
 492         int ret;
 493
 494         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 495
 496         for (id = start_btree_id;
 497              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 498              id++) {
 499                 stats->btree_id = id;
 500
 501                 if (id != BTREE_ID_extents &&
 502                     id != BTREE_ID_reflink)
 503                         continue;
 504
 505                 ret = __bch2_move_data(&ctxt,
 506                                        id == start_btree_id ? start_pos : POS_MIN,
 507                                        id == end_btree_id   ? end_pos   : POS_MAX,
 508                                        pred, arg, id);
 509                 if (ret)
 510                         break;
 511         }
 512
 513         bch2_moving_ctxt_exit(&ctxt);
 514
 515         return ret;
 516 }
 517
 518 static int verify_bucket_evacuated(struct btree_trans *trans, struct bpos bucket, int gen)
 519 {
 520         struct bch_fs *c = trans->c;
 521         struct btree_iter iter;
 522         struct bkey_s_c k;
 523         int ret;
 524
 525         bch2_trans_iter_init(trans, &iter, BTREE_ID_alloc,
 526                              bucket, BTREE_ITER_CACHED);
 527 again:
 528         k = bch2_btree_iter_peek_slot(&iter);
 529         ret = bkey_err(k);
 530
 531         if (!ret && k.k->type == KEY_TYPE_alloc_v4) {
 532                 struct bkey_s_c_alloc_v4 a = bkey_s_c_to_alloc_v4(k);
 533
 534                 if (a.v->gen == gen &&
 535                     a.v->dirty_sectors) {
 536                         struct printbuf buf = PRINTBUF;
 537
 538                         if (a.v->data_type == BCH_DATA_btree) {
 539                                 bch2_trans_unlock(trans);
 540                                 if (bch2_btree_interior_updates_flush(c))
 541                                         goto again;
 542                         }
 543
 544                         prt_str(&buf, "failed to evacuate bucket ");
 545                         bch2_bkey_val_to_text(&buf, c, k);
 546
 547                         bch2_trans_inconsistent(trans, "%s", buf.buf);
 548                         printbuf_exit(&buf);
 549                 }
 550         }
 551
 552         bch2_trans_iter_exit(trans, &iter);
 553         return ret;
 554 }
 555
 556 int __bch2_evacuate_bucket(struct moving_context *ctxt,
 557                            struct bpos bucket, int gen,
 558                            struct data_update_opts _data_opts)
 559 {
 560         struct bch_fs *c = ctxt->c;
 561         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 562         struct btree_trans trans;
 563         struct btree_iter iter;
 564         struct bkey_buf sk;
 565         struct bch_backpointer bp;
 566         struct data_update_opts data_opts;
 567         u64 bp_offset = 0, cur_inum = U64_MAX;
 568         int ret = 0;
 569
 570         bch2_bkey_buf_init(&sk);
 571         bch2_trans_init(&trans, c, 0, 0);
 572
 573         while (!(ret = move_ratelimit(&trans, ctxt))) {
 574                 bch2_trans_begin(&trans);
 575
 576                 ret = bch2_get_next_backpointer(&trans, bucket, gen,
 577                                                 &bp_offset, &bp);
 578                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 579                         continue;
 580                 if (ret)
 581                         goto err;
 582                 if (bp_offset == U64_MAX)
 583                         break;
 584
 585                 if (!bp.level) {
 586                         const struct bch_extent_ptr *ptr;
 587                         struct bkey_s_c k;
 588                         unsigned i = 0;
 589
 590                         k = bch2_backpointer_get_key(&trans, &iter,
 591                                                 bucket, bp_offset, bp);
 592                         ret = bkey_err(k);
 593                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 594                                 continue;
 595                         if (ret)
 596                                 goto err;
 597                         if (!k.k)
 598                                 continue;
 599
 600                         bch2_bkey_buf_reassemble(&sk, c, k);
 601                         k = bkey_i_to_s_c(sk.k);
 602                         bch2_trans_iter_exit(&trans, &iter);
 603
 604                         ret = move_get_io_opts(&trans, &io_opts, k, &cur_inum);
 605                         if (ret)
 606                                 continue;
 607
 608                         data_opts = _data_opts;
 609                         data_opts.target        = io_opts.background_target;
 610                         data_opts.rewrite_ptrs = 0;
 611
 612                         bkey_for_each_ptr(bch2_bkey_ptrs_c(k), ptr) {
 613                                 if (ptr->dev == bucket.inode)
 614                                         data_opts.rewrite_ptrs |= 1U << i;
 615                                 i++;
 616                         }
 617
 618                         ret = bch2_move_extent(&trans, ctxt, io_opts,
 619                                                bp.btree_id, k, data_opts);
 620                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 621                                 continue;
 622                         if (ret == -ENOMEM) {
 623                                 /* memory allocation failure, wait for some IO to finish */
 624                                 bch2_move_ctxt_wait_for_io(ctxt, &trans);
 625                                 continue;
 626                         }
 627                         if (ret)
 628                                 goto err;
 629
 630                         if (ctxt->rate)
 631                                 bch2_ratelimit_increment(ctxt->rate, k.k->size);
 632                         atomic64_add(k.k->size, &ctxt->stats->sectors_seen);
 633                 } else {
 634                         struct btree *b;
 635
 636                         b = bch2_backpointer_get_node(&trans, &iter,
 637                                                 bucket, bp_offset, bp);
 638                         ret = PTR_ERR_OR_ZERO(b);
 639                         if (ret == -BCH_ERR_backpointer_to_overwritten_btree_node)
 640                                 continue;
 641                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 642                                 continue;
 643                         if (ret)
 644                                 goto err;
 645                         if (!b)
 646                                 continue;
 647
 648                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0);
 649                         bch2_trans_iter_exit(&trans, &iter);
 650
 651                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 652                                 continue;
 653                         if (ret)
 654                                 goto err;
 655
 656                         if (ctxt->rate)
 657                                 bch2_ratelimit_increment(ctxt->rate,
 658                                                          c->opts.btree_node_size >> 9);
 659                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_seen);
 660                         atomic64_add(c->opts.btree_node_size >> 9, &ctxt->stats->sectors_moved);
 661                 }
 662
 663                 bp_offset++;
 664         }
 665
 666         if (IS_ENABLED(CONFIG_BCACHEFS_DEBUG) && gen >= 0) {
 667                 bch2_trans_unlock(&trans);
 668                 move_ctxt_wait_event(ctxt, NULL, list_empty(&ctxt->reads));
 669                 closure_sync(&ctxt->cl);
 670                 if (!ctxt->write_error)
 671                         lockrestart_do(&trans, verify_bucket_evacuated(&trans, bucket, gen));
 672         }
 673 err:
 674         bch2_trans_exit(&trans);
 675         bch2_bkey_buf_exit(&sk, c);
 676         return ret;
 677 }
 678
 679 int bch2_evacuate_bucket(struct bch_fs *c,
 680                          struct bpos bucket, int gen,
 681                          struct data_update_opts data_opts,
 682                          struct bch_ratelimit *rate,
 683                          struct bch_move_stats *stats,
 684                          struct write_point_specifier wp,
 685                          bool wait_on_copygc)
 686 {
 687         struct moving_context ctxt;
 688         int ret;
 689
 690         bch2_moving_ctxt_init(&ctxt, c, rate, stats, wp, wait_on_copygc);
 691         ret = __bch2_evacuate_bucket(&ctxt, bucket, gen, data_opts);
 692         bch2_moving_ctxt_exit(&ctxt);
 693
 694         return ret;
 695 }
 696
 697 typedef bool (*move_btree_pred)(struct bch_fs *, void *,
 698                                 struct btree *, struct bch_io_opts *,
 699                                 struct data_update_opts *);
 700
 701 static int bch2_move_btree(struct bch_fs *c,
 702                            enum btree_id start_btree_id, struct bpos start_pos,
 703                            enum btree_id end_btree_id,   struct bpos end_pos,
 704                            move_btree_pred pred, void *arg,
 705                            struct bch_move_stats *stats)
 706 {
 707         bool kthread = (current->flags & PF_KTHREAD) != 0;
 708         struct bch_io_opts io_opts = bch2_opts_to_inode_opts(c->opts);
 709         struct btree_trans trans;
 710         struct btree_iter iter;
 711         struct btree *b;
 712         enum btree_id id;
 713         struct data_update_opts data_opts;
 714         int ret = 0;
 715
 716         bch2_trans_init(&trans, c, 0, 0);
 717         progress_list_add(c, stats);
 718
 719         stats->data_type = BCH_DATA_btree;
 720
 721         for (id = start_btree_id;
 722              id <= min_t(unsigned, end_btree_id, BTREE_ID_NR - 1);
 723              id++) {
 724                 stats->btree_id = id;
 725
 726                 bch2_trans_node_iter_init(&trans, &iter, id, POS_MIN, 0, 0,
 727                                           BTREE_ITER_PREFETCH);
 728 retry:
 729                 ret = 0;
 730                 while (bch2_trans_begin(&trans),
 731                        (b = bch2_btree_iter_peek_node(&iter)) &&
 732                        !(ret = PTR_ERR_OR_ZERO(b))) {
 733                         if (kthread && kthread_should_stop())
 734                                 break;
 735
 736                         if ((cmp_int(id, end_btree_id) ?:
 737                              bpos_cmp(b->key.k.p, end_pos)) > 0)
 738                                 break;
 739
 740                         stats->pos = iter.pos;
 741
 742                         if (!pred(c, arg, b, &io_opts, &data_opts))
 743                                 goto next;
 744
 745                         ret = bch2_btree_node_rewrite(&trans, &iter, b, 0) ?: ret;
 746                         if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 747                                 continue;
 748                         if (ret)
 749                                 break;
 750 next:
 751                         bch2_btree_iter_next_node(&iter);
 752                 }
 753                 if (bch2_err_matches(ret, BCH_ERR_transaction_restart))
 754                         goto retry;
 755
 756                 bch2_trans_iter_exit(&trans, &iter);
 757
 758                 if (kthread && kthread_should_stop())
 759                         break;
 760         }
 761
 762         bch2_trans_exit(&trans);
 763
 764         if (ret)
 765                 bch_err(c, "error in %s(): %s", __func__, bch2_err_str(ret));
 766
 767         bch2_btree_interior_updates_flush(c);
 768
 769         progress_list_del(c, stats);
 770         return ret;
 771 }
 772
 773 static bool rereplicate_pred(struct bch_fs *c, void *arg,
 774                              struct bkey_s_c k,
 775                              struct bch_io_opts *io_opts,
 776                              struct data_update_opts *data_opts)
 777 {
 778         unsigned nr_good = bch2_bkey_durability(c, k);
 779         unsigned replicas = bkey_is_btree_ptr(k.k)
 780                 ? c->opts.metadata_replicas
 781                 : io_opts->data_replicas;
 782
 783         if (!nr_good || nr_good >= replicas)
 784                 return false;
 785
 786         data_opts->target               = 0;
 787         data_opts->extra_replicas       = replicas - nr_good;
 788         data_opts->btree_insert_flags   = 0;
 789         return true;
 790 }
 791
 792 static bool migrate_pred(struct bch_fs *c, void *arg,
 793                          struct bkey_s_c k,
 794                          struct bch_io_opts *io_opts,
 795                          struct data_update_opts *data_opts)
 796 {
 797         struct bkey_ptrs_c ptrs = bch2_bkey_ptrs_c(k);
 798         const struct bch_extent_ptr *ptr;
 799         struct bch_ioctl_data *op = arg;
 800         unsigned i = 0;
 801
 802         data_opts->rewrite_ptrs         = 0;
 803         data_opts->target               = 0;
 804         data_opts->extra_replicas       = 0;
 805         data_opts->btree_insert_flags   = 0;
 806
 807         bkey_for_each_ptr(ptrs, ptr) {
 808                 if (ptr->dev == op->migrate.dev)
 809                         data_opts->rewrite_ptrs |= 1U << i;
 810                 i++;
 811         }
 812
 813         return data_opts->rewrite_ptrs != 0;;
 814 }
 815
 816 static bool rereplicate_btree_pred(struct bch_fs *c, void *arg,
 817                                    struct btree *b,
 818                                    struct bch_io_opts *io_opts,
 819                                    struct data_update_opts *data_opts)
 820 {
 821         return rereplicate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 822 }
 823
 824 static bool migrate_btree_pred(struct bch_fs *c, void *arg,
 825                                struct btree *b,
 826                                struct bch_io_opts *io_opts,
 827                                struct data_update_opts *data_opts)
 828 {
 829         return migrate_pred(c, arg, bkey_i_to_s_c(&b->key), io_opts, data_opts);
 830 }
 831
 832 static bool bformat_needs_redo(struct bkey_format *f)
 833 {
 834         unsigned i;
 835
 836         for (i = 0; i < f->nr_fields; i++) {
 837                 unsigned unpacked_bits = bch2_bkey_format_current.bits_per_field[i];
 838                 u64 unpacked_mask = ~((~0ULL << 1) << (unpacked_bits - 1));
 839                 u64 field_offset = le64_to_cpu(f->field_offset[i]);
 840
 841                 if (f->bits_per_field[i] > unpacked_bits)
 842                         return true;
 843
 844                 if ((f->bits_per_field[i] == unpacked_bits) && field_offset)
 845                         return true;
 846
 847                 if (((field_offset + ((1ULL << f->bits_per_field[i]) - 1)) &
 848                      unpacked_mask) <
 849                     field_offset)
 850                         return true;
 851         }
 852
 853         return false;
 854 }
 855
 856 static bool rewrite_old_nodes_pred(struct bch_fs *c, void *arg,
 857                                    struct btree *b,
 858                                    struct bch_io_opts *io_opts,
 859                                    struct data_update_opts *data_opts)
 860 {
 861         if (b->version_ondisk != c->sb.version ||
 862             btree_node_need_rewrite(b) ||
 863             bformat_needs_redo(&b->format)) {
 864                 data_opts->target               = 0;
 865                 data_opts->extra_replicas       = 0;
 866                 data_opts->btree_insert_flags   = 0;
 867                 return true;
 868         }
 869
 870         return false;
 871 }
 872
 873 int bch2_scan_old_btree_nodes(struct bch_fs *c, struct bch_move_stats *stats)
 874 {
 875         int ret;
 876
 877         ret = bch2_move_btree(c,
 878                               0,                POS_MIN,
 879                               BTREE_ID_NR,      SPOS_MAX,
 880                               rewrite_old_nodes_pred, c, stats);
 881         if (!ret) {
 882                 mutex_lock(&c->sb_lock);
 883                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_extents_above_btree_updates_done);
 884                 c->disk_sb.sb->compat[0] |= cpu_to_le64(1ULL << BCH_COMPAT_bformat_overflow_done);
 885                 c->disk_sb.sb->version_min = c->disk_sb.sb->version;
 886                 bch2_write_super(c);
 887                 mutex_unlock(&c->sb_lock);
 888         }
 889
 890         return ret;
 891 }
 892
 893 int bch2_data_job(struct bch_fs *c,
 894                   struct bch_move_stats *stats,
 895                   struct bch_ioctl_data op)
 896 {
 897         int ret = 0;
 898
 899         switch (op.op) {
 900         case BCH_DATA_OP_REREPLICATE:
 901                 bch_move_stats_init(stats, "rereplicate");
 902                 stats->data_type = BCH_DATA_journal;
 903                 ret = bch2_journal_flush_device_pins(&c->journal, -1);
 904
 905                 ret = bch2_move_btree(c,
 906                                       op.start_btree,   op.start_pos,
 907                                       op.end_btree,     op.end_pos,
 908                                       rereplicate_btree_pred, c, stats) ?: ret;
 909                 ret = bch2_replicas_gc2(c) ?: ret;
 910
 911                 ret = bch2_move_data(c,
 912                                      op.start_btree,    op.start_pos,
 913                                      op.end_btree,      op.end_pos,
 914                                      NULL,
 915                                      stats,
 916                                      writepoint_hashed((unsigned long) current),
 917                                      true,
 918                                      rereplicate_pred, c) ?: ret;
 919                 ret = bch2_replicas_gc2(c) ?: ret;
 920                 break;
 921         case BCH_DATA_OP_MIGRATE:
 922                 if (op.migrate.dev >= c->sb.nr_devices)
 923                         return -EINVAL;
 924
 925                 bch_move_stats_init(stats, "migrate");
 926                 stats->data_type = BCH_DATA_journal;
 927                 ret = bch2_journal_flush_device_pins(&c->journal, op.migrate.dev);
 928
 929                 ret = bch2_move_btree(c,
 930                                       op.start_btree,   op.start_pos,
 931                                       op.end_btree,     op.end_pos,
 932                                       migrate_btree_pred, &op, stats) ?: ret;
 933                 ret = bch2_replicas_gc2(c) ?: ret;
 934
 935                 ret = bch2_move_data(c,
 936                                      op.start_btree,    op.start_pos,
 937                                      op.end_btree,      op.end_pos,
 938                                      NULL,
 939                                      stats,
 940                                      writepoint_hashed((unsigned long) current),
 941                                      true,
 942                                      migrate_pred, &op) ?: ret;
 943                 ret = bch2_replicas_gc2(c) ?: ret;
 944                 break;
 945         case BCH_DATA_OP_REWRITE_OLD_NODES:
 946                 bch_move_stats_init(stats, "rewrite_old_nodes");
 947                 ret = bch2_scan_old_btree_nodes(c, stats);
 948                 break;
 949         default:
 950                 ret = -EINVAL;
 951         }
 952
 953         return ret;
 954 }